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/bio.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
17 #include "ctree.h"
18 #include "extent_map.h"
19 #include "disk-io.h"
20 #include "transaction.h"
21 #include "print-tree.h"
22 #include "volumes.h"
23 #include "raid56.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
27 #include "math.h"
28 #include "dev-replace.h"
29 #include "sysfs.h"
30 #include "tree-checker.h"
31
32 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
33 [BTRFS_RAID_RAID10] = {
34 .sub_stripes = 2,
35 .dev_stripes = 1,
36 .devs_max = 0, /* 0 == as many as possible */
37 .devs_min = 4,
38 .tolerated_failures = 1,
39 .devs_increment = 2,
40 .ncopies = 2,
41 .nparity = 0,
42 .raid_name = "raid10",
43 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
44 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
45 },
46 [BTRFS_RAID_RAID1] = {
47 .sub_stripes = 1,
48 .dev_stripes = 1,
49 .devs_max = 2,
50 .devs_min = 2,
51 .tolerated_failures = 1,
52 .devs_increment = 2,
53 .ncopies = 2,
54 .nparity = 0,
55 .raid_name = "raid1",
56 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
57 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
58 },
59 [BTRFS_RAID_DUP] = {
60 .sub_stripes = 1,
61 .dev_stripes = 2,
62 .devs_max = 1,
63 .devs_min = 1,
64 .tolerated_failures = 0,
65 .devs_increment = 1,
66 .ncopies = 2,
67 .nparity = 0,
68 .raid_name = "dup",
69 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
70 .mindev_error = 0,
71 },
72 [BTRFS_RAID_RAID0] = {
73 .sub_stripes = 1,
74 .dev_stripes = 1,
75 .devs_max = 0,
76 .devs_min = 2,
77 .tolerated_failures = 0,
78 .devs_increment = 1,
79 .ncopies = 1,
80 .nparity = 0,
81 .raid_name = "raid0",
82 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
83 .mindev_error = 0,
84 },
85 [BTRFS_RAID_SINGLE] = {
86 .sub_stripes = 1,
87 .dev_stripes = 1,
88 .devs_max = 1,
89 .devs_min = 1,
90 .tolerated_failures = 0,
91 .devs_increment = 1,
92 .ncopies = 1,
93 .nparity = 0,
94 .raid_name = "single",
95 .bg_flag = 0,
96 .mindev_error = 0,
97 },
98 [BTRFS_RAID_RAID5] = {
99 .sub_stripes = 1,
100 .dev_stripes = 1,
101 .devs_max = 0,
102 .devs_min = 2,
103 .tolerated_failures = 1,
104 .devs_increment = 1,
105 .ncopies = 1,
106 .nparity = 1,
107 .raid_name = "raid5",
108 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
109 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
110 },
111 [BTRFS_RAID_RAID6] = {
112 .sub_stripes = 1,
113 .dev_stripes = 1,
114 .devs_max = 0,
115 .devs_min = 3,
116 .tolerated_failures = 2,
117 .devs_increment = 1,
118 .ncopies = 1,
119 .nparity = 2,
120 .raid_name = "raid6",
121 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
122 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
123 },
124 };
125
126 const char *get_raid_name(enum btrfs_raid_types type)
127 {
128 if (type >= BTRFS_NR_RAID_TYPES)
129 return NULL;
130
131 return btrfs_raid_array[type].raid_name;
132 }
133
134 /*
135 * Fill @buf with textual description of @bg_flags, no more than @size_buf
136 * bytes including terminating null byte.
137 */
138 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
139 {
140 int i;
141 int ret;
142 char *bp = buf;
143 u64 flags = bg_flags;
144 u32 size_bp = size_buf;
145
146 if (!flags) {
147 strcpy(bp, "NONE");
148 return;
149 }
150
151 #define DESCRIBE_FLAG(flag, desc) \
152 do { \
153 if (flags & (flag)) { \
154 ret = snprintf(bp, size_bp, "%s|", (desc)); \
155 if (ret < 0 || ret >= size_bp) \
156 goto out_overflow; \
157 size_bp -= ret; \
158 bp += ret; \
159 flags &= ~(flag); \
160 } \
161 } while (0)
162
163 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
164 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
165 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
166
167 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
168 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
169 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
170 btrfs_raid_array[i].raid_name);
171 #undef DESCRIBE_FLAG
172
173 if (flags) {
174 ret = snprintf(bp, size_bp, "0x%llx|", flags);
175 size_bp -= ret;
176 }
177
178 if (size_bp < size_buf)
179 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
180
181 /*
182 * The text is trimmed, it's up to the caller to provide sufficiently
183 * large buffer
184 */
185 out_overflow:;
186 }
187
188 static int init_first_rw_device(struct btrfs_trans_handle *trans);
189 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
190 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
191 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
192 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
193 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
194 enum btrfs_map_op op,
195 u64 logical, u64 *length,
196 struct btrfs_bio **bbio_ret,
197 int mirror_num, int need_raid_map);
198
199 /*
200 * Device locking
201 * ==============
202 *
203 * There are several mutexes that protect manipulation of devices and low-level
204 * structures like chunks but not block groups, extents or files
205 *
206 * uuid_mutex (global lock)
207 * ------------------------
208 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
209 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
210 * device) or requested by the device= mount option
211 *
212 * the mutex can be very coarse and can cover long-running operations
213 *
214 * protects: updates to fs_devices counters like missing devices, rw devices,
215 * seeding, structure cloning, opening/closing devices at mount/umount time
216 *
217 * global::fs_devs - add, remove, updates to the global list
218 *
219 * does not protect: manipulation of the fs_devices::devices list!
220 *
221 * btrfs_device::name - renames (write side), read is RCU
222 *
223 * fs_devices::device_list_mutex (per-fs, with RCU)
224 * ------------------------------------------------
225 * protects updates to fs_devices::devices, ie. adding and deleting
226 *
227 * simple list traversal with read-only actions can be done with RCU protection
228 *
229 * may be used to exclude some operations from running concurrently without any
230 * modifications to the list (see write_all_supers)
231 *
232 * balance_mutex
233 * -------------
234 * protects balance structures (status, state) and context accessed from
235 * several places (internally, ioctl)
236 *
237 * chunk_mutex
238 * -----------
239 * protects chunks, adding or removing during allocation, trim or when a new
240 * device is added/removed
241 *
242 * cleaner_mutex
243 * -------------
244 * a big lock that is held by the cleaner thread and prevents running subvolume
245 * cleaning together with relocation or delayed iputs
246 *
247 *
248 * Lock nesting
249 * ============
250 *
251 * uuid_mutex
252 * volume_mutex
253 * device_list_mutex
254 * chunk_mutex
255 * balance_mutex
256 *
257 *
258 * Exclusive operations, BTRFS_FS_EXCL_OP
259 * ======================================
260 *
261 * Maintains the exclusivity of the following operations that apply to the
262 * whole filesystem and cannot run in parallel.
263 *
264 * - Balance (*)
265 * - Device add
266 * - Device remove
267 * - Device replace (*)
268 * - Resize
269 *
270 * The device operations (as above) can be in one of the following states:
271 *
272 * - Running state
273 * - Paused state
274 * - Completed state
275 *
276 * Only device operations marked with (*) can go into the Paused state for the
277 * following reasons:
278 *
279 * - ioctl (only Balance can be Paused through ioctl)
280 * - filesystem remounted as read-only
281 * - filesystem unmounted and mounted as read-only
282 * - system power-cycle and filesystem mounted as read-only
283 * - filesystem or device errors leading to forced read-only
284 *
285 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
286 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
287 * A device operation in Paused or Running state can be canceled or resumed
288 * either by ioctl (Balance only) or when remounted as read-write.
289 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
290 * completed.
291 */
292
293 DEFINE_MUTEX(uuid_mutex);
294 static LIST_HEAD(fs_uuids);
295 struct list_head *btrfs_get_fs_uuids(void)
296 {
297 return &fs_uuids;
298 }
299
300 /*
301 * alloc_fs_devices - allocate struct btrfs_fs_devices
302 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
303 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
304 *
305 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
306 * The returned struct is not linked onto any lists and can be destroyed with
307 * kfree() right away.
308 */
309 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
310 const u8 *metadata_fsid)
311 {
312 struct btrfs_fs_devices *fs_devs;
313
314 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
315 if (!fs_devs)
316 return ERR_PTR(-ENOMEM);
317
318 mutex_init(&fs_devs->device_list_mutex);
319
320 INIT_LIST_HEAD(&fs_devs->devices);
321 INIT_LIST_HEAD(&fs_devs->alloc_list);
322 INIT_LIST_HEAD(&fs_devs->fs_list);
323 if (fsid)
324 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
325
326 if (metadata_fsid)
327 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
328 else if (fsid)
329 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
330
331 return fs_devs;
332 }
333
334 void btrfs_free_device(struct btrfs_device *device)
335 {
336 WARN_ON(!list_empty(&device->post_commit_list));
337 rcu_string_free(device->name);
338 extent_io_tree_release(&device->alloc_state);
339 bio_put(device->flush_bio);
340 kfree(device);
341 }
342
343 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
344 {
345 struct btrfs_device *device;
346 WARN_ON(fs_devices->opened);
347 while (!list_empty(&fs_devices->devices)) {
348 device = list_entry(fs_devices->devices.next,
349 struct btrfs_device, dev_list);
350 list_del(&device->dev_list);
351 btrfs_free_device(device);
352 }
353 kfree(fs_devices);
354 }
355
356 static void btrfs_kobject_uevent(struct block_device *bdev,
357 enum kobject_action action)
358 {
359 int ret;
360
361 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
362 if (ret)
363 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
364 action,
365 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
366 &disk_to_dev(bdev->bd_disk)->kobj);
367 }
368
369 void __exit btrfs_cleanup_fs_uuids(void)
370 {
371 struct btrfs_fs_devices *fs_devices;
372
373 while (!list_empty(&fs_uuids)) {
374 fs_devices = list_entry(fs_uuids.next,
375 struct btrfs_fs_devices, fs_list);
376 list_del(&fs_devices->fs_list);
377 free_fs_devices(fs_devices);
378 }
379 }
380
381 /*
382 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
383 * Returned struct is not linked onto any lists and must be destroyed using
384 * btrfs_free_device.
385 */
386 static struct btrfs_device *__alloc_device(void)
387 {
388 struct btrfs_device *dev;
389
390 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
391 if (!dev)
392 return ERR_PTR(-ENOMEM);
393
394 /*
395 * Preallocate a bio that's always going to be used for flushing device
396 * barriers and matches the device lifespan
397 */
398 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
399 if (!dev->flush_bio) {
400 kfree(dev);
401 return ERR_PTR(-ENOMEM);
402 }
403
404 INIT_LIST_HEAD(&dev->dev_list);
405 INIT_LIST_HEAD(&dev->dev_alloc_list);
406 INIT_LIST_HEAD(&dev->post_commit_list);
407
408 spin_lock_init(&dev->io_lock);
409
410 atomic_set(&dev->reada_in_flight, 0);
411 atomic_set(&dev->dev_stats_ccnt, 0);
412 btrfs_device_data_ordered_init(dev);
413 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
414 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
415 extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL);
416
417 return dev;
418 }
419
420 static noinline struct btrfs_fs_devices *find_fsid(
421 const u8 *fsid, const u8 *metadata_fsid)
422 {
423 struct btrfs_fs_devices *fs_devices;
424
425 ASSERT(fsid);
426
427 if (metadata_fsid) {
428 /*
429 * Handle scanned device having completed its fsid change but
430 * belonging to a fs_devices that was created by first scanning
431 * a device which didn't have its fsid/metadata_uuid changed
432 * at all and the CHANGING_FSID_V2 flag set.
433 */
434 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
435 if (fs_devices->fsid_change &&
436 memcmp(metadata_fsid, fs_devices->fsid,
437 BTRFS_FSID_SIZE) == 0 &&
438 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
439 BTRFS_FSID_SIZE) == 0) {
440 return fs_devices;
441 }
442 }
443 /*
444 * Handle scanned device having completed its fsid change but
445 * belonging to a fs_devices that was created by a device that
446 * has an outdated pair of fsid/metadata_uuid and
447 * CHANGING_FSID_V2 flag set.
448 */
449 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
450 if (fs_devices->fsid_change &&
451 memcmp(fs_devices->metadata_uuid,
452 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
453 memcmp(metadata_fsid, fs_devices->metadata_uuid,
454 BTRFS_FSID_SIZE) == 0) {
455 return fs_devices;
456 }
457 }
458 }
459
460 /* Handle non-split brain cases */
461 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
462 if (metadata_fsid) {
463 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
464 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
465 BTRFS_FSID_SIZE) == 0)
466 return fs_devices;
467 } else {
468 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
469 return fs_devices;
470 }
471 }
472 return NULL;
473 }
474
475 static int
476 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
477 int flush, struct block_device **bdev,
478 struct buffer_head **bh)
479 {
480 int ret;
481
482 *bdev = blkdev_get_by_path(device_path, flags, holder);
483
484 if (IS_ERR(*bdev)) {
485 ret = PTR_ERR(*bdev);
486 goto error;
487 }
488
489 if (flush)
490 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
491 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
492 if (ret) {
493 blkdev_put(*bdev, flags);
494 goto error;
495 }
496 invalidate_bdev(*bdev);
497 *bh = btrfs_read_dev_super(*bdev);
498 if (IS_ERR(*bh)) {
499 ret = PTR_ERR(*bh);
500 blkdev_put(*bdev, flags);
501 goto error;
502 }
503
504 return 0;
505
506 error:
507 *bdev = NULL;
508 *bh = NULL;
509 return ret;
510 }
511
512 static void requeue_list(struct btrfs_pending_bios *pending_bios,
513 struct bio *head, struct bio *tail)
514 {
515
516 struct bio *old_head;
517
518 old_head = pending_bios->head;
519 pending_bios->head = head;
520 if (pending_bios->tail)
521 tail->bi_next = old_head;
522 else
523 pending_bios->tail = tail;
524 }
525
526 /*
527 * we try to collect pending bios for a device so we don't get a large
528 * number of procs sending bios down to the same device. This greatly
529 * improves the schedulers ability to collect and merge the bios.
530 *
531 * But, it also turns into a long list of bios to process and that is sure
532 * to eventually make the worker thread block. The solution here is to
533 * make some progress and then put this work struct back at the end of
534 * the list if the block device is congested. This way, multiple devices
535 * can make progress from a single worker thread.
536 */
537 static noinline void run_scheduled_bios(struct btrfs_device *device)
538 {
539 struct btrfs_fs_info *fs_info = device->fs_info;
540 struct bio *pending;
541 struct backing_dev_info *bdi;
542 struct btrfs_pending_bios *pending_bios;
543 struct bio *tail;
544 struct bio *cur;
545 int again = 0;
546 unsigned long num_run;
547 unsigned long batch_run = 0;
548 unsigned long last_waited = 0;
549 int force_reg = 0;
550 int sync_pending = 0;
551 struct blk_plug plug;
552
553 /*
554 * this function runs all the bios we've collected for
555 * a particular device. We don't want to wander off to
556 * another device without first sending all of these down.
557 * So, setup a plug here and finish it off before we return
558 */
559 blk_start_plug(&plug);
560
561 bdi = device->bdev->bd_bdi;
562
563 loop:
564 spin_lock(&device->io_lock);
565
566 loop_lock:
567 num_run = 0;
568
569 /* take all the bios off the list at once and process them
570 * later on (without the lock held). But, remember the
571 * tail and other pointers so the bios can be properly reinserted
572 * into the list if we hit congestion
573 */
574 if (!force_reg && device->pending_sync_bios.head) {
575 pending_bios = &device->pending_sync_bios;
576 force_reg = 1;
577 } else {
578 pending_bios = &device->pending_bios;
579 force_reg = 0;
580 }
581
582 pending = pending_bios->head;
583 tail = pending_bios->tail;
584 WARN_ON(pending && !tail);
585
586 /*
587 * if pending was null this time around, no bios need processing
588 * at all and we can stop. Otherwise it'll loop back up again
589 * and do an additional check so no bios are missed.
590 *
591 * device->running_pending is used to synchronize with the
592 * schedule_bio code.
593 */
594 if (device->pending_sync_bios.head == NULL &&
595 device->pending_bios.head == NULL) {
596 again = 0;
597 device->running_pending = 0;
598 } else {
599 again = 1;
600 device->running_pending = 1;
601 }
602
603 pending_bios->head = NULL;
604 pending_bios->tail = NULL;
605
606 spin_unlock(&device->io_lock);
607
608 while (pending) {
609
610 rmb();
611 /* we want to work on both lists, but do more bios on the
612 * sync list than the regular list
613 */
614 if ((num_run > 32 &&
615 pending_bios != &device->pending_sync_bios &&
616 device->pending_sync_bios.head) ||
617 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
618 device->pending_bios.head)) {
619 spin_lock(&device->io_lock);
620 requeue_list(pending_bios, pending, tail);
621 goto loop_lock;
622 }
623
624 cur = pending;
625 pending = pending->bi_next;
626 cur->bi_next = NULL;
627
628 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
629
630 /*
631 * if we're doing the sync list, record that our
632 * plug has some sync requests on it
633 *
634 * If we're doing the regular list and there are
635 * sync requests sitting around, unplug before
636 * we add more
637 */
638 if (pending_bios == &device->pending_sync_bios) {
639 sync_pending = 1;
640 } else if (sync_pending) {
641 blk_finish_plug(&plug);
642 blk_start_plug(&plug);
643 sync_pending = 0;
644 }
645
646 btrfsic_submit_bio(cur);
647 num_run++;
648 batch_run++;
649
650 cond_resched();
651
652 /*
653 * we made progress, there is more work to do and the bdi
654 * is now congested. Back off and let other work structs
655 * run instead
656 */
657 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
658 fs_info->fs_devices->open_devices > 1) {
659 struct io_context *ioc;
660
661 ioc = current->io_context;
662
663 /*
664 * the main goal here is that we don't want to
665 * block if we're going to be able to submit
666 * more requests without blocking.
667 *
668 * This code does two great things, it pokes into
669 * the elevator code from a filesystem _and_
670 * it makes assumptions about how batching works.
671 */
672 if (ioc && ioc->nr_batch_requests > 0 &&
673 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
674 (last_waited == 0 ||
675 ioc->last_waited == last_waited)) {
676 /*
677 * we want to go through our batch of
678 * requests and stop. So, we copy out
679 * the ioc->last_waited time and test
680 * against it before looping
681 */
682 last_waited = ioc->last_waited;
683 cond_resched();
684 continue;
685 }
686 spin_lock(&device->io_lock);
687 requeue_list(pending_bios, pending, tail);
688 device->running_pending = 1;
689
690 spin_unlock(&device->io_lock);
691 btrfs_queue_work(fs_info->submit_workers,
692 &device->work);
693 goto done;
694 }
695 }
696
697 cond_resched();
698 if (again)
699 goto loop;
700
701 spin_lock(&device->io_lock);
702 if (device->pending_bios.head || device->pending_sync_bios.head)
703 goto loop_lock;
704 spin_unlock(&device->io_lock);
705
706 done:
707 blk_finish_plug(&plug);
708 }
709
710 static void pending_bios_fn(struct btrfs_work *work)
711 {
712 struct btrfs_device *device;
713
714 device = container_of(work, struct btrfs_device, work);
715 run_scheduled_bios(device);
716 }
717
718 static bool device_path_matched(const char *path, struct btrfs_device *device)
719 {
720 int found;
721
722 rcu_read_lock();
723 found = strcmp(rcu_str_deref(device->name), path);
724 rcu_read_unlock();
725
726 return found == 0;
727 }
728
729 /*
730 * Search and remove all stale (devices which are not mounted) devices.
731 * When both inputs are NULL, it will search and release all stale devices.
732 * path: Optional. When provided will it release all unmounted devices
733 * matching this path only.
734 * skip_dev: Optional. Will skip this device when searching for the stale
735 * devices.
736 * Return: 0 for success or if @path is NULL.
737 * -EBUSY if @path is a mounted device.
738 * -ENOENT if @path does not match any device in the list.
