]>
Commit | Line | Data |
---|---|---|
0b86a832 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8a4b83cc | 21 | #include <linux/buffer_head.h> |
f2d8d74d | 22 | #include <linux/blkdev.h> |
b765ead5 | 23 | #include <linux/iocontext.h> |
6f88a440 | 24 | #include <linux/capability.h> |
442a4f63 | 25 | #include <linux/ratelimit.h> |
59641015 | 26 | #include <linux/kthread.h> |
53b381b3 | 27 | #include <linux/raid/pq.h> |
803b2f54 | 28 | #include <linux/semaphore.h> |
8da4b8c4 | 29 | #include <linux/uuid.h> |
53b381b3 | 30 | #include <asm/div64.h> |
0b86a832 CM |
31 | #include "ctree.h" |
32 | #include "extent_map.h" | |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "print-tree.h" | |
36 | #include "volumes.h" | |
53b381b3 | 37 | #include "raid56.h" |
8b712842 | 38 | #include "async-thread.h" |
21adbd5c | 39 | #include "check-integrity.h" |
606686ee | 40 | #include "rcu-string.h" |
3fed40cc | 41 | #include "math.h" |
8dabb742 | 42 | #include "dev-replace.h" |
99994cde | 43 | #include "sysfs.h" |
0b86a832 | 44 | |
af902047 ZL |
45 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
46 | [BTRFS_RAID_RAID10] = { | |
47 | .sub_stripes = 2, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 0, /* 0 == as many as possible */ | |
50 | .devs_min = 4, | |
8789f4fe | 51 | .tolerated_failures = 1, |
af902047 ZL |
52 | .devs_increment = 2, |
53 | .ncopies = 2, | |
54 | }, | |
55 | [BTRFS_RAID_RAID1] = { | |
56 | .sub_stripes = 1, | |
57 | .dev_stripes = 1, | |
58 | .devs_max = 2, | |
59 | .devs_min = 2, | |
8789f4fe | 60 | .tolerated_failures = 1, |
af902047 ZL |
61 | .devs_increment = 2, |
62 | .ncopies = 2, | |
63 | }, | |
64 | [BTRFS_RAID_DUP] = { | |
65 | .sub_stripes = 1, | |
66 | .dev_stripes = 2, | |
67 | .devs_max = 1, | |
68 | .devs_min = 1, | |
8789f4fe | 69 | .tolerated_failures = 0, |
af902047 ZL |
70 | .devs_increment = 1, |
71 | .ncopies = 2, | |
72 | }, | |
73 | [BTRFS_RAID_RAID0] = { | |
74 | .sub_stripes = 1, | |
75 | .dev_stripes = 1, | |
76 | .devs_max = 0, | |
77 | .devs_min = 2, | |
8789f4fe | 78 | .tolerated_failures = 0, |
af902047 ZL |
79 | .devs_increment = 1, |
80 | .ncopies = 1, | |
81 | }, | |
82 | [BTRFS_RAID_SINGLE] = { | |
83 | .sub_stripes = 1, | |
84 | .dev_stripes = 1, | |
85 | .devs_max = 1, | |
86 | .devs_min = 1, | |
8789f4fe | 87 | .tolerated_failures = 0, |
af902047 ZL |
88 | .devs_increment = 1, |
89 | .ncopies = 1, | |
90 | }, | |
91 | [BTRFS_RAID_RAID5] = { | |
92 | .sub_stripes = 1, | |
93 | .dev_stripes = 1, | |
94 | .devs_max = 0, | |
95 | .devs_min = 2, | |
8789f4fe | 96 | .tolerated_failures = 1, |
af902047 ZL |
97 | .devs_increment = 1, |
98 | .ncopies = 2, | |
99 | }, | |
100 | [BTRFS_RAID_RAID6] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 3, | |
8789f4fe | 105 | .tolerated_failures = 2, |
af902047 ZL |
106 | .devs_increment = 1, |
107 | .ncopies = 3, | |
108 | }, | |
109 | }; | |
110 | ||
fb75d857 | 111 | const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { |
af902047 ZL |
112 | [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10, |
113 | [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1, | |
114 | [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP, | |
115 | [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0, | |
116 | [BTRFS_RAID_SINGLE] = 0, | |
117 | [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5, | |
118 | [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6, | |
119 | }; | |
120 | ||
621292ba DS |
121 | /* |
122 | * Table to convert BTRFS_RAID_* to the error code if minimum number of devices | |
123 | * condition is not met. Zero means there's no corresponding | |
124 | * BTRFS_ERROR_DEV_*_NOT_MET value. | |
125 | */ | |
126 | const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = { | |
127 | [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, | |
128 | [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, | |
129 | [BTRFS_RAID_DUP] = 0, | |
130 | [BTRFS_RAID_RAID0] = 0, | |
131 | [BTRFS_RAID_SINGLE] = 0, | |
132 | [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, | |
133 | [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, | |
134 | }; | |
135 | ||
2b82032c | 136 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 137 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 138 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 139 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 140 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 141 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
142 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
143 | enum btrfs_map_op op, | |
144 | u64 logical, u64 *length, | |
145 | struct btrfs_bio **bbio_ret, | |
146 | int mirror_num, int need_raid_map); | |
2b82032c | 147 | |
67a2c45e | 148 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 149 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
150 | struct list_head *btrfs_get_fs_uuids(void) |
151 | { | |
152 | return &fs_uuids; | |
153 | } | |
8a4b83cc | 154 | |
2208a378 ID |
155 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
156 | { | |
157 | struct btrfs_fs_devices *fs_devs; | |
158 | ||
78f2c9e6 | 159 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
160 | if (!fs_devs) |
161 | return ERR_PTR(-ENOMEM); | |
162 | ||
163 | mutex_init(&fs_devs->device_list_mutex); | |
164 | ||
165 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 166 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
167 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
168 | INIT_LIST_HEAD(&fs_devs->list); | |
169 | ||
170 | return fs_devs; | |
171 | } | |
172 | ||
173 | /** | |
174 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
175 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
176 | * generated. | |
177 | * | |
178 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
179 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
180 | * can be destroyed with kfree() right away. | |
181 | */ | |
182 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
183 | { | |
184 | struct btrfs_fs_devices *fs_devs; | |
185 | ||
186 | fs_devs = __alloc_fs_devices(); | |
187 | if (IS_ERR(fs_devs)) | |
188 | return fs_devs; | |
189 | ||
190 | if (fsid) | |
191 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
192 | else | |
193 | generate_random_uuid(fs_devs->fsid); | |
194 | ||
195 | return fs_devs; | |
196 | } | |
197 | ||
e4404d6e YZ |
198 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
199 | { | |
200 | struct btrfs_device *device; | |
201 | WARN_ON(fs_devices->opened); | |
202 | while (!list_empty(&fs_devices->devices)) { | |
203 | device = list_entry(fs_devices->devices.next, | |
204 | struct btrfs_device, dev_list); | |
205 | list_del(&device->dev_list); | |
606686ee | 206 | rcu_string_free(device->name); |
e4404d6e YZ |
207 | kfree(device); |
208 | } | |
209 | kfree(fs_devices); | |
210 | } | |
211 | ||
b8b8ff59 LC |
212 | static void btrfs_kobject_uevent(struct block_device *bdev, |
213 | enum kobject_action action) | |
214 | { | |
215 | int ret; | |
216 | ||
217 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
218 | if (ret) | |
efe120a0 | 219 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
220 | action, |
221 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
222 | &disk_to_dev(bdev->bd_disk)->kobj); | |
223 | } | |
224 | ||
143bede5 | 225 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
226 | { |
227 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 228 | |
2b82032c YZ |
229 | while (!list_empty(&fs_uuids)) { |
230 | fs_devices = list_entry(fs_uuids.next, | |
231 | struct btrfs_fs_devices, list); | |
232 | list_del(&fs_devices->list); | |
e4404d6e | 233 | free_fs_devices(fs_devices); |
8a4b83cc | 234 | } |
8a4b83cc CM |
235 | } |
236 | ||
12bd2fc0 ID |
237 | static struct btrfs_device *__alloc_device(void) |
238 | { | |
239 | struct btrfs_device *dev; | |
240 | ||
78f2c9e6 | 241 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
242 | if (!dev) |
243 | return ERR_PTR(-ENOMEM); | |
244 | ||
245 | INIT_LIST_HEAD(&dev->dev_list); | |
246 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 247 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
248 | |
249 | spin_lock_init(&dev->io_lock); | |
250 | ||
251 | spin_lock_init(&dev->reada_lock); | |
252 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 253 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 254 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 255 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 256 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
257 | |
258 | return dev; | |
259 | } | |
260 | ||
a1b32a59 CM |
261 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
262 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
263 | { |
264 | struct btrfs_device *dev; | |
8a4b83cc | 265 | |
c6e30871 | 266 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 267 | if (dev->devid == devid && |
8f18cf13 | 268 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 269 | return dev; |
a443755f | 270 | } |
8a4b83cc CM |
271 | } |
272 | return NULL; | |
273 | } | |
274 | ||
a1b32a59 | 275 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 276 | { |
8a4b83cc CM |
277 | struct btrfs_fs_devices *fs_devices; |
278 | ||
c6e30871 | 279 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
280 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
281 | return fs_devices; | |
282 | } | |
283 | return NULL; | |
284 | } | |
285 | ||
beaf8ab3 SB |
286 | static int |
287 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
288 | int flush, struct block_device **bdev, | |
289 | struct buffer_head **bh) | |
290 | { | |
291 | int ret; | |
292 | ||
293 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
294 | ||
295 | if (IS_ERR(*bdev)) { | |
296 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
297 | goto error; |
298 | } | |
299 | ||
300 | if (flush) | |
301 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
302 | ret = set_blocksize(*bdev, 4096); | |
303 | if (ret) { | |
304 | blkdev_put(*bdev, flags); | |
305 | goto error; | |
306 | } | |
307 | invalidate_bdev(*bdev); | |
308 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
309 | if (IS_ERR(*bh)) { |
310 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
311 | blkdev_put(*bdev, flags); |
312 | goto error; | |
313 | } | |
314 | ||
315 | return 0; | |
316 | ||
317 | error: | |
318 | *bdev = NULL; | |
319 | *bh = NULL; | |
320 | return ret; | |
321 | } | |
322 | ||
ffbd517d CM |
323 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
324 | struct bio *head, struct bio *tail) | |
325 | { | |
326 | ||
327 | struct bio *old_head; | |
328 | ||
329 | old_head = pending_bios->head; | |
330 | pending_bios->head = head; | |
331 | if (pending_bios->tail) | |
332 | tail->bi_next = old_head; | |
333 | else | |
334 | pending_bios->tail = tail; | |
335 | } | |
336 | ||
8b712842 CM |
337 | /* |
338 | * we try to collect pending bios for a device so we don't get a large | |
339 | * number of procs sending bios down to the same device. This greatly | |
340 | * improves the schedulers ability to collect and merge the bios. | |
341 | * | |
342 | * But, it also turns into a long list of bios to process and that is sure | |
343 | * to eventually make the worker thread block. The solution here is to | |
344 | * make some progress and then put this work struct back at the end of | |
345 | * the list if the block device is congested. This way, multiple devices | |
346 | * can make progress from a single worker thread. | |
347 | */ | |
143bede5 | 348 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 349 | { |
0b246afa | 350 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
351 | struct bio *pending; |
352 | struct backing_dev_info *bdi; | |
ffbd517d | 353 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
354 | struct bio *tail; |
355 | struct bio *cur; | |
356 | int again = 0; | |
ffbd517d | 357 | unsigned long num_run; |
d644d8a1 | 358 | unsigned long batch_run = 0; |
b64a2851 | 359 | unsigned long limit; |
b765ead5 | 360 | unsigned long last_waited = 0; |
d84275c9 | 361 | int force_reg = 0; |
0e588859 | 362 | int sync_pending = 0; |
211588ad CM |
363 | struct blk_plug plug; |
364 | ||
365 | /* | |
366 | * this function runs all the bios we've collected for | |
367 | * a particular device. We don't want to wander off to | |
368 | * another device without first sending all of these down. | |
369 | * So, setup a plug here and finish it off before we return | |
370 | */ | |
371 | blk_start_plug(&plug); | |
8b712842 | 372 | |
efa7c9f9 | 373 | bdi = device->bdev->bd_bdi; |
b64a2851 CM |
374 | limit = btrfs_async_submit_limit(fs_info); |
375 | limit = limit * 2 / 3; | |
376 | ||
8b712842 CM |
377 | loop: |
378 | spin_lock(&device->io_lock); | |
379 | ||
a6837051 | 380 | loop_lock: |
d84275c9 | 381 | num_run = 0; |
ffbd517d | 382 | |
8b712842 CM |
383 | /* take all the bios off the list at once and process them |
384 | * later on (without the lock held). But, remember the | |
385 | * tail and other pointers so the bios can be properly reinserted | |
386 | * into the list if we hit congestion | |
387 | */ | |
d84275c9 | 388 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 389 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
390 | force_reg = 1; |
391 | } else { | |
ffbd517d | 392 | pending_bios = &device->pending_bios; |
d84275c9 CM |
393 | force_reg = 0; |
394 | } | |
ffbd517d CM |
395 | |
396 | pending = pending_bios->head; | |
397 | tail = pending_bios->tail; | |
8b712842 | 398 | WARN_ON(pending && !tail); |
8b712842 CM |
399 | |
400 | /* | |
401 | * if pending was null this time around, no bios need processing | |
402 | * at all and we can stop. Otherwise it'll loop back up again | |
403 | * and do an additional check so no bios are missed. | |
404 | * | |
405 | * device->running_pending is used to synchronize with the | |
406 | * schedule_bio code. | |
407 | */ | |
ffbd517d CM |
408 | if (device->pending_sync_bios.head == NULL && |
409 | device->pending_bios.head == NULL) { | |
8b712842 CM |
410 | again = 0; |
411 | device->running_pending = 0; | |
ffbd517d CM |
412 | } else { |
413 | again = 1; | |
414 | device->running_pending = 1; | |
8b712842 | 415 | } |
ffbd517d CM |
416 | |
417 | pending_bios->head = NULL; | |
418 | pending_bios->tail = NULL; | |
419 | ||
8b712842 CM |
420 | spin_unlock(&device->io_lock); |
421 | ||
d397712b | 422 | while (pending) { |
ffbd517d CM |
423 | |
424 | rmb(); | |
d84275c9 CM |
425 | /* we want to work on both lists, but do more bios on the |
426 | * sync list than the regular list | |
427 | */ | |
428 | if ((num_run > 32 && | |
429 | pending_bios != &device->pending_sync_bios && | |
430 | device->pending_sync_bios.head) || | |
431 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
432 | device->pending_bios.head)) { | |
ffbd517d CM |
433 | spin_lock(&device->io_lock); |
434 | requeue_list(pending_bios, pending, tail); | |
435 | goto loop_lock; | |
436 | } | |
437 | ||
8b712842 CM |
438 | cur = pending; |
439 | pending = pending->bi_next; | |
440 | cur->bi_next = NULL; | |
b64a2851 | 441 | |
ee863954 DS |
442 | /* |
443 | * atomic_dec_return implies a barrier for waitqueue_active | |
444 | */ | |
66657b31 | 445 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
446 | waitqueue_active(&fs_info->async_submit_wait)) |
447 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 448 | |
dac56212 | 449 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 450 | |
2ab1ba68 CM |
451 | /* |
452 | * if we're doing the sync list, record that our | |
453 | * plug has some sync requests on it | |
454 | * | |
455 | * If we're doing the regular list and there are | |
456 | * sync requests sitting around, unplug before | |
457 | * we add more | |
458 | */ | |
459 | if (pending_bios == &device->pending_sync_bios) { | |
460 | sync_pending = 1; | |
461 | } else if (sync_pending) { | |
462 | blk_finish_plug(&plug); | |
463 | blk_start_plug(&plug); | |
464 | sync_pending = 0; | |
465 | } | |
466 | ||
4e49ea4a | 467 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
468 | num_run++; |
469 | batch_run++; | |
853d8ec4 DS |
470 | |
471 | cond_resched(); | |
8b712842 CM |
472 | |
473 | /* | |
474 | * we made progress, there is more work to do and the bdi | |
475 | * is now congested. Back off and let other work structs | |
476 | * run instead | |
477 | */ | |
57fd5a5f | 478 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 479 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 480 | struct io_context *ioc; |
8b712842 | 481 | |
b765ead5 CM |
482 | ioc = current->io_context; |
483 | ||
484 | /* | |
485 | * the main goal here is that we don't want to | |
486 | * block if we're going to be able to submit | |
487 | * more requests without blocking. | |
488 | * | |
489 | * This code does two great things, it pokes into | |
490 | * the elevator code from a filesystem _and_ | |
491 | * it makes assumptions about how batching works. | |
492 | */ | |
493 | if (ioc && ioc->nr_batch_requests > 0 && | |
494 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
495 | (last_waited == 0 || | |
496 | ioc->last_waited == last_waited)) { | |
497 | /* | |
498 | * we want to go through our batch of | |
499 | * requests and stop. So, we copy out | |
500 | * the ioc->last_waited time and test | |
501 | * against it before looping | |
502 | */ | |
503 | last_waited = ioc->last_waited; | |
853d8ec4 | 504 | cond_resched(); |
b765ead5 CM |
505 | continue; |
506 | } | |
8b712842 | 507 | spin_lock(&device->io_lock); |
ffbd517d | 508 | requeue_list(pending_bios, pending, tail); |
a6837051 | 509 | device->running_pending = 1; |
8b712842 CM |
510 | |
511 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
512 | btrfs_queue_work(fs_info->submit_workers, |
513 | &device->work); | |
8b712842 CM |
514 | goto done; |
515 | } | |
d85c8a6f CM |
516 | /* unplug every 64 requests just for good measure */ |
517 | if (batch_run % 64 == 0) { | |
518 | blk_finish_plug(&plug); | |
519 | blk_start_plug(&plug); | |
520 | sync_pending = 0; | |
521 | } | |
8b712842 | 522 | } |
ffbd517d | 523 | |
51684082 CM |
524 | cond_resched(); |
525 | if (again) | |
526 | goto loop; | |
527 | ||
528 | spin_lock(&device->io_lock); | |
529 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
530 | goto loop_lock; | |
531 | spin_unlock(&device->io_lock); | |
532 | ||
8b712842 | 533 | done: |
211588ad | 534 | blk_finish_plug(&plug); |
8b712842 CM |
535 | } |
536 | ||
b2950863 | 537 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
538 | { |
539 | struct btrfs_device *device; | |
540 | ||
541 | device = container_of(work, struct btrfs_device, work); | |
542 | run_scheduled_bios(device); | |
543 | } | |
544 | ||
4fde46f0 AJ |
545 | |
546 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
547 | { | |
548 | struct btrfs_fs_devices *fs_devs; | |
549 | struct btrfs_device *dev; | |
550 | ||
551 | if (!cur_dev->name) | |
552 | return; | |
553 | ||
554 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
555 | int del = 1; | |
556 | ||
557 | if (fs_devs->opened) | |
558 | continue; | |
559 | if (fs_devs->seeding) | |
560 | continue; | |
561 | ||
562 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
563 | ||
564 | if (dev == cur_dev) | |
565 | continue; | |
566 | if (!dev->name) | |
567 | continue; | |
568 | ||
569 | /* | |
570 | * Todo: This won't be enough. What if the same device | |
571 | * comes back (with new uuid and) with its mapper path? | |
572 | * But for now, this does help as mostly an admin will | |
573 | * either use mapper or non mapper path throughout. | |
574 | */ | |
575 | rcu_read_lock(); | |
576 | del = strcmp(rcu_str_deref(dev->name), | |
577 | rcu_str_deref(cur_dev->name)); | |
578 | rcu_read_unlock(); | |
579 | if (!del) | |
580 | break; | |
581 | } | |
582 | ||
583 | if (!del) { | |
584 | /* delete the stale device */ | |
585 | if (fs_devs->num_devices == 1) { | |
586 | btrfs_sysfs_remove_fsid(fs_devs); | |
587 | list_del(&fs_devs->list); | |
588 | free_fs_devices(fs_devs); | |
589 | } else { | |
590 | fs_devs->num_devices--; | |
591 | list_del(&dev->dev_list); | |
592 | rcu_string_free(dev->name); | |
593 | kfree(dev); | |
594 | } | |
595 | break; | |
596 | } | |
597 | } | |
598 | } | |
599 | ||
60999ca4 DS |
600 | /* |
601 | * Add new device to list of registered devices | |
602 | * | |
603 | * Returns: | |
604 | * 1 - first time device is seen | |
605 | * 0 - device already known | |
606 | * < 0 - error | |
607 | */ | |
a1b32a59 | 608 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
609 | struct btrfs_super_block *disk_super, |
610 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
611 | { | |
612 | struct btrfs_device *device; | |
613 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 614 | struct rcu_string *name; |
60999ca4 | 615 | int ret = 0; |
8a4b83cc CM |
616 | u64 found_transid = btrfs_super_generation(disk_super); |
617 | ||
618 | fs_devices = find_fsid(disk_super->fsid); | |
619 | if (!fs_devices) { | |
2208a378 ID |
620 | fs_devices = alloc_fs_devices(disk_super->fsid); |
621 | if (IS_ERR(fs_devices)) | |
622 | return PTR_ERR(fs_devices); | |
623 | ||
8a4b83cc | 624 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 625 | |
8a4b83cc CM |
626 | device = NULL; |
627 | } else { | |
a443755f CM |
628 | device = __find_device(&fs_devices->devices, devid, |
629 | disk_super->dev_item.uuid); | |
8a4b83cc | 630 | } |
443f24fe | 631 | |
8a4b83cc | 632 | if (!device) { |
2b82032c YZ |
633 | if (fs_devices->opened) |
634 | return -EBUSY; | |
635 | ||
12bd2fc0 ID |
636 | device = btrfs_alloc_device(NULL, &devid, |
637 | disk_super->dev_item.uuid); | |
638 | if (IS_ERR(device)) { | |
8a4b83cc | 639 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 640 | return PTR_ERR(device); |
8a4b83cc | 641 | } |
606686ee JB |
642 | |
643 | name = rcu_string_strdup(path, GFP_NOFS); | |
644 | if (!name) { | |
8a4b83cc CM |
645 | kfree(device); |
646 | return -ENOMEM; | |
647 | } | |
606686ee | 648 | rcu_assign_pointer(device->name, name); |
90519d66 | 649 | |
e5e9a520 | 650 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 651 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 652 | fs_devices->num_devices++; |
e5e9a520 CM |
653 | mutex_unlock(&fs_devices->device_list_mutex); |
654 | ||
60999ca4 | 655 | ret = 1; |
2b82032c | 656 | device->fs_devices = fs_devices; |
606686ee | 657 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
658 | /* |
659 | * When FS is already mounted. | |
660 | * 1. If you are here and if the device->name is NULL that | |
661 | * means this device was missing at time of FS mount. | |
662 | * 2. If you are here and if the device->name is different | |
663 | * from 'path' that means either | |
664 | * a. The same device disappeared and reappeared with | |
665 | * different name. or | |
666 | * b. The missing-disk-which-was-replaced, has | |
667 | * reappeared now. | |
668 | * | |
669 | * We must allow 1 and 2a above. But 2b would be a spurious | |
670 | * and unintentional. | |
671 | * | |
672 | * Further in case of 1 and 2a above, the disk at 'path' | |
673 | * would have missed some transaction when it was away and | |
674 | * in case of 2a the stale bdev has to be updated as well. | |
675 | * 2b must not be allowed at all time. | |
676 | */ | |
677 | ||
678 | /* | |
0f23ae74 CM |
679 | * For now, we do allow update to btrfs_fs_device through the |
680 | * btrfs dev scan cli after FS has been mounted. We're still | |
681 | * tracking a problem where systems fail mount by subvolume id | |
682 | * when we reject replacement on a mounted FS. | |
b96de000 | 683 | */ |
0f23ae74 | 684 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
685 | /* |
686 | * That is if the FS is _not_ mounted and if you | |
687 | * are here, that means there is more than one | |
688 | * disk with same uuid and devid.We keep the one | |
689 | * with larger generation number or the last-in if | |
690 | * generation are equal. | |
691 | */ | |
0f23ae74 | 692 | return -EEXIST; |
77bdae4d | 693 | } |
b96de000 | 694 | |
606686ee | 695 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
696 | if (!name) |
697 | return -ENOMEM; | |
606686ee JB |
698 | rcu_string_free(device->name); |
699 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
700 | if (device->missing) { |
701 | fs_devices->missing_devices--; | |
702 | device->missing = 0; | |
703 | } | |
8a4b83cc CM |
704 | } |
705 | ||
77bdae4d AJ |
706 | /* |
707 | * Unmount does not free the btrfs_device struct but would zero | |
708 | * generation along with most of the other members. So just update | |
709 | * it back. We need it to pick the disk with largest generation | |
710 | * (as above). | |
711 | */ | |
712 | if (!fs_devices->opened) | |
713 | device->generation = found_transid; | |
714 | ||
4fde46f0 AJ |
715 | /* |
716 | * if there is new btrfs on an already registered device, | |
717 | * then remove the stale device entry. | |
718 | */ | |
02feae3c AJ |
719 | if (ret > 0) |
720 | btrfs_free_stale_device(device); | |
4fde46f0 | 721 | |
8a4b83cc | 722 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
723 | |
724 | return ret; | |
8a4b83cc CM |
725 | } |
726 | ||
e4404d6e YZ |
727 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
728 | { | |
729 | struct btrfs_fs_devices *fs_devices; | |
730 | struct btrfs_device *device; | |
731 | struct btrfs_device *orig_dev; | |
732 | ||
2208a378 ID |
733 | fs_devices = alloc_fs_devices(orig->fsid); |
734 | if (IS_ERR(fs_devices)) | |
735 | return fs_devices; | |
e4404d6e | 736 | |
adbbb863 | 737 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 738 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 739 | |
46224705 | 740 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 741 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
742 | struct rcu_string *name; |
743 | ||
12bd2fc0 ID |
744 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
745 | orig_dev->uuid); | |
746 | if (IS_ERR(device)) | |
e4404d6e YZ |
747 | goto error; |
748 | ||
606686ee JB |
749 | /* |
750 | * This is ok to do without rcu read locked because we hold the | |
751 | * uuid mutex so nothing we touch in here is going to disappear. | |
752 | */ | |
e755f780 | 753 | if (orig_dev->name) { |
78f2c9e6 DS |
754 | name = rcu_string_strdup(orig_dev->name->str, |
755 | GFP_KERNEL); | |
e755f780 AJ |
756 | if (!name) { |
757 | kfree(device); | |
758 | goto error; | |
759 | } | |
760 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 761 | } |
e4404d6e | 762 | |
e4404d6e YZ |
763 | list_add(&device->dev_list, &fs_devices->devices); |
764 | device->fs_devices = fs_devices; | |
765 | fs_devices->num_devices++; | |
766 | } | |
adbbb863 | 767 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
768 | return fs_devices; |
769 | error: | |
adbbb863 | 770 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
771 | free_fs_devices(fs_devices); |
772 | return ERR_PTR(-ENOMEM); | |
773 | } | |
774 | ||
9eaed21e | 775 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 776 | { |
c6e30871 | 777 | struct btrfs_device *device, *next; |
443f24fe | 778 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 779 | |
dfe25020 CM |
780 | mutex_lock(&uuid_mutex); |
781 | again: | |
46224705 | 782 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 783 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 784 | if (device->in_fs_metadata) { |
63a212ab | 785 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
786 | (!latest_dev || |
787 | device->generation > latest_dev->generation)) { | |
788 | latest_dev = device; | |
a6b0d5c8 | 789 | } |
2b82032c | 790 | continue; |
a6b0d5c8 | 791 | } |
2b82032c | 792 | |
8dabb742 SB |
793 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
794 | /* | |
795 | * In the first step, keep the device which has | |
796 | * the correct fsid and the devid that is used | |
797 | * for the dev_replace procedure. | |
798 | * In the second step, the dev_replace state is | |
799 | * read from the device tree and it is known | |
800 | * whether the procedure is really active or | |
801 | * not, which means whether this device is | |
802 | * used or whether it should be removed. | |
803 | */ | |
804 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
805 | continue; | |
806 | } | |
807 | } | |
2b82032c | 808 | if (device->bdev) { |
d4d77629 | 809 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
810 | device->bdev = NULL; |
811 | fs_devices->open_devices--; | |
812 | } | |
813 | if (device->writeable) { | |
814 | list_del_init(&device->dev_alloc_list); | |
815 | device->writeable = 0; | |
8dabb742 SB |
816 | if (!device->is_tgtdev_for_dev_replace) |
817 | fs_devices->rw_devices--; | |
2b82032c | 818 | } |
e4404d6e YZ |
819 | list_del_init(&device->dev_list); |
820 | fs_devices->num_devices--; | |
606686ee | 821 | rcu_string_free(device->name); |
e4404d6e | 822 | kfree(device); |
dfe25020 | 823 | } |
2b82032c YZ |
824 | |
825 | if (fs_devices->seed) { | |
826 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
827 | goto again; |
828 | } | |
829 | ||
443f24fe | 830 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 831 | |
dfe25020 | 832 | mutex_unlock(&uuid_mutex); |
dfe25020 | 833 | } |
a0af469b | 834 | |
1f78160c XG |
835 | static void __free_device(struct work_struct *work) |
836 | { | |
837 | struct btrfs_device *device; | |
838 | ||
839 | device = container_of(work, struct btrfs_device, rcu_work); | |
606686ee | 840 | rcu_string_free(device->name); |
1f78160c XG |
841 | kfree(device); |
842 | } | |
843 | ||
844 | static void free_device(struct rcu_head *head) | |
845 | { | |
846 | struct btrfs_device *device; | |
847 | ||
848 | device = container_of(head, struct btrfs_device, rcu); | |
849 | ||
850 | INIT_WORK(&device->rcu_work, __free_device); | |
851 | schedule_work(&device->rcu_work); | |
852 | } | |
853 | ||
14238819 AJ |
854 | static void btrfs_close_bdev(struct btrfs_device *device) |
855 | { | |
856 | if (device->bdev && device->writeable) { | |
857 | sync_blockdev(device->bdev); | |
858 | invalidate_bdev(device->bdev); | |
859 | } | |
860 | ||
861 | if (device->bdev) | |
862 | blkdev_put(device->bdev, device->mode); | |
863 | } | |
864 | ||
0ccd0528 | 865 | static void btrfs_prepare_close_one_device(struct btrfs_device *device) |
f448341a AJ |
866 | { |
867 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
868 | struct btrfs_device *new_device; | |
869 | struct rcu_string *name; | |
870 | ||
871 | if (device->bdev) | |
872 | fs_devices->open_devices--; | |
873 | ||
874 | if (device->writeable && | |
875 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
876 | list_del_init(&device->dev_alloc_list); | |
877 | fs_devices->rw_devices--; | |
878 | } | |
879 | ||
880 | if (device->missing) | |
881 | fs_devices->missing_devices--; | |
882 | ||
883 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
884 | device->uuid); | |
885 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
886 | ||
887 | /* Safe because we are under uuid_mutex */ | |
888 | if (device->name) { | |
889 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
890 | BUG_ON(!name); /* -ENOMEM */ | |
891 | rcu_assign_pointer(new_device->name, name); | |
892 | } | |
893 | ||
894 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
895 | new_device->fs_devices = device->fs_devices; | |
f448341a AJ |
896 | } |
897 | ||
2b82032c | 898 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 899 | { |
2037a093 | 900 | struct btrfs_device *device, *tmp; |
0ccd0528 AJ |
901 | struct list_head pending_put; |
902 | ||
903 | INIT_LIST_HEAD(&pending_put); | |
e4404d6e | 904 | |
2b82032c YZ |
905 | if (--fs_devices->opened > 0) |
906 | return 0; | |
8a4b83cc | 907 | |
c9513edb | 908 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 909 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
0ccd0528 AJ |
910 | btrfs_prepare_close_one_device(device); |
911 | list_add(&device->dev_list, &pending_put); | |
8a4b83cc | 912 | } |
c9513edb XG |
913 | mutex_unlock(&fs_devices->device_list_mutex); |
914 | ||
0ccd0528 AJ |
915 | /* |
916 | * btrfs_show_devname() is using the device_list_mutex, | |
917 | * sometimes call to blkdev_put() leads vfs calling | |
918 | * into this func. So do put outside of device_list_mutex, | |
919 | * as of now. | |
920 | */ | |
921 | while (!list_empty(&pending_put)) { | |
922 | device = list_first_entry(&pending_put, | |
923 | struct btrfs_device, dev_list); | |
924 | list_del(&device->dev_list); | |
925 | btrfs_close_bdev(device); | |
926 | call_rcu(&device->rcu, free_device); | |
927 | } | |
928 | ||
e4404d6e YZ |
929 | WARN_ON(fs_devices->open_devices); |
930 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
931 | fs_devices->opened = 0; |
932 | fs_devices->seeding = 0; | |
2b82032c | 933 | |
8a4b83cc CM |
934 | return 0; |
935 | } | |
936 | ||
2b82032c YZ |
937 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
938 | { | |
e4404d6e | 939 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
940 | int ret; |
941 | ||
942 | mutex_lock(&uuid_mutex); | |
943 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
944 | if (!fs_devices->opened) { |
945 | seed_devices = fs_devices->seed; | |
946 | fs_devices->seed = NULL; | |
947 | } | |
2b82032c | 948 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
949 | |
950 | while (seed_devices) { | |
951 | fs_devices = seed_devices; | |
952 | seed_devices = fs_devices->seed; | |
953 | __btrfs_close_devices(fs_devices); | |
954 | free_fs_devices(fs_devices); | |
955 | } | |
bc178622 ES |
956 | /* |
957 | * Wait for rcu kworkers under __btrfs_close_devices | |
958 | * to finish all blkdev_puts so device is really | |
959 | * free when umount is done. | |
960 | */ | |
961 | rcu_barrier(); | |
2b82032c YZ |
962 | return ret; |
963 | } | |
964 | ||
e4404d6e YZ |
965 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
966 | fmode_t flags, void *holder) | |
8a4b83cc | 967 | { |
d5e2003c | 968 | struct request_queue *q; |
8a4b83cc CM |
969 | struct block_device *bdev; |
970 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 971 | struct btrfs_device *device; |
443f24fe | 972 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
973 | struct buffer_head *bh; |
974 | struct btrfs_super_block *disk_super; | |
a0af469b | 975 | u64 devid; |
2b82032c | 976 | int seeding = 1; |
a0af469b | 977 | int ret = 0; |