739 */
740 static int btrfs_free_stale_devices(const char *path,
741 struct btrfs_device *skip_device)
742 {
743 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
744 struct btrfs_device *device, *tmp_device;
745 int ret = 0;
746
747 if (path)
748 ret = -ENOENT;
749
750 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
751
752 mutex_lock(&fs_devices->device_list_mutex);
753 list_for_each_entry_safe(device, tmp_device,
754 &fs_devices->devices, dev_list) {
755 if (skip_device && skip_device == device)
756 continue;
757 if (path && !device->name)
758 continue;
759 if (path && !device_path_matched(path, device))
760 continue;
761 if (fs_devices->opened) {
762 /* for an already deleted device return 0 */
763 if (path && ret != 0)
764 ret = -EBUSY;
765 break;
766 }
767
768 /* delete the stale device */
769 fs_devices->num_devices--;
770 list_del(&device->dev_list);
771 btrfs_free_device(device);
772
773 ret = 0;
774 if (fs_devices->num_devices == 0)
775 break;
776 }
777 mutex_unlock(&fs_devices->device_list_mutex);
778
779 if (fs_devices->num_devices == 0) {
780 btrfs_sysfs_remove_fsid(fs_devices);
781 list_del(&fs_devices->fs_list);
782 free_fs_devices(fs_devices);
783 }
784 }
785
786 return ret;
787 }
788
789 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
790 struct btrfs_device *device, fmode_t flags,
791 void *holder)
792 {
793 struct request_queue *q;
794 struct block_device *bdev;
795 struct buffer_head *bh;
796 struct btrfs_super_block *disk_super;
797 u64 devid;
798 int ret;
799
800 if (device->bdev)
801 return -EINVAL;
802 if (!device->name)
803 return -EINVAL;
804
805 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
806 &bdev, &bh);
807 if (ret)
808 return ret;
809
810 disk_super = (struct btrfs_super_block *)bh->b_data;
811 devid = btrfs_stack_device_id(&disk_super->dev_item);
812 if (devid != device->devid)
813 goto error_brelse;
814
815 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
816 goto error_brelse;
817
818 device->generation = btrfs_super_generation(disk_super);
819
820 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
821 if (btrfs_super_incompat_flags(disk_super) &
822 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
823 pr_err(
824 "BTRFS: Invalid seeding and uuid-changed device detected\n");
825 goto error_brelse;
826 }
827
828 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
829 fs_devices->seeding = 1;
830 } else {
831 if (bdev_read_only(bdev))
832 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
833 else
834 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
835 }
836
837 q = bdev_get_queue(bdev);
838 if (!blk_queue_nonrot(q))
839 fs_devices->rotating = 1;
840
841 device->bdev = bdev;
842 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
843 device->mode = flags;
844
845 fs_devices->open_devices++;
846 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
847 device->devid != BTRFS_DEV_REPLACE_DEVID) {
848 fs_devices->rw_devices++;
849 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
850 }
851 brelse(bh);
852
853 return 0;
854
855 error_brelse:
856 brelse(bh);
857 blkdev_put(bdev, flags);
858
859 return -EINVAL;
860 }
861
862 /*
863 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
864 * being created with a disk that has already completed its fsid change.
865 */
866 static struct btrfs_fs_devices *find_fsid_inprogress(
867 struct btrfs_super_block *disk_super)
868 {
869 struct btrfs_fs_devices *fs_devices;
870
871 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
872 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
873 BTRFS_FSID_SIZE) != 0 &&
874 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
875 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
876 return fs_devices;
877 }
878 }
879
880 return NULL;
881 }
882
883
884 static struct btrfs_fs_devices *find_fsid_changed(
885 struct btrfs_super_block *disk_super)
886 {
887 struct btrfs_fs_devices *fs_devices;
888
889 /*
890 * Handles the case where scanned device is part of an fs that had
891 * multiple successful changes of FSID but curently device didn't
892 * observe it. Meaning our fsid will be different than theirs.
893 */
894 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
895 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
896 BTRFS_FSID_SIZE) != 0 &&
897 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
898 BTRFS_FSID_SIZE) == 0 &&
899 memcmp(fs_devices->fsid, disk_super->fsid,
900 BTRFS_FSID_SIZE) != 0) {
901 return fs_devices;
902 }
903 }
904
905 return NULL;
906 }
907 /*
908 * Add new device to list of registered devices
909 *
910 * Returns:
911 * device pointer which was just added or updated when successful
912 * error pointer when failed
913 */
914 static noinline struct btrfs_device *device_list_add(const char *path,
915 struct btrfs_super_block *disk_super,
916 bool *new_device_added)
917 {
918 struct btrfs_device *device;
919 struct btrfs_fs_devices *fs_devices = NULL;
920 struct rcu_string *name;
921 u64 found_transid = btrfs_super_generation(disk_super);
922 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
923 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
924 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
925 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
926 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
927
928 if (fsid_change_in_progress) {
929 if (!has_metadata_uuid) {
930 /*
931 * When we have an image which has CHANGING_FSID_V2 set
932 * it might belong to either a filesystem which has
933 * disks with completed fsid change or it might belong
934 * to fs with no UUID changes in effect, handle both.
935 */
936 fs_devices = find_fsid_inprogress(disk_super);
937 if (!fs_devices)
938 fs_devices = find_fsid(disk_super->fsid, NULL);
939 } else {
940 fs_devices = find_fsid_changed(disk_super);
941 }
942 } else if (has_metadata_uuid) {
943 fs_devices = find_fsid(disk_super->fsid,
944 disk_super->metadata_uuid);
945 } else {
946 fs_devices = find_fsid(disk_super->fsid, NULL);
947 }
948
949
950 if (!fs_devices) {
951 if (has_metadata_uuid)
952 fs_devices = alloc_fs_devices(disk_super->fsid,
953 disk_super->metadata_uuid);
954 else
955 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
956
957 if (IS_ERR(fs_devices))
958 return ERR_CAST(fs_devices);
959
960 fs_devices->fsid_change = fsid_change_in_progress;
961
962 mutex_lock(&fs_devices->device_list_mutex);
963 list_add(&fs_devices->fs_list, &fs_uuids);
964
965 device = NULL;
966 } else {
967 mutex_lock(&fs_devices->device_list_mutex);
968 device = btrfs_find_device(fs_devices, devid,
969 disk_super->dev_item.uuid, NULL, false);
970
971 /*
972 * If this disk has been pulled into an fs devices created by
973 * a device which had the CHANGING_FSID_V2 flag then replace the
974 * metadata_uuid/fsid values of the fs_devices.
975 */
976 if (has_metadata_uuid && fs_devices->fsid_change &&
977 found_transid > fs_devices->latest_generation) {
978 memcpy(fs_devices->fsid, disk_super->fsid,
979 BTRFS_FSID_SIZE);
980 memcpy(fs_devices->metadata_uuid,
981 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
982
983 fs_devices->fsid_change = false;
984 }
985 }
986
987 if (!device) {
988 if (fs_devices->opened) {
989 mutex_unlock(&fs_devices->device_list_mutex);
990 return ERR_PTR(-EBUSY);
991 }
992
993 device = btrfs_alloc_device(NULL, &devid,
994 disk_super->dev_item.uuid);
995 if (IS_ERR(device)) {
996 mutex_unlock(&fs_devices->device_list_mutex);
997 /* we can safely leave the fs_devices entry around */
998 return device;
999 }
1000
1001 name = rcu_string_strdup(path, GFP_NOFS);
1002 if (!name) {
1003 btrfs_free_device(device);
1004 mutex_unlock(&fs_devices->device_list_mutex);
1005 return ERR_PTR(-ENOMEM);
1006 }
1007 rcu_assign_pointer(device->name, name);
1008
1009 list_add_rcu(&device->dev_list, &fs_devices->devices);
1010 fs_devices->num_devices++;
1011
1012 device->fs_devices = fs_devices;
1013 *new_device_added = true;
1014
1015 if (disk_super->label[0])
1016 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
1017 disk_super->label, devid, found_transid, path);
1018 else
1019 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
1020 disk_super->fsid, devid, found_transid, path);
1021
1022 } else if (!device->name || strcmp(device->name->str, path)) {
1023 /*
1024 * When FS is already mounted.
1025 * 1. If you are here and if the device->name is NULL that
1026 * means this device was missing at time of FS mount.
1027 * 2. If you are here and if the device->name is different
1028 * from 'path' that means either
1029 * a. The same device disappeared and reappeared with
1030 * different name. or
1031 * b. The missing-disk-which-was-replaced, has
1032 * reappeared now.
1033 *
1034 * We must allow 1 and 2a above. But 2b would be a spurious
1035 * and unintentional.
1036 *
1037 * Further in case of 1 and 2a above, the disk at 'path'
1038 * would have missed some transaction when it was away and
1039 * in case of 2a the stale bdev has to be updated as well.
1040 * 2b must not be allowed at all time.
1041 */
1042
1043 /*
1044 * For now, we do allow update to btrfs_fs_device through the
1045 * btrfs dev scan cli after FS has been mounted. We're still
1046 * tracking a problem where systems fail mount by subvolume id
1047 * when we reject replacement on a mounted FS.
1048 */
1049 if (!fs_devices->opened && found_transid < device->generation) {
1050 /*
1051 * That is if the FS is _not_ mounted and if you
1052 * are here, that means there is more than one
1053 * disk with same uuid and devid.We keep the one
1054 * with larger generation number or the last-in if
1055 * generation are equal.
1056 */
1057 mutex_unlock(&fs_devices->device_list_mutex);
1058 return ERR_PTR(-EEXIST);
1059 }
1060
1061 /*
1062 * We are going to replace the device path for a given devid,
1063 * make sure it's the same device if the device is mounted
1064 */
1065 if (device->bdev) {
1066 struct block_device *path_bdev;
1067
1068 path_bdev = lookup_bdev(path);
1069 if (IS_ERR(path_bdev)) {
1070 mutex_unlock(&fs_devices->device_list_mutex);
1071 return ERR_CAST(path_bdev);
1072 }
1073
1074 if (device->bdev != path_bdev) {
1075 bdput(path_bdev);
1076 mutex_unlock(&fs_devices->device_list_mutex);
1077 btrfs_warn_in_rcu(device->fs_info,
1078 "duplicate device fsid:devid for %pU:%llu old:%s new:%s",
1079 disk_super->fsid, devid,
1080 rcu_str_deref(device->name), path);
1081 return ERR_PTR(-EEXIST);
1082 }
1083 bdput(path_bdev);
1084 btrfs_info_in_rcu(device->fs_info,
1085 "device fsid %pU devid %llu moved old:%s new:%s",
1086 disk_super->fsid, devid,
1087 rcu_str_deref(device->name), path);
1088 }
1089
1090 name = rcu_string_strdup(path, GFP_NOFS);
1091 if (!name) {
1092 mutex_unlock(&fs_devices->device_list_mutex);
1093 return ERR_PTR(-ENOMEM);
1094 }
1095 rcu_string_free(device->name);
1096 rcu_assign_pointer(device->name, name);
1097 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1098 fs_devices->missing_devices--;
1099 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1100 }
1101 }
1102
1103 /*
1104 * Unmount does not free the btrfs_device struct but would zero
1105 * generation along with most of the other members. So just update
1106 * it back. We need it to pick the disk with largest generation
1107 * (as above).
1108 */
1109 if (!fs_devices->opened) {
1110 device->generation = found_transid;
1111 fs_devices->latest_generation = max_t(u64, found_transid,
1112 fs_devices->latest_generation);
1113 }
1114
1115 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
1116
1117 mutex_unlock(&fs_devices->device_list_mutex);
1118 return device;
1119 }
1120
1121 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1122 {
1123 struct btrfs_fs_devices *fs_devices;
1124 struct btrfs_device *device;
1125 struct btrfs_device *orig_dev;
1126
1127 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1128 if (IS_ERR(fs_devices))
1129 return fs_devices;
1130
1131 mutex_lock(&orig->device_list_mutex);
1132 fs_devices->total_devices = orig->total_devices;
1133
1134 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1135 struct rcu_string *name;
1136
1137 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1138 orig_dev->uuid);
1139 if (IS_ERR(device))
1140 goto error;
1141
1142 /*
1143 * This is ok to do without rcu read locked because we hold the
1144 * uuid mutex so nothing we touch in here is going to disappear.
1145 */
1146 if (orig_dev->name) {
1147 name = rcu_string_strdup(orig_dev->name->str,
1148 GFP_KERNEL);
1149 if (!name) {
1150 btrfs_free_device(device);
1151 goto error;
1152 }
1153 rcu_assign_pointer(device->name, name);
1154 }
1155
1156 list_add(&device->dev_list, &fs_devices->devices);
1157 device->fs_devices = fs_devices;
1158 fs_devices->num_devices++;
1159 }
1160 mutex_unlock(&orig->device_list_mutex);
1161 return fs_devices;
1162 error:
1163 mutex_unlock(&orig->device_list_mutex);
1164 free_fs_devices(fs_devices);
1165 return ERR_PTR(-ENOMEM);
1166 }
1167
1168 /*
1169 * After we have read the system tree and know devids belonging to
1170 * this filesystem, remove the device which does not belong there.
1171 */
1172 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1173 {
1174 struct btrfs_device *device, *next;
1175 struct btrfs_device *latest_dev = NULL;
1176
1177 mutex_lock(&uuid_mutex);
1178 again:
1179 /* This is the initialized path, it is safe to release the devices. */
1180 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1181 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1182 &device->dev_state)) {
1183 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1184 &device->dev_state) &&
1185 (!latest_dev ||
1186 device->generation > latest_dev->generation)) {
1187 latest_dev = device;
1188 }
1189 continue;
1190 }
1191
1192 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1193 /*
1194 * In the first step, keep the device which has
1195 * the correct fsid and the devid that is used
1196 * for the dev_replace procedure.
1197 * In the second step, the dev_replace state is
1198 * read from the device tree and it is known
1199 * whether the procedure is really active or
1200 * not, which means whether this device is
1201 * used or whether it should be removed.
1202 */
1203 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1204 &device->dev_state)) {
1205 continue;
1206 }
1207 }
1208 if (device->bdev) {
1209 blkdev_put(device->bdev, device->mode);
1210 device->bdev = NULL;
1211 fs_devices->open_devices--;
1212 }
1213 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1214 list_del_init(&device->dev_alloc_list);
1215 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1216 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1217 &device->dev_state))
1218 fs_devices->rw_devices--;
1219 }
1220 list_del_init(&device->dev_list);
1221 fs_devices->num_devices--;
1222 btrfs_free_device(device);
1223 }
1224
1225 if (fs_devices->seed) {
1226 fs_devices = fs_devices->seed;
1227 goto again;
1228 }
1229
1230 fs_devices->latest_bdev = latest_dev->bdev;
1231
1232 mutex_unlock(&uuid_mutex);
1233 }
1234
1235 static void btrfs_close_bdev(struct btrfs_device *device)
1236 {
1237 if (!device->bdev)
1238 return;
1239
1240 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1241 sync_blockdev(device->bdev);
1242 invalidate_bdev(device->bdev);
1243 }
1244
1245 blkdev_put(device->bdev, device->mode);
1246 }
1247
1248 static void btrfs_close_one_device(struct btrfs_device *device)
1249 {
1250 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1251 struct btrfs_device *new_device;
1252 struct rcu_string *name;
1253
1254 if (device->bdev)
1255 fs_devices->open_devices--;
1256
1257 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1258 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1259 list_del_init(&device->dev_alloc_list);
1260 fs_devices->rw_devices--;
1261 }
1262
1263 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1264 fs_devices->missing_devices--;
1265
1266 btrfs_close_bdev(device);
1267
1268 new_device = btrfs_alloc_device(NULL, &device->devid,
1269 device->uuid);
1270 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1271
1272 /* Safe because we are under uuid_mutex */
1273 if (device->name) {
1274 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1275 BUG_ON(!name); /* -ENOMEM */
1276 rcu_assign_pointer(new_device->name, name);
1277 }
1278
1279 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1280 new_device->fs_devices = device->fs_devices;
1281
1282 synchronize_rcu();
1283 btrfs_free_device(device);
1284 }
1285
1286 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1287 {
1288 struct btrfs_device *device, *tmp;
1289
1290 if (--fs_devices->opened > 0)
1291 return 0;
1292
1293 mutex_lock(&fs_devices->device_list_mutex);
1294 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1295 btrfs_close_one_device(device);
1296 }
1297 mutex_unlock(&fs_devices->device_list_mutex);
1298
1299 WARN_ON(fs_devices->open_devices);
1300 WARN_ON(fs_devices->rw_devices);
1301 fs_devices->opened = 0;
1302 fs_devices->seeding = 0;
1303
1304 return 0;
1305 }
1306
1307 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1308 {
1309 struct btrfs_fs_devices *seed_devices = NULL;
1310 int ret;
1311
1312 mutex_lock(&uuid_mutex);
1313 ret = close_fs_devices(fs_devices);
1314 if (!fs_devices->opened) {
1315 seed_devices = fs_devices->seed;
1316 fs_devices->seed = NULL;
1317 }
1318 mutex_unlock(&uuid_mutex);
1319
1320 while (seed_devices) {
1321 fs_devices = seed_devices;
1322 seed_devices = fs_devices->seed;
1323 close_fs_devices(fs_devices);
1324 free_fs_devices(fs_devices);
1325 }
1326 return ret;
1327 }
1328
1329 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1330 fmode_t flags, void *holder)
1331 {
1332 struct btrfs_device *device;
1333 struct btrfs_device *latest_dev = NULL;
1334 int ret = 0;
1335
1336 flags |= FMODE_EXCL;
1337
1338 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1339 /* Just open everything we can; ignore failures here */
1340 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1341 continue;
1342
1343 if (!latest_dev ||
1344 device->generation > latest_dev->generation)
1345 latest_dev = device;
1346 }
1347 if (fs_devices->open_devices == 0) {
1348 ret = -EINVAL;
1349 goto out;
1350 }
1351 fs_devices->opened = 1;
1352 fs_devices->latest_bdev = latest_dev->bdev;
1353 fs_devices->total_rw_bytes = 0;
1354 out:
1355 return ret;
1356 }
1357
1358 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1359 {
1360 struct btrfs_device *dev1, *dev2;
1361
1362 dev1 = list_entry(a, struct btrfs_device, dev_list);
1363 dev2 = list_entry(b, struct btrfs_device, dev_list);
1364
1365 if (dev1->devid < dev2->devid)
1366 return -1;
1367 else if (dev1->devid > dev2->devid)
1368 return 1;
1369 return 0;
1370 }
1371
1372 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1373 fmode_t flags, void *holder)
1374 {
1375 int ret;
1376
1377 lockdep_assert_held(&uuid_mutex);
1378
1379 mutex_lock(&fs_devices->device_list_mutex);
1380 if (fs_devices->opened) {
1381 fs_devices->opened++;
1382 ret = 0;
1383 } else {
1384 list_sort(NULL, &fs_devices->devices, devid_cmp);
1385 ret = open_fs_devices(fs_devices, flags, holder);
1386 }
1387 mutex_unlock(&fs_devices->device_list_mutex);
1388
1389 return ret;
1390 }
1391
1392 static void btrfs_release_disk_super(struct page *page)
1393 {
1394 kunmap(page);
1395 put_page(page);
1396 }
1397
1398 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1399 struct page **page,
1400 struct btrfs_super_block **disk_super)
1401 {
1402 void *p;
1403 pgoff_t index;
1404
1405 /* make sure our super fits in the device */
1406 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1407 return 1;
1408
1409 /* make sure our super fits in the page */
1410 if (sizeof(**disk_super) > PAGE_SIZE)
1411 return 1;
1412
1413 /* make sure our super doesn't straddle pages on disk */
1414 index = bytenr >> PAGE_SHIFT;
1415 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1416 return 1;
1417
1418 /* pull in the page with our super */
1419 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1420 index, GFP_KERNEL);
1421
1422 if (IS_ERR_OR_NULL(*page))
1423 return 1;
1424
1425 p = kmap(*page);
1426
1427 /* align our pointer to the offset of the super block */
1428 *disk_super = p + offset_in_page(bytenr);
1429
1430 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1431 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1432 btrfs_release_disk_super(*page);
1433 return 1;
1434 }
1435
1436 if ((*disk_super)->label[0] &&
1437 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1438 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1439
1440 return 0;
1441 }
1442
1443 int btrfs_forget_devices(const char *path)
1444 {
1445 int ret;
1446
1447 mutex_lock(&uuid_mutex);
1448 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1449 mutex_unlock(&uuid_mutex);
1450
1451 return ret;
1452 }
1453
1454 /*
1455 * Look for a btrfs signature on a device. This may be called out of the mount path
1456 * and we are not allowed to call set_blocksize during the scan. The superblock
1457 * is read via pagecache
1458 */
1459 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1460 void *holder)
1461 {
1462 struct btrfs_super_block *disk_super;
1463 bool new_device_added = false;
1464 struct btrfs_device *device = NULL;
1465 struct block_device *bdev;
1466 struct page *page;
1467 u64 bytenr;
1468
1469 lockdep_assert_held(&uuid_mutex);
1470
1471 /*
1472 * we would like to check all the supers, but that would make
1473 * a btrfs mount succeed after a mkfs from a different FS.