8a4b83cc | 978 | |
d4d77629 TH |
979 | flags |= FMODE_EXCL; |
980 | ||
c6e30871 | 981 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
982 | if (device->bdev) |
983 | continue; | |
dfe25020 CM |
984 | if (!device->name) |
985 | continue; | |
986 | ||
f63e0cca ES |
987 | /* Just open everything we can; ignore failures here */ |
988 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
989 | &bdev, &bh)) | |
beaf8ab3 | 990 | continue; |
a0af469b CM |
991 | |
992 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 993 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
994 | if (devid != device->devid) |
995 | goto error_brelse; | |
996 | ||
2b82032c YZ |
997 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
998 | BTRFS_UUID_SIZE)) | |
999 | goto error_brelse; | |
1000 | ||
1001 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
1002 | if (!latest_dev || |
1003 | device->generation > latest_dev->generation) | |
1004 | latest_dev = device; | |
a0af469b | 1005 | |
2b82032c YZ |
1006 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
1007 | device->writeable = 0; | |
1008 | } else { | |
1009 | device->writeable = !bdev_read_only(bdev); | |
1010 | seeding = 0; | |
1011 | } | |
1012 | ||
d5e2003c | 1013 | q = bdev_get_queue(bdev); |
90180da4 | 1014 | if (blk_queue_discard(q)) |
d5e2003c | 1015 | device->can_discard = 1; |
e884f4f0 AJ |
1016 | if (!blk_queue_nonrot(q)) |
1017 | fs_devices->rotating = 1; | |
d5e2003c | 1018 | |
8a4b83cc | 1019 | device->bdev = bdev; |
dfe25020 | 1020 | device->in_fs_metadata = 0; |
15916de8 CM |
1021 | device->mode = flags; |
1022 | ||
a0af469b | 1023 | fs_devices->open_devices++; |
55e50e45 ID |
1024 | if (device->writeable && |
1025 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
1026 | fs_devices->rw_devices++; |
1027 | list_add(&device->dev_alloc_list, | |
1028 | &fs_devices->alloc_list); | |
1029 | } | |
4f6c9328 | 1030 | brelse(bh); |
a0af469b | 1031 | continue; |
a061fc8d | 1032 | |
a0af469b CM |
1033 | error_brelse: |
1034 | brelse(bh); | |
d4d77629 | 1035 | blkdev_put(bdev, flags); |
a0af469b | 1036 | continue; |
8a4b83cc | 1037 | } |
a0af469b | 1038 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1039 | ret = -EINVAL; |
a0af469b CM |
1040 | goto out; |
1041 | } | |
2b82032c YZ |
1042 | fs_devices->seeding = seeding; |
1043 | fs_devices->opened = 1; | |
443f24fe | 1044 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1045 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1046 | out: |
2b82032c YZ |
1047 | return ret; |
1048 | } | |
1049 | ||
1050 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 1051 | fmode_t flags, void *holder) |
2b82032c YZ |
1052 | { |
1053 | int ret; | |
1054 | ||
1055 | mutex_lock(&uuid_mutex); | |
1056 | if (fs_devices->opened) { | |
e4404d6e YZ |
1057 | fs_devices->opened++; |
1058 | ret = 0; | |
2b82032c | 1059 | } else { |
15916de8 | 1060 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1061 | } |
8a4b83cc | 1062 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1063 | return ret; |
1064 | } | |
1065 | ||
6cf86a00 AJ |
1066 | void btrfs_release_disk_super(struct page *page) |
1067 | { | |
1068 | kunmap(page); | |
1069 | put_page(page); | |
1070 | } | |
1071 | ||
1072 | int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, | |
1073 | struct page **page, struct btrfs_super_block **disk_super) | |
1074 | { | |
1075 | void *p; | |
1076 | pgoff_t index; | |
1077 | ||
1078 | /* make sure our super fits in the device */ | |
1079 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1080 | return 1; | |
1081 | ||
1082 | /* make sure our super fits in the page */ | |
1083 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1084 | return 1; | |
1085 | ||
1086 | /* make sure our super doesn't straddle pages on disk */ | |
1087 | index = bytenr >> PAGE_SHIFT; | |
1088 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1089 | return 1; | |
1090 | ||
1091 | /* pull in the page with our super */ | |
1092 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1093 | index, GFP_KERNEL); | |
1094 | ||
1095 | if (IS_ERR_OR_NULL(*page)) | |
1096 | return 1; | |
1097 | ||
1098 | p = kmap(*page); | |
1099 | ||
1100 | /* align our pointer to the offset of the super block */ | |
1101 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1102 | ||
1103 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1104 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1105 | btrfs_release_disk_super(*page); | |
1106 | return 1; | |
1107 | } | |
1108 | ||
1109 | if ((*disk_super)->label[0] && | |
1110 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1111 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1112 | ||
1113 | return 0; | |
1114 | } | |
1115 | ||
6f60cbd3 DS |
1116 | /* |
1117 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1118 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1119 | * is read via pagecache | |
1120 | */ | |
97288f2c | 1121 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1122 | struct btrfs_fs_devices **fs_devices_ret) |
1123 | { | |
1124 | struct btrfs_super_block *disk_super; | |
1125 | struct block_device *bdev; | |
6f60cbd3 | 1126 | struct page *page; |
6f60cbd3 | 1127 | int ret = -EINVAL; |
8a4b83cc | 1128 | u64 devid; |
f2984462 | 1129 | u64 transid; |
02db0844 | 1130 | u64 total_devices; |
6f60cbd3 | 1131 | u64 bytenr; |
8a4b83cc | 1132 | |
6f60cbd3 DS |
1133 | /* |
1134 | * we would like to check all the supers, but that would make | |
1135 | * a btrfs mount succeed after a mkfs from a different FS. | |
1136 | * So, we need to add a special mount option to scan for | |
1137 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1138 | */ | |
1139 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1140 | flags |= FMODE_EXCL; |
10f6327b | 1141 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1142 | |
1143 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1144 | if (IS_ERR(bdev)) { |
1145 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1146 | goto error; |
6f60cbd3 DS |
1147 | } |
1148 | ||
6cf86a00 | 1149 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) |
6f60cbd3 DS |
1150 | goto error_bdev_put; |
1151 | ||
a343832f | 1152 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1153 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1154 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1155 | |
8a4b83cc | 1156 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1157 | if (ret > 0) { |
1158 | if (disk_super->label[0]) { | |
62e85577 | 1159 | pr_info("BTRFS: device label %s ", disk_super->label); |
60999ca4 | 1160 | } else { |
62e85577 | 1161 | pr_info("BTRFS: device fsid %pU ", disk_super->fsid); |
60999ca4 DS |
1162 | } |
1163 | ||
62e85577 | 1164 | pr_cont("devid %llu transid %llu %s\n", devid, transid, path); |
60999ca4 DS |
1165 | ret = 0; |
1166 | } | |
02db0844 JB |
1167 | if (!ret && fs_devices_ret) |
1168 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 | 1169 | |
6cf86a00 | 1170 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1171 | |
1172 | error_bdev_put: | |
d4d77629 | 1173 | blkdev_put(bdev, flags); |
8a4b83cc | 1174 | error: |
beaf8ab3 | 1175 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1176 | return ret; |
1177 | } | |
0b86a832 | 1178 | |
6d07bcec MX |
1179 | /* helper to account the used device space in the range */ |
1180 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1181 | u64 end, u64 *length) | |
1182 | { | |
1183 | struct btrfs_key key; | |
fb456252 | 1184 | struct btrfs_root *root = device->fs_info->dev_root; |
6d07bcec MX |
1185 | struct btrfs_dev_extent *dev_extent; |
1186 | struct btrfs_path *path; | |
1187 | u64 extent_end; | |
1188 | int ret; | |
1189 | int slot; | |
1190 | struct extent_buffer *l; | |
1191 | ||
1192 | *length = 0; | |
1193 | ||
63a212ab | 1194 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1195 | return 0; |
1196 | ||
1197 | path = btrfs_alloc_path(); | |
1198 | if (!path) | |
1199 | return -ENOMEM; | |
e4058b54 | 1200 | path->reada = READA_FORWARD; |
6d07bcec MX |
1201 | |
1202 | key.objectid = device->devid; | |
1203 | key.offset = start; | |
1204 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1205 | ||
1206 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1207 | if (ret < 0) | |
1208 | goto out; | |
1209 | if (ret > 0) { | |
1210 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1211 | if (ret < 0) | |
1212 | goto out; | |
1213 | } | |
1214 | ||
1215 | while (1) { | |
1216 | l = path->nodes[0]; | |
1217 | slot = path->slots[0]; | |
1218 | if (slot >= btrfs_header_nritems(l)) { | |
1219 | ret = btrfs_next_leaf(root, path); | |
1220 | if (ret == 0) | |
1221 | continue; | |
1222 | if (ret < 0) | |
1223 | goto out; | |
1224 | ||
1225 | break; | |
1226 | } | |
1227 | btrfs_item_key_to_cpu(l, &key, slot); | |
1228 | ||
1229 | if (key.objectid < device->devid) | |
1230 | goto next; | |
1231 | ||
1232 | if (key.objectid > device->devid) | |
1233 | break; | |
1234 | ||
962a298f | 1235 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1236 | goto next; |
1237 | ||
1238 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1239 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1240 | dev_extent); | |
1241 | if (key.offset <= start && extent_end > end) { | |
1242 | *length = end - start + 1; | |
1243 | break; | |
1244 | } else if (key.offset <= start && extent_end > start) | |
1245 | *length += extent_end - start; | |
1246 | else if (key.offset > start && extent_end <= end) | |
1247 | *length += extent_end - key.offset; | |
1248 | else if (key.offset > start && key.offset <= end) { | |
1249 | *length += end - key.offset + 1; | |
1250 | break; | |
1251 | } else if (key.offset > end) | |
1252 | break; | |
1253 | ||
1254 | next: | |
1255 | path->slots[0]++; | |
1256 | } | |
1257 | ret = 0; | |
1258 | out: | |
1259 | btrfs_free_path(path); | |
1260 | return ret; | |
1261 | } | |
1262 | ||
499f377f | 1263 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1264 | struct btrfs_device *device, |
1265 | u64 *start, u64 len) | |
1266 | { | |
fb456252 | 1267 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1268 | struct extent_map *em; |
499f377f | 1269 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1270 | int ret = 0; |
1b984508 | 1271 | u64 physical_start = *start; |
6df9a95e | 1272 | |
499f377f JM |
1273 | if (transaction) |
1274 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1275 | again: |
1276 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1277 | struct map_lookup *map; |
1278 | int i; | |
1279 | ||
95617d69 | 1280 | map = em->map_lookup; |
6df9a95e | 1281 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1282 | u64 end; |
1283 | ||
6df9a95e JB |
1284 | if (map->stripes[i].dev != device) |
1285 | continue; | |
1b984508 | 1286 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1287 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1288 | physical_start) |
6df9a95e | 1289 | continue; |
c152b63e FM |
1290 | /* |
1291 | * Make sure that while processing the pinned list we do | |
1292 | * not override our *start with a lower value, because | |
1293 | * we can have pinned chunks that fall within this | |
1294 | * device hole and that have lower physical addresses | |
1295 | * than the pending chunks we processed before. If we | |
1296 | * do not take this special care we can end up getting | |
1297 | * 2 pending chunks that start at the same physical | |
1298 | * device offsets because the end offset of a pinned | |
1299 | * chunk can be equal to the start offset of some | |
1300 | * pending chunk. | |
1301 | */ | |
1302 | end = map->stripes[i].physical + em->orig_block_len; | |
1303 | if (end > *start) { | |
1304 | *start = end; | |
1305 | ret = 1; | |
1306 | } | |
6df9a95e JB |
1307 | } |
1308 | } | |
499f377f JM |
1309 | if (search_list != &fs_info->pinned_chunks) { |
1310 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1311 | goto again; |
1312 | } | |
6df9a95e JB |
1313 | |
1314 | return ret; | |
1315 | } | |
1316 | ||
1317 | ||
0b86a832 | 1318 | /* |
499f377f JM |
1319 | * find_free_dev_extent_start - find free space in the specified device |
1320 | * @device: the device which we search the free space in | |
1321 | * @num_bytes: the size of the free space that we need | |
1322 | * @search_start: the position from which to begin the search | |
1323 | * @start: store the start of the free space. | |
1324 | * @len: the size of the free space. that we find, or the size | |
1325 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1326 | * |
0b86a832 CM |
1327 | * this uses a pretty simple search, the expectation is that it is |
1328 | * called very infrequently and that a given device has a small number | |
1329 | * of extents | |
7bfc837d MX |
1330 | * |
1331 | * @start is used to store the start of the free space if we find. But if we | |
1332 | * don't find suitable free space, it will be used to store the start position | |
1333 | * of the max free space. | |
1334 | * | |
1335 | * @len is used to store the size of the free space that we find. | |
1336 | * But if we don't find suitable free space, it is used to store the size of | |
1337 | * the max free space. | |
0b86a832 | 1338 | */ |
499f377f JM |
1339 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1340 | struct btrfs_device *device, u64 num_bytes, | |
1341 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1342 | { |
0b246afa JM |
1343 | struct btrfs_fs_info *fs_info = device->fs_info; |
1344 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1345 | struct btrfs_key key; |
7bfc837d | 1346 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1347 | struct btrfs_path *path; |
7bfc837d MX |
1348 | u64 hole_size; |
1349 | u64 max_hole_start; | |
1350 | u64 max_hole_size; | |
1351 | u64 extent_end; | |
0b86a832 CM |
1352 | u64 search_end = device->total_bytes; |
1353 | int ret; | |
7bfc837d | 1354 | int slot; |
0b86a832 | 1355 | struct extent_buffer *l; |
8cdc7c5b FM |
1356 | u64 min_search_start; |
1357 | ||
1358 | /* | |
1359 | * We don't want to overwrite the superblock on the drive nor any area | |
1360 | * used by the boot loader (grub for example), so we make sure to start | |
1361 | * at an offset of at least 1MB. | |
1362 | */ | |
0b246afa | 1363 | min_search_start = max(fs_info->alloc_start, 1024ull * 1024); |
8cdc7c5b | 1364 | search_start = max(search_start, min_search_start); |
0b86a832 | 1365 | |
6df9a95e JB |
1366 | path = btrfs_alloc_path(); |
1367 | if (!path) | |
1368 | return -ENOMEM; | |
f2ab7618 | 1369 | |
7bfc837d MX |
1370 | max_hole_start = search_start; |
1371 | max_hole_size = 0; | |
1372 | ||
f2ab7618 | 1373 | again: |
63a212ab | 1374 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1375 | ret = -ENOSPC; |
6df9a95e | 1376 | goto out; |
7bfc837d MX |
1377 | } |
1378 | ||
e4058b54 | 1379 | path->reada = READA_FORWARD; |
6df9a95e JB |
1380 | path->search_commit_root = 1; |
1381 | path->skip_locking = 1; | |
7bfc837d | 1382 | |
0b86a832 CM |
1383 | key.objectid = device->devid; |
1384 | key.offset = search_start; | |
1385 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1386 | |
125ccb0a | 1387 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1388 | if (ret < 0) |
7bfc837d | 1389 | goto out; |
1fcbac58 YZ |
1390 | if (ret > 0) { |
1391 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1392 | if (ret < 0) | |
7bfc837d | 1393 | goto out; |
1fcbac58 | 1394 | } |
7bfc837d | 1395 | |
0b86a832 CM |
1396 | while (1) { |
1397 | l = path->nodes[0]; | |
1398 | slot = path->slots[0]; | |
1399 | if (slot >= btrfs_header_nritems(l)) { | |
1400 | ret = btrfs_next_leaf(root, path); | |
1401 | if (ret == 0) | |
1402 | continue; | |
1403 | if (ret < 0) | |
7bfc837d MX |
1404 | goto out; |
1405 | ||
1406 | break; | |
0b86a832 CM |
1407 | } |
1408 | btrfs_item_key_to_cpu(l, &key, slot); | |
1409 | ||
1410 | if (key.objectid < device->devid) | |
1411 | goto next; | |
1412 | ||
1413 | if (key.objectid > device->devid) | |
7bfc837d | 1414 | break; |
0b86a832 | 1415 | |
962a298f | 1416 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1417 | goto next; |
9779b72f | 1418 | |
7bfc837d MX |
1419 | if (key.offset > search_start) { |
1420 | hole_size = key.offset - search_start; | |
9779b72f | 1421 | |
6df9a95e JB |
1422 | /* |
1423 | * Have to check before we set max_hole_start, otherwise | |
1424 | * we could end up sending back this offset anyway. | |
1425 | */ | |
499f377f | 1426 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1427 | &search_start, |
1b984508 FL |
1428 | hole_size)) { |
1429 | if (key.offset >= search_start) { | |
1430 | hole_size = key.offset - search_start; | |
1431 | } else { | |
1432 | WARN_ON_ONCE(1); | |
1433 | hole_size = 0; | |
1434 | } | |
1435 | } | |
6df9a95e | 1436 | |
7bfc837d MX |
1437 | if (hole_size > max_hole_size) { |
1438 | max_hole_start = search_start; | |
1439 | max_hole_size = hole_size; | |
1440 | } | |
9779b72f | 1441 | |
7bfc837d MX |
1442 | /* |
1443 | * If this free space is greater than which we need, | |
1444 | * it must be the max free space that we have found | |
1445 | * until now, so max_hole_start must point to the start | |
1446 | * of this free space and the length of this free space | |
1447 | * is stored in max_hole_size. Thus, we return | |
1448 | * max_hole_start and max_hole_size and go back to the | |
1449 | * caller. | |
1450 | */ | |
1451 | if (hole_size >= num_bytes) { | |
1452 | ret = 0; | |
1453 | goto out; | |
0b86a832 CM |
1454 | } |
1455 | } | |
0b86a832 | 1456 | |
0b86a832 | 1457 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1458 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1459 | dev_extent); | |
1460 | if (extent_end > search_start) | |
1461 | search_start = extent_end; | |
0b86a832 CM |
1462 | next: |
1463 | path->slots[0]++; | |
1464 | cond_resched(); | |
1465 | } | |
0b86a832 | 1466 | |
38c01b96 | 1467 | /* |
1468 | * At this point, search_start should be the end of | |
1469 | * allocated dev extents, and when shrinking the device, | |
1470 | * search_end may be smaller than search_start. | |
1471 | */ | |
f2ab7618 | 1472 | if (search_end > search_start) { |
38c01b96 | 1473 | hole_size = search_end - search_start; |
1474 | ||
499f377f | 1475 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1476 | hole_size)) { |
1477 | btrfs_release_path(path); | |
1478 | goto again; | |
1479 | } | |
0b86a832 | 1480 | |
f2ab7618 ZL |
1481 | if (hole_size > max_hole_size) { |
1482 | max_hole_start = search_start; | |
1483 | max_hole_size = hole_size; | |
1484 | } | |
6df9a95e JB |
1485 | } |
1486 | ||
7bfc837d | 1487 | /* See above. */ |
f2ab7618 | 1488 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1489 | ret = -ENOSPC; |
1490 | else | |
1491 | ret = 0; | |
1492 | ||
1493 | out: | |
2b82032c | 1494 | btrfs_free_path(path); |
7bfc837d | 1495 | *start = max_hole_start; |
b2117a39 | 1496 | if (len) |
7bfc837d | 1497 | *len = max_hole_size; |
0b86a832 CM |
1498 | return ret; |
1499 | } | |
1500 | ||
499f377f JM |
1501 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1502 | struct btrfs_device *device, u64 num_bytes, | |
1503 | u64 *start, u64 *len) | |
1504 | { | |
499f377f | 1505 | /* FIXME use last free of some kind */ |
499f377f | 1506 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1507 | num_bytes, 0, start, len); |
499f377f JM |
1508 | } |
1509 | ||
b2950863 | 1510 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1511 | struct btrfs_device *device, |
2196d6e8 | 1512 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1513 | { |
0b246afa JM |
1514 | struct btrfs_fs_info *fs_info = device->fs_info; |
1515 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1516 | int ret; |
1517 | struct btrfs_path *path; | |
8f18cf13 | 1518 | struct btrfs_key key; |
a061fc8d CM |
1519 | struct btrfs_key found_key; |
1520 | struct extent_buffer *leaf = NULL; | |
1521 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1522 | |
1523 | path = btrfs_alloc_path(); | |
1524 | if (!path) | |
1525 | return -ENOMEM; | |
1526 | ||
1527 | key.objectid = device->devid; | |
1528 | key.offset = start; | |
1529 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1530 | again: |
8f18cf13 | 1531 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1532 | if (ret > 0) { |
1533 | ret = btrfs_previous_item(root, path, key.objectid, | |
1534 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1535 | if (ret) |
1536 | goto out; | |
a061fc8d CM |
1537 | leaf = path->nodes[0]; |
1538 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1539 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1540 | struct btrfs_dev_extent); | |
1541 | BUG_ON(found_key.offset > start || found_key.offset + | |
1542 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1543 | key = found_key; |
1544 | btrfs_release_path(path); | |
1545 | goto again; | |
a061fc8d CM |
1546 | } else if (ret == 0) { |
1547 | leaf = path->nodes[0]; | |
1548 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1549 | struct btrfs_dev_extent); | |
79787eaa | 1550 | } else { |
0b246afa | 1551 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1552 | goto out; |
a061fc8d | 1553 | } |
8f18cf13 | 1554 | |
2196d6e8 MX |
1555 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1556 | ||
8f18cf13 | 1557 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1558 | if (ret) { |
0b246afa JM |
1559 | btrfs_handle_fs_error(fs_info, ret, |
1560 | "Failed to remove dev extent item"); | |
13212b54 | 1561 | } else { |
3204d33c | 1562 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1563 | } |
b0b802d7 | 1564 | out: |
8f18cf13 CM |
1565 | btrfs_free_path(path); |
1566 | return ret; | |
1567 | } | |
1568 | ||
48a3b636 ES |
1569 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1570 | struct btrfs_device *device, | |
1571 | u64 chunk_tree, u64 chunk_objectid, | |
1572 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1573 | { |
1574 | int ret; | |
1575 | struct btrfs_path *path; | |
0b246afa JM |
1576 | struct btrfs_fs_info *fs_info = device->fs_info; |
1577 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1578 | struct btrfs_dev_extent *extent; |
1579 | struct extent_buffer *leaf; | |
1580 | struct btrfs_key key; | |
1581 | ||
dfe25020 | 1582 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1583 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1584 | path = btrfs_alloc_path(); |
1585 | if (!path) | |
1586 | return -ENOMEM; | |
1587 | ||
0b86a832 | 1588 | key.objectid = device->devid; |
2b82032c | 1589 | key.offset = start; |
0b86a832 CM |
1590 | key.type = BTRFS_DEV_EXTENT_KEY; |
1591 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1592 | sizeof(*extent)); | |
2cdcecbc MF |
1593 | if (ret) |
1594 | goto out; | |
0b86a832 CM |
1595 | |
1596 | leaf = path->nodes[0]; | |
1597 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1598 | struct btrfs_dev_extent); | |
e17cade2 CM |
1599 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1600 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1601 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1602 | ||
0b246afa | 1603 | write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid); |
e17cade2 | 1604 | |
0b86a832 CM |
1605 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1606 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1607 | out: |
0b86a832 CM |
1608 | btrfs_free_path(path); |
1609 | return ret; | |
1610 | } | |
1611 | ||
6df9a95e | 1612 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1613 | { |
6df9a95e JB |
1614 | struct extent_map_tree *em_tree; |
1615 | struct extent_map *em; | |
1616 | struct rb_node *n; | |
1617 | u64 ret = 0; | |
0b86a832 | 1618 | |
6df9a95e JB |
1619 | em_tree = &fs_info->mapping_tree.map_tree; |
1620 | read_lock(&em_tree->lock); | |
1621 | n = rb_last(&em_tree->map); | |
1622 | if (n) { | |
1623 | em = rb_entry(n, struct extent_map, rb_node); | |
1624 | ret = em->start + em->len; | |
0b86a832 | 1625 | } |
6df9a95e JB |
1626 | read_unlock(&em_tree->lock); |
1627 | ||
0b86a832 CM |
1628 | return ret; |
1629 | } | |
1630 | ||
53f10659 ID |
1631 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1632 | u64 *devid_ret) | |
0b86a832 CM |
1633 | { |
1634 | int ret; | |
1635 | struct btrfs_key key; | |
1636 | struct btrfs_key found_key; | |
2b82032c YZ |
1637 | struct btrfs_path *path; |
1638 | ||
2b82032c YZ |
1639 | path = btrfs_alloc_path(); |
1640 | if (!path) | |
1641 | return -ENOMEM; | |
0b86a832 CM |
1642 | |
1643 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1644 | key.type = BTRFS_DEV_ITEM_KEY; | |
1645 | key.offset = (u64)-1; | |
1646 | ||
53f10659 | 1647 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1648 | if (ret < 0) |
1649 | goto error; | |
1650 | ||
79787eaa | 1651 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1652 | |
53f10659 ID |
1653 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1654 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1655 | BTRFS_DEV_ITEM_KEY); |
1656 | if (ret) { | |
53f10659 | 1657 | *devid_ret = 1; |
0b86a832 CM |
1658 | } else { |
1659 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1660 | path->slots[0]); | |
53f10659 | 1661 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1662 | } |
1663 | ret = 0; | |
1664 | error: | |
2b82032c | 1665 | btrfs_free_path(path); |
0b86a832 CM |
1666 | return ret; |
1667 | } | |
1668 | ||
1669 | /* | |
1670 | * the device information is stored in the chunk root | |
1671 | * the btrfs_device struct should be fully filled in | |
1672 | */ | |
48a3b636 | 1673 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
5b4aacef | 1674 | struct btrfs_fs_info *fs_info, |
48a3b636 | 1675 | struct btrfs_device *device) |
0b86a832 | 1676 | { |
5b4aacef | 1677 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
1678 | int ret; |
1679 | struct btrfs_path *path; | |
1680 | struct btrfs_dev_item *dev_item; | |
1681 | struct extent_buffer *leaf; | |
1682 | struct btrfs_key key; | |
1683 | unsigned long ptr; | |
0b86a832 | 1684 | |
0b86a832 CM |
1685 | path = btrfs_alloc_path(); |
1686 | if (!path) | |
1687 | return -ENOMEM; | |
1688 | ||
0b86a832 CM |
1689 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1690 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1691 | key.offset = device->devid; |
0b86a832 CM |
1692 | |
1693 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1694 | sizeof(*dev_item)); |
0b86a832 CM |
1695 | if (ret) |
1696 | goto out; | |
1697 | ||
1698 | leaf = path->nodes[0]; | |
1699 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1700 | ||
1701 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1702 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1703 | btrfs_set_device_type(leaf, dev_item, device->type); |
1704 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1705 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1706 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1707 | btrfs_set_device_total_bytes(leaf, dev_item, |
1708 | btrfs_device_get_disk_total_bytes(device)); | |
1709 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1710 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1711 | btrfs_set_device_group(leaf, dev_item, 0); |
1712 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1713 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1714 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1715 | |
410ba3a2 | 1716 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1717 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1718 | ptr = btrfs_device_fsid(dev_item); |
0b246afa | 1719 | write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1720 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1721 | |
2b82032c | 1722 | ret = 0; |
0b86a832 CM |
1723 | out: |
1724 | btrfs_free_path(path); | |
1725 | return ret; | |
1726 | } | |
8f18cf13 | 1727 | |
5a1972bd QW |
1728 | /* |
1729 | * Function to update ctime/mtime for a given device path. | |
1730 | * Mainly used for ctime/mtime based probe like libblkid. | |
1731 | */ | |
da353f6b | 1732 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1733 | { |
1734 | struct file *filp; | |
1735 | ||
1736 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1737 | if (IS_ERR(filp)) |
5a1972bd QW |
1738 | return; |
1739 | file_update_time(filp); | |
1740 | filp_close(filp, NULL); | |
5a1972bd QW |
1741 | } |
1742 | ||
5b4aacef | 1743 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1744 | struct btrfs_device *device) |
1745 | { | |
5b4aacef | 1746 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1747 | int ret; |
1748 | struct btrfs_path *path; | |
a061fc8d | 1749 | struct btrfs_key key; |
a061fc8d CM |
1750 | struct btrfs_trans_handle *trans; |
1751 | ||
a061fc8d CM |
1752 | path = btrfs_alloc_path(); |
1753 | if (!path) | |
1754 | return -ENOMEM; | |
1755 | ||
a22285a6 | 1756 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1757 | if (IS_ERR(trans)) { |
1758 | btrfs_free_path(path); | |
1759 | return PTR_ERR(trans); | |
1760 | } | |
a061fc8d CM |
1761 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1762 | key.type = BTRFS_DEV_ITEM_KEY; | |
1763 | key.offset = device->devid; | |
1764 | ||
1765 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1766 | if (ret < 0) | |
1767 | goto out; | |
1768 | ||
1769 | if (ret > 0) { | |
1770 | ret = -ENOENT; | |
1771 | goto out; | |
1772 | } | |
1773 | ||
1774 | ret = btrfs_del_item(trans, root, path); | |
1775 | if (ret) | |
1776 | goto out; | |
a061fc8d CM |
1777 | out: |
1778 | btrfs_free_path(path); | |
3a45bb20 | 1779 | btrfs_commit_transaction(trans); |
a061fc8d CM |
1780 | return ret; |
1781 | } | |
1782 | ||
3cc31a0d DS |
1783 | /* |
1784 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1785 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1786 | * replace. | |
1787 | */ | |
1788 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1789 | u64 num_devices) | |
a061fc8d | 1790 | { |
a061fc8d | 1791 | u64 all_avail; |
de98ced9 | 1792 | unsigned seq; |
418775a2 | 1793 | int i; |
a061fc8d | 1794 | |
de98ced9 | 1795 | do { |
bd45ffbc | 1796 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1797 | |
bd45ffbc AJ |
1798 | all_avail = fs_info->avail_data_alloc_bits | |
1799 | fs_info->avail_system_alloc_bits | | |
1800 | fs_info->avail_metadata_alloc_bits; | |
1801 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1802 | |
418775a2 DS |
1803 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1804 | if (!(all_avail & btrfs_raid_group[i])) | |
1805 | continue; | |
a061fc8d | 1806 | |
418775a2 DS |
1807 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1808 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1809 | |
418775a2 DS |
1810 | if (ret) |
1811 | return ret; | |
1812 | } | |
53b381b3 DW |
1813 | } |
1814 | ||
bd45ffbc | 1815 | return 0; |
f1fa7f26 AJ |
1816 | } |
1817 | ||
88acff64 AJ |
1818 | struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs, |
1819 | struct btrfs_device *device) | |
a061fc8d | 1820 | { |
2b82032c | 1821 | struct btrfs_device *next_device; |
88acff64 AJ |
1822 | |
1823 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1824 | if (next_device != device && | |
1825 | !next_device->missing && next_device->bdev) | |
1826 | return next_device; | |
1827 | } | |
1828 | ||
1829 | return NULL; | |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1834 | * and replace it with the provided or the next active device, in the context | |
1835 | * where this function called, there should be always be another device (or | |
1836 | * this_dev) which is active. | |
1837 | */ | |
1838 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1839 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1840 | { | |
1841 | struct btrfs_device *next_device; | |
1842 | ||
1843 | if (this_dev) | |
1844 | next_device = this_dev; | |
1845 | else | |
1846 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1847 | device); | |
1848 | ASSERT(next_device); | |
1849 | ||
1850 | if (fs_info->sb->s_bdev && | |
1851 | (fs_info->sb->s_bdev == device->bdev)) | |
1852 | fs_info->sb->s_bdev = next_device->bdev; | |
1853 | ||
1854 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1855 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1856 | } | |
1857 | ||
da353f6b DS |
1858 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1859 | u64 devid) | |
f1fa7f26 AJ |
1860 | { |
1861 | struct btrfs_device *device; | |
1f78160c | 1862 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1863 | u64 num_devices; |
a061fc8d | 1864 | int ret = 0; |
1f78160c | 1865 | bool clear_super = false; |
a061fc8d | 1866 | |
a061fc8d CM |
1867 | mutex_lock(&uuid_mutex); |
1868 | ||
0b246afa JM |
1869 | num_devices = fs_info->fs_devices->num_devices; |
1870 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
1871 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
8dabb742 SB |
1872 | WARN_ON(num_devices < 1); |
1873 | num_devices--; | |
1874 | } | |
0b246afa | 1875 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
8dabb742 | 1876 | |
0b246afa | 1877 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1878 | if (ret) |
a061fc8d | 1879 | goto out; |
a061fc8d | 1880 | |
2ff7e61e JM |
1881 | ret = btrfs_find_device_by_devspec(fs_info, devid, device_path, |
1882 | &device); | |
24fc572f | 1883 | if (ret) |
53b381b3 | 1884 | goto out; |
dfe25020 | 1885 | |
63a212ab | 1886 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1887 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1888 | goto out; |
63a212ab SB |
1889 | } |
1890 | ||
0b246afa | 1891 | if (device->writeable && fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1892 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1893 | goto out; |
2b82032c YZ |
1894 | } |
1895 | ||
1896 | if (device->writeable) { | |
34441361 | 1897 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1898 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1899 | device->fs_devices->rw_devices--; |
34441361 | 1900 | mutex_unlock(&fs_info->chunk_mutex); |
1f78160c | 1901 | clear_super = true; |
dfe25020 | 1902 | } |
a061fc8d | 1903 | |
d7901554 | 1904 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1905 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1906 | mutex_lock(&uuid_mutex); |
a061fc8d | 1907 | if (ret) |
9b3517e9 | 1908 | goto error_undo; |
a061fc8d | 1909 | |
63a212ab SB |
1910 | /* |
1911 | * TODO: the superblock still includes this device in its num_devices | |