1474 * So, we need to add a special mount option to scan for
1475 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1476 */
1477 bytenr = btrfs_sb_offset(0);
1478 flags |= FMODE_EXCL;
1479
1480 bdev = blkdev_get_by_path(path, flags, holder);
1481 if (IS_ERR(bdev))
1482 return ERR_CAST(bdev);
1483
1484 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1485 device = ERR_PTR(-EINVAL);
1486 goto error_bdev_put;
1487 }
1488
1489 device = device_list_add(path, disk_super, &new_device_added);
1490 if (!IS_ERR(device)) {
1491 if (new_device_added)
1492 btrfs_free_stale_devices(path, device);
1493 }
1494
1495 btrfs_release_disk_super(page);
1496
1497 error_bdev_put:
1498 blkdev_put(bdev, flags);
1499
1500 return device;
1501 }
1502
1503 /*
1504 * Try to find a chunk that intersects [start, start + len] range and when one
1505 * such is found, record the end of it in *start
1506 */
1507 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1508 u64 len)
1509 {
1510 u64 physical_start, physical_end;
1511
1512 lockdep_assert_held(&device->fs_info->chunk_mutex);
1513
1514 if (!find_first_extent_bit(&device->alloc_state, *start,
1515 &physical_start, &physical_end,
1516 CHUNK_ALLOCATED, NULL)) {
1517
1518 if (in_range(physical_start, *start, len) ||
1519 in_range(*start, physical_start,
1520 physical_end - physical_start)) {
1521 *start = physical_end + 1;
1522 return true;
1523 }
1524 }
1525 return false;
1526 }
1527
1528
1529 /*
1530 * find_free_dev_extent_start - find free space in the specified device
1531 * @device: the device which we search the free space in
1532 * @num_bytes: the size of the free space that we need
1533 * @search_start: the position from which to begin the search
1534 * @start: store the start of the free space.
1535 * @len: the size of the free space. that we find, or the size
1536 * of the max free space if we don't find suitable free space
1537 *
1538 * this uses a pretty simple search, the expectation is that it is
1539 * called very infrequently and that a given device has a small number
1540 * of extents
1541 *
1542 * @start is used to store the start of the free space if we find. But if we
1543 * don't find suitable free space, it will be used to store the start position
1544 * of the max free space.
1545 *
1546 * @len is used to store the size of the free space that we find.
1547 * But if we don't find suitable free space, it is used to store the size of
1548 * the max free space.
1549 */
1550 int find_free_dev_extent_start(struct btrfs_device *device, u64 num_bytes,
1551 u64 search_start, u64 *start, u64 *len)
1552 {
1553 struct btrfs_fs_info *fs_info = device->fs_info;
1554 struct btrfs_root *root = fs_info->dev_root;
1555 struct btrfs_key key;
1556 struct btrfs_dev_extent *dev_extent;
1557 struct btrfs_path *path;
1558 u64 hole_size;
1559 u64 max_hole_start;
1560 u64 max_hole_size;
1561 u64 extent_end;
1562 u64 search_end = device->total_bytes;
1563 int ret;
1564 int slot;
1565 struct extent_buffer *l;
1566
1567 /*
1568 * We don't want to overwrite the superblock on the drive nor any area
1569 * used by the boot loader (grub for example), so we make sure to start
1570 * at an offset of at least 1MB.
1571 */
1572 search_start = max_t(u64, search_start, SZ_1M);
1573
1574 path = btrfs_alloc_path();
1575 if (!path)
1576 return -ENOMEM;
1577
1578 max_hole_start = search_start;
1579 max_hole_size = 0;
1580
1581 again:
1582 if (search_start >= search_end ||
1583 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1584 ret = -ENOSPC;
1585 goto out;
1586 }
1587
1588 path->reada = READA_FORWARD;
1589 path->search_commit_root = 1;
1590 path->skip_locking = 1;
1591
1592 key.objectid = device->devid;
1593 key.offset = search_start;
1594 key.type = BTRFS_DEV_EXTENT_KEY;
1595
1596 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1597 if (ret < 0)
1598 goto out;
1599 if (ret > 0) {
1600 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1601 if (ret < 0)
1602 goto out;
1603 }
1604
1605 while (1) {
1606 l = path->nodes[0];
1607 slot = path->slots[0];
1608 if (slot >= btrfs_header_nritems(l)) {
1609 ret = btrfs_next_leaf(root, path);
1610 if (ret == 0)
1611 continue;
1612 if (ret < 0)
1613 goto out;
1614
1615 break;
1616 }
1617 btrfs_item_key_to_cpu(l, &key, slot);
1618
1619 if (key.objectid < device->devid)
1620 goto next;
1621
1622 if (key.objectid > device->devid)
1623 break;
1624
1625 if (key.type != BTRFS_DEV_EXTENT_KEY)
1626 goto next;
1627
1628 if (key.offset > search_start) {
1629 hole_size = key.offset - search_start;
1630
1631 /*
1632 * Have to check before we set max_hole_start, otherwise
1633 * we could end up sending back this offset anyway.
1634 */
1635 if (contains_pending_extent(device, &search_start,
1636 hole_size)) {
1637 if (key.offset >= search_start)
1638 hole_size = key.offset - search_start;
1639 else
1640 hole_size = 0;
1641 }
1642
1643 if (hole_size > max_hole_size) {
1644 max_hole_start = search_start;
1645 max_hole_size = hole_size;
1646 }
1647
1648 /*
1649 * If this free space is greater than which we need,
1650 * it must be the max free space that we have found
1651 * until now, so max_hole_start must point to the start
1652 * of this free space and the length of this free space
1653 * is stored in max_hole_size. Thus, we return
1654 * max_hole_start and max_hole_size and go back to the
1655 * caller.
1656 */
1657 if (hole_size >= num_bytes) {
1658 ret = 0;
1659 goto out;
1660 }
1661 }
1662
1663 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1664 extent_end = key.offset + btrfs_dev_extent_length(l,
1665 dev_extent);
1666 if (extent_end > search_start)
1667 search_start = extent_end;
1668 next:
1669 path->slots[0]++;
1670 cond_resched();
1671 }
1672
1673 /*
1674 * At this point, search_start should be the end of
1675 * allocated dev extents, and when shrinking the device,
1676 * search_end may be smaller than search_start.
1677 */
1678 if (search_end > search_start) {
1679 hole_size = search_end - search_start;
1680
1681 if (contains_pending_extent(device, &search_start, hole_size)) {
1682 btrfs_release_path(path);
1683 goto again;
1684 }
1685
1686 if (hole_size > max_hole_size) {
1687 max_hole_start = search_start;
1688 max_hole_size = hole_size;
1689 }
1690 }
1691
1692 /* See above. */
1693 if (max_hole_size < num_bytes)
1694 ret = -ENOSPC;
1695 else
1696 ret = 0;
1697
1698 out:
1699 btrfs_free_path(path);
1700 *start = max_hole_start;
1701 if (len)
1702 *len = max_hole_size;
1703 return ret;
1704 }
1705
1706 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1707 u64 *start, u64 *len)
1708 {
1709 /* FIXME use last free of some kind */
1710 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1711 }
1712
1713 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1714 struct btrfs_device *device,
1715 u64 start, u64 *dev_extent_len)
1716 {
1717 struct btrfs_fs_info *fs_info = device->fs_info;
1718 struct btrfs_root *root = fs_info->dev_root;
1719 int ret;
1720 struct btrfs_path *path;
1721 struct btrfs_key key;
1722 struct btrfs_key found_key;
1723 struct extent_buffer *leaf = NULL;
1724 struct btrfs_dev_extent *extent = NULL;
1725
1726 path = btrfs_alloc_path();
1727 if (!path)
1728 return -ENOMEM;
1729
1730 key.objectid = device->devid;
1731 key.offset = start;
1732 key.type = BTRFS_DEV_EXTENT_KEY;
1733 again:
1734 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1735 if (ret > 0) {
1736 ret = btrfs_previous_item(root, path, key.objectid,
1737 BTRFS_DEV_EXTENT_KEY);
1738 if (ret)
1739 goto out;
1740 leaf = path->nodes[0];
1741 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1742 extent = btrfs_item_ptr(leaf, path->slots[0],
1743 struct btrfs_dev_extent);
1744 BUG_ON(found_key.offset > start || found_key.offset +
1745 btrfs_dev_extent_length(leaf, extent) < start);
1746 key = found_key;
1747 btrfs_release_path(path);
1748 goto again;
1749 } else if (ret == 0) {
1750 leaf = path->nodes[0];
1751 extent = btrfs_item_ptr(leaf, path->slots[0],
1752 struct btrfs_dev_extent);
1753 } else {
1754 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1755 goto out;
1756 }
1757
1758 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1759
1760 ret = btrfs_del_item(trans, root, path);
1761 if (ret) {
1762 btrfs_handle_fs_error(fs_info, ret,
1763 "Failed to remove dev extent item");
1764 } else {
1765 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1766 }
1767 out:
1768 btrfs_free_path(path);
1769 return ret;
1770 }
1771
1772 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1773 struct btrfs_device *device,
1774 u64 chunk_offset, u64 start, u64 num_bytes)
1775 {
1776 int ret;
1777 struct btrfs_path *path;
1778 struct btrfs_fs_info *fs_info = device->fs_info;
1779 struct btrfs_root *root = fs_info->dev_root;
1780 struct btrfs_dev_extent *extent;
1781 struct extent_buffer *leaf;
1782 struct btrfs_key key;
1783
1784 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1785 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1786 path = btrfs_alloc_path();
1787 if (!path)
1788 return -ENOMEM;
1789
1790 key.objectid = device->devid;
1791 key.offset = start;
1792 key.type = BTRFS_DEV_EXTENT_KEY;
1793 ret = btrfs_insert_empty_item(trans, root, path, &key,
1794 sizeof(*extent));
1795 if (ret)
1796 goto out;
1797
1798 leaf = path->nodes[0];
1799 extent = btrfs_item_ptr(leaf, path->slots[0],
1800 struct btrfs_dev_extent);
1801 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1802 BTRFS_CHUNK_TREE_OBJECTID);
1803 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1804 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1805 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1806
1807 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1808 btrfs_mark_buffer_dirty(leaf);
1809 out:
1810 btrfs_free_path(path);
1811 return ret;
1812 }
1813
1814 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1815 {
1816 struct extent_map_tree *em_tree;
1817 struct extent_map *em;
1818 struct rb_node *n;
1819 u64 ret = 0;
1820
1821 em_tree = &fs_info->mapping_tree.map_tree;
1822 read_lock(&em_tree->lock);
1823 n = rb_last(&em_tree->map.rb_root);
1824 if (n) {
1825 em = rb_entry(n, struct extent_map, rb_node);
1826 ret = em->start + em->len;
1827 }
1828 read_unlock(&em_tree->lock);
1829
1830 return ret;
1831 }
1832
1833 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1834 u64 *devid_ret)
1835 {
1836 int ret;
1837 struct btrfs_key key;
1838 struct btrfs_key found_key;
1839 struct btrfs_path *path;
1840
1841 path = btrfs_alloc_path();
1842 if (!path)
1843 return -ENOMEM;
1844
1845 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1846 key.type = BTRFS_DEV_ITEM_KEY;
1847 key.offset = (u64)-1;
1848
1849 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1850 if (ret < 0)
1851 goto error;
1852
1853 BUG_ON(ret == 0); /* Corruption */
1854
1855 ret = btrfs_previous_item(fs_info->chunk_root, path,
1856 BTRFS_DEV_ITEMS_OBJECTID,
1857 BTRFS_DEV_ITEM_KEY);
1858 if (ret) {
1859 *devid_ret = 1;
1860 } else {
1861 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1862 path->slots[0]);
1863 *devid_ret = found_key.offset + 1;
1864 }
1865 ret = 0;
1866 error:
1867 btrfs_free_path(path);
1868 return ret;
1869 }
1870
1871 /*
1872 * the device information is stored in the chunk root
1873 * the btrfs_device struct should be fully filled in
1874 */
1875 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1876 struct btrfs_device *device)
1877 {
1878 int ret;
1879 struct btrfs_path *path;
1880 struct btrfs_dev_item *dev_item;
1881 struct extent_buffer *leaf;
1882 struct btrfs_key key;
1883 unsigned long ptr;
1884
1885 path = btrfs_alloc_path();
1886 if (!path)
1887 return -ENOMEM;
1888
1889 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1890 key.type = BTRFS_DEV_ITEM_KEY;
1891 key.offset = device->devid;
1892
1893 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1894 &key, sizeof(*dev_item));
1895 if (ret)
1896 goto out;
1897
1898 leaf = path->nodes[0];
1899 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1900
1901 btrfs_set_device_id(leaf, dev_item, device->devid);
1902 btrfs_set_device_generation(leaf, dev_item, 0);
1903 btrfs_set_device_type(leaf, dev_item, device->type);
1904 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1905 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1906 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1907 btrfs_set_device_total_bytes(leaf, dev_item,
1908 btrfs_device_get_disk_total_bytes(device));
1909 btrfs_set_device_bytes_used(leaf, dev_item,
1910 btrfs_device_get_bytes_used(device));
1911 btrfs_set_device_group(leaf, dev_item, 0);
1912 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1913 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1914 btrfs_set_device_start_offset(leaf, dev_item, 0);
1915
1916 ptr = btrfs_device_uuid(dev_item);
1917 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1918 ptr = btrfs_device_fsid(dev_item);
1919 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1920 ptr, BTRFS_FSID_SIZE);
1921 btrfs_mark_buffer_dirty(leaf);
1922
1923 ret = 0;
1924 out:
1925 btrfs_free_path(path);
1926 return ret;
1927 }
1928
1929 /*
1930 * Function to update ctime/mtime for a given device path.
1931 * Mainly used for ctime/mtime based probe like libblkid.
1932 */
1933 static void update_dev_time(const char *path_name)
1934 {
1935 struct file *filp;
1936
1937 filp = filp_open(path_name, O_RDWR, 0);
1938 if (IS_ERR(filp))
1939 return;
1940 file_update_time(filp);
1941 filp_close(filp, NULL);
1942 }
1943
1944 static int btrfs_rm_dev_item(struct btrfs_device *device)
1945 {
1946 struct btrfs_root *root = device->fs_info->chunk_root;
1947 int ret;
1948 struct btrfs_path *path;
1949 struct btrfs_key key;
1950 struct btrfs_trans_handle *trans;
1951
1952 path = btrfs_alloc_path();
1953 if (!path)
1954 return -ENOMEM;
1955
1956 trans = btrfs_start_transaction(root, 0);
1957 if (IS_ERR(trans)) {
1958 btrfs_free_path(path);
1959 return PTR_ERR(trans);
1960 }
1961 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1962 key.type = BTRFS_DEV_ITEM_KEY;
1963 key.offset = device->devid;
1964
1965 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1966 if (ret) {
1967 if (ret > 0)
1968 ret = -ENOENT;
1969 btrfs_abort_transaction(trans, ret);
1970 btrfs_end_transaction(trans);
1971 goto out;
1972 }
1973
1974 ret = btrfs_del_item(trans, root, path);
1975 if (ret) {
1976 btrfs_abort_transaction(trans, ret);
1977 btrfs_end_transaction(trans);
1978 }
1979
1980 out:
1981 btrfs_free_path(path);
1982 if (!ret)
1983 ret = btrfs_commit_transaction(trans);
1984 return ret;
1985 }
1986
1987 /*
1988 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1989 * filesystem. It's up to the caller to adjust that number regarding eg. device
1990 * replace.
1991 */
1992 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1993 u64 num_devices)
1994 {
1995 u64 all_avail;
1996 unsigned seq;
1997 int i;
1998
1999 do {
2000 seq = read_seqbegin(&fs_info->profiles_lock);
2001
2002 all_avail = fs_info->avail_data_alloc_bits |
2003 fs_info->avail_system_alloc_bits |
2004 fs_info->avail_metadata_alloc_bits;
2005 } while (read_seqretry(&fs_info->profiles_lock, seq));
2006
2007 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2008 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2009 continue;
2010
2011 if (num_devices < btrfs_raid_array[i].devs_min) {
2012 int ret = btrfs_raid_array[i].mindev_error;
2013
2014 if (ret)
2015 return ret;
2016 }
2017 }
2018
2019 return 0;
2020 }
2021
2022 static struct btrfs_device * btrfs_find_next_active_device(
2023 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2024 {
2025 struct btrfs_device *next_device;
2026
2027 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2028 if (next_device != device &&
2029 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2030 && next_device->bdev)
2031 return next_device;
2032 }
2033
2034 return NULL;
2035 }
2036
2037 /*
2038 * Helper function to check if the given device is part of s_bdev / latest_bdev
2039 * and replace it with the provided or the next active device, in the context
2040 * where this function called, there should be always be another device (or
2041 * this_dev) which is active.
2042 */
2043 void btrfs_assign_next_active_device(struct btrfs_device *device,
2044 struct btrfs_device *this_dev)
2045 {
2046 struct btrfs_fs_info *fs_info = device->fs_info;
2047 struct btrfs_device *next_device;
2048
2049 if (this_dev)
2050 next_device = this_dev;
2051 else
2052 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2053 device);
2054 ASSERT(next_device);
2055
2056 if (fs_info->sb->s_bdev &&
2057 (fs_info->sb->s_bdev == device->bdev))
2058 fs_info->sb->s_bdev = next_device->bdev;
2059
2060 if (fs_info->fs_devices->latest_bdev == device->bdev)
2061 fs_info->fs_devices->latest_bdev = next_device->bdev;
2062 }
2063
2064 /*
2065 * Return btrfs_fs_devices::num_devices excluding the device that's being
2066 * currently replaced.
2067 */
2068 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2069 {
2070 u64 num_devices = fs_info->fs_devices->num_devices;
2071
2072 down_read(&fs_info->dev_replace.rwsem);
2073 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2074 ASSERT(num_devices > 1);
2075 num_devices--;
2076 }
2077 up_read(&fs_info->dev_replace.rwsem);
2078
2079 return num_devices;
2080 }
2081
2082 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2083 u64 devid)
2084 {
2085 struct btrfs_device *device;
2086 struct btrfs_fs_devices *cur_devices;
2087 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2088 u64 num_devices;
2089 int ret = 0;
2090
2091 mutex_lock(&uuid_mutex);
2092
2093 num_devices = btrfs_num_devices(fs_info);
2094
2095 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2096 if (ret)
2097 goto out;
2098
2099 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2100
2101 if (IS_ERR(device)) {
2102 if (PTR_ERR(device) == -ENOENT &&
2103 strcmp(device_path, "missing") == 0)
2104 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2105 else
2106 ret = PTR_ERR(device);
2107 goto out;
2108 }
2109
2110 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2111 btrfs_warn_in_rcu(fs_info,
2112 "cannot remove device %s (devid %llu) due to active swapfile",
2113 rcu_str_deref(device->name), device->devid);
2114 ret = -ETXTBSY;
2115 goto out;
2116 }
2117
2118 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2119 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2120 goto out;
2121 }
2122
2123 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2124 fs_info->fs_devices->rw_devices == 1) {
2125 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2126 goto out;
2127 }
2128
2129 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2130 mutex_lock(&fs_info->chunk_mutex);
2131 list_del_init(&device->dev_alloc_list);
2132 device->fs_devices->rw_devices--;
2133 mutex_unlock(&fs_info->chunk_mutex);
2134 }
2135
2136 mutex_unlock(&uuid_mutex);
2137 ret = btrfs_shrink_device(device, 0);
2138 mutex_lock(&uuid_mutex);
2139 if (ret)
2140 goto error_undo;
2141
2142 /*
2143 * TODO: the superblock still includes this device in its num_devices
2144 * counter although write_all_supers() is not locked out. This
2145 * could give a filesystem state which requires a degraded mount.
2146 */
2147 ret = btrfs_rm_dev_item(device);
2148 if (ret)
2149 goto error_undo;
2150
2151 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2152 btrfs_scrub_cancel_dev(device);
2153
2154 /*
2155 * the device list mutex makes sure that we don't change
2156 * the device list while someone else is writing out all
2157 * the device supers. Whoever is writing all supers, should
2158 * lock the device list mutex before getting the number of
2159 * devices in the super block (super_copy). Conversely,
2160 * whoever updates the number of devices in the super block
2161 * (super_copy) should hold the device list mutex.
2162 */
2163
2164 /*
2165 * In normal cases the cur_devices == fs_devices. But in case
2166 * of deleting a seed device, the cur_devices should point to
2167 * its own fs_devices listed under the fs_devices->seed.
2168 */
2169 cur_devices = device->fs_devices;
2170 mutex_lock(&fs_devices->device_list_mutex);
2171 list_del_rcu(&device->dev_list);
2172
2173 cur_devices->num_devices--;
2174 cur_devices->total_devices--;
2175 /* Update total_devices of the parent fs_devices if it's seed */
2176 if (cur_devices != fs_devices)
2177 fs_devices->total_devices--;
2178
2179 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2180 cur_devices->missing_devices--;
2181
2182 btrfs_assign_next_active_device(device, NULL);
2183
2184 if (device->bdev) {
2185 cur_devices->open_devices--;
2186 /* remove sysfs entry */
2187 btrfs_sysfs_rm_device_link(fs_devices, device);
2188 }
2189
2190 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2191 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2192 mutex_unlock(&fs_devices->device_list_mutex);
2193
2194 /*
2195 * at this point, the device is zero sized and detached from
2196 * the devices list. All that's left is to zero out the old
2197 * supers and free the device.
2198 */
2199 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2200 btrfs_scratch_superblocks(device->bdev, device->name->str);
2201
2202 btrfs_close_bdev(device);
2203 synchronize_rcu();
2204 btrfs_free_device(device);
2205
2206 if (cur_devices->open_devices == 0) {
2207 while (fs_devices) {
2208 if (fs_devices->seed == cur_devices) {
2209 fs_devices->seed = cur_devices->seed;
2210 break;
2211 }
2212 fs_devices = fs_devices->seed;
2213 }
2214 cur_devices->seed = NULL;
2215 close_fs_devices(cur_devices);
2216 free_fs_devices(cur_devices);
2217 }
2218
2219 out:
2220 mutex_unlock(&uuid_mutex);
2221 return ret;
2222
2223 error_undo:
2224 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2225 mutex_lock(&fs_info->chunk_mutex);
2226 list_add(&device->dev_alloc_list,
2227 &fs_devices->alloc_list);
2228 device->fs_devices->rw_devices++;
2229 mutex_unlock(&fs_info->chunk_mutex);
2230 }
2231 goto out;
2232 }
2233
2234 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2235 {
2236 struct btrfs_fs_devices *fs_devices;
2237
2238 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2239
2240 /*
2241 * in case of fs with no seed, srcdev->fs_devices will point
2242 * to fs_devices of fs_info. However when the dev being replaced is
2243 * a seed dev it will point to the seed's local fs_devices. In short
2244 * srcdev will have its correct fs_devices in both the cases.
2245 */
2246 fs_devices = srcdev->fs_devices;
2247
2248 list_del_rcu(&srcdev->dev_list);
2249 list_del(&srcdev->dev_alloc_list);
2250 fs_devices->num_devices--;
2251 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2252 fs_devices->missing_devices--;
2253
2254 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2255 fs_devices->rw_devices--;
2256
2257 if (srcdev->bdev)
2258 fs_devices->open_devices--;
2259 }
2260
2261 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2262 {
2263 struct btrfs_fs_info *fs_info = srcdev->fs_info;
2264 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2265
2266 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2267 /* zero out the old super if it is writable */
2268 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2269 }
2270
2271 btrfs_close_bdev(srcdev);
2272 synchronize_rcu();
2273 btrfs_free_device(srcdev);
2274
2275 /* if this is no devs we rather delete the fs_devices */
2276 if (!fs_devices->num_devices) {
2277 struct btrfs_fs_devices *tmp_fs_devices;
2278
2279 /*
2280 * On a mounted FS, num_devices can't be zero unless it's a
2281 * seed. In case of a seed device being replaced, the replace
2282 * target added to the sprout FS, so there will be no more
2283 * device left under the seed FS.
2284 */
2285 ASSERT(fs_devices->seeding);
2286
2287 tmp_fs_devices = fs_info->fs_devices;
2288 while (tmp_fs_devices) {
2289 if (tmp_fs_devices->seed == fs_devices) {
2290 tmp_fs_devices->seed = fs_devices->seed;
2291 break;
2292 }
2293 tmp_fs_devices = tmp_fs_devices->seed;
2294 }
2295 fs_devices->seed = NULL;
2296 close_fs_devices(fs_devices);
2297 free_fs_devices(fs_devices);
2298 }
2299 }
2300
2301 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2302 {
2303 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2304
2305 WARN_ON(!tgtdev);
2306 mutex_lock(&fs_devices->device_list_mutex);
2307
2308 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2309
2310 if (tgtdev->bdev)
2311 fs_devices->open_devices--;
2312
2313 fs_devices->num_devices--;
2314
2315 btrfs_assign_next_active_device(tgtdev, NULL);
2316
2317 list_del_rcu(&tgtdev->dev_list);
2318
2319 mutex_unlock(&fs_devices->device_list_mutex);
2320
2321 /*
2322 * The update_dev_time() with in btrfs_scratch_superblocks()
2323 * may lead to a call to btrfs_show_devname() which will try
2324 * to hold device_list_mutex. And here this device
2325 * is already out of device list, so we don't have to hold
2326 * the device_list_mutex lock.
2327 */
2328 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2329
2330 btrfs_close_bdev(tgtdev);
2331 synchronize_rcu();
2332 btrfs_free_device(tgtdev);
2333 }
2334
2335 static struct btrfs_device *btrfs_find_device_by_path(
2336 struct btrfs_fs_info *fs_info, const char *device_path)
2337 {
2338 int ret = 0;
2339 struct btrfs_super_block *disk_super;
2340 u64 devid;
2341 u8 *dev_uuid;
2342 struct block_device *bdev;
2343 struct buffer_head *bh;
2344 struct btrfs_device *device;
2345
2346 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2347 fs_info->bdev_holder, 0, &bdev, &bh);
2348 if (ret)
2349 return ERR_PTR(ret);
2350 disk_super = (struct btrfs_super_block *)bh->b_data;
2351 devid = btrfs_stack_device_id(&disk_super->dev_item);
2352 dev_uuid = disk_super->dev_item.uuid;
2353 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2354 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2355 disk_super->metadata_uuid, true);
2356 else
2357 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2358 disk_super->fsid, true);
2359
2360 brelse(bh);
2361 if (!device)
2362 device = ERR_PTR(-ENOENT);
2363 blkdev_put(bdev, FMODE_READ);
2364 return device;
2365 }
2366
2367 /*
2368 * Lookup a device given by device id, or the path if the id is 0.
2369 */
2370 struct btrfs_device *btrfs_find_device_by_devspec(
2371 struct btrfs_fs_info *fs_info, u64 devid,
2372 const char *device_path)
2373 {
2374 struct btrfs_device *device;
2375
2376 if (devid) {
2377 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2378 NULL, true);
2379 if (!device)
2380 return ERR_PTR(-ENOENT);
2381 return device;
2382 }
2383
2384 if (!device_path || !device_path[0])
2385 return ERR_PTR(-EINVAL);
2386
2387 if (strcmp(device_path, "missing") == 0) {
2388 /* Find first missing device */
2389 list_for_each_entry(device, &fs_info->fs_devices->devices,
2390 dev_list) {
2391 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2392 &device->dev_state) && !device->bdev)
2393 return device;
2394 }
2395 return ERR_PTR(-ENOENT);
2396 }
2397
2398 return btrfs_find_device_by_path(fs_info, device_path);
2399 }
2400
2401 /*
2402 * does all the dirty work required for changing file system's UUID.
2403 */
2404 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2405 {
2406 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2407 struct btrfs_fs_devices *old_devices;
2408 struct btrfs_fs_devices *seed_devices;
2409 struct btrfs_super_block *disk_super = fs_info->super_copy;
2410 struct btrfs_device *device;
2411 u64 super_flags;
2412
2413 lockdep_assert_held(&uuid_mutex);
2414 if (!fs_devices->seeding)
2415 return -EINVAL;
2416
2417 seed_devices = alloc_fs_devices(NULL, NULL);
2418 if (IS_ERR(seed_devices))
2419 return PTR_ERR(seed_devices);
2420
2421 old_devices = clone_fs_devices(fs_devices);
2422 if (IS_ERR(old_devices)) {
2423 kfree(seed_devices);
2424 return PTR_ERR(old_devices);
2425 }
2426
2427 list_add(&old_devices->fs_list, &fs_uuids);
2428
2429 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2430 seed_devices->opened = 1;
2431 INIT_LIST_HEAD(&seed_devices->devices);
2432 INIT_LIST_HEAD(&seed_devices->alloc_list);
2433 mutex_init(&seed_devices->device_list_mutex);
2434
2435 mutex_lock(&fs_devices->device_list_mutex);
2436 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2437 synchronize_rcu);
2438 list_for_each_entry(device, &seed_devices->devices, dev_list)
2439 device->fs_devices = seed_devices;
2440
2441 mutex_lock(&fs_info->chunk_mutex);
2442 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2443 mutex_unlock(&fs_info->chunk_mutex);
2444
2445 fs_devices->seeding = 0;
2446 fs_devices->num_devices = 0;
2447 fs_devices->open_devices = 0;
2448 fs_devices->missing_devices = 0;
2449 fs_devices->rotating = 0;
2450 fs_devices->seed = seed_devices;
2451
2452 generate_random_uuid(fs_devices->fsid);
2453 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2454 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2455 mutex_unlock(&fs_devices->device_list_mutex);
2456
2457 super_flags = btrfs_super_flags(disk_super) &
2458 ~BTRFS_SUPER_FLAG_SEEDING;
2459 btrfs_set_super_flags(disk_super, super_flags);
2460
2461 return 0;
2462 }
2463
2464 /*
2465 * Store the expected generation for seed devices in device items.
2466 */
2467 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2468 {
2469 struct btrfs_fs_info *fs_info = trans->fs_info;
2470 struct btrfs_root *root = fs_info->chunk_root;
2471 struct btrfs_path *path;
2472 struct extent_buffer *leaf;
2473 struct btrfs_dev_item *dev_item;
2474 struct btrfs_device *device;
2475 struct btrfs_key key;
2476 u8 fs_uuid[BTRFS_FSID_SIZE];
2477 u8 dev_uuid[BTRFS_UUID_SIZE];
2478 u64 devid;
2479 int ret;
2480
2481 path = btrfs_alloc_path();
2482 if (!path)
2483 return -ENOMEM;
2484
2485 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2486 key.offset = 0;
2487 key.type = BTRFS_DEV_ITEM_KEY;
2488
2489 while (1) {
2490 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2491 if (ret < 0)
2492 goto error;
2493
2494 leaf = path->nodes[0];
2495 next_slot:
2496 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2497 ret = btrfs_next_leaf(root, path);
2498 if (ret > 0)
2499 break;
2500 if (ret < 0)
2501 goto error;
2502 leaf = path->nodes[0];
2503 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2504 btrfs_release_path(path);
2505 continue;
2506 }
2507
2508 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2509 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2510 key.type != BTRFS_DEV_ITEM_KEY)
2511 break;
2512
2513 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2514 struct btrfs_dev_item);
2515 devid = btrfs_device_id(leaf, dev_item);
2516 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2517 BTRFS_UUID_SIZE);
2518 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2519 BTRFS_FSID_SIZE);
2520 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2521 fs_uuid, true);
2522 BUG_ON(!device); /* Logic error */
2523
2524 if (device->fs_devices->seeding) {
2525 btrfs_set_device_generation(leaf, dev_item,
2526 device->generation);
2527 btrfs_mark_buffer_dirty(leaf);
2528 }
2529
2530 path->slots[0]++;
2531 goto next_slot;
2532 }
2533 ret = 0;
2534 error:
2535 btrfs_free_path(path);
2536 return ret;
2537 }
2538
2539 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2540 {
2541 struct btrfs_root *root = fs_info->dev_root;
2542 struct request_queue *q;
2543 struct btrfs_trans_handle *trans;
2544 struct btrfs_device *device;
2545 struct block_device *bdev;
2546 struct super_block *sb = fs_info->sb;
2547 struct rcu_string *name;
2548 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2549 u64 orig_super_total_bytes;
2550 u64 orig_super_num_devices;
2551 int seeding_dev = 0;
2552 int ret = 0;
2553 bool unlocked = false;
2554
2555 if (sb_rdonly(sb) && !fs_devices->seeding)
2556 return -EROFS;
2557
2558 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2559 fs_info->bdev_holder);
2560 if (IS_ERR(bdev))
2561 return PTR_ERR(bdev);
2562
2563 if (fs_devices->seeding) {
2564 seeding_dev = 1;
2565 down_write(&sb->s_umount);
2566 mutex_lock(&uuid_mutex);
2567 }
2568
2569 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2570
2571 mutex_lock(&fs_devices->device_list_mutex);
2572 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2573 if (device->bdev == bdev) {
2574 ret = -EEXIST;
2575 mutex_unlock(
2576 &fs_devices->device_list_mutex);
2577 goto error;
2578 }
2579 }
2580 mutex_unlock(&fs_devices->device_list_mutex);
2581
2582 device = btrfs_alloc_device(fs_info, NULL, NULL);
2583 if (IS_ERR(device)) {
2584 /* we can safely leave the fs_devices entry around */
2585 ret = PTR_ERR(device);
2586 goto error;
2587 }
2588
2589 name = rcu_string_strdup(device_path, GFP_KERNEL);
2590 if (!name) {
2591 ret = -ENOMEM;
2592 goto error_free_device;
2593 }
2594 rcu_assign_pointer(device->name, name);
2595
2596 trans = btrfs_start_transaction(root, 0);
2597 if (IS_ERR(trans)) {
2598 ret = PTR_ERR(trans);
2599 goto error_free_device;
2600 }
2601
2602 q = bdev_get_queue(bdev);
2603 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2604 device->generation = trans->transid;
2605 device->io_width = fs_info->sectorsize;
2606 device->io_align = fs_info->sectorsize;
2607 device->sector_size = fs_info->sectorsize;
2608 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2609 fs_info->sectorsize);
2610 device->disk_total_bytes = device->total_bytes;
2611 device->commit_total_bytes = device->total_bytes;
2612 device->fs_info = fs_info;
2613 device->bdev = bdev;
2614 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2615 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2616 device->mode = FMODE_EXCL;
2617 device->dev_stats_valid = 1;
2618 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2619
2620 if (seeding_dev) {
2621 sb->s_flags &= ~SB_RDONLY;
2622 ret = btrfs_prepare_sprout(fs_info);
2623 if (ret) {
2624 btrfs_abort_transaction(trans, ret);
2625 goto error_trans;
2626 }
2627 }
2628
2629 device->fs_devices = fs_devices;
2630
2631 mutex_lock(&fs_devices->device_list_mutex);
2632 mutex_lock(&fs_info->chunk_mutex);
2633 list_add_rcu(&device->dev_list, &fs_devices->devices);
2634 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2635 fs_devices->num_devices++;
2636 fs_devices->open_devices++;
2637 fs_devices->rw_devices++;
2638 fs_devices->total_devices++;
2639 fs_devices->total_rw_bytes += device->total_bytes;
2640
2641 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2642
2643 if (!blk_queue_nonrot(q))
2644 fs_devices->rotating = 1;
2645
2646 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2647 btrfs_set_super_total_bytes(fs_info->super_copy,
2648 round_down(orig_super_total_bytes + device->total_bytes,
2649 fs_info->sectorsize));
2650
2651 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2652 btrfs_set_super_num_devices(fs_info->super_copy,
2653 orig_super_num_devices + 1);
2654
2655 /* add sysfs device entry */
2656 btrfs_sysfs_add_device_link(fs_devices, device);
2657
2658 /*
2659 * we've got more storage, clear any full flags on the space
2660 * infos
2661 */
2662 btrfs_clear_space_info_full(fs_info);
2663
2664 mutex_unlock(&fs_info->chunk_mutex);
2665 mutex_unlock(&fs_devices->device_list_mutex);
2666
2667 if (seeding_dev) {
2668 mutex_lock(&fs_info->chunk_mutex);
2669 ret = init_first_rw_device(trans);
2670 mutex_unlock(&fs_info->chunk_mutex);
2671 if (ret) {
2672 btrfs_abort_transaction(trans, ret);
2673 goto error_sysfs;
2674 }
2675 }
2676
2677 ret = btrfs_add_dev_item(trans, device);
2678 if (ret) {
2679 btrfs_abort_transaction(trans, ret);
2680 goto error_sysfs;
2681 }
2682
2683 if (seeding_dev) {
2684 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2685
2686 ret = btrfs_finish_sprout(trans);
2687 if (ret) {
2688 btrfs_abort_transaction(trans, ret);
2689 goto error_sysfs;
2690 }
2691
2692 /* Sprouting would change fsid of the mounted root,
2693 * so rename the fsid on the sysfs
2694 */
2695 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2696 fs_info->fs_devices->fsid);
2697 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2698 btrfs_warn(fs_info,
2699 "sysfs: failed to create fsid for sprout");
2700 }
2701
2702 ret = btrfs_commit_transaction(trans);
2703
2704 if (seeding_dev) {
2705 mutex_unlock(&uuid_mutex);
2706 up_write(&sb->s_umount);
2707 unlocked = true;
2708
2709 if (ret) /* transaction commit */
2710 return ret;
2711
2712 ret = btrfs_relocate_sys_chunks(fs_info);
2713 if (ret < 0)
2714 btrfs_handle_fs_error(fs_info, ret,
2715 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2716 trans = btrfs_attach_transaction(root);
2717 if (IS_ERR(trans)) {
2718 if (PTR_ERR(trans) == -ENOENT)
2719 return 0;
2720 ret = PTR_ERR(trans);
2721 trans = NULL;
2722 goto error_sysfs;
2723 }
2724 ret = btrfs_commit_transaction(trans);
2725 }
2726
2727 /* Update ctime/mtime for libblkid */
2728 update_dev_time(device_path);
2729 return ret;
2730
2731 error_sysfs:
2732 btrfs_sysfs_rm_device_link(fs_devices, device);
2733 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2734 mutex_lock(&fs_info->chunk_mutex);
2735 list_del_rcu(&device->dev_list);
2736 list_del(&device->dev_alloc_list);
2737 fs_info->fs_devices->num_devices--;
2738 fs_info->fs_devices->open_devices--;
2739 fs_info->fs_devices->rw_devices--;
2740 fs_info->fs_devices->total_devices--;
2741 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2742 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2743 btrfs_set_super_total_bytes(fs_info->super_copy,
2744 orig_super_total_bytes);
2745 btrfs_set_super_num_devices(fs_info->super_copy,
2746 orig_super_num_devices);
2747 mutex_unlock(&fs_info->chunk_mutex);
2748 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2749 error_trans:
2750 if (seeding_dev)
2751 sb->s_flags |= SB_RDONLY;
2752 if (trans)
2753 btrfs_end_transaction(trans);
2754 error_free_device:
2755 btrfs_free_device(device);
2756 error:
2757 blkdev_put(bdev, FMODE_EXCL);
2758 if (seeding_dev && !unlocked) {
2759 mutex_unlock(&uuid_mutex);
2760 up_write(&sb->s_umount);
2761 }
2762 return ret;
2763 }
2764
2765 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2766 struct btrfs_device *device)
2767 {
2768 int ret;
2769 struct btrfs_path *path;
2770 struct btrfs_root *root = device->fs_info->chunk_root;
2771 struct btrfs_dev_item *dev_item;
2772 struct extent_buffer *leaf;
2773 struct btrfs_key key;
2774
2775 path = btrfs_alloc_path();
2776 if (!path)
2777 return -ENOMEM;
2778
2779 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2780 key.type = BTRFS_DEV_ITEM_KEY;
2781 key.offset = device->devid;
2782
2783 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2784 if (ret < 0)
2785 goto out;
2786
2787 if (ret > 0) {
2788 ret = -ENOENT;
2789 goto out;
2790 }
2791
2792 leaf = path->nodes[0];
2793 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2794
2795 btrfs_set_device_id(leaf, dev_item, device->devid);
2796 btrfs_set_device_type(leaf, dev_item, device->type);
2797 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2798 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2799 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2800 btrfs_set_device_total_bytes(leaf, dev_item,
2801 btrfs_device_get_disk_total_bytes(device));
2802 btrfs_set_device_bytes_used(leaf, dev_item,
2803 btrfs_device_get_bytes_used(device));
2804 btrfs_mark_buffer_dirty(leaf);
2805
2806 out:
2807 btrfs_free_path(path);
2808 return ret;
2809 }
2810
2811 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2812 struct btrfs_device *device, u64 new_size)
2813 {
2814 struct btrfs_fs_info *fs_info = device->fs_info;
2815 struct btrfs_super_block *super_copy = fs_info->super_copy;
2816 u64 old_total;
2817 u64 diff;
2818
2819 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2820 return -EACCES;
2821
2822 new_size = round_down(new_size, fs_info->sectorsize);
2823
2824 mutex_lock(&fs_info->chunk_mutex);
2825 old_total = btrfs_super_total_bytes(super_copy);
2826 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2827
2828 if (new_size <= device->total_bytes ||
2829 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2830 mutex_unlock(&fs_info->chunk_mutex);
2831 return -EINVAL;
2832 }
2833
2834 btrfs_set_super_total_bytes(super_copy,
2835 round_down(old_total + diff, fs_info->sectorsize));
2836 device->fs_devices->total_rw_bytes += diff;
2837
2838 btrfs_device_set_total_bytes(device, new_size);
2839 btrfs_device_set_disk_total_bytes(device, new_size);
2840 btrfs_clear_space_info_full(device->fs_info);
2841 if (list_empty(&device->post_commit_list))
2842 list_add_tail(&device->post_commit_list,
2843 &trans->transaction->dev_update_list);
2844 mutex_unlock(&fs_info->chunk_mutex);
2845
2846 return btrfs_update_device(trans, device);
2847 }
2848
2849 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2850 {
2851 struct btrfs_fs_info *fs_info = trans->fs_info;
2852 struct btrfs_root *root = fs_info->chunk_root;
2853 int ret;
2854 struct btrfs_path *path;
2855 struct btrfs_key key;
2856
2857 path = btrfs_alloc_path();
2858 if (!path)
2859 return -ENOMEM;
2860
2861 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2862 key.offset = chunk_offset;
2863 key.type = BTRFS_CHUNK_ITEM_KEY;
2864
2865 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2866 if (ret < 0)
2867 goto out;
2868 else if (ret > 0) { /* Logic error or corruption */
2869 btrfs_handle_fs_error(fs_info, -ENOENT,
2870 "Failed lookup while freeing chunk.");
2871 ret = -ENOENT;
2872 goto out;
2873 }
2874
2875 ret = btrfs_del_item(trans, root, path);
2876 if (ret < 0)
2877 btrfs_handle_fs_error(fs_info, ret,
2878 "Failed to delete chunk item.");
2879 out:
2880 btrfs_free_path(path);
2881 return ret;
2882 }
2883
2884 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2885 {
2886 struct btrfs_super_block *super_copy = fs_info->super_copy;
2887 struct btrfs_disk_key *disk_key;
2888 struct btrfs_chunk *chunk;
2889 u8 *ptr;
2890 int ret = 0;
2891 u32 num_stripes;
2892 u32 array_size;
2893 u32 len = 0;
2894 u32 cur;
2895 struct btrfs_key key;
2896
2897 mutex_lock(&fs_info->chunk_mutex);
2898 array_size = btrfs_super_sys_array_size(super_copy);
2899
2900 ptr = super_copy->sys_chunk_array;
2901 cur = 0;
2902
2903 while (cur < array_size) {
2904 disk_key = (struct btrfs_disk_key *)ptr;
2905 btrfs_disk_key_to_cpu(&key, disk_key);
2906
2907 len = sizeof(*disk_key);
2908
2909 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2910 chunk = (struct btrfs_chunk *)(ptr + len);
2911 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2912 len += btrfs_chunk_item_size(num_stripes);
2913 } else {
2914 ret = -EIO;
2915 break;
2916 }
2917 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2918 key.offset == chunk_offset) {
2919 memmove(ptr, ptr + len, array_size - (cur + len));
2920 array_size -= len;
2921 btrfs_set_super_sys_array_size(super_copy, array_size);
2922 } else {
2923 ptr += len;
2924 cur += len;
2925 }
2926 }
2927 mutex_unlock(&fs_info->chunk_mutex);
2928 return ret;
2929 }
2930
2931 /*
2932 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2933 * @logical: Logical block offset in bytes.