1912 | * counter although write_all_supers() is not locked out. This | |
1913 | * could give a filesystem state which requires a degraded mount. | |
1914 | */ | |
0b246afa | 1915 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1916 | if (ret) |
9b3517e9 | 1917 | goto error_undo; |
a061fc8d | 1918 | |
2b82032c | 1919 | device->in_fs_metadata = 0; |
0b246afa | 1920 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1921 | |
1922 | /* | |
1923 | * the device list mutex makes sure that we don't change | |
1924 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1925 | * the device supers. Whoever is writing all supers, should |
1926 | * lock the device list mutex before getting the number of | |
1927 | * devices in the super block (super_copy). Conversely, | |
1928 | * whoever updates the number of devices in the super block | |
1929 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1930 | */ |
1f78160c XG |
1931 | |
1932 | cur_devices = device->fs_devices; | |
0b246afa | 1933 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c | 1934 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1935 | |
e4404d6e | 1936 | device->fs_devices->num_devices--; |
02db0844 | 1937 | device->fs_devices->total_devices--; |
2b82032c | 1938 | |
cd02dca5 | 1939 | if (device->missing) |
3a7d55c8 | 1940 | device->fs_devices->missing_devices--; |
cd02dca5 | 1941 | |
0b246afa | 1942 | btrfs_assign_next_active_device(fs_info, device, NULL); |
2b82032c | 1943 | |
0bfaa9c5 | 1944 | if (device->bdev) { |
e4404d6e | 1945 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1946 | /* remove sysfs entry */ |
0b246afa | 1947 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
0bfaa9c5 | 1948 | } |
99994cde | 1949 | |
0b246afa JM |
1950 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
1951 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
1952 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2b82032c | 1953 | |
cea67ab9 JM |
1954 | /* |
1955 | * at this point, the device is zero sized and detached from | |
1956 | * the devices list. All that's left is to zero out the old | |
1957 | * supers and free the device. | |
1958 | */ | |
1959 | if (device->writeable) | |
1960 | btrfs_scratch_superblocks(device->bdev, device->name->str); | |
1961 | ||
1962 | btrfs_close_bdev(device); | |
1963 | call_rcu(&device->rcu, free_device); | |
1964 | ||
1f78160c | 1965 | if (cur_devices->open_devices == 0) { |
e4404d6e | 1966 | struct btrfs_fs_devices *fs_devices; |
0b246afa | 1967 | fs_devices = fs_info->fs_devices; |
e4404d6e | 1968 | while (fs_devices) { |
8321cf25 RS |
1969 | if (fs_devices->seed == cur_devices) { |
1970 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1971 | break; |
8321cf25 | 1972 | } |
e4404d6e | 1973 | fs_devices = fs_devices->seed; |
2b82032c | 1974 | } |
1f78160c | 1975 | cur_devices->seed = NULL; |
1f78160c | 1976 | __btrfs_close_devices(cur_devices); |
1f78160c | 1977 | free_fs_devices(cur_devices); |
2b82032c YZ |
1978 | } |
1979 | ||
0b246afa JM |
1980 | fs_info->num_tolerated_disk_barrier_failures = |
1981 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
5af3e8cc | 1982 | |
a061fc8d CM |
1983 | out: |
1984 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1985 | return ret; |
24fc572f | 1986 | |
9b3517e9 ID |
1987 | error_undo: |
1988 | if (device->writeable) { | |
34441361 | 1989 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 1990 | list_add(&device->dev_alloc_list, |
0b246afa | 1991 | &fs_info->fs_devices->alloc_list); |
c3929c36 | 1992 | device->fs_devices->rw_devices++; |
34441361 | 1993 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 1994 | } |
24fc572f | 1995 | goto out; |
a061fc8d CM |
1996 | } |
1997 | ||
084b6e7c QW |
1998 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1999 | struct btrfs_device *srcdev) | |
e93c89c1 | 2000 | { |
d51908ce AJ |
2001 | struct btrfs_fs_devices *fs_devices; |
2002 | ||
e93c89c1 | 2003 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 2004 | |
25e8e911 AJ |
2005 | /* |
2006 | * in case of fs with no seed, srcdev->fs_devices will point | |
2007 | * to fs_devices of fs_info. However when the dev being replaced is | |
2008 | * a seed dev it will point to the seed's local fs_devices. In short | |
2009 | * srcdev will have its correct fs_devices in both the cases. | |
2010 | */ | |
2011 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2012 | |
e93c89c1 SB |
2013 | list_del_rcu(&srcdev->dev_list); |
2014 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 2015 | fs_devices->num_devices--; |
82372bc8 | 2016 | if (srcdev->missing) |
d51908ce | 2017 | fs_devices->missing_devices--; |
e93c89c1 | 2018 | |
48b3b9d4 | 2019 | if (srcdev->writeable) |
82372bc8 | 2020 | fs_devices->rw_devices--; |
1357272f | 2021 | |
82372bc8 | 2022 | if (srcdev->bdev) |
d51908ce | 2023 | fs_devices->open_devices--; |
084b6e7c QW |
2024 | } |
2025 | ||
2026 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2027 | struct btrfs_device *srcdev) | |
2028 | { | |
2029 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2030 | |
48b3b9d4 AJ |
2031 | if (srcdev->writeable) { |
2032 | /* zero out the old super if it is writable */ | |
2033 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2034 | } | |
14238819 AJ |
2035 | |
2036 | btrfs_close_bdev(srcdev); | |
2037 | ||
e93c89c1 | 2038 | call_rcu(&srcdev->rcu, free_device); |
94d5f0c2 AJ |
2039 | |
2040 | /* | |
2041 | * unless fs_devices is seed fs, num_devices shouldn't go | |
2042 | * zero | |
2043 | */ | |
2044 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
2045 | ||
2046 | /* if this is no devs we rather delete the fs_devices */ | |
2047 | if (!fs_devices->num_devices) { | |
2048 | struct btrfs_fs_devices *tmp_fs_devices; | |
2049 | ||
2050 | tmp_fs_devices = fs_info->fs_devices; | |
2051 | while (tmp_fs_devices) { | |
2052 | if (tmp_fs_devices->seed == fs_devices) { | |
2053 | tmp_fs_devices->seed = fs_devices->seed; | |
2054 | break; | |
2055 | } | |
2056 | tmp_fs_devices = tmp_fs_devices->seed; | |
2057 | } | |
2058 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2059 | __btrfs_close_devices(fs_devices); |
2060 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2061 | } |
e93c89c1 SB |
2062 | } |
2063 | ||
2064 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2065 | struct btrfs_device *tgtdev) | |
2066 | { | |
67a2c45e | 2067 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2068 | WARN_ON(!tgtdev); |
2069 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2070 | |
32576040 | 2071 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2072 | |
779bf3fe | 2073 | if (tgtdev->bdev) |
e93c89c1 | 2074 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2075 | |
e93c89c1 | 2076 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2077 | |
88acff64 | 2078 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2079 | |
e93c89c1 | 2080 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2081 | |
2082 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2083 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2084 | |
2085 | /* | |
2086 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2087 | * may lead to a call to btrfs_show_devname() which will try | |
2088 | * to hold device_list_mutex. And here this device | |
2089 | * is already out of device list, so we don't have to hold | |
2090 | * the device_list_mutex lock. | |
2091 | */ | |
2092 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2093 | |
2094 | btrfs_close_bdev(tgtdev); | |
779bf3fe | 2095 | call_rcu(&tgtdev->rcu, free_device); |
e93c89c1 SB |
2096 | } |
2097 | ||
2ff7e61e | 2098 | static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2099 | const char *device_path, |
48a3b636 | 2100 | struct btrfs_device **device) |
7ba15b7d SB |
2101 | { |
2102 | int ret = 0; | |
2103 | struct btrfs_super_block *disk_super; | |
2104 | u64 devid; | |
2105 | u8 *dev_uuid; | |
2106 | struct block_device *bdev; | |
2107 | struct buffer_head *bh; | |
2108 | ||
2109 | *device = NULL; | |
2110 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2111 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2112 | if (ret) |
2113 | return ret; | |
2114 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2115 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2116 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2117 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2118 | brelse(bh); |
2119 | if (!*device) | |
2120 | ret = -ENOENT; | |
2121 | blkdev_put(bdev, FMODE_READ); | |
2122 | return ret; | |
2123 | } | |
2124 | ||
2ff7e61e | 2125 | int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2126 | const char *device_path, |
7ba15b7d SB |
2127 | struct btrfs_device **device) |
2128 | { | |
2129 | *device = NULL; | |
2130 | if (strcmp(device_path, "missing") == 0) { | |
2131 | struct list_head *devices; | |
2132 | struct btrfs_device *tmp; | |
2133 | ||
0b246afa | 2134 | devices = &fs_info->fs_devices->devices; |
7ba15b7d SB |
2135 | /* |
2136 | * It is safe to read the devices since the volume_mutex | |
2137 | * is held by the caller. | |
2138 | */ | |
2139 | list_for_each_entry(tmp, devices, dev_list) { | |
2140 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2141 | *device = tmp; | |
2142 | break; | |
2143 | } | |
2144 | } | |
2145 | ||
d74a6259 AJ |
2146 | if (!*device) |
2147 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2148 | |
2149 | return 0; | |
2150 | } else { | |
2ff7e61e | 2151 | return btrfs_find_device_by_path(fs_info, device_path, device); |
7ba15b7d SB |
2152 | } |
2153 | } | |
2154 | ||
5c5c0df0 DS |
2155 | /* |
2156 | * Lookup a device given by device id, or the path if the id is 0. | |
2157 | */ | |
2ff7e61e | 2158 | int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid, |
da353f6b DS |
2159 | const char *devpath, |
2160 | struct btrfs_device **device) | |
24e0474b AJ |
2161 | { |
2162 | int ret; | |
2163 | ||
5c5c0df0 | 2164 | if (devid) { |
24e0474b | 2165 | ret = 0; |
0b246afa | 2166 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2167 | if (!*device) |
2168 | ret = -ENOENT; | |
2169 | } else { | |
5c5c0df0 | 2170 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2171 | return -EINVAL; |
2172 | ||
2ff7e61e | 2173 | ret = btrfs_find_device_missing_or_by_path(fs_info, devpath, |
24e0474b AJ |
2174 | device); |
2175 | } | |
2176 | return ret; | |
2177 | } | |
2178 | ||
2b82032c YZ |
2179 | /* |
2180 | * does all the dirty work required for changing file system's UUID. | |
2181 | */ | |
2ff7e61e | 2182 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2183 | { |
0b246afa | 2184 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2185 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2186 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2187 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2188 | struct btrfs_device *device; |
2189 | u64 super_flags; | |
2190 | ||
2191 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2192 | if (!fs_devices->seeding) |
2b82032c YZ |
2193 | return -EINVAL; |
2194 | ||
2208a378 ID |
2195 | seed_devices = __alloc_fs_devices(); |
2196 | if (IS_ERR(seed_devices)) | |
2197 | return PTR_ERR(seed_devices); | |
2b82032c | 2198 | |
e4404d6e YZ |
2199 | old_devices = clone_fs_devices(fs_devices); |
2200 | if (IS_ERR(old_devices)) { | |
2201 | kfree(seed_devices); | |
2202 | return PTR_ERR(old_devices); | |
2b82032c | 2203 | } |
e4404d6e | 2204 | |
2b82032c YZ |
2205 | list_add(&old_devices->list, &fs_uuids); |
2206 | ||
e4404d6e YZ |
2207 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2208 | seed_devices->opened = 1; | |
2209 | INIT_LIST_HEAD(&seed_devices->devices); | |
2210 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2211 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2212 | |
0b246afa | 2213 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c XG |
2214 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2215 | synchronize_rcu); | |
2196d6e8 MX |
2216 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2217 | device->fs_devices = seed_devices; | |
c9513edb | 2218 | |
34441361 | 2219 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2220 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2221 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2222 | |
2b82032c YZ |
2223 | fs_devices->seeding = 0; |
2224 | fs_devices->num_devices = 0; | |
2225 | fs_devices->open_devices = 0; | |
69611ac8 | 2226 | fs_devices->missing_devices = 0; |
69611ac8 | 2227 | fs_devices->rotating = 0; |
e4404d6e | 2228 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2229 | |
2230 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2231 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2232 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
0b246afa | 2233 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
f7171750 | 2234 | |
2b82032c YZ |
2235 | super_flags = btrfs_super_flags(disk_super) & |
2236 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2237 | btrfs_set_super_flags(disk_super, super_flags); | |
2238 | ||
2239 | return 0; | |
2240 | } | |
2241 | ||
2242 | /* | |
01327610 | 2243 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2244 | */ |
2245 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2246 | struct btrfs_fs_info *fs_info) |
2b82032c | 2247 | { |
5b4aacef | 2248 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2249 | struct btrfs_path *path; |
2250 | struct extent_buffer *leaf; | |
2251 | struct btrfs_dev_item *dev_item; | |
2252 | struct btrfs_device *device; | |
2253 | struct btrfs_key key; | |
2254 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2255 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2256 | u64 devid; | |
2257 | int ret; | |
2258 | ||
2259 | path = btrfs_alloc_path(); | |
2260 | if (!path) | |
2261 | return -ENOMEM; | |
2262 | ||
2b82032c YZ |
2263 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2264 | key.offset = 0; | |
2265 | key.type = BTRFS_DEV_ITEM_KEY; | |
2266 | ||
2267 | while (1) { | |
2268 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2269 | if (ret < 0) | |
2270 | goto error; | |
2271 | ||
2272 | leaf = path->nodes[0]; | |
2273 | next_slot: | |
2274 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2275 | ret = btrfs_next_leaf(root, path); | |
2276 | if (ret > 0) | |
2277 | break; | |
2278 | if (ret < 0) | |
2279 | goto error; | |
2280 | leaf = path->nodes[0]; | |
2281 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2282 | btrfs_release_path(path); |
2b82032c YZ |
2283 | continue; |
2284 | } | |
2285 | ||
2286 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2287 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2288 | key.type != BTRFS_DEV_ITEM_KEY) | |
2289 | break; | |
2290 | ||
2291 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2292 | struct btrfs_dev_item); | |
2293 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2294 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2295 | BTRFS_UUID_SIZE); |
1473b24e | 2296 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2297 | BTRFS_UUID_SIZE); |
0b246afa | 2298 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2299 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2300 | |
2301 | if (device->fs_devices->seeding) { | |
2302 | btrfs_set_device_generation(leaf, dev_item, | |
2303 | device->generation); | |
2304 | btrfs_mark_buffer_dirty(leaf); | |
2305 | } | |
2306 | ||
2307 | path->slots[0]++; | |
2308 | goto next_slot; | |
2309 | } | |
2310 | ret = 0; | |
2311 | error: | |
2312 | btrfs_free_path(path); | |
2313 | return ret; | |
2314 | } | |
2315 | ||
da353f6b | 2316 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2317 | { |
5112febb | 2318 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2319 | struct request_queue *q; |
788f20eb CM |
2320 | struct btrfs_trans_handle *trans; |
2321 | struct btrfs_device *device; | |
2322 | struct block_device *bdev; | |
788f20eb | 2323 | struct list_head *devices; |
0b246afa | 2324 | struct super_block *sb = fs_info->sb; |
606686ee | 2325 | struct rcu_string *name; |
3c1dbdf5 | 2326 | u64 tmp; |
2b82032c | 2327 | int seeding_dev = 0; |
788f20eb CM |
2328 | int ret = 0; |
2329 | ||
0b246afa | 2330 | if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding) |
f8c5d0b4 | 2331 | return -EROFS; |
788f20eb | 2332 | |
a5d16333 | 2333 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2334 | fs_info->bdev_holder); |
7f59203a JB |
2335 | if (IS_ERR(bdev)) |
2336 | return PTR_ERR(bdev); | |
a2135011 | 2337 | |
0b246afa | 2338 | if (fs_info->fs_devices->seeding) { |
2b82032c YZ |
2339 | seeding_dev = 1; |
2340 | down_write(&sb->s_umount); | |
2341 | mutex_lock(&uuid_mutex); | |
2342 | } | |
2343 | ||
8c8bee1d | 2344 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2345 | |
0b246afa | 2346 | devices = &fs_info->fs_devices->devices; |
d25628bd | 2347 | |
0b246afa | 2348 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
c6e30871 | 2349 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2350 | if (device->bdev == bdev) { |
2351 | ret = -EEXIST; | |
d25628bd | 2352 | mutex_unlock( |
0b246afa | 2353 | &fs_info->fs_devices->device_list_mutex); |