2934 * @length: Length of extent in bytes.
2935 *
2936 * Return: Chunk mapping or ERR_PTR.
2937 */
2938 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2939 u64 logical, u64 length)
2940 {
2941 struct extent_map_tree *em_tree;
2942 struct extent_map *em;
2943
2944 em_tree = &fs_info->mapping_tree.map_tree;
2945 read_lock(&em_tree->lock);
2946 em = lookup_extent_mapping(em_tree, logical, length);
2947 read_unlock(&em_tree->lock);
2948
2949 if (!em) {
2950 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2951 logical, length);
2952 return ERR_PTR(-EINVAL);
2953 }
2954
2955 if (em->start > logical || em->start + em->len < logical) {
2956 btrfs_crit(fs_info,
2957 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2958 logical, length, em->start, em->start + em->len);
2959 free_extent_map(em);
2960 return ERR_PTR(-EINVAL);
2961 }
2962
2963 /* callers are responsible for dropping em's ref. */
2964 return em;
2965 }
2966
2967 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2968 {
2969 struct btrfs_fs_info *fs_info = trans->fs_info;
2970 struct extent_map *em;
2971 struct map_lookup *map;
2972 u64 dev_extent_len = 0;
2973 int i, ret = 0;
2974 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2975
2976 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2977 if (IS_ERR(em)) {
2978 /*
2979 * This is a logic error, but we don't want to just rely on the
2980 * user having built with ASSERT enabled, so if ASSERT doesn't
2981 * do anything we still error out.
2982 */
2983 ASSERT(0);
2984 return PTR_ERR(em);
2985 }
2986 map = em->map_lookup;
2987 mutex_lock(&fs_info->chunk_mutex);
2988 check_system_chunk(trans, map->type);
2989 mutex_unlock(&fs_info->chunk_mutex);
2990
2991 /*
2992 * Take the device list mutex to prevent races with the final phase of
2993 * a device replace operation that replaces the device object associated
2994 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2995 */
2996 mutex_lock(&fs_devices->device_list_mutex);
2997 for (i = 0; i < map->num_stripes; i++) {
2998 struct btrfs_device *device = map->stripes[i].dev;
2999 ret = btrfs_free_dev_extent(trans, device,
3000 map->stripes[i].physical,
3001 &dev_extent_len);
3002 if (ret) {
3003 mutex_unlock(&fs_devices->device_list_mutex);
3004 btrfs_abort_transaction(trans, ret);
3005 goto out;
3006 }
3007
3008 if (device->bytes_used > 0) {
3009 mutex_lock(&fs_info->chunk_mutex);
3010 btrfs_device_set_bytes_used(device,
3011 device->bytes_used - dev_extent_len);
3012 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3013 btrfs_clear_space_info_full(fs_info);
3014 mutex_unlock(&fs_info->chunk_mutex);
3015 }
3016
3017 ret = btrfs_update_device(trans, device);
3018 if (ret) {
3019 mutex_unlock(&fs_devices->device_list_mutex);
3020 btrfs_abort_transaction(trans, ret);
3021 goto out;
3022 }
3023 }
3024 mutex_unlock(&fs_devices->device_list_mutex);
3025
3026 ret = btrfs_free_chunk(trans, chunk_offset);
3027 if (ret) {
3028 btrfs_abort_transaction(trans, ret);
3029 goto out;
3030 }
3031
3032 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3033
3034 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3035 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3036 if (ret) {
3037 btrfs_abort_transaction(trans, ret);
3038 goto out;
3039 }
3040 }
3041
3042 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3043 if (ret) {
3044 btrfs_abort_transaction(trans, ret);
3045 goto out;
3046 }
3047
3048 out:
3049 /* once for us */
3050 free_extent_map(em);
3051 return ret;
3052 }
3053
3054 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3055 {
3056 struct btrfs_root *root = fs_info->chunk_root;
3057 struct btrfs_trans_handle *trans;
3058 int ret;
3059
3060 /*
3061 * Prevent races with automatic removal of unused block groups.
3062 * After we relocate and before we remove the chunk with offset
3063 * chunk_offset, automatic removal of the block group can kick in,
3064 * resulting in a failure when calling btrfs_remove_chunk() below.
3065 *
3066 * Make sure to acquire this mutex before doing a tree search (dev
3067 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3068 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3069 * we release the path used to search the chunk/dev tree and before
3070 * the current task acquires this mutex and calls us.
3071 */
3072 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3073
3074 ret = btrfs_can_relocate(fs_info, chunk_offset);
3075 if (ret)
3076 return -ENOSPC;
3077
3078 /* step one, relocate all the extents inside this chunk */
3079 btrfs_scrub_pause(fs_info);
3080 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3081 btrfs_scrub_continue(fs_info);
3082 if (ret)
3083 return ret;
3084
3085 /*
3086 * We add the kobjects here (and after forcing data chunk creation)
3087 * since relocation is the only place we'll create chunks of a new
3088 * type at runtime. The only place where we'll remove the last
3089 * chunk of a type is the call immediately below this one. Even
3090 * so, we're protected against races with the cleaner thread since
3091 * we're covered by the delete_unused_bgs_mutex.
3092 */
3093 btrfs_add_raid_kobjects(fs_info);
3094
3095 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3096 chunk_offset);
3097 if (IS_ERR(trans)) {
3098 ret = PTR_ERR(trans);
3099 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3100 return ret;
3101 }
3102
3103 /*
3104 * step two, delete the device extents and the
3105 * chunk tree entries
3106 */
3107 ret = btrfs_remove_chunk(trans, chunk_offset);
3108 btrfs_end_transaction(trans);
3109 return ret;
3110 }
3111
3112 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3113 {
3114 struct btrfs_root *chunk_root = fs_info->chunk_root;
3115 struct btrfs_path *path;
3116 struct extent_buffer *leaf;
3117 struct btrfs_chunk *chunk;
3118 struct btrfs_key key;
3119 struct btrfs_key found_key;
3120 u64 chunk_type;
3121 bool retried = false;
3122 int failed = 0;
3123 int ret;
3124
3125 path = btrfs_alloc_path();
3126 if (!path)
3127 return -ENOMEM;
3128
3129 again:
3130 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3131 key.offset = (u64)-1;
3132 key.type = BTRFS_CHUNK_ITEM_KEY;
3133
3134 while (1) {
3135 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3136 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3137 if (ret < 0) {
3138 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3139 goto error;
3140 }
3141 BUG_ON(ret == 0); /* Corruption */
3142
3143 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3144 key.type);
3145 if (ret)
3146 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3147 if (ret < 0)
3148 goto error;
3149 if (ret > 0)
3150 break;
3151
3152 leaf = path->nodes[0];
3153 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3154
3155 chunk = btrfs_item_ptr(leaf, path->slots[0],
3156 struct btrfs_chunk);
3157 chunk_type = btrfs_chunk_type(leaf, chunk);
3158 btrfs_release_path(path);
3159
3160 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3161 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3162 if (ret == -ENOSPC)
3163 failed++;
3164 else
3165 BUG_ON(ret);
3166 }
3167 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3168
3169 if (found_key.offset == 0)
3170 break;
3171 key.offset = found_key.offset - 1;
3172 }
3173 ret = 0;
3174 if (failed && !retried) {
3175 failed = 0;
3176 retried = true;
3177 goto again;
3178 } else if (WARN_ON(failed && retried)) {
3179 ret = -ENOSPC;
3180 }
3181 error:
3182 btrfs_free_path(path);
3183 return ret;
3184 }
3185
3186 /*
3187 * return 1 : allocate a data chunk successfully,
3188 * return <0: errors during allocating a data chunk,
3189 * return 0 : no need to allocate a data chunk.
3190 */
3191 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3192 u64 chunk_offset)
3193 {
3194 struct btrfs_block_group_cache *cache;
3195 u64 bytes_used;
3196 u64 chunk_type;
3197
3198 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3199 ASSERT(cache);
3200 chunk_type = cache->flags;
3201 btrfs_put_block_group(cache);
3202
3203 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3204 spin_lock(&fs_info->data_sinfo->lock);
3205 bytes_used = fs_info->data_sinfo->bytes_used;
3206 spin_unlock(&fs_info->data_sinfo->lock);
3207
3208 if (!bytes_used) {
3209 struct btrfs_trans_handle *trans;
3210 int ret;
3211
3212 trans = btrfs_join_transaction(fs_info->tree_root);
3213 if (IS_ERR(trans))
3214 return PTR_ERR(trans);
3215
3216 ret = btrfs_force_chunk_alloc(trans,
3217 BTRFS_BLOCK_GROUP_DATA);
3218 btrfs_end_transaction(trans);
3219 if (ret < 0)
3220 return ret;
3221
3222 btrfs_add_raid_kobjects(fs_info);
3223
3224 return 1;
3225 }
3226 }
3227 return 0;
3228 }
3229
3230 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3231 struct btrfs_balance_control *bctl)
3232 {
3233 struct btrfs_root *root = fs_info->tree_root;
3234 struct btrfs_trans_handle *trans;
3235 struct btrfs_balance_item *item;
3236 struct btrfs_disk_balance_args disk_bargs;
3237 struct btrfs_path *path;
3238 struct extent_buffer *leaf;
3239 struct btrfs_key key;
3240 int ret, err;
3241
3242 path = btrfs_alloc_path();
3243 if (!path)
3244 return -ENOMEM;
3245
3246 trans = btrfs_start_transaction(root, 0);
3247 if (IS_ERR(trans)) {
3248 btrfs_free_path(path);
3249 return PTR_ERR(trans);
3250 }
3251
3252 key.objectid = BTRFS_BALANCE_OBJECTID;
3253 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3254 key.offset = 0;
3255
3256 ret = btrfs_insert_empty_item(trans, root, path, &key,
3257 sizeof(*item));
3258 if (ret)
3259 goto out;
3260
3261 leaf = path->nodes[0];
3262 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3263
3264 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3265
3266 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3267 btrfs_set_balance_data(leaf, item, &disk_bargs);
3268 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3269 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3270 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3271 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3272
3273 btrfs_set_balance_flags(leaf, item, bctl->flags);
3274
3275 btrfs_mark_buffer_dirty(leaf);
3276 out:
3277 btrfs_free_path(path);
3278 err = btrfs_commit_transaction(trans);
3279 if (err && !ret)
3280 ret = err;
3281 return ret;
3282 }
3283
3284 static int del_balance_item(struct btrfs_fs_info *fs_info)
3285 {
3286 struct btrfs_root *root = fs_info->tree_root;
3287 struct btrfs_trans_handle *trans;
3288 struct btrfs_path *path;
3289 struct btrfs_key key;
3290 int ret, err;
3291
3292 path = btrfs_alloc_path();
3293 if (!path)
3294 return -ENOMEM;
3295
3296 trans = btrfs_start_transaction(root, 0);
3297 if (IS_ERR(trans)) {
3298 btrfs_free_path(path);
3299 return PTR_ERR(trans);
3300 }
3301
3302 key.objectid = BTRFS_BALANCE_OBJECTID;
3303 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3304 key.offset = 0;
3305
3306 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3307 if (ret < 0)
3308 goto out;
3309 if (ret > 0) {
3310 ret = -ENOENT;
3311 goto out;
3312 }
3313
3314 ret = btrfs_del_item(trans, root, path);
3315 out:
3316 btrfs_free_path(path);
3317 err = btrfs_commit_transaction(trans);
3318 if (err && !ret)
3319 ret = err;
3320 return ret;
3321 }
3322
3323 /*
3324 * This is a heuristic used to reduce the number of chunks balanced on
3325 * resume after balance was interrupted.
3326 */
3327 static void update_balance_args(struct btrfs_balance_control *bctl)
3328 {
3329 /*
3330 * Turn on soft mode for chunk types that were being converted.
3331 */
3332 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3333 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3334 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3335 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3336 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3337 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3338
3339 /*
3340 * Turn on usage filter if is not already used. The idea is
3341 * that chunks that we have already balanced should be
3342 * reasonably full. Don't do it for chunks that are being
3343 * converted - that will keep us from relocating unconverted
3344 * (albeit full) chunks.
3345 */
3346 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3347 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3348 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3349 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3350 bctl->data.usage = 90;
3351 }
3352 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3353 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3354 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3355 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3356 bctl->sys.usage = 90;
3357 }
3358 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3359 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3360 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3361 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3362 bctl->meta.usage = 90;
3363 }
3364 }
3365
3366 /*
3367 * Clear the balance status in fs_info and delete the balance item from disk.
3368 */
3369 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3370 {
3371 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3372 int ret;
3373
3374 BUG_ON(!fs_info->balance_ctl);
3375
3376 spin_lock(&fs_info->balance_lock);
3377 fs_info->balance_ctl = NULL;
3378 spin_unlock(&fs_info->balance_lock);
3379
3380 kfree(bctl);
3381 ret = del_balance_item(fs_info);
3382 if (ret)
3383 btrfs_handle_fs_error(fs_info, ret, NULL);
3384 }
3385
3386 /*
3387 * Balance filters. Return 1 if chunk should be filtered out
3388 * (should not be balanced).
3389 */
3390 static int chunk_profiles_filter(u64 chunk_type,
3391 struct btrfs_balance_args *bargs)
3392 {
3393 chunk_type = chunk_to_extended(chunk_type) &
3394 BTRFS_EXTENDED_PROFILE_MASK;
3395
3396 if (bargs->profiles & chunk_type)
3397 return 0;
3398
3399 return 1;
3400 }
3401
3402 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3403 struct btrfs_balance_args *bargs)
3404 {
3405 struct btrfs_block_group_cache *cache;
3406 u64 chunk_used;
3407 u64 user_thresh_min;
3408 u64 user_thresh_max;
3409 int ret = 1;
3410
3411 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3412 chunk_used = btrfs_block_group_used(&cache->item);
3413
3414 if (bargs->usage_min == 0)
3415 user_thresh_min = 0;
3416 else
3417 user_thresh_min = div_factor_fine(cache->key.offset,
3418 bargs->usage_min);
3419
3420 if (bargs->usage_max == 0)
3421 user_thresh_max = 1;
3422 else if (bargs->usage_max > 100)
3423 user_thresh_max = cache->key.offset;
3424 else
3425 user_thresh_max = div_factor_fine(cache->key.offset,
3426 bargs->usage_max);
3427
3428 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3429 ret = 0;
3430
3431 btrfs_put_block_group(cache);
3432 return ret;
3433 }
3434
3435 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3436 u64 chunk_offset, struct btrfs_balance_args *bargs)
3437 {
3438 struct btrfs_block_group_cache *cache;
3439 u64 chunk_used, user_thresh;
3440 int ret = 1;
3441
3442 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3443 chunk_used = btrfs_block_group_used(&cache->item);
3444
3445 if (bargs->usage_min == 0)
3446 user_thresh = 1;
3447 else if (bargs->usage > 100)
3448 user_thresh = cache->key.offset;
3449 else
3450 user_thresh = div_factor_fine(cache->key.offset,
3451 bargs->usage);
3452
3453 if (chunk_used < user_thresh)
3454 ret = 0;
3455
3456 btrfs_put_block_group(cache);
3457 return ret;
3458 }
3459
3460 static int chunk_devid_filter(struct extent_buffer *leaf,
3461 struct btrfs_chunk *chunk,
3462 struct btrfs_balance_args *bargs)
3463 {
3464 struct btrfs_stripe *stripe;
3465 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3466 int i;
3467
3468 for (i = 0; i < num_stripes; i++) {
3469 stripe = btrfs_stripe_nr(chunk, i);
3470 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3471 return 0;
3472 }
3473
3474 return 1;
3475 }
3476
3477 /* [pstart, pend) */
3478 static int chunk_drange_filter(struct extent_buffer *leaf,
3479 struct btrfs_chunk *chunk,
3480 struct btrfs_balance_args *bargs)
3481 {
3482 struct btrfs_stripe *stripe;
3483 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3484 u64 stripe_offset;
3485 u64 stripe_length;
3486 int factor;
3487 int i;
3488
3489 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3490 return 0;
3491
3492 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3493 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3494 factor = num_stripes / 2;
3495 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3496 factor = num_stripes - 1;
3497 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3498 factor = num_stripes - 2;
3499 } else {
3500 factor = num_stripes;
3501 }
3502
3503 for (i = 0; i < num_stripes; i++) {
3504 stripe = btrfs_stripe_nr(chunk, i);
3505 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3506 continue;
3507
3508 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3509 stripe_length = btrfs_chunk_length(leaf, chunk);
3510 stripe_length = div_u64(stripe_length, factor);
3511
3512 if (stripe_offset < bargs->pend &&
3513 stripe_offset + stripe_length > bargs->pstart)
3514 return 0;
3515 }
3516
3517 return 1;
3518 }
3519
3520 /* [vstart, vend) */
3521 static int chunk_vrange_filter(struct extent_buffer *leaf,
3522 struct btrfs_chunk *chunk,
3523 u64 chunk_offset,
3524 struct btrfs_balance_args *bargs)
3525 {
3526 if (chunk_offset < bargs->vend &&
3527 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3528 /* at least part of the chunk is inside this vrange */
3529 return 0;
3530
3531 return 1;
3532 }
3533
3534 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3535 struct btrfs_chunk *chunk,
3536 struct btrfs_balance_args *bargs)
3537 {
3538 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3539
3540 if (bargs->stripes_min <= num_stripes
3541 && num_stripes <= bargs->stripes_max)
3542 return 0;
3543
3544 return 1;
3545 }
3546
3547 static int chunk_soft_convert_filter(u64 chunk_type,
3548 struct btrfs_balance_args *bargs)
3549 {
3550 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3551 return 0;
3552
3553 chunk_type = chunk_to_extended(chunk_type) &
3554 BTRFS_EXTENDED_PROFILE_MASK;
3555
3556 if (bargs->target == chunk_type)
3557 return 1;
3558
3559 return 0;
3560 }
3561
3562 static int should_balance_chunk(struct extent_buffer *leaf,
3563 struct btrfs_chunk *chunk, u64 chunk_offset)
3564 {
3565 struct btrfs_fs_info *fs_info = leaf->fs_info;
3566 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3567 struct btrfs_balance_args *bargs = NULL;
3568 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3569
3570 /* type filter */
3571 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3572 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3573 return 0;
3574 }
3575
3576 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3577 bargs = &bctl->data;
3578 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3579 bargs = &bctl->sys;
3580 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3581 bargs = &bctl->meta;
3582
3583 /* profiles filter */
3584 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3585 chunk_profiles_filter(chunk_type, bargs)) {
3586 return 0;
3587 }
3588
3589 /* usage filter */
3590 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3591 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3592 return 0;
3593 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3594 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3595 return 0;
3596 }
3597
3598 /* devid filter */
3599 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3600 chunk_devid_filter(leaf, chunk, bargs)) {
3601 return 0;
3602 }
3603
3604 /* drange filter, makes sense only with devid filter */
3605 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3606 chunk_drange_filter(leaf, chunk, bargs)) {
3607 return 0;
3608 }
3609
3610 /* vrange filter */
3611 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3612 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3613 return 0;
3614 }
3615
3616 /* stripes filter */
3617 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3618 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3619 return 0;
3620 }
3621
3622 /* soft profile changing mode */
3623 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3624 chunk_soft_convert_filter(chunk_type, bargs)) {
3625 return 0;
3626 }
3627
3628 /*
3629 * limited by count, must be the last filter
3630 */
3631 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3632 if (bargs->limit == 0)
3633 return 0;
3634 else
3635 bargs->limit--;
3636 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3637 /*
3638 * Same logic as the 'limit' filter; the minimum cannot be
3639 * determined here because we do not have the global information
3640 * about the count of all chunks that satisfy the filters.
3641 */
3642 if (bargs->limit_max == 0)
3643 return 0;
3644 else
3645 bargs->limit_max--;
3646 }
3647
3648 return 1;
3649 }
3650
3651 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3652 {
3653 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3654 struct btrfs_root *chunk_root = fs_info->chunk_root;
3655 u64 chunk_type;
3656 struct btrfs_chunk *chunk;
3657 struct btrfs_path *path = NULL;
3658 struct btrfs_key key;
3659 struct btrfs_key found_key;
3660 struct extent_buffer *leaf;
3661 int slot;
3662 int ret;
3663 int enospc_errors = 0;
3664 bool counting = true;
3665 /* The single value limit and min/max limits use the same bytes in the */
3666 u64 limit_data = bctl->data.limit;
3667 u64 limit_meta = bctl->meta.limit;
3668 u64 limit_sys = bctl->sys.limit;
3669 u32 count_data = 0;
3670 u32 count_meta = 0;
3671 u32 count_sys = 0;
3672 int chunk_reserved = 0;
3673
3674 path = btrfs_alloc_path();
3675 if (!path) {
3676 ret = -ENOMEM;
3677 goto error;
3678 }
3679
3680 /* zero out stat counters */
3681 spin_lock(&fs_info->balance_lock);
3682 memset(&bctl->stat, 0, sizeof(bctl->stat));
3683 spin_unlock(&fs_info->balance_lock);
3684 again:
3685 if (!counting) {
3686 /*
3687 * The single value limit and min/max limits use the same bytes
3688 * in the
3689 */
3690 bctl->data.limit = limit_data;
3691 bctl->meta.limit = limit_meta;
3692 bctl->sys.limit = limit_sys;
3693 }
3694 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3695 key.offset = (u64)-1;
3696 key.type = BTRFS_CHUNK_ITEM_KEY;
3697
3698 while (1) {
3699 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3700 atomic_read(&fs_info->balance_cancel_req)) {
3701 ret = -ECANCELED;
3702 goto error;
3703 }
3704
3705 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3706 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3707 if (ret < 0) {
3708 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3709 goto error;
3710 }
3711
3712 /*
3713 * this shouldn't happen, it means the last relocate
3714 * failed
3715 */
3716 if (ret == 0)
3717 BUG(); /* FIXME break ? */
3718
3719 ret = btrfs_previous_item(chunk_root, path, 0,
3720 BTRFS_CHUNK_ITEM_KEY);
3721 if (ret) {
3722 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3723 ret = 0;
3724 break;
3725 }
3726
3727 leaf = path->nodes[0];
3728 slot = path->slots[0];
3729 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3730
3731 if (found_key.objectid != key.objectid) {
3732 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3733 break;
3734 }
3735
3736 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3737 chunk_type = btrfs_chunk_type(leaf, chunk);
3738
3739 if (!counting) {
3740 spin_lock(&fs_info->balance_lock);
3741 bctl->stat.considered++;
3742 spin_unlock(&fs_info->balance_lock);
3743 }
3744
3745 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3746
3747 btrfs_release_path(path);
3748 if (!ret) {
3749 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3750 goto loop;
3751 }
3752
3753 if (counting) {
3754 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3755 spin_lock(&fs_info->balance_lock);
3756 bctl->stat.expected++;
3757 spin_unlock(&fs_info->balance_lock);
3758
3759 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3760 count_data++;
3761 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3762 count_sys++;
3763 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3764 count_meta++;
3765
3766 goto loop;
3767 }
3768
3769 /*
3770 * Apply limit_min filter, no need to check if the LIMITS
3771 * filter is used, limit_min is 0 by default
3772 */
3773 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3774 count_data < bctl->data.limit_min)
3775 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3776 count_meta < bctl->meta.limit_min)
3777 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3778 count_sys < bctl->sys.limit_min)) {
3779 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3780 goto loop;
3781 }
3782
3783 if (!chunk_reserved) {
3784 /*
3785 * We may be relocating the only data chunk we have,
3786 * which could potentially end up with losing data's
3787 * raid profile, so lets allocate an empty one in
3788 * advance.
3789 */
3790 ret = btrfs_may_alloc_data_chunk(fs_info,
3791 found_key.offset);
3792 if (ret < 0) {
3793 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3794 goto error;
3795 } else if (ret == 1) {
3796 chunk_reserved = 1;
3797 }
3798 }
3799
3800 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3801 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3802 if (ret == -ENOSPC) {
3803 enospc_errors++;
3804 } else if (ret == -ETXTBSY) {
3805 btrfs_info(fs_info,
3806 "skipping relocation of block group %llu due to active swapfile",
3807 found_key.offset);
3808 ret = 0;
3809 } else if (ret) {
3810 goto error;
3811 } else {
3812 spin_lock(&fs_info->balance_lock);
3813 bctl->stat.completed++;
3814 spin_unlock(&fs_info->balance_lock);
3815 }
3816 loop:
3817 if (found_key.offset == 0)
3818 break;
3819 key.offset = found_key.offset - 1;
3820 }
3821
3822 if (counting) {
3823 btrfs_release_path(path);
3824 counting = false;
3825 goto again;
3826 }
3827 error:
3828 btrfs_free_path(path);
3829 if (enospc_errors) {
3830 btrfs_info(fs_info, "%d enospc errors during balance",
3831 enospc_errors);
3832 if (!ret)
3833 ret = -ENOSPC;
3834 }
3835
3836 return ret;
3837 }
3838
3839 /**
3840 * alloc_profile_is_valid - see if a given profile is valid and reduced
3841 * @flags: profile to validate
3842 * @extended: if true @flags is treated as an extended profile
3843 */
3844 static int alloc_profile_is_valid(u64 flags, int extended)
3845 {
3846 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3847 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3848
3849 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3850
3851 /* 1) check that all other bits are zeroed */
3852 if (flags & ~mask)
3853 return 0;
3854
3855 /* 2) see if profile is reduced */
3856 if (flags == 0)
3857 return !extended; /* "" is valid for usual profiles */
3858
3859 /* true if exactly one bit set */
3860 return is_power_of_2(flags);
3861 }
3862
3863 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3864 {
3865 /* cancel requested || normal exit path */
3866 return atomic_read(&fs_info->balance_cancel_req) ||
3867 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3868 atomic_read(&fs_info->balance_cancel_req) == 0);
3869 }
3870
3871 /* Non-zero return value signifies invalidity */
3872 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3873 u64 allowed)
3874 {
3875 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3876 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3877 (bctl_arg->target & ~allowed)));
3878 }
3879
3880 /*
3881 * Fill @buf with textual description of balance filter flags @bargs, up to
3882 * @size_buf including the terminating null. The output may be trimmed if it
3883 * does not fit into the provided buffer.
3884 */
3885 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3886 u32 size_buf)
3887 {
3888 int ret;
3889 u32 size_bp = size_buf;
3890 char *bp = buf;
3891 u64 flags = bargs->flags;
3892 char tmp_buf[128] = {'\0'};
3893
3894 if (!flags)
3895 return;
3896
3897 #define CHECK_APPEND_NOARG(a) \
3898 do { \
3899 ret = snprintf(bp, size_bp, (a)); \
3900 if (ret < 0 || ret >= size_bp) \
3901 goto out_overflow; \
3902 size_bp -= ret; \
3903 bp += ret; \
3904 } while (0)
3905
3906 #define CHECK_APPEND_1ARG(a, v1) \
3907 do { \
3908 ret = snprintf(bp, size_bp, (a), (v1)); \
3909 if (ret < 0 || ret >= size_bp) \
3910 goto out_overflow; \
3911 size_bp -= ret; \
3912 bp += ret; \
3913 } while (0)
3914
3915 #define CHECK_APPEND_2ARG(a, v1, v2) \
3916 do { \
3917 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3918 if (ret < 0 || ret >= size_bp) \
3919 goto out_overflow; \
3920 size_bp -= ret; \
3921 bp += ret; \
3922 } while (0)
3923
3924 if (flags & BTRFS_BALANCE_ARGS_CONVERT) {
3925 int index = btrfs_bg_flags_to_raid_index(bargs->target);
3926
3927 CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index));
3928 }
3929
3930 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3931 CHECK_APPEND_NOARG("soft,");
3932
3933 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3934 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3935 sizeof(tmp_buf));
3936 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3937 }
3938
3939 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3940 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3941
3942 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3943 CHECK_APPEND_2ARG("usage=%u..%u,",
3944 bargs->usage_min, bargs->usage_max);
3945
3946 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3947 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3948
3949 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3950 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3951 bargs->pstart, bargs->pend);
3952
3953 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3954 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3955 bargs->vstart, bargs->vend);
3956
3957 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3958 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3959
3960 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3961 CHECK_APPEND_2ARG("limit=%u..%u,",
3962 bargs->limit_min, bargs->limit_max);
3963
3964 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3965 CHECK_APPEND_2ARG("stripes=%u..%u,",
3966 bargs->stripes_min, bargs->stripes_max);
3967
3968 #undef CHECK_APPEND_2ARG
3969 #undef CHECK_APPEND_1ARG
3970 #undef CHECK_APPEND_NOARG
3971
3972 out_overflow:
3973
3974 if (size_bp < size_buf)
3975 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3976 else
3977 buf[0] = '\0';
3978 }
3979
3980 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3981 {
3982 u32 size_buf = 1024;
3983 char tmp_buf[192] = {'\0'};
3984 char *buf;
3985 char *bp;
3986 u32 size_bp = size_buf;
3987 int ret;
3988 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3989
3990 buf = kzalloc(size_buf, GFP_KERNEL);
3991 if (!buf)
3992 return;
3993
3994 bp = buf;
3995
3996 #define CHECK_APPEND_1ARG(a, v1) \
3997 do { \
3998 ret = snprintf(bp, size_bp, (a), (v1)); \
3999 if (ret < 0 || ret >= size_bp) \
4000 goto out_overflow; \
4001 size_bp -= ret; \
4002 bp += ret; \
4003 } while (0)
4004
4005 if (bctl->flags & BTRFS_BALANCE_FORCE)
4006 CHECK_APPEND_1ARG("%s", "-f ");
4007
4008 if (bctl->flags & BTRFS_BALANCE_DATA) {
4009 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4010 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4011 }
4012
4013 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4014 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4015 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4016 }
4017
4018 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4019 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4020 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4021 }
4022
4023 #undef CHECK_APPEND_1ARG
4024
4025 out_overflow:
4026
4027 if (size_bp < size_buf)
4028 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4029 btrfs_info(fs_info, "balance: %s %s",
4030 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4031 "resume" : "start", buf);
4032
4033 kfree(buf);
4034 }
4035
4036 /*
4037 * Should be called with balance mutexe held
4038 */
4039 int btrfs_balance(struct btrfs_fs_info *fs_info,
4040 struct btrfs_balance_control *bctl,
4041 struct btrfs_ioctl_balance_args *bargs)
4042 {
4043 u64 meta_target, data_target;
4044 u64 allowed;
4045 int mixed = 0;
4046 int ret;
4047 u64 num_devices;
4048 unsigned seq;
4049 bool reducing_integrity;
4050
4051 if (btrfs_fs_closing(fs_info) ||
4052 atomic_read(&fs_info->balance_pause_req) ||
4053 atomic_read(&fs_info->balance_cancel_req)) {
4054 ret = -EINVAL;
4055 goto out;
4056 }
4057
4058 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4059 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4060 mixed = 1;
4061
4062 /*
4063 * In case of mixed groups both data and meta should be picked,
4064 * and identical options should be given for both of them.
4065 */
4066 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4067 if (mixed && (bctl->flags & allowed)) {
4068 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4069 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4070 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4071 btrfs_err(fs_info,
4072 "balance: mixed groups data and metadata options must be the same");
4073 ret = -EINVAL;
4074 goto out;
4075 }
4076 }
4077
4078 num_devices = btrfs_num_devices(fs_info);
4079
4080 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
4081 if (num_devices > 1)
4082 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
4083 if (num_devices > 2)
4084 allowed |= BTRFS_BLOCK_GROUP_RAID5;
4085 if (num_devices > 3)
4086 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
4087 BTRFS_BLOCK_GROUP_RAID6);
4088 if (validate_convert_profile(&bctl->data, allowed)) {
4089 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
4090
4091 btrfs_err(fs_info,
4092 "balance: invalid convert data profile %s",
4093 get_raid_name(index));
4094 ret = -EINVAL;
4095 goto out;
4096 }
4097 if (validate_convert_profile(&bctl->meta, allowed)) {
4098 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
4099
4100 btrfs_err(fs_info,
4101 "balance: invalid convert metadata profile %s",
4102 get_raid_name(index));
4103 ret = -EINVAL;
4104 goto out;
4105 }
4106 if (validate_convert_profile(&bctl->sys, allowed)) {
4107 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
4108
4109 btrfs_err(fs_info,
4110 "balance: invalid convert system profile %s",
4111 get_raid_name(index));
4112 ret = -EINVAL;
4113 goto out;
4114 }
4115
4116 /* allow to reduce meta or sys integrity only if force set */
4117 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4118 BTRFS_BLOCK_GROUP_RAID10 |
4119 BTRFS_BLOCK_GROUP_RAID5 |
4120 BTRFS_BLOCK_GROUP_RAID6;
4121 do {
4122 seq = read_seqbegin(&fs_info->profiles_lock);
4123
4124 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4125 (fs_info->avail_system_alloc_bits & allowed) &&
4126 !(bctl->sys.target & allowed)) ||
4127 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4128 (fs_info->avail_metadata_alloc_bits & allowed) &&
4129 !(bctl->meta.target & allowed)))
4130 reducing_integrity = true;
4131 else
4132 reducing_integrity = false;
4133
4134 /* if we're not converting, the target field is uninitialized */
4135 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4136 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4137 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4138 bctl->data.target : fs_info->avail_data_alloc_bits;
4139 } while (read_seqretry(&fs_info->profiles_lock, seq));
4140
4141 if (reducing_integrity) {
4142 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4143 btrfs_info(fs_info,
4144 "balance: force reducing metadata integrity");
4145 } else {
4146 btrfs_err(fs_info,
4147 "balance: reduces metadata integrity, use --force if you want this");
4148 ret = -EINVAL;
4149 goto out;
4150 }
4151 }
4152
4153 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4154 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4155 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
4156 int data_index = btrfs_bg_flags_to_raid_index(data_target);
4157
4158 btrfs_warn(fs_info,
4159 "balance: metadata profile %s has lower redundancy than data profile %s",
4160 get_raid_name(meta_index), get_raid_name(data_index));
4161 }
4162
4163 ret = insert_balance_item(fs_info, bctl);
4164 if (ret && ret != -EEXIST)
4165 goto out;
4166
4167 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4168 BUG_ON(ret == -EEXIST);
4169 BUG_ON(fs_info->balance_ctl);
4170 spin_lock(&fs_info->balance_lock);
4171 fs_info->balance_ctl = bctl;
4172 spin_unlock(&fs_info->balance_lock);
4173 } else {
4174 BUG_ON(ret != -EEXIST);
4175 spin_lock(&fs_info->balance_lock);
4176 update_balance_args(bctl);
4177 spin_unlock(&fs_info->balance_lock);
4178 }
4179
4180 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4181 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4182 describe_balance_start_or_resume(fs_info);
4183 mutex_unlock(&fs_info->balance_mutex);
4184
4185 ret = __btrfs_balance(fs_info);
4186
4187 mutex_lock(&fs_info->balance_mutex);
4188 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4189 btrfs_info(fs_info, "balance: paused");
4190 else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4191 btrfs_info(fs_info, "balance: canceled");
4192 else
4193 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4194
4195 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4196
4197 if (bargs) {
4198 memset(bargs, 0, sizeof(*bargs));
4199 btrfs_update_ioctl_balance_args(fs_info, bargs);
4200 }
4201
4202 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4203 balance_need_close(fs_info)) {
4204 reset_balance_state(fs_info);
4205 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4206 }
4207
4208 wake_up(&fs_info->balance_wait_q);
4209
4210 return ret;
4211 out:
4212 if (bctl->flags & BTRFS_BALANCE_RESUME)
4213 reset_balance_state(fs_info);
4214 else
4215 kfree(bctl);
4216 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4217
4218 return ret;
4219 }
4220
4221 static int balance_kthread(void *data)
4222 {
4223 struct btrfs_fs_info *fs_info = data;
4224 int ret = 0;
4225
4226 mutex_lock(&fs_info->balance_mutex);
4227 if (fs_info->balance_ctl)
4228 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4229 mutex_unlock(&fs_info->balance_mutex);
4230
4231 return ret;
4232 }
4233
4234 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4235 {
4236 struct task_struct *tsk;
4237
4238 mutex_lock(&fs_info->balance_mutex);
4239 if (!fs_info->balance_ctl) {
4240 mutex_unlock(&fs_info->balance_mutex);
4241 return 0;
4242 }
4243 mutex_unlock(&fs_info->balance_mutex);
4244
4245 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4246 btrfs_info(fs_info, "balance: resume skipped");
4247 return 0;
4248 }
4249
4250 /*
4251 * A ro->rw remount sequence should continue with the paused balance
4252 * regardless of who pauses it, system or the user as of now, so set
4253 * the resume flag.
4254 */
4255 spin_lock(&fs_info->balance_lock);
4256 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4257 spin_unlock(&fs_info->balance_lock);
4258
4259 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4260 return PTR_ERR_OR_ZERO(tsk);
4261 }
4262
4263 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4264 {
4265 struct btrfs_balance_control *bctl;
4266 struct btrfs_balance_item *item;
4267 struct btrfs_disk_balance_args disk_bargs;
4268 struct btrfs_path *path;
4269 struct extent_buffer *leaf;
4270 struct btrfs_key key;
4271 int ret;
4272
4273 path = btrfs_alloc_path();
4274 if (!path)
4275 return -ENOMEM;
4276
4277 key.objectid = BTRFS_BALANCE_OBJECTID;
4278 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4279 key.offset = 0;
4280
4281 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4282 if (ret < 0)
4283 goto out;
4284 if (ret > 0) { /* ret = -ENOENT; */
4285 ret = 0;
4286 goto out;
4287 }
4288
4289 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4290 if (!bctl) {
4291 ret = -ENOMEM;
4292 goto out;
4293 }
4294
4295 leaf = path->nodes[0];
4296 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4297
4298 bctl->flags = btrfs_balance_flags(leaf, item);
4299 bctl->flags |= BTRFS_BALANCE_RESUME;
4300
4301 btrfs_balance_data(leaf, item, &disk_bargs);
4302 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4303 btrfs_balance_meta(leaf, item, &disk_bargs);
4304 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4305 btrfs_balance_sys(leaf, item, &disk_bargs);
4306 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4307
4308 /*
4309 * This should never happen, as the paused balance state is recovered
4310 * during mount without any chance of other exclusive ops to collide.
4311 *
4312 * This gives the exclusive op status to balance and keeps in paused
4313 * state until user intervention (cancel or umount). If the ownership
4314 * cannot be assigned, show a message but do not fail. The balance
4315 * is in a paused state and must have fs_info::balance_ctl properly
4316 * set up.
4317 */
4318 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4319 btrfs_warn(fs_info,
4320 "balance: cannot set exclusive op status, resume manually");
4321
4322 mutex_lock(&fs_info->balance_mutex);
4323 BUG_ON(fs_info->balance_ctl);
4324 spin_lock(&fs_info->balance_lock);
4325 fs_info->balance_ctl = bctl;
4326 spin_unlock(&fs_info->balance_lock);
4327 mutex_unlock(&fs_info->balance_mutex);
4328 out:
4329 btrfs_free_path(path);
4330 return ret;
4331 }
4332
4333 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4334 {
4335 int ret = 0;
4336
4337 mutex_lock(&fs_info->balance_mutex);
4338 if (!fs_info->balance_ctl) {
4339 mutex_unlock(&fs_info->balance_mutex);
4340 return -ENOTCONN;
4341 }
4342
4343 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4344 atomic_inc(&fs_info->balance_pause_req);
4345 mutex_unlock(&fs_info->balance_mutex);
4346
4347 wait_event(fs_info->balance_wait_q,
4348 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4349
4350 mutex_lock(&fs_info->balance_mutex);
4351 /* we are good with balance_ctl ripped off from under us */
4352 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4353 atomic_dec(&fs_info->balance_pause_req);
4354 } else {
4355 ret = -ENOTCONN;
4356 }
4357
4358 mutex_unlock(&fs_info->balance_mutex);
4359 return ret;
4360 }
4361
4362 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4363 {
4364 mutex_lock(&fs_info->balance_mutex);
4365 if (!fs_info->balance_ctl) {
4366 mutex_unlock(&fs_info->balance_mutex);
4367 return -ENOTCONN;
4368 }
4369
4370 /*
4371 * A paused balance with the item stored on disk can be resumed at
4372 * mount time if the mount is read-write. Otherwise it's still paused
4373 * and we must not allow cancelling as it deletes the item.