2b82032c | 2354 | goto error; |
788f20eb CM |
2355 | } |
2356 | } | |
0b246afa | 2357 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2358 | |
0b246afa | 2359 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2360 | if (IS_ERR(device)) { |
788f20eb | 2361 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2362 | ret = PTR_ERR(device); |
2b82032c | 2363 | goto error; |
788f20eb CM |
2364 | } |
2365 | ||
78f2c9e6 | 2366 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2367 | if (!name) { |
788f20eb | 2368 | kfree(device); |
2b82032c YZ |
2369 | ret = -ENOMEM; |
2370 | goto error; | |
788f20eb | 2371 | } |
606686ee | 2372 | rcu_assign_pointer(device->name, name); |
2b82032c | 2373 | |
a22285a6 | 2374 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2375 | if (IS_ERR(trans)) { |
606686ee | 2376 | rcu_string_free(device->name); |
98d5dc13 TI |
2377 | kfree(device); |
2378 | ret = PTR_ERR(trans); | |
2379 | goto error; | |
2380 | } | |
2381 | ||
d5e2003c JB |
2382 | q = bdev_get_queue(bdev); |
2383 | if (blk_queue_discard(q)) | |
2384 | device->can_discard = 1; | |
2b82032c | 2385 | device->writeable = 1; |
2b82032c | 2386 | device->generation = trans->transid; |
0b246afa JM |
2387 | device->io_width = fs_info->sectorsize; |
2388 | device->io_align = fs_info->sectorsize; | |
2389 | device->sector_size = fs_info->sectorsize; | |
788f20eb | 2390 | device->total_bytes = i_size_read(bdev->bd_inode); |
2cc3c559 | 2391 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2392 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2393 | device->fs_info = fs_info; |
788f20eb | 2394 | device->bdev = bdev; |
dfe25020 | 2395 | device->in_fs_metadata = 1; |
63a212ab | 2396 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2397 | device->mode = FMODE_EXCL; |
27087f37 | 2398 | device->dev_stats_valid = 1; |
2b82032c | 2399 | set_blocksize(device->bdev, 4096); |
788f20eb | 2400 | |
2b82032c YZ |
2401 | if (seeding_dev) { |
2402 | sb->s_flags &= ~MS_RDONLY; | |
2ff7e61e | 2403 | ret = btrfs_prepare_sprout(fs_info); |
79787eaa | 2404 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2405 | } |
788f20eb | 2406 | |
0b246afa | 2407 | device->fs_devices = fs_info->fs_devices; |
e5e9a520 | 2408 | |
0b246afa | 2409 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
34441361 | 2410 | mutex_lock(&fs_info->chunk_mutex); |
0b246afa | 2411 | list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices); |
2b82032c | 2412 | list_add(&device->dev_alloc_list, |
0b246afa JM |
2413 | &fs_info->fs_devices->alloc_list); |
2414 | fs_info->fs_devices->num_devices++; | |
2415 | fs_info->fs_devices->open_devices++; | |
2416 | fs_info->fs_devices->rw_devices++; | |
2417 | fs_info->fs_devices->total_devices++; | |
2418 | fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2419 | |
a5ed45f8 | 2420 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2421 | |
e884f4f0 | 2422 | if (!blk_queue_nonrot(q)) |
0b246afa | 2423 | fs_info->fs_devices->rotating = 1; |
c289811c | 2424 | |
0b246afa JM |
2425 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2426 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
3c1dbdf5 | 2427 | tmp + device->total_bytes); |
788f20eb | 2428 | |
0b246afa JM |
2429 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2430 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2431 | |
2432 | /* add sysfs device entry */ | |
0b246afa | 2433 | btrfs_sysfs_add_device_link(fs_info->fs_devices, device); |
0d39376a | 2434 | |
2196d6e8 MX |
2435 | /* |
2436 | * we've got more storage, clear any full flags on the space | |
2437 | * infos | |
2438 | */ | |
0b246afa | 2439 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2440 | |
34441361 | 2441 | mutex_unlock(&fs_info->chunk_mutex); |
0b246afa | 2442 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2443 | |
2b82032c | 2444 | if (seeding_dev) { |
34441361 | 2445 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2446 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2447 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2448 | if (ret) { |
66642832 | 2449 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2450 | goto error_trans; |
005d6427 | 2451 | } |
2196d6e8 MX |
2452 | } |
2453 | ||
0b246afa | 2454 | ret = btrfs_add_device(trans, fs_info, device); |
2196d6e8 | 2455 | if (ret) { |
66642832 | 2456 | btrfs_abort_transaction(trans, ret); |
2196d6e8 MX |
2457 | goto error_trans; |
2458 | } | |
2459 | ||
2460 | if (seeding_dev) { | |
2461 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2462 | ||
0b246afa | 2463 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2464 | if (ret) { |
66642832 | 2465 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2466 | goto error_trans; |
005d6427 | 2467 | } |
b2373f25 AJ |
2468 | |
2469 | /* Sprouting would change fsid of the mounted root, | |
2470 | * so rename the fsid on the sysfs | |
2471 | */ | |
2472 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa JM |
2473 | fs_info->fsid); |
2474 | if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf)) | |
2475 | btrfs_warn(fs_info, | |
2476 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2477 | } |
2478 | ||
0b246afa JM |
2479 | fs_info->num_tolerated_disk_barrier_failures = |
2480 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3a45bb20 | 2481 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2482 | |
2b82032c YZ |
2483 | if (seeding_dev) { |
2484 | mutex_unlock(&uuid_mutex); | |
2485 | up_write(&sb->s_umount); | |
788f20eb | 2486 | |
79787eaa JM |
2487 | if (ret) /* transaction commit */ |
2488 | return ret; | |
2489 | ||
2ff7e61e | 2490 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2491 | if (ret < 0) |
0b246afa | 2492 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2493 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2494 | trans = btrfs_attach_transaction(root); |
2495 | if (IS_ERR(trans)) { | |
2496 | if (PTR_ERR(trans) == -ENOENT) | |
2497 | return 0; | |
2498 | return PTR_ERR(trans); | |
2499 | } | |
3a45bb20 | 2500 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2501 | } |
c9e9f97b | 2502 | |
5a1972bd QW |
2503 | /* Update ctime/mtime for libblkid */ |
2504 | update_dev_time(device_path); | |
2b82032c | 2505 | return ret; |
79787eaa JM |
2506 | |
2507 | error_trans: | |
3a45bb20 | 2508 | btrfs_end_transaction(trans); |
606686ee | 2509 | rcu_string_free(device->name); |
0b246afa | 2510 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
79787eaa | 2511 | kfree(device); |
2b82032c | 2512 | error: |
e525fd89 | 2513 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2514 | if (seeding_dev) { |
2515 | mutex_unlock(&uuid_mutex); | |
2516 | up_write(&sb->s_umount); | |
2517 | } | |
c9e9f97b | 2518 | return ret; |
788f20eb CM |
2519 | } |
2520 | ||
2ff7e61e | 2521 | int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, |
da353f6b | 2522 | const char *device_path, |
1c43366d | 2523 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2524 | struct btrfs_device **device_out) |
2525 | { | |
2526 | struct request_queue *q; | |
2527 | struct btrfs_device *device; | |
2528 | struct block_device *bdev; | |
e93c89c1 SB |
2529 | struct list_head *devices; |
2530 | struct rcu_string *name; | |
12bd2fc0 | 2531 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2532 | int ret = 0; |
2533 | ||
2534 | *device_out = NULL; | |
1c43366d MX |
2535 | if (fs_info->fs_devices->seeding) { |
2536 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2537 | return -EINVAL; |
1c43366d | 2538 | } |
e93c89c1 SB |
2539 | |
2540 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2541 | fs_info->bdev_holder); | |
1c43366d MX |
2542 | if (IS_ERR(bdev)) { |
2543 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2544 | return PTR_ERR(bdev); |
1c43366d | 2545 | } |
e93c89c1 SB |
2546 | |
2547 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2548 | ||
2549 | devices = &fs_info->fs_devices->devices; | |
2550 | list_for_each_entry(device, devices, dev_list) { | |
2551 | if (device->bdev == bdev) { | |
5d163e0e JM |
2552 | btrfs_err(fs_info, |
2553 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2554 | ret = -EEXIST; |
2555 | goto error; | |
2556 | } | |
2557 | } | |
2558 | ||
1c43366d | 2559 | |
7cc8e58d MX |
2560 | if (i_size_read(bdev->bd_inode) < |
2561 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2562 | btrfs_err(fs_info, |
2563 | "target device is smaller than source device!"); | |
1c43366d MX |
2564 | ret = -EINVAL; |
2565 | goto error; | |
2566 | } | |
2567 | ||
2568 | ||
12bd2fc0 ID |
2569 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2570 | if (IS_ERR(device)) { | |
2571 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2572 | goto error; |
2573 | } | |
2574 | ||
2575 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2576 | if (!name) { | |
2577 | kfree(device); | |
2578 | ret = -ENOMEM; | |
2579 | goto error; | |
2580 | } | |
2581 | rcu_assign_pointer(device->name, name); | |
2582 | ||
2583 | q = bdev_get_queue(bdev); | |
2584 | if (blk_queue_discard(q)) | |
2585 | device->can_discard = 1; | |
0b246afa | 2586 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 | 2587 | device->writeable = 1; |
e93c89c1 | 2588 | device->generation = 0; |
0b246afa JM |
2589 | device->io_width = fs_info->sectorsize; |
2590 | device->io_align = fs_info->sectorsize; | |
2591 | device->sector_size = fs_info->sectorsize; | |
7cc8e58d MX |
2592 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2593 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2594 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2595 | ASSERT(list_empty(&srcdev->resized_list)); |
2596 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2597 | device->commit_bytes_used = device->bytes_used; |
fb456252 | 2598 | device->fs_info = fs_info; |
e93c89c1 SB |
2599 | device->bdev = bdev; |
2600 | device->in_fs_metadata = 1; | |
2601 | device->is_tgtdev_for_dev_replace = 1; | |
2602 | device->mode = FMODE_EXCL; | |
27087f37 | 2603 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2604 | set_blocksize(device->bdev, 4096); |
2605 | device->fs_devices = fs_info->fs_devices; | |
2606 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2607 | fs_info->fs_devices->num_devices++; | |
2608 | fs_info->fs_devices->open_devices++; | |
0b246afa | 2609 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 SB |
2610 | |
2611 | *device_out = device; | |
2612 | return ret; | |
2613 | ||
2614 | error: | |
2615 | blkdev_put(bdev, FMODE_EXCL); | |
2616 | return ret; | |
2617 | } | |
2618 | ||
2619 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2620 | struct btrfs_device *tgtdev) | |
2621 | { | |
da17066c JM |
2622 | u32 sectorsize = fs_info->sectorsize; |
2623 | ||
e93c89c1 | 2624 | WARN_ON(fs_info->fs_devices->rw_devices == 0); |
da17066c JM |
2625 | tgtdev->io_width = sectorsize; |
2626 | tgtdev->io_align = sectorsize; | |
2627 | tgtdev->sector_size = sectorsize; | |
fb456252 | 2628 | tgtdev->fs_info = fs_info; |
e93c89c1 SB |
2629 | tgtdev->in_fs_metadata = 1; |
2630 | } | |
2631 | ||
d397712b CM |
2632 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2633 | struct btrfs_device *device) | |
0b86a832 CM |
2634 | { |
2635 | int ret; | |
2636 | struct btrfs_path *path; | |
0b246afa | 2637 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2638 | struct btrfs_dev_item *dev_item; |
2639 | struct extent_buffer *leaf; | |
2640 | struct btrfs_key key; | |
2641 | ||
0b86a832 CM |
2642 | path = btrfs_alloc_path(); |
2643 | if (!path) | |
2644 | return -ENOMEM; | |
2645 | ||
2646 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2647 | key.type = BTRFS_DEV_ITEM_KEY; | |
2648 | key.offset = device->devid; | |
2649 | ||
2650 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2651 | if (ret < 0) | |
2652 | goto out; | |
2653 | ||
2654 | if (ret > 0) { | |
2655 | ret = -ENOENT; | |
2656 | goto out; | |
2657 | } | |
2658 | ||
2659 | leaf = path->nodes[0]; | |
2660 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2661 | ||
2662 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2663 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2664 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2665 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2666 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2667 | btrfs_set_device_total_bytes(leaf, dev_item, |
2668 | btrfs_device_get_disk_total_bytes(device)); | |
2669 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2670 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2671 | btrfs_mark_buffer_dirty(leaf); |
2672 | ||
2673 | out: | |
2674 | btrfs_free_path(path); | |
2675 | return ret; | |
2676 | } | |
2677 | ||
2196d6e8 | 2678 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2679 | struct btrfs_device *device, u64 new_size) |
2680 | { | |
0b246afa JM |
2681 | struct btrfs_fs_info *fs_info = device->fs_info; |
2682 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2683 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2684 | u64 old_total; |
2685 | u64 diff; | |
8f18cf13 | 2686 | |
2b82032c YZ |
2687 | if (!device->writeable) |
2688 | return -EACCES; | |
2196d6e8 | 2689 | |
34441361 | 2690 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2691 | old_total = btrfs_super_total_bytes(super_copy); |
2692 | diff = new_size - device->total_bytes; | |
2693 | ||
63a212ab | 2694 | if (new_size <= device->total_bytes || |
2196d6e8 | 2695 | device->is_tgtdev_for_dev_replace) { |
34441361 | 2696 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2697 | return -EINVAL; |
2196d6e8 | 2698 | } |
2b82032c | 2699 | |
0b246afa | 2700 | fs_devices = fs_info->fs_devices; |
2b82032c | 2701 | |
8f18cf13 | 2702 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2703 | device->fs_devices->total_rw_bytes += diff; |
2704 | ||
7cc8e58d MX |
2705 | btrfs_device_set_total_bytes(device, new_size); |
2706 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2707 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2708 | if (list_empty(&device->resized_list)) |
2709 | list_add_tail(&device->resized_list, | |
2710 | &fs_devices->resized_devices); | |
34441361 | 2711 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2712 | |
8f18cf13 CM |
2713 | return btrfs_update_device(trans, device); |
2714 | } | |
2715 | ||
2716 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
5b4aacef | 2717 | struct btrfs_fs_info *fs_info, u64 chunk_objectid, |
8f18cf13 CM |
2718 | u64 chunk_offset) |
2719 | { | |
5b4aacef | 2720 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2721 | int ret; |
2722 | struct btrfs_path *path; | |
2723 | struct btrfs_key key; | |
2724 | ||
8f18cf13 CM |
2725 | path = btrfs_alloc_path(); |
2726 | if (!path) | |
2727 | return -ENOMEM; | |
2728 | ||
2729 | key.objectid = chunk_objectid; | |
2730 | key.offset = chunk_offset; | |
2731 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2732 | ||
2733 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2734 | if (ret < 0) |
2735 | goto out; | |
2736 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2737 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2738 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2739 | ret = -ENOENT; |
2740 | goto out; | |
2741 | } | |
8f18cf13 CM |
2742 | |
2743 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2744 | if (ret < 0) |
0b246afa JM |
2745 | btrfs_handle_fs_error(fs_info, ret, |
2746 | "Failed to delete chunk item."); | |
79787eaa | 2747 | out: |
8f18cf13 | 2748 | btrfs_free_path(path); |
65a246c5 | 2749 | return ret; |
8f18cf13 CM |
2750 | } |
2751 | ||
6bccf3ab JM |
2752 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, |
2753 | u64 chunk_objectid, u64 chunk_offset) | |
8f18cf13 | 2754 | { |
0b246afa | 2755 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2756 | struct btrfs_disk_key *disk_key; |
2757 | struct btrfs_chunk *chunk; | |
2758 | u8 *ptr; | |
2759 | int ret = 0; | |
2760 | u32 num_stripes; | |
2761 | u32 array_size; | |
2762 | u32 len = 0; | |
2763 | u32 cur; | |
2764 | struct btrfs_key key; | |
2765 | ||
34441361 | 2766 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2767 | array_size = btrfs_super_sys_array_size(super_copy); |
2768 | ||
2769 | ptr = super_copy->sys_chunk_array; | |
2770 | cur = 0; | |
2771 | ||
2772 | while (cur < array_size) { | |
2773 | disk_key = (struct btrfs_disk_key *)ptr; | |
2774 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2775 | ||
2776 | len = sizeof(*disk_key); | |
2777 | ||
2778 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2779 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2780 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2781 | len += btrfs_chunk_item_size(num_stripes); | |
2782 | } else { | |
2783 | ret = -EIO; | |
2784 | break; | |
2785 | } | |
2786 | if (key.objectid == chunk_objectid && | |
2787 | key.offset == chunk_offset) { | |
2788 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2789 | array_size -= len; | |
2790 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2791 | } else { | |
2792 | ptr += len; | |
2793 | cur += len; | |
2794 | } | |
2795 | } | |
34441361 | 2796 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2797 | return ret; |
2798 | } | |
2799 | ||