4374 */
4375 if (sb_rdonly(fs_info->sb)) {
4376 mutex_unlock(&fs_info->balance_mutex);
4377 return -EROFS;
4378 }
4379
4380 atomic_inc(&fs_info->balance_cancel_req);
4381 /*
4382 * if we are running just wait and return, balance item is
4383 * deleted in btrfs_balance in this case
4384 */
4385 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4386 mutex_unlock(&fs_info->balance_mutex);
4387 wait_event(fs_info->balance_wait_q,
4388 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4389 mutex_lock(&fs_info->balance_mutex);
4390 } else {
4391 mutex_unlock(&fs_info->balance_mutex);
4392 /*
4393 * Lock released to allow other waiters to continue, we'll
4394 * reexamine the status again.
4395 */
4396 mutex_lock(&fs_info->balance_mutex);
4397
4398 if (fs_info->balance_ctl) {
4399 reset_balance_state(fs_info);
4400 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4401 btrfs_info(fs_info, "balance: canceled");
4402 }
4403 }
4404
4405 BUG_ON(fs_info->balance_ctl ||
4406 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4407 atomic_dec(&fs_info->balance_cancel_req);
4408 mutex_unlock(&fs_info->balance_mutex);
4409 return 0;
4410 }
4411
4412 static int btrfs_uuid_scan_kthread(void *data)
4413 {
4414 struct btrfs_fs_info *fs_info = data;
4415 struct btrfs_root *root = fs_info->tree_root;
4416 struct btrfs_key key;
4417 struct btrfs_path *path = NULL;
4418 int ret = 0;
4419 struct extent_buffer *eb;
4420 int slot;
4421 struct btrfs_root_item root_item;
4422 u32 item_size;
4423 struct btrfs_trans_handle *trans = NULL;
4424
4425 path = btrfs_alloc_path();
4426 if (!path) {
4427 ret = -ENOMEM;
4428 goto out;
4429 }
4430
4431 key.objectid = 0;
4432 key.type = BTRFS_ROOT_ITEM_KEY;
4433 key.offset = 0;
4434
4435 while (1) {
4436 ret = btrfs_search_forward(root, &key, path,
4437 BTRFS_OLDEST_GENERATION);
4438 if (ret) {
4439 if (ret > 0)
4440 ret = 0;
4441 break;
4442 }
4443
4444 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4445 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4446 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4447 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4448 goto skip;
4449
4450 eb = path->nodes[0];
4451 slot = path->slots[0];
4452 item_size = btrfs_item_size_nr(eb, slot);
4453 if (item_size < sizeof(root_item))
4454 goto skip;
4455
4456 read_extent_buffer(eb, &root_item,
4457 btrfs_item_ptr_offset(eb, slot),
4458 (int)sizeof(root_item));
4459 if (btrfs_root_refs(&root_item) == 0)
4460 goto skip;
4461
4462 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4463 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4464 if (trans)
4465 goto update_tree;
4466
4467 btrfs_release_path(path);
4468 /*
4469 * 1 - subvol uuid item
4470 * 1 - received_subvol uuid item
4471 */
4472 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4473 if (IS_ERR(trans)) {
4474 ret = PTR_ERR(trans);
4475 break;
4476 }
4477 continue;
4478 } else {
4479 goto skip;
4480 }
4481 update_tree:
4482 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4483 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4484 BTRFS_UUID_KEY_SUBVOL,
4485 key.objectid);
4486 if (ret < 0) {
4487 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4488 ret);
4489 break;
4490 }
4491 }
4492
4493 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4494 ret = btrfs_uuid_tree_add(trans,
4495 root_item.received_uuid,
4496 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4497 key.objectid);
4498 if (ret < 0) {
4499 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4500 ret);
4501 break;
4502 }
4503 }
4504
4505 skip:
4506 if (trans) {
4507 ret = btrfs_end_transaction(trans);
4508 trans = NULL;
4509 if (ret)
4510 break;
4511 }
4512
4513 btrfs_release_path(path);
4514 if (key.offset < (u64)-1) {
4515 key.offset++;
4516 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4517 key.offset = 0;
4518 key.type = BTRFS_ROOT_ITEM_KEY;
4519 } else if (key.objectid < (u64)-1) {
4520 key.offset = 0;
4521 key.type = BTRFS_ROOT_ITEM_KEY;
4522 key.objectid++;
4523 } else {
4524 break;
4525 }
4526 cond_resched();
4527 }
4528
4529 out:
4530 btrfs_free_path(path);
4531 if (trans && !IS_ERR(trans))
4532 btrfs_end_transaction(trans);
4533 if (ret)
4534 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4535 else
4536 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4537 up(&fs_info->uuid_tree_rescan_sem);
4538 return 0;
4539 }
4540
4541 /*
4542 * Callback for btrfs_uuid_tree_iterate().
4543 * returns:
4544 * 0 check succeeded, the entry is not outdated.
4545 * < 0 if an error occurred.
4546 * > 0 if the check failed, which means the caller shall remove the entry.
4547 */
4548 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4549 u8 *uuid, u8 type, u64 subid)
4550 {
4551 struct btrfs_key key;
4552 int ret = 0;
4553 struct btrfs_root *subvol_root;
4554
4555 if (type != BTRFS_UUID_KEY_SUBVOL &&
4556 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4557 goto out;
4558
4559 key.objectid = subid;
4560 key.type = BTRFS_ROOT_ITEM_KEY;
4561 key.offset = (u64)-1;
4562 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4563 if (IS_ERR(subvol_root)) {
4564 ret = PTR_ERR(subvol_root);
4565 if (ret == -ENOENT)
4566 ret = 1;
4567 goto out;
4568 }
4569
4570 switch (type) {
4571 case BTRFS_UUID_KEY_SUBVOL:
4572 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4573 ret = 1;
4574 break;
4575 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4576 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4577 BTRFS_UUID_SIZE))
4578 ret = 1;
4579 break;
4580 }
4581
4582 out:
4583 return ret;
4584 }
4585
4586 static int btrfs_uuid_rescan_kthread(void *data)
4587 {
4588 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4589 int ret;
4590
4591 /*
4592 * 1st step is to iterate through the existing UUID tree and
4593 * to delete all entries that contain outdated data.
4594 * 2nd step is to add all missing entries to the UUID tree.
4595 */
4596 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4597 if (ret < 0) {
4598 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4599 up(&fs_info->uuid_tree_rescan_sem);
4600 return ret;
4601 }
4602 return btrfs_uuid_scan_kthread(data);
4603 }
4604
4605 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4606 {
4607 struct btrfs_trans_handle *trans;
4608 struct btrfs_root *tree_root = fs_info->tree_root;
4609 struct btrfs_root *uuid_root;
4610 struct task_struct *task;
4611 int ret;
4612
4613 /*
4614 * 1 - root node
4615 * 1 - root item
4616 */
4617 trans = btrfs_start_transaction(tree_root, 2);
4618 if (IS_ERR(trans))
4619 return PTR_ERR(trans);
4620
4621 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4622 if (IS_ERR(uuid_root)) {
4623 ret = PTR_ERR(uuid_root);
4624 btrfs_abort_transaction(trans, ret);
4625 btrfs_end_transaction(trans);
4626 return ret;
4627 }
4628
4629 fs_info->uuid_root = uuid_root;
4630
4631 ret = btrfs_commit_transaction(trans);
4632 if (ret)
4633 return ret;
4634
4635 down(&fs_info->uuid_tree_rescan_sem);
4636 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4637 if (IS_ERR(task)) {
4638 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4639 btrfs_warn(fs_info, "failed to start uuid_scan task");
4640 up(&fs_info->uuid_tree_rescan_sem);
4641 return PTR_ERR(task);
4642 }
4643
4644 return 0;
4645 }
4646
4647 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4648 {
4649 struct task_struct *task;
4650
4651 down(&fs_info->uuid_tree_rescan_sem);
4652 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4653 if (IS_ERR(task)) {
4654 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4655 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4656 up(&fs_info->uuid_tree_rescan_sem);
4657 return PTR_ERR(task);
4658 }
4659
4660 return 0;
4661 }
4662
4663 /*
4664 * shrinking a device means finding all of the device extents past
4665 * the new size, and then following the back refs to the chunks.
4666 * The chunk relocation code actually frees the device extent
4667 */
4668 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4669 {
4670 struct btrfs_fs_info *fs_info = device->fs_info;
4671 struct btrfs_root *root = fs_info->dev_root;
4672 struct btrfs_trans_handle *trans;
4673 struct btrfs_dev_extent *dev_extent = NULL;
4674 struct btrfs_path *path;
4675 u64 length;
4676 u64 chunk_offset;
4677 int ret;
4678 int slot;
4679 int failed = 0;
4680 bool retried = false;
4681 struct extent_buffer *l;
4682 struct btrfs_key key;
4683 struct btrfs_super_block *super_copy = fs_info->super_copy;
4684 u64 old_total = btrfs_super_total_bytes(super_copy);
4685 u64 old_size = btrfs_device_get_total_bytes(device);
4686 u64 diff;
4687 u64 start;
4688
4689 new_size = round_down(new_size, fs_info->sectorsize);
4690 start = new_size;
4691 diff = round_down(old_size - new_size, fs_info->sectorsize);
4692
4693 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4694 return -EINVAL;
4695
4696 path = btrfs_alloc_path();
4697 if (!path)
4698 return -ENOMEM;
4699
4700 path->reada = READA_BACK;
4701
4702 trans = btrfs_start_transaction(root, 0);
4703 if (IS_ERR(trans)) {
4704 btrfs_free_path(path);
4705 return PTR_ERR(trans);
4706 }
4707
4708 mutex_lock(&fs_info->chunk_mutex);
4709
4710 btrfs_device_set_total_bytes(device, new_size);
4711 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4712 device->fs_devices->total_rw_bytes -= diff;
4713 atomic64_sub(diff, &fs_info->free_chunk_space);
4714 }
4715
4716 /*
4717 * Once the device's size has been set to the new size, ensure all
4718 * in-memory chunks are synced to disk so that the loop below sees them
4719 * and relocates them accordingly.
4720 */
4721 if (contains_pending_extent(device, &start, diff)) {
4722 mutex_unlock(&fs_info->chunk_mutex);
4723 ret = btrfs_commit_transaction(trans);
4724 if (ret)
4725 goto done;
4726 } else {
4727 mutex_unlock(&fs_info->chunk_mutex);
4728 btrfs_end_transaction(trans);
4729 }
4730
4731 again:
4732 key.objectid = device->devid;
4733 key.offset = (u64)-1;
4734 key.type = BTRFS_DEV_EXTENT_KEY;
4735
4736 do {
4737 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4738 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4739 if (ret < 0) {
4740 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4741 goto done;
4742 }
4743
4744 ret = btrfs_previous_item(root, path, 0, key.type);
4745 if (ret)
4746 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4747 if (ret < 0)
4748 goto done;
4749 if (ret) {
4750 ret = 0;
4751 btrfs_release_path(path);
4752 break;
4753 }
4754
4755 l = path->nodes[0];
4756 slot = path->slots[0];
4757 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4758
4759 if (key.objectid != device->devid) {
4760 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4761 btrfs_release_path(path);
4762 break;
4763 }
4764
4765 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4766 length = btrfs_dev_extent_length(l, dev_extent);
4767
4768 if (key.offset + length <= new_size) {
4769 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4770 btrfs_release_path(path);
4771 break;
4772 }
4773
4774 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4775 btrfs_release_path(path);
4776
4777 /*
4778 * We may be relocating the only data chunk we have,
4779 * which could potentially end up with losing data's
4780 * raid profile, so lets allocate an empty one in
4781 * advance.
4782 */
4783 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4784 if (ret < 0) {
4785 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4786 goto done;
4787 }
4788
4789 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4790 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4791 if (ret == -ENOSPC) {
4792 failed++;
4793 } else if (ret) {
4794 if (ret == -ETXTBSY) {
4795 btrfs_warn(fs_info,
4796 "could not shrink block group %llu due to active swapfile",
4797 chunk_offset);
4798 }
4799 goto done;
4800 }
4801 } while (key.offset-- > 0);
4802
4803 if (failed && !retried) {
4804 failed = 0;
4805 retried = true;
4806 goto again;
4807 } else if (failed && retried) {
4808 ret = -ENOSPC;
4809 goto done;
4810 }
4811
4812 /* Shrinking succeeded, else we would be at "done". */
4813 trans = btrfs_start_transaction(root, 0);
4814 if (IS_ERR(trans)) {
4815 ret = PTR_ERR(trans);
4816 goto done;
4817 }
4818
4819 mutex_lock(&fs_info->chunk_mutex);
4820 btrfs_device_set_disk_total_bytes(device, new_size);
4821 if (list_empty(&device->post_commit_list))
4822 list_add_tail(&device->post_commit_list,
4823 &trans->transaction->dev_update_list);
4824
4825 WARN_ON(diff > old_total);
4826 btrfs_set_super_total_bytes(super_copy,
4827 round_down(old_total - diff, fs_info->sectorsize));
4828 mutex_unlock(&fs_info->chunk_mutex);
4829
4830 /* Now btrfs_update_device() will change the on-disk size. */
4831 ret = btrfs_update_device(trans, device);
4832 if (ret < 0) {
4833 btrfs_abort_transaction(trans, ret);
4834 btrfs_end_transaction(trans);
4835 } else {
4836 ret = btrfs_commit_transaction(trans);
4837 }
4838 done:
4839 btrfs_free_path(path);
4840 if (ret) {
4841 mutex_lock(&fs_info->chunk_mutex);
4842 btrfs_device_set_total_bytes(device, old_size);
4843 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4844 device->fs_devices->total_rw_bytes += diff;
4845 atomic64_add(diff, &fs_info->free_chunk_space);
4846 mutex_unlock(&fs_info->chunk_mutex);
4847 }
4848 return ret;
4849 }
4850
4851 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4852 struct btrfs_key *key,
4853 struct btrfs_chunk *chunk, int item_size)
4854 {
4855 struct btrfs_super_block *super_copy = fs_info->super_copy;
4856 struct btrfs_disk_key disk_key;
4857 u32 array_size;
4858 u8 *ptr;
4859
4860 mutex_lock(&fs_info->chunk_mutex);
4861 array_size = btrfs_super_sys_array_size(super_copy);
4862 if (array_size + item_size + sizeof(disk_key)
4863 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4864 mutex_unlock(&fs_info->chunk_mutex);
4865 return -EFBIG;
4866 }
4867
4868 ptr = super_copy->sys_chunk_array + array_size;
4869 btrfs_cpu_key_to_disk(&disk_key, key);
4870 memcpy(ptr, &disk_key, sizeof(disk_key));
4871 ptr += sizeof(disk_key);
4872 memcpy(ptr, chunk, item_size);
4873 item_size += sizeof(disk_key);
4874 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4875 mutex_unlock(&fs_info->chunk_mutex);
4876
4877 return 0;
4878 }
4879
4880 /*
4881 * sort the devices in descending order by max_avail, total_avail
4882 */
4883 static int btrfs_cmp_device_info(const void *a, const void *b)
4884 {
4885 const struct btrfs_device_info *di_a = a;
4886 const struct btrfs_device_info *di_b = b;
4887
4888 if (di_a->max_avail > di_b->max_avail)
4889 return -1;
4890 if (di_a->max_avail < di_b->max_avail)
4891 return 1;
4892 if (di_a->total_avail > di_b->total_avail)
4893 return -1;
4894 if (di_a->total_avail < di_b->total_avail)
4895 return 1;
4896 return 0;
4897 }
4898
4899 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4900 {
4901 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4902 return;
4903
4904 btrfs_set_fs_incompat(info, RAID56);
4905 }
4906
4907 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4908 u64 start, u64 type)
4909 {
4910 struct btrfs_fs_info *info = trans->fs_info;
4911 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4912 struct btrfs_device *device;
4913 struct map_lookup *map = NULL;
4914 struct extent_map_tree *em_tree;
4915 struct extent_map *em;
4916 struct btrfs_device_info *devices_info = NULL;
4917 u64 total_avail;
4918 int num_stripes; /* total number of stripes to allocate */
4919 int data_stripes; /* number of stripes that count for
4920 block group size */
4921 int sub_stripes; /* sub_stripes info for map */
4922 int dev_stripes; /* stripes per dev */
4923 int devs_max; /* max devs to use */
4924 int devs_min; /* min devs needed */
4925 int devs_increment; /* ndevs has to be a multiple of this */
4926 int ncopies; /* how many copies to data has */
4927 int nparity; /* number of stripes worth of bytes to
4928 store parity information */
4929 int ret;
4930 u64 max_stripe_size;
4931 u64 max_chunk_size;
4932 u64 stripe_size;
4933 u64 chunk_size;
4934 int ndevs;
4935 int i;
4936 int j;
4937 int index;
4938
4939 BUG_ON(!alloc_profile_is_valid(type, 0));
4940
4941 if (list_empty(&fs_devices->alloc_list)) {
4942 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4943 btrfs_debug(info, "%s: no writable device", __func__);
4944 return -ENOSPC;
4945 }
4946
4947 index = btrfs_bg_flags_to_raid_index(type);
4948
4949 sub_stripes = btrfs_raid_array[index].sub_stripes;
4950 dev_stripes = btrfs_raid_array[index].dev_stripes;
4951 devs_max = btrfs_raid_array[index].devs_max;
4952 devs_min = btrfs_raid_array[index].devs_min;
4953 devs_increment = btrfs_raid_array[index].devs_increment;
4954 ncopies = btrfs_raid_array[index].ncopies;
4955 nparity = btrfs_raid_array[index].nparity;
4956
4957 if (type & BTRFS_BLOCK_GROUP_DATA) {
4958 max_stripe_size = SZ_1G;
4959 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4960 if (!devs_max)
4961 devs_max = BTRFS_MAX_DEVS(info);
4962 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4963 /* for larger filesystems, use larger metadata chunks */
4964 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4965 max_stripe_size = SZ_1G;
4966 else
4967 max_stripe_size = SZ_256M;
4968 max_chunk_size = max_stripe_size;
4969 if (!devs_max)
4970 devs_max = BTRFS_MAX_DEVS(info);
4971 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4972 max_stripe_size = SZ_32M;
4973 max_chunk_size = 2 * max_stripe_size;
4974 if (!devs_max)
4975 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4976 } else {
4977 btrfs_err(info, "invalid chunk type 0x%llx requested",
4978 type);
4979 BUG();
4980 }
4981
4982 /* We don't want a chunk larger than 10% of writable space */
4983 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4984 max_chunk_size);
4985
4986 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4987 GFP_NOFS);
4988 if (!devices_info)
4989 return -ENOMEM;
4990
4991 /*
4992 * in the first pass through the devices list, we gather information
4993 * about the available holes on each device.
4994 */
4995 ndevs = 0;
4996 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4997 u64 max_avail;
4998 u64 dev_offset;
4999
5000 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5001 WARN(1, KERN_ERR
5002 "BTRFS: read-only device in alloc_list\n");
5003 continue;
5004 }
5005
5006 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5007 &device->dev_state) ||
5008 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5009 continue;
5010
5011 if (device->total_bytes > device->bytes_used)
5012 total_avail = device->total_bytes - device->bytes_used;
5013 else
5014 total_avail = 0;
5015
5016 /* If there is no space on this device, skip it. */
5017 if (total_avail == 0)
5018 continue;
5019
5020 ret = find_free_dev_extent(device,
5021 max_stripe_size * dev_stripes,
5022 &dev_offset, &max_avail);
5023 if (ret && ret != -ENOSPC)
5024 goto error;
5025
5026 if (ret == 0)
5027 max_avail = max_stripe_size * dev_stripes;
5028
5029 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
5030 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5031 btrfs_debug(info,
5032 "%s: devid %llu has no free space, have=%llu want=%u",
5033 __func__, device->devid, max_avail,
5034 BTRFS_STRIPE_LEN * dev_stripes);
5035 continue;
5036 }
5037
5038 if (ndevs == fs_devices->rw_devices) {
5039 WARN(1, "%s: found more than %llu devices\n",
5040 __func__, fs_devices->rw_devices);
5041 break;
5042 }
5043 devices_info[ndevs].dev_offset = dev_offset;
5044 devices_info[ndevs].max_avail = max_avail;
5045 devices_info[ndevs].total_avail = total_avail;
5046 devices_info[ndevs].dev = device;
5047 ++ndevs;
5048 }
5049
5050 /*
5051 * now sort the devices by hole size / available space
5052 */
5053 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5054 btrfs_cmp_device_info, NULL);
5055
5056 /* round down to number of usable stripes */
5057 ndevs = round_down(ndevs, devs_increment);
5058
5059 if (ndevs < devs_min) {
5060 ret = -ENOSPC;
5061 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5062 btrfs_debug(info,
5063 "%s: not enough devices with free space: have=%d minimum required=%d",
5064 __func__, ndevs, devs_min);
5065 }
5066 goto error;
5067 }
5068
5069 ndevs = min(ndevs, devs_max);
5070
5071 /*
5072 * The primary goal is to maximize the number of stripes, so use as
5073 * many devices as possible, even if the stripes are not maximum sized.