592d92ee LB |
2800 | static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info, |
2801 | u64 logical, u64 length) | |
2802 | { | |
2803 | struct extent_map_tree *em_tree; | |
2804 | struct extent_map *em; | |
2805 | ||
2806 | em_tree = &fs_info->mapping_tree.map_tree; | |
2807 | read_lock(&em_tree->lock); | |
2808 | em = lookup_extent_mapping(em_tree, logical, length); | |
2809 | read_unlock(&em_tree->lock); | |
2810 | ||
2811 | if (!em) { | |
2812 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2813 | logical, length); | |
2814 | return ERR_PTR(-EINVAL); | |
2815 | } | |
2816 | ||
2817 | if (em->start > logical || em->start + em->len < logical) { | |
2818 | btrfs_crit(fs_info, | |
2819 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2820 | logical, length, em->start, em->start + em->len); | |
2821 | free_extent_map(em); | |
2822 | return ERR_PTR(-EINVAL); | |
2823 | } | |
2824 | ||
2825 | /* callers are responsible for dropping em's ref. */ | |
2826 | return em; | |
2827 | } | |
2828 | ||
47ab2a6c | 2829 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
5b4aacef | 2830 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2831 | { |
8f18cf13 CM |
2832 | struct extent_map *em; |
2833 | struct map_lookup *map; | |
2196d6e8 | 2834 | u64 dev_extent_len = 0; |
47ab2a6c | 2835 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2836 | int i, ret = 0; |
0b246afa | 2837 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2838 | |
592d92ee LB |
2839 | em = get_chunk_map(fs_info, chunk_offset, 1); |
2840 | if (IS_ERR(em)) { | |
47ab2a6c JB |
2841 | /* |
2842 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2843 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2844 | * do anything we still error out. |
2845 | */ | |
2846 | ASSERT(0); | |
592d92ee | 2847 | return PTR_ERR(em); |
47ab2a6c | 2848 | } |
95617d69 | 2849 | map = em->map_lookup; |
34441361 | 2850 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 2851 | check_system_chunk(trans, fs_info, map->type); |
34441361 | 2852 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2853 | |
57ba4cb8 FM |
2854 | /* |
2855 | * Take the device list mutex to prevent races with the final phase of | |
2856 | * a device replace operation that replaces the device object associated | |
2857 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2858 | */ | |
2859 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2860 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2861 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2862 | ret = btrfs_free_dev_extent(trans, device, |
2863 | map->stripes[i].physical, | |
2864 | &dev_extent_len); | |
47ab2a6c | 2865 | if (ret) { |
57ba4cb8 | 2866 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2867 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2868 | goto out; |
2869 | } | |
a061fc8d | 2870 | |
2196d6e8 | 2871 | if (device->bytes_used > 0) { |
34441361 | 2872 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2873 | btrfs_device_set_bytes_used(device, |
2874 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2875 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2876 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2877 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2878 | } |
a061fc8d | 2879 | |
dfe25020 CM |
2880 | if (map->stripes[i].dev) { |
2881 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2882 | if (ret) { |
57ba4cb8 | 2883 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2884 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2885 | goto out; |
2886 | } | |
dfe25020 | 2887 | } |
8f18cf13 | 2888 | } |
57ba4cb8 FM |
2889 | mutex_unlock(&fs_devices->device_list_mutex); |
2890 | ||
0b246afa | 2891 | ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset); |
47ab2a6c | 2892 | if (ret) { |
66642832 | 2893 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2894 | goto out; |
2895 | } | |
8f18cf13 | 2896 | |
6bccf3ab | 2897 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2898 | |
8f18cf13 | 2899 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
6bccf3ab JM |
2900 | ret = btrfs_del_sys_chunk(fs_info, chunk_objectid, |
2901 | chunk_offset); | |
47ab2a6c | 2902 | if (ret) { |
66642832 | 2903 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2904 | goto out; |
2905 | } | |
8f18cf13 CM |
2906 | } |
2907 | ||
6bccf3ab | 2908 | ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em); |
47ab2a6c | 2909 | if (ret) { |
66642832 | 2910 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2911 | goto out; |
2912 | } | |
2b82032c | 2913 | |
47ab2a6c | 2914 | out: |
2b82032c YZ |
2915 | /* once for us */ |
2916 | free_extent_map(em); | |
47ab2a6c JB |
2917 | return ret; |
2918 | } | |
2b82032c | 2919 | |
5b4aacef | 2920 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2921 | { |
5b4aacef | 2922 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2923 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2924 | int ret; |
2b82032c | 2925 | |
67c5e7d4 FM |
2926 | /* |
2927 | * Prevent races with automatic removal of unused block groups. | |
2928 | * After we relocate and before we remove the chunk with offset | |
2929 | * chunk_offset, automatic removal of the block group can kick in, | |
2930 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2931 | * | |
2932 | * Make sure to acquire this mutex before doing a tree search (dev | |
2933 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2934 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2935 | * we release the path used to search the chunk/dev tree and before | |
2936 | * the current task acquires this mutex and calls us. | |
2937 | */ | |
0b246afa | 2938 | ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex)); |
67c5e7d4 | 2939 | |
0b246afa | 2940 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2941 | if (ret) |
2942 | return -ENOSPC; | |
2943 | ||
2944 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 2945 | btrfs_scrub_pause(fs_info); |
0b246afa | 2946 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2947 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2948 | if (ret) |
2949 | return ret; | |
2950 | ||
19c4d2f9 CM |
2951 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2952 | chunk_offset); | |
2953 | if (IS_ERR(trans)) { | |
2954 | ret = PTR_ERR(trans); | |
2955 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2956 | return ret; | |
2957 | } | |
2958 | ||
47ab2a6c | 2959 | /* |
19c4d2f9 CM |
2960 | * step two, delete the device extents and the |
2961 | * chunk tree entries | |
47ab2a6c | 2962 | */ |
5b4aacef | 2963 | ret = btrfs_remove_chunk(trans, fs_info, chunk_offset); |
3a45bb20 | 2964 | btrfs_end_transaction(trans); |
19c4d2f9 | 2965 | return ret; |
2b82032c YZ |
2966 | } |
2967 | ||
2ff7e61e | 2968 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2969 | { |
0b246afa | 2970 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2971 | struct btrfs_path *path; |
2972 | struct extent_buffer *leaf; | |
2973 | struct btrfs_chunk *chunk; | |
2974 | struct btrfs_key key; | |
2975 | struct btrfs_key found_key; | |
2b82032c | 2976 | u64 chunk_type; |
ba1bf481 JB |
2977 | bool retried = false; |
2978 | int failed = 0; | |
2b82032c YZ |
2979 | int ret; |
2980 | ||
2981 | path = btrfs_alloc_path(); | |
2982 | if (!path) | |
2983 | return -ENOMEM; | |
2984 | ||
ba1bf481 | 2985 | again: |
2b82032c YZ |
2986 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2987 | key.offset = (u64)-1; | |
2988 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2989 | ||
2990 | while (1) { | |
0b246afa | 2991 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2992 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2993 | if (ret < 0) { |
0b246afa | 2994 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2995 | goto error; |
67c5e7d4 | 2996 | } |
79787eaa | 2997 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2998 | |
2999 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3000 | key.type); | |
67c5e7d4 | 3001 | if (ret) |
0b246afa | 3002 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3003 | if (ret < 0) |
3004 | goto error; | |
3005 | if (ret > 0) | |
3006 | break; | |
1a40e23b | 3007 | |
2b82032c YZ |
3008 | leaf = path->nodes[0]; |
3009 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3010 | |
2b82032c YZ |
3011 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3012 | struct btrfs_chunk); | |
3013 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3014 | btrfs_release_path(path); |
8f18cf13 | 3015 | |
2b82032c | 3016 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3017 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3018 | if (ret == -ENOSPC) |
3019 | failed++; | |
14586651 HS |
3020 | else |
3021 | BUG_ON(ret); | |
2b82032c | 3022 | } |
0b246afa | 3023 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3024 | |
2b82032c YZ |
3025 | if (found_key.offset == 0) |
3026 | break; | |
3027 | key.offset = found_key.offset - 1; | |
3028 | } | |
3029 | ret = 0; | |
ba1bf481 JB |
3030 | if (failed && !retried) { |
3031 | failed = 0; | |
3032 | retried = true; | |
3033 | goto again; | |
fae7f21c | 3034 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3035 | ret = -ENOSPC; |
3036 | } | |
2b82032c YZ |
3037 | error: |
3038 | btrfs_free_path(path); | |
3039 | return ret; | |
8f18cf13 CM |
3040 | } |
3041 | ||
6bccf3ab | 3042 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3043 | struct btrfs_balance_control *bctl) |
3044 | { | |
6bccf3ab | 3045 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3046 | struct btrfs_trans_handle *trans; |
3047 | struct btrfs_balance_item *item; | |
3048 | struct btrfs_disk_balance_args disk_bargs; | |
3049 | struct btrfs_path *path; | |
3050 | struct extent_buffer *leaf; | |
3051 | struct btrfs_key key; | |
3052 | int ret, err; | |
3053 | ||
3054 | path = btrfs_alloc_path(); | |
3055 | if (!path) | |
3056 | return -ENOMEM; | |
3057 | ||
3058 | trans = btrfs_start_transaction(root, 0); | |
3059 | if (IS_ERR(trans)) { | |
3060 | btrfs_free_path(path); | |
3061 | return PTR_ERR(trans); | |
3062 | } | |
3063 | ||
3064 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3065 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3066 | key.offset = 0; |
3067 | ||
3068 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3069 | sizeof(*item)); | |
3070 | if (ret) | |
3071 | goto out; | |
3072 | ||
3073 | leaf = path->nodes[0]; | |
3074 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3075 | ||
b159fa28 | 3076 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3077 | |
3078 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3079 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3080 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3081 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3082 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3083 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3084 | ||
3085 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3086 | ||
3087 | btrfs_mark_buffer_dirty(leaf); | |
3088 | out: | |
3089 | btrfs_free_path(path); | |
3a45bb20 | 3090 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3091 | if (err && !ret) |
3092 | ret = err; | |
3093 | return ret; | |
3094 | } | |
3095 | ||
6bccf3ab | 3096 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3097 | { |
6bccf3ab | 3098 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3099 | struct btrfs_trans_handle *trans; |
3100 | struct btrfs_path *path; | |
3101 | struct btrfs_key key; | |
3102 | int ret, err; | |
3103 | ||
3104 | path = btrfs_alloc_path(); | |
3105 | if (!path) | |
3106 | return -ENOMEM; | |
3107 | ||
3108 | trans = btrfs_start_transaction(root, 0); | |
3109 | if (IS_ERR(trans)) { | |
3110 | btrfs_free_path(path); | |
3111 | return PTR_ERR(trans); | |
3112 | } | |
3113 | ||
3114 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3115 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3116 | key.offset = 0; |
3117 | ||
3118 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3119 | if (ret < 0) | |
3120 | goto out; | |
3121 | if (ret > 0) { | |
3122 | ret = -ENOENT; | |
3123 | goto out; | |
3124 | } | |
3125 | ||
3126 | ret = btrfs_del_item(trans, root, path); | |
3127 | out: | |
3128 | btrfs_free_path(path); | |
3a45bb20 | 3129 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3130 | if (err && !ret) |
3131 | ret = err; | |
3132 | return ret; | |
3133 | } | |
3134 | ||
59641015 ID |
3135 | /* |
3136 | * This is a heuristic used to reduce the number of chunks balanced on | |
3137 | * resume after balance was interrupted. | |
3138 | */ | |
3139 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3140 | { | |
3141 | /* | |
3142 | * Turn on soft mode for chunk types that were being converted. | |
3143 | */ | |
3144 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3145 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3146 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3147 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3148 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3149 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3150 | ||
3151 | /* | |
3152 | * Turn on usage filter if is not already used. The idea is | |
3153 | * that chunks that we have already balanced should be | |
3154 | * reasonably full. Don't do it for chunks that are being | |
3155 | * converted - that will keep us from relocating unconverted | |
3156 | * (albeit full) chunks. | |
3157 | */ | |
3158 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3159 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3160 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3161 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3162 | bctl->data.usage = 90; | |
3163 | } | |
3164 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3165 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3166 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3167 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3168 | bctl->sys.usage = 90; | |
3169 | } | |
3170 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3171 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3172 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3173 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3174 | bctl->meta.usage = 90; | |
3175 | } | |
3176 | } | |
3177 | ||
c9e9f97b ID |
3178 | /* |
3179 | * Should be called with both balance and volume mutexes held to | |
3180 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3181 | * restriper. Same goes for unset_balance_control. | |
3182 | */ | |
3183 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3184 | { | |
3185 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3186 | ||
3187 | BUG_ON(fs_info->balance_ctl); | |
3188 | ||
3189 | spin_lock(&fs_info->balance_lock); | |
3190 | fs_info->balance_ctl = bctl; | |
3191 | spin_unlock(&fs_info->balance_lock); | |
3192 | } | |
3193 | ||
3194 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3195 | { | |
3196 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3197 | ||
3198 | BUG_ON(!fs_info->balance_ctl); | |
3199 | ||
3200 | spin_lock(&fs_info->balance_lock); | |
3201 | fs_info->balance_ctl = NULL; | |
3202 | spin_unlock(&fs_info->balance_lock); | |
3203 | ||
3204 | kfree(bctl); | |
3205 | } | |
3206 | ||
ed25e9b2 ID |
3207 | /* |
3208 | * Balance filters. Return 1 if chunk should be filtered out | |
3209 | * (should not be balanced). | |
3210 | */ | |
899c81ea | 3211 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3212 | struct btrfs_balance_args *bargs) |
3213 | { | |
899c81ea ID |
3214 | chunk_type = chunk_to_extended(chunk_type) & |
3215 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3216 | |
899c81ea | 3217 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3218 | return 0; |
3219 | ||
3220 | return 1; | |
3221 | } | |
3222 | ||
dba72cb3 | 3223 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3224 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3225 | { |
3226 | struct btrfs_block_group_cache *cache; | |
3227 | u64 chunk_used; | |
3228 | u64 user_thresh_min; | |
3229 | u64 user_thresh_max; | |
3230 | int ret = 1; | |
3231 | ||
3232 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3233 | chunk_used = btrfs_block_group_used(&cache->item); | |
3234 | ||
3235 | if (bargs->usage_min == 0) | |
3236 | user_thresh_min = 0; | |
3237 | else | |
3238 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3239 | bargs->usage_min); | |
3240 | ||
3241 | if (bargs->usage_max == 0) | |
3242 | user_thresh_max = 1; | |
3243 | else if (bargs->usage_max > 100) | |
3244 | user_thresh_max = cache->key.offset; | |
3245 | else | |
3246 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3247 | bargs->usage_max); | |
3248 | ||
3249 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3250 | ret = 0; | |
3251 | ||
3252 | btrfs_put_block_group(cache); | |
3253 | return ret; | |
3254 | } | |
3255 | ||
dba72cb3 | 3256 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3257 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3258 | { |
3259 | struct btrfs_block_group_cache *cache; | |
3260 | u64 chunk_used, user_thresh; | |
3261 | int ret = 1; | |
3262 | ||