5074 *
5075 * The DUP profile stores more than one stripe per device, the
5076 * max_avail is the total size so we have to adjust.
5077 */
5078 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
5079 num_stripes = ndevs * dev_stripes;
5080
5081 /*
5082 * this will have to be fixed for RAID1 and RAID10 over
5083 * more drives
5084 */
5085 data_stripes = (num_stripes - nparity) / ncopies;
5086
5087 /*
5088 * Use the number of data stripes to figure out how big this chunk
5089 * is really going to be in terms of logical address space,
5090 * and compare that answer with the max chunk size. If it's higher,
5091 * we try to reduce stripe_size.
5092 */
5093 if (stripe_size * data_stripes > max_chunk_size) {
5094 /*
5095 * Reduce stripe_size, round it up to a 16MB boundary again and
5096 * then use it, unless it ends up being even bigger than the
5097 * previous value we had already.
5098 */
5099 stripe_size = min(round_up(div_u64(max_chunk_size,
5100 data_stripes), SZ_16M),
5101 stripe_size);
5102 }
5103
5104 /* align to BTRFS_STRIPE_LEN */
5105 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
5106
5107 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5108 if (!map) {
5109 ret = -ENOMEM;
5110 goto error;
5111 }
5112 map->num_stripes = num_stripes;
5113
5114 for (i = 0; i < ndevs; ++i) {
5115 for (j = 0; j < dev_stripes; ++j) {
5116 int s = i * dev_stripes + j;
5117 map->stripes[s].dev = devices_info[i].dev;
5118 map->stripes[s].physical = devices_info[i].dev_offset +
5119 j * stripe_size;
5120 }
5121 }
5122 map->stripe_len = BTRFS_STRIPE_LEN;
5123 map->io_align = BTRFS_STRIPE_LEN;
5124 map->io_width = BTRFS_STRIPE_LEN;
5125 map->type = type;
5126 map->sub_stripes = sub_stripes;
5127
5128 chunk_size = stripe_size * data_stripes;
5129
5130 trace_btrfs_chunk_alloc(info, map, start, chunk_size);
5131
5132 em = alloc_extent_map();
5133 if (!em) {
5134 kfree(map);
5135 ret = -ENOMEM;
5136 goto error;
5137 }
5138 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5139 em->map_lookup = map;
5140 em->start = start;
5141 em->len = chunk_size;
5142 em->block_start = 0;
5143 em->block_len = em->len;
5144 em->orig_block_len = stripe_size;
5145
5146 em_tree = &info->mapping_tree.map_tree;
5147 write_lock(&em_tree->lock);
5148 ret = add_extent_mapping(em_tree, em, 0);
5149 if (ret) {
5150 write_unlock(&em_tree->lock);
5151 free_extent_map(em);
5152 goto error;
5153 }
5154 write_unlock(&em_tree->lock);
5155
5156 ret = btrfs_make_block_group(trans, 0, type, start, chunk_size);
5157 if (ret)
5158 goto error_del_extent;
5159
5160 for (i = 0; i < map->num_stripes; i++) {
5161 struct btrfs_device *dev = map->stripes[i].dev;
5162
5163 btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size);
5164 if (list_empty(&dev->post_commit_list))
5165 list_add_tail(&dev->post_commit_list,
5166 &trans->transaction->dev_update_list);
5167 }
5168
5169 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
5170
5171 free_extent_map(em);
5172 check_raid56_incompat_flag(info, type);
5173
5174 kfree(devices_info);
5175 return 0;
5176
5177 error_del_extent:
5178 write_lock(&em_tree->lock);
5179 remove_extent_mapping(em_tree, em);
5180 write_unlock(&em_tree->lock);
5181
5182 /* One for our allocation */
5183 free_extent_map(em);
5184 /* One for the tree reference */
5185 free_extent_map(em);
5186 error:
5187 kfree(devices_info);
5188 return ret;
5189 }
5190
5191 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5192 u64 chunk_offset, u64 chunk_size)
5193 {
5194 struct btrfs_fs_info *fs_info = trans->fs_info;
5195 struct btrfs_root *extent_root = fs_info->extent_root;
5196 struct btrfs_root *chunk_root = fs_info->chunk_root;
5197 struct btrfs_key key;
5198 struct btrfs_device *device;
5199 struct btrfs_chunk *chunk;
5200 struct btrfs_stripe *stripe;
5201 struct extent_map *em;
5202 struct map_lookup *map;
5203 size_t item_size;
5204 u64 dev_offset;
5205 u64 stripe_size;
5206 int i = 0;
5207 int ret = 0;
5208
5209 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5210 if (IS_ERR(em))
5211 return PTR_ERR(em);
5212
5213 map = em->map_lookup;
5214 item_size = btrfs_chunk_item_size(map->num_stripes);
5215 stripe_size = em->orig_block_len;
5216
5217 chunk = kzalloc(item_size, GFP_NOFS);
5218 if (!chunk) {
5219 ret = -ENOMEM;
5220 goto out;
5221 }
5222
5223 /*
5224 * Take the device list mutex to prevent races with the final phase of
5225 * a device replace operation that replaces the device object associated
5226 * with the map's stripes, because the device object's id can change
5227 * at any time during that final phase of the device replace operation
5228 * (dev-replace.c:btrfs_dev_replace_finishing()).
5229 */
5230 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5231 for (i = 0; i < map->num_stripes; i++) {
5232 device = map->stripes[i].dev;
5233 dev_offset = map->stripes[i].physical;
5234
5235 ret = btrfs_update_device(trans, device);
5236 if (ret)
5237 break;
5238 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5239 dev_offset, stripe_size);
5240 if (ret)
5241 break;
5242 }
5243 if (ret) {
5244 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5245 goto out;
5246 }
5247
5248 stripe = &chunk->stripe;
5249 for (i = 0; i < map->num_stripes; i++) {
5250 device = map->stripes[i].dev;
5251 dev_offset = map->stripes[i].physical;
5252
5253 btrfs_set_stack_stripe_devid(stripe, device->devid);
5254 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5255 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5256 stripe++;
5257 }
5258 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5259
5260 btrfs_set_stack_chunk_length(chunk, chunk_size);
5261 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5262 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5263 btrfs_set_stack_chunk_type(chunk, map->type);
5264 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5265 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5266 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5267 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5268 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5269
5270 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5271 key.type = BTRFS_CHUNK_ITEM_KEY;
5272 key.offset = chunk_offset;
5273
5274 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5275 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5276 /*
5277 * TODO: Cleanup of inserted chunk root in case of
5278 * failure.
5279 */
5280 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5281 }
5282
5283 out:
5284 kfree(chunk);
5285 free_extent_map(em);
5286 return ret;
5287 }
5288
5289 /*
5290 * Chunk allocation falls into two parts. The first part does work
5291 * that makes the new allocated chunk usable, but does not do any operation
5292 * that modifies the chunk tree. The second part does the work that
5293 * requires modifying the chunk tree. This division is important for the
5294 * bootstrap process of adding storage to a seed btrfs.
5295 */
5296 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5297 {
5298 u64 chunk_offset;
5299
5300 lockdep_assert_held(&trans->fs_info->chunk_mutex);
5301 chunk_offset = find_next_chunk(trans->fs_info);
5302 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5303 }
5304
5305 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5306 {
5307 struct btrfs_fs_info *fs_info = trans->fs_info;
5308 u64 chunk_offset;
5309 u64 sys_chunk_offset;
5310 u64 alloc_profile;
5311 int ret;
5312
5313 chunk_offset = find_next_chunk(fs_info);
5314 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5315 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5316 if (ret)
5317 return ret;
5318
5319 sys_chunk_offset = find_next_chunk(fs_info);
5320 alloc_profile = btrfs_system_alloc_profile(fs_info);
5321 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5322 return ret;
5323 }
5324
5325 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5326 {
5327 int max_errors;
5328
5329 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5330 BTRFS_BLOCK_GROUP_RAID10 |
5331 BTRFS_BLOCK_GROUP_RAID5)) {
5332 max_errors = 1;
5333 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5334 max_errors = 2;
5335 } else {
5336 max_errors = 0;
5337 }
5338
5339 return max_errors;
5340 }
5341
5342 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5343 {
5344 struct extent_map *em;
5345 struct map_lookup *map;
5346 int readonly = 0;
5347 int miss_ndevs = 0;
5348 int i;
5349
5350 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5351 if (IS_ERR(em))
5352 return 1;
5353
5354 map = em->map_lookup;
5355 for (i = 0; i < map->num_stripes; i++) {
5356 if (test_bit(BTRFS_DEV_STATE_MISSING,
5357 &map->stripes[i].dev->dev_state)) {
5358 miss_ndevs++;
5359 continue;
5360 }
5361 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5362 &map->stripes[i].dev->dev_state)) {
5363 readonly = 1;
5364 goto end;
5365 }
5366 }
5367
5368 /*
5369 * If the number of missing devices is larger than max errors,
5370 * we can not write the data into that chunk successfully, so
5371 * set it readonly.
5372 */
5373 if (miss_ndevs > btrfs_chunk_max_errors(map))
5374 readonly = 1;
5375 end:
5376 free_extent_map(em);
5377 return readonly;
5378 }
5379
5380 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5381 {
5382 extent_map_tree_init(&tree->map_tree);
5383 }
5384
5385 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5386 {
5387 struct extent_map *em;
5388
5389 while (1) {
5390 write_lock(&tree->map_tree.lock);
5391 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5392 if (em)
5393 remove_extent_mapping(&tree->map_tree, em);
5394 write_unlock(&tree->map_tree.lock);
5395 if (!em)
5396 break;
5397 /* once for us */
5398 free_extent_map(em);
5399 /* once for the tree */
5400 free_extent_map(em);
5401 }
5402 }
5403
5404 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5405 {
5406 struct extent_map *em;
5407 struct map_lookup *map;
5408 int ret;
5409
5410 em = btrfs_get_chunk_map(fs_info, logical, len);
5411 if (IS_ERR(em))
5412 /*
5413 * We could return errors for these cases, but that could get
5414 * ugly and we'd probably do the same thing which is just not do
5415 * anything else and exit, so return 1 so the callers don't try
5416 * to use other copies.
5417 */
5418 return 1;
5419
5420 map = em->map_lookup;
5421 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5422 ret = map->num_stripes;
5423 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5424 ret = map->sub_stripes;
5425 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5426 ret = 2;
5427 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5428 /*
5429 * There could be two corrupted data stripes, we need
5430 * to loop retry in order to rebuild the correct data.
5431 *
5432 * Fail a stripe at a time on every retry except the
5433 * stripe under reconstruction.
5434 */
5435 ret = map->num_stripes;
5436 else
5437 ret = 1;
5438 free_extent_map(em);
5439
5440 down_read(&fs_info->dev_replace.rwsem);
5441 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5442 fs_info->dev_replace.tgtdev)
5443 ret++;
5444 up_read(&fs_info->dev_replace.rwsem);
5445
5446 return ret;
5447 }
5448
5449 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5450 u64 logical)
5451 {
5452 struct extent_map *em;
5453 struct map_lookup *map;
5454 unsigned long len = fs_info->sectorsize;
5455
5456 em = btrfs_get_chunk_map(fs_info, logical, len);
5457
5458 if (!WARN_ON(IS_ERR(em))) {
5459 map = em->map_lookup;
5460 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5461 len = map->stripe_len * nr_data_stripes(map);
5462 free_extent_map(em);
5463 }
5464 return len;
5465 }
5466
5467 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5468 {
5469 struct extent_map *em;
5470 struct map_lookup *map;
5471 int ret = 0;
5472
5473 em = btrfs_get_chunk_map(fs_info, logical, len);
5474
5475 if(!WARN_ON(IS_ERR(em))) {
5476 map = em->map_lookup;
5477 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5478 ret = 1;
5479 free_extent_map(em);
5480 }
5481 return ret;
5482 }
5483
5484 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5485 struct map_lookup *map, int first,
5486 int dev_replace_is_ongoing)
5487 {
5488 int i;
5489 int num_stripes;
5490 int preferred_mirror;
5491 int tolerance;
5492 struct btrfs_device *srcdev;
5493
5494 ASSERT((map->type &
5495 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5496
5497 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5498 num_stripes = map->sub_stripes;
5499 else
5500 num_stripes = map->num_stripes;
5501
5502 preferred_mirror = first + current->pid % num_stripes;
5503
5504 if (dev_replace_is_ongoing &&
5505 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5506 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5507 srcdev = fs_info->dev_replace.srcdev;
5508 else
5509 srcdev = NULL;
5510
5511 /*
5512 * try to avoid the drive that is the source drive for a
5513 * dev-replace procedure, only choose it if no other non-missing
5514 * mirror is available
5515 */
5516 for (tolerance = 0; tolerance < 2; tolerance++) {
5517 if (map->stripes[preferred_mirror].dev->bdev &&
5518 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5519 return preferred_mirror;
5520 for (i = first; i < first + num_stripes; i++) {
5521 if (map->stripes[i].dev->bdev &&
5522 (tolerance || map->stripes[i].dev != srcdev))
5523 return i;
5524 }
5525 }
5526
5527 /* we couldn't find one that doesn't fail. Just return something
5528 * and the io error handling code will clean up eventually
5529 */
5530 return preferred_mirror;
5531 }
5532
5533 static inline int parity_smaller(u64 a, u64 b)
5534 {
5535 return a > b;
5536 }
5537
5538 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5539 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5540 {
5541 struct btrfs_bio_stripe s;
5542 int i;
5543 u64 l;
5544 int again = 1;
5545
5546 while (again) {
5547 again = 0;
5548 for (i = 0; i < num_stripes - 1; i++) {
5549 if (parity_smaller(bbio->raid_map[i],
5550 bbio->raid_map[i+1])) {
5551 s = bbio->stripes[i];
5552 l = bbio->raid_map[i];
5553 bbio->stripes[i] = bbio->stripes[i+1];
5554 bbio->raid_map[i] = bbio->raid_map[i+1];
5555 bbio->stripes[i+1] = s;
5556 bbio->raid_map[i+1] = l;
5557
5558 again = 1;
5559 }
5560 }
5561 }
5562 }
5563
5564 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5565 {
5566 struct btrfs_bio *bbio = kzalloc(
5567 /* the size of the btrfs_bio */
5568 sizeof(struct btrfs_bio) +
5569 /* plus the variable array for the stripes */
5570 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5571 /* plus the variable array for the tgt dev */
5572 sizeof(int) * (real_stripes) +
5573 /*
5574 * plus the raid_map, which includes both the tgt dev
5575 * and the stripes
5576 */
5577 sizeof(u64) * (total_stripes),
5578 GFP_NOFS|__GFP_NOFAIL);
5579
5580 atomic_set(&bbio->error, 0);
5581 refcount_set(&bbio->refs, 1);
5582
5583 return bbio;
5584 }
5585
5586 void btrfs_get_bbio(struct btrfs_bio *bbio)
5587 {
5588 WARN_ON(!refcount_read(&bbio->refs));
5589 refcount_inc(&bbio->refs);
5590 }
5591
5592 void btrfs_put_bbio(struct btrfs_bio *bbio)
5593 {
5594 if (!bbio)
5595 return;
5596 if (refcount_dec_and_test(&bbio->refs))
5597 kfree(bbio);
5598 }
5599
5600 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5601 /*
5602 * Please note that, discard won't be sent to target device of device
5603 * replace.
5604 */
5605 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5606 u64 logical, u64 length,
5607 struct btrfs_bio **bbio_ret)
5608 {
5609 struct extent_map *em;
5610 struct map_lookup *map;
5611 struct btrfs_bio *bbio;
5612 u64 offset;
5613 u64 stripe_nr;
5614 u64 stripe_nr_end;
5615 u64 stripe_end_offset;
5616 u64 stripe_cnt;
5617 u64 stripe_len;
5618 u64 stripe_offset;
5619 u64 num_stripes;
5620 u32 stripe_index;
5621 u32 factor = 0;
5622 u32 sub_stripes = 0;
5623 u64 stripes_per_dev = 0;
5624 u32 remaining_stripes = 0;
5625 u32 last_stripe = 0;
5626 int ret = 0;
5627 int i;
5628
5629 /* discard always return a bbio */
5630 ASSERT(bbio_ret);
5631
5632 em = btrfs_get_chunk_map(fs_info, logical, length);
5633 if (IS_ERR(em))
5634 return PTR_ERR(em);
5635
5636 map = em->map_lookup;
5637 /* we don't discard raid56 yet */
5638 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5639 ret = -EOPNOTSUPP;
5640 goto out;
5641 }
5642
5643 offset = logical - em->start;
5644 length = min_t(u64, em->len - offset, length);
5645
5646 stripe_len = map->stripe_len;
5647 /*
5648 * stripe_nr counts the total number of stripes we have to stride
5649 * to get to this block
5650 */
5651 stripe_nr = div64_u64(offset, stripe_len);
5652
5653 /* stripe_offset is the offset of this block in its stripe */
5654 stripe_offset = offset - stripe_nr * stripe_len;
5655
5656 stripe_nr_end = round_up(offset + length, map->stripe_len);
5657 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5658 stripe_cnt = stripe_nr_end - stripe_nr;
5659 stripe_end_offset = stripe_nr_end * map->stripe_len -
5660 (offset + length);
5661 /*
5662 * after this, stripe_nr is the number of stripes on this
5663 * device we have to walk to find the data, and stripe_index is
5664 * the number of our device in the stripe array
5665 */
5666 num_stripes = 1;
5667 stripe_index = 0;
5668 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5669 BTRFS_BLOCK_GROUP_RAID10)) {
5670 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5671 sub_stripes = 1;
5672 else
5673 sub_stripes = map->sub_stripes;
5674
5675 factor = map->num_stripes / sub_stripes;
5676 num_stripes = min_t(u64, map->num_stripes,
5677 sub_stripes * stripe_cnt);
5678 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5679 stripe_index *= sub_stripes;
5680 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5681 &remaining_stripes);
5682 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5683 last_stripe *= sub_stripes;
5684 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5685 BTRFS_BLOCK_GROUP_DUP)) {
5686 num_stripes = map->num_stripes;
5687 } else {
5688 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5689 &stripe_index);
5690 }
5691
5692 bbio = alloc_btrfs_bio(num_stripes, 0);
5693 if (!bbio) {
5694 ret = -ENOMEM;
5695 goto out;
5696 }
5697
5698 for (i = 0; i < num_stripes; i++) {
5699 bbio->stripes[i].physical =
5700 map->stripes[stripe_index].physical +
5701 stripe_offset + stripe_nr * map->stripe_len;
5702 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5703
5704 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5705 BTRFS_BLOCK_GROUP_RAID10)) {
5706 bbio->stripes[i].length = stripes_per_dev *
5707 map->stripe_len;
5708
5709 if (i / sub_stripes < remaining_stripes)
5710 bbio->stripes[i].length +=
5711 map->stripe_len;
5712
5713 /*
5714 * Special for the first stripe and
5715 * the last stripe:
5716 *
5717 * |-------|...|-------|
5718 * |----------|
5719 * off end_off
5720 */
5721 if (i < sub_stripes)
5722 bbio->stripes[i].length -=
5723 stripe_offset;
5724
5725 if (stripe_index >= last_stripe &&
5726 stripe_index <= (last_stripe +
5727 sub_stripes - 1))
5728 bbio->stripes[i].length -=
5729 stripe_end_offset;
5730
5731 if (i == sub_stripes - 1)
5732 stripe_offset = 0;
5733 } else {
5734 bbio->stripes[i].length = length;
5735 }
5736
5737 stripe_index++;
5738 if (stripe_index == map->num_stripes) {
5739 stripe_index = 0;
5740 stripe_nr++;
5741 }
5742 }
5743
5744 *bbio_ret = bbio;
5745 bbio->map_type = map->type;
5746 bbio->num_stripes = num_stripes;
5747 out:
5748 free_extent_map(em);
5749 return ret;
5750 }
5751
5752 /*
5753 * In dev-replace case, for repair case (that's the only case where the mirror
5754 * is selected explicitly when calling btrfs_map_block), blocks left of the
5755 * left cursor can also be read from the target drive.
5756 *
5757 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5758 * array of stripes.
5759 * For READ, it also needs to be supported using the same mirror number.
5760 *
5761 * If the requested block is not left of the left cursor, EIO is returned. This
5762 * can happen because btrfs_num_copies() returns one more in the dev-replace
5763 * case.
5764 */
5765 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5766 u64 logical, u64 length,
5767 u64 srcdev_devid, int *mirror_num,
5768 u64 *physical)
5769 {
5770 struct btrfs_bio *bbio = NULL;
5771 int num_stripes;
5772 int index_srcdev = 0;
5773 int found = 0;
5774 u64 physical_of_found = 0;
5775 int i;
5776 int ret = 0;
5777
5778 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5779 logical, &length, &bbio, 0, 0);
5780 if (ret) {
5781 ASSERT(bbio == NULL);
5782 return ret;
5783 }
5784
5785 num_stripes = bbio->num_stripes;
5786 if (*mirror_num > num_stripes) {
5787 /*
5788 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5789 * that means that the requested area is not left