3263 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3264 | chunk_used = btrfs_block_group_used(&cache->item); | |
3265 | ||
bc309467 | 3266 | if (bargs->usage_min == 0) |
3e39cea6 | 3267 | user_thresh = 1; |
a105bb88 ID |
3268 | else if (bargs->usage > 100) |
3269 | user_thresh = cache->key.offset; | |
3270 | else | |
3271 | user_thresh = div_factor_fine(cache->key.offset, | |
3272 | bargs->usage); | |
3273 | ||
5ce5b3c0 ID |
3274 | if (chunk_used < user_thresh) |
3275 | ret = 0; | |
3276 | ||
3277 | btrfs_put_block_group(cache); | |
3278 | return ret; | |
3279 | } | |
3280 | ||
409d404b ID |
3281 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3282 | struct btrfs_chunk *chunk, | |
3283 | struct btrfs_balance_args *bargs) | |
3284 | { | |
3285 | struct btrfs_stripe *stripe; | |
3286 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3287 | int i; | |
3288 | ||
3289 | for (i = 0; i < num_stripes; i++) { | |
3290 | stripe = btrfs_stripe_nr(chunk, i); | |
3291 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3292 | return 0; | |
3293 | } | |
3294 | ||
3295 | return 1; | |
3296 | } | |
3297 | ||
94e60d5a ID |
3298 | /* [pstart, pend) */ |
3299 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3300 | struct btrfs_chunk *chunk, | |
3301 | u64 chunk_offset, | |
3302 | struct btrfs_balance_args *bargs) | |
3303 | { | |
3304 | struct btrfs_stripe *stripe; | |
3305 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3306 | u64 stripe_offset; | |
3307 | u64 stripe_length; | |
3308 | int factor; | |
3309 | int i; | |
3310 | ||
3311 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3312 | return 0; | |
3313 | ||
3314 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3315 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3316 | factor = num_stripes / 2; | |
3317 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3318 | factor = num_stripes - 1; | |
3319 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3320 | factor = num_stripes - 2; | |
3321 | } else { | |
3322 | factor = num_stripes; | |
3323 | } | |
94e60d5a ID |
3324 | |
3325 | for (i = 0; i < num_stripes; i++) { | |
3326 | stripe = btrfs_stripe_nr(chunk, i); | |
3327 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3328 | continue; | |
3329 | ||
3330 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3331 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3332 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3333 | |
3334 | if (stripe_offset < bargs->pend && | |
3335 | stripe_offset + stripe_length > bargs->pstart) | |
3336 | return 0; | |
3337 | } | |
3338 | ||
3339 | return 1; | |
3340 | } | |
3341 | ||
ea67176a ID |
3342 | /* [vstart, vend) */ |
3343 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3344 | struct btrfs_chunk *chunk, | |
3345 | u64 chunk_offset, | |
3346 | struct btrfs_balance_args *bargs) | |
3347 | { | |
3348 | if (chunk_offset < bargs->vend && | |
3349 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3350 | /* at least part of the chunk is inside this vrange */ | |
3351 | return 0; | |
3352 | ||
3353 | return 1; | |
3354 | } | |
3355 | ||
dee32d0a GAP |
3356 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3357 | struct btrfs_chunk *chunk, | |
3358 | struct btrfs_balance_args *bargs) | |
3359 | { | |
3360 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3361 | ||
3362 | if (bargs->stripes_min <= num_stripes | |
3363 | && num_stripes <= bargs->stripes_max) | |
3364 | return 0; | |
3365 | ||
3366 | return 1; | |
3367 | } | |
3368 | ||
899c81ea | 3369 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3370 | struct btrfs_balance_args *bargs) |
3371 | { | |
3372 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3373 | return 0; | |
3374 | ||
899c81ea ID |
3375 | chunk_type = chunk_to_extended(chunk_type) & |
3376 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3377 | |
899c81ea | 3378 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3379 | return 1; |
3380 | ||
3381 | return 0; | |
3382 | } | |
3383 | ||
2ff7e61e | 3384 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3385 | struct extent_buffer *leaf, |
3386 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3387 | { | |
0b246afa | 3388 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3389 | struct btrfs_balance_args *bargs = NULL; |
3390 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3391 | ||
3392 | /* type filter */ | |
3393 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3394 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3395 | return 0; | |
3396 | } | |
3397 | ||
3398 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3399 | bargs = &bctl->data; | |
3400 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3401 | bargs = &bctl->sys; | |
3402 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3403 | bargs = &bctl->meta; | |
3404 | ||
ed25e9b2 ID |
3405 | /* profiles filter */ |
3406 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3407 | chunk_profiles_filter(chunk_type, bargs)) { | |
3408 | return 0; | |
5ce5b3c0 ID |
3409 | } |
3410 | ||
3411 | /* usage filter */ | |
3412 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3413 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3414 | return 0; |
bc309467 | 3415 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3416 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3417 | return 0; |
409d404b ID |
3418 | } |
3419 | ||
3420 | /* devid filter */ | |
3421 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3422 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3423 | return 0; | |
94e60d5a ID |
3424 | } |
3425 | ||
3426 | /* drange filter, makes sense only with devid filter */ | |
3427 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3428 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3429 | return 0; | |
ea67176a ID |
3430 | } |
3431 | ||
3432 | /* vrange filter */ | |
3433 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3434 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3435 | return 0; | |
ed25e9b2 ID |
3436 | } |
3437 | ||
dee32d0a GAP |
3438 | /* stripes filter */ |
3439 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3440 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3441 | return 0; | |
3442 | } | |
3443 | ||
cfa4c961 ID |
3444 | /* soft profile changing mode */ |
3445 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3446 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3447 | return 0; | |
3448 | } | |
3449 | ||
7d824b6f DS |
3450 | /* |
3451 | * limited by count, must be the last filter | |
3452 | */ | |
3453 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3454 | if (bargs->limit == 0) | |
3455 | return 0; | |
3456 | else | |
3457 | bargs->limit--; | |
12907fc7 DS |
3458 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3459 | /* | |
3460 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3461 | * determined here because we do not have the global information |
12907fc7 DS |
3462 | * about the count of all chunks that satisfy the filters. |
3463 | */ | |
3464 | if (bargs->limit_max == 0) | |
3465 | return 0; | |
3466 | else | |
3467 | bargs->limit_max--; | |
7d824b6f DS |
3468 | } |
3469 | ||
f43ffb60 ID |
3470 | return 1; |
3471 | } | |
3472 | ||
c9e9f97b | 3473 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3474 | { |
19a39dce | 3475 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3476 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3477 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3478 | struct list_head *devices; | |
ec44a35c CM |
3479 | struct btrfs_device *device; |
3480 | u64 old_size; | |
3481 | u64 size_to_free; | |
12907fc7 | 3482 | u64 chunk_type; |
f43ffb60 | 3483 | struct btrfs_chunk *chunk; |
5a488b9d | 3484 | struct btrfs_path *path = NULL; |
ec44a35c | 3485 | struct btrfs_key key; |
ec44a35c | 3486 | struct btrfs_key found_key; |
c9e9f97b | 3487 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3488 | struct extent_buffer *leaf; |
3489 | int slot; | |
c9e9f97b ID |
3490 | int ret; |
3491 | int enospc_errors = 0; | |
19a39dce | 3492 | bool counting = true; |
12907fc7 | 3493 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3494 | u64 limit_data = bctl->data.limit; |
3495 | u64 limit_meta = bctl->meta.limit; | |
3496 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3497 | u32 count_data = 0; |
3498 | u32 count_meta = 0; | |
3499 | u32 count_sys = 0; | |
2c9fe835 | 3500 | int chunk_reserved = 0; |
cf25ce51 | 3501 | u64 bytes_used = 0; |
ec44a35c | 3502 | |
ec44a35c | 3503 | /* step one make some room on all the devices */ |
c9e9f97b | 3504 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3505 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3506 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3507 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3508 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
2b82032c | 3509 | if (!device->writeable || |
7cc8e58d MX |
3510 | btrfs_device_get_total_bytes(device) - |
3511 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3512 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3513 | continue; |
3514 | ||
3515 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3516 | if (ret == -ENOSPC) |
3517 | break; | |
5a488b9d LB |
3518 | if (ret) { |
3519 | /* btrfs_shrink_device never returns ret > 0 */ | |
3520 | WARN_ON(ret > 0); | |
3521 | goto error; | |
3522 | } | |
ec44a35c | 3523 | |
a22285a6 | 3524 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3525 | if (IS_ERR(trans)) { |
3526 | ret = PTR_ERR(trans); | |
3527 | btrfs_info_in_rcu(fs_info, | |
3528 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3529 | rcu_str_deref(device->name), ret, | |
3530 | old_size, old_size - size_to_free); | |
3531 | goto error; | |
3532 | } | |
ec44a35c CM |
3533 | |
3534 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3535 | if (ret) { |
3a45bb20 | 3536 | btrfs_end_transaction(trans); |
5a488b9d LB |
3537 | /* btrfs_grow_device never returns ret > 0 */ |
3538 | WARN_ON(ret > 0); | |
3539 | btrfs_info_in_rcu(fs_info, | |
3540 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3541 | rcu_str_deref(device->name), ret, | |
3542 | old_size, old_size - size_to_free); | |
3543 | goto error; | |
3544 | } | |
ec44a35c | 3545 | |
3a45bb20 | 3546 | btrfs_end_transaction(trans); |
ec44a35c CM |
3547 | } |
3548 | ||
3549 | /* step two, relocate all the chunks */ | |
3550 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3551 | if (!path) { |
3552 | ret = -ENOMEM; | |
3553 | goto error; | |
3554 | } | |
19a39dce ID |
3555 | |
3556 | /* zero out stat counters */ | |
3557 | spin_lock(&fs_info->balance_lock); | |
3558 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3559 | spin_unlock(&fs_info->balance_lock); | |
3560 | again: | |
7d824b6f | 3561 | if (!counting) { |
12907fc7 DS |
3562 | /* |
3563 | * The single value limit and min/max limits use the same bytes | |
3564 | * in the | |
3565 | */ | |
7d824b6f DS |
3566 | bctl->data.limit = limit_data; |
3567 | bctl->meta.limit = limit_meta; | |
3568 | bctl->sys.limit = limit_sys; | |
3569 | } | |
ec44a35c CM |
3570 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3571 | key.offset = (u64)-1; | |
3572 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3573 | ||
d397712b | 3574 | while (1) { |
19a39dce | 3575 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3576 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3577 | ret = -ECANCELED; |
3578 | goto error; | |
3579 | } | |
3580 | ||
67c5e7d4 | 3581 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3582 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3583 | if (ret < 0) { |
3584 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3585 | goto error; |
67c5e7d4 | 3586 | } |
ec44a35c CM |
3587 | |
3588 | /* | |
3589 | * this shouldn't happen, it means the last relocate | |
3590 | * failed | |
3591 | */ | |
3592 | if (ret == 0) | |
c9e9f97b | 3593 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3594 | |
3595 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3596 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3597 | if (ret) { |
67c5e7d4 | 3598 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3599 | ret = 0; |
ec44a35c | 3600 | break; |
c9e9f97b | 3601 | } |
7d9eb12c | 3602 | |
f43ffb60 ID |
3603 | leaf = path->nodes[0]; |
3604 | slot = path->slots[0]; | |
3605 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3606 | |
67c5e7d4 FM |
3607 | if (found_key.objectid != key.objectid) { |
3608 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3609 | break; |
67c5e7d4 | 3610 | } |
7d9eb12c | 3611 | |
f43ffb60 | 3612 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3613 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3614 | |
19a39dce ID |
3615 | if (!counting) { |
3616 | spin_lock(&fs_info->balance_lock); | |
3617 | bctl->stat.considered++; | |
3618 | spin_unlock(&fs_info->balance_lock); | |
3619 | } | |
3620 | ||
2ff7e61e | 3621 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3622 | found_key.offset); |
2c9fe835 | 3623 | |
b3b4aa74 | 3624 | btrfs_release_path(path); |
67c5e7d4 FM |
3625 | if (!ret) { |
3626 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3627 | goto loop; |
67c5e7d4 | 3628 | } |
f43ffb60 | 3629 | |
19a39dce | 3630 | if (counting) { |
67c5e7d4 | 3631 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3632 | spin_lock(&fs_info->balance_lock); |
3633 | bctl->stat.expected++; | |
3634 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3635 | |
3636 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3637 | count_data++; | |
3638 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3639 | count_sys++; | |
3640 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3641 | count_meta++; | |
3642 | ||
3643 | goto loop; | |
3644 | } | |
3645 | ||
3646 | /* | |
3647 | * Apply limit_min filter, no need to check if the LIMITS | |
3648 | * filter is used, limit_min is 0 by default | |
3649 | */ | |
3650 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3651 | count_data < bctl->data.limit_min) | |
3652 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3653 | count_meta < bctl->meta.limit_min) | |
3654 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3655 | count_sys < bctl->sys.limit_min)) { | |
3656 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3657 | goto loop; |
3658 | } | |
3659 | ||
cf25ce51 LB |
3660 | ASSERT(fs_info->data_sinfo); |
3661 | spin_lock(&fs_info->data_sinfo->lock); | |
3662 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3663 | spin_unlock(&fs_info->data_sinfo->lock); | |
3664 | ||
3665 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3666 | !chunk_reserved && !bytes_used) { | |
2c9fe835 ZL |
3667 | trans = btrfs_start_transaction(chunk_root, 0); |
3668 | if (IS_ERR(trans)) { | |
3669 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3670 | ret = PTR_ERR(trans); | |
3671 | goto error; | |
3672 | } | |
3673 | ||
2ff7e61e | 3674 | ret = btrfs_force_chunk_alloc(trans, fs_info, |
2c9fe835 | 3675 | BTRFS_BLOCK_GROUP_DATA); |
3a45bb20 | 3676 | btrfs_end_transaction(trans); |
2c9fe835 ZL |
3677 | if (ret < 0) { |
3678 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3679 | goto error; | |
3680 | } | |
2c9fe835 ZL |
3681 | chunk_reserved = 1; |
3682 | } | |
3683 | ||
5b4aacef | 3684 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3685 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3686 | if (ret && ret != -ENOSPC) |
3687 | goto error; | |
19a39dce | 3688 | if (ret == -ENOSPC) { |
c9e9f97b | 3689 | enospc_errors++; |
19a39dce ID |
3690 | } else { |
3691 | spin_lock(&fs_info->balance_lock); | |
3692 | bctl->stat.completed++; | |
3693 | spin_unlock(&fs_info->balance_lock); | |
3694 | } | |
f43ffb60 | 3695 | loop: |
795a3321 ID |
3696 | if (found_key.offset == 0) |
3697 | break; | |
ba1bf481 | 3698 | key.offset = found_key.offset - 1; |
ec44a35c | 3699 | } |
c9e9f97b | 3700 | |
19a39dce ID |
3701 | if (counting) { |
3702 | btrfs_release_path(path); | |
3703 | counting = false; | |
3704 | goto again; | |
3705 | } | |
ec44a35c CM |
3706 | error: |
3707 | btrfs_free_path(path); | |
c9e9f97b | 3708 | if (enospc_errors) { |
efe120a0 | 3709 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3710 | enospc_errors); |
c9e9f97b ID |
3711 | if (!ret) |
3712 | ret = -ENOSPC; | |
3713 | } | |
3714 | ||
ec44a35c CM |
3715 | return ret; |
3716 | } | |
3717 | ||
0c460c0d ID |
3718 | /** |
3719 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3720 | * @flags: profile to validate | |
3721 | * @extended: if true @flags is treated as an extended profile | |
3722 | */ | |
3723 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3724 | { | |
3725 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3726 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3727 | ||
3728 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3729 | ||
3730 | /* 1) check that all other bits are zeroed */ | |
3731 | if (flags & ~mask) | |
3732 | return 0; | |
3733 | ||
3734 | /* 2) see if profile is reduced */ | |
3735 | if (flags == 0) | |
3736 | return !extended; /* "0" is valid for usual profiles */ | |
3737 | ||
3738 | /* true if exactly one bit set */ | |
3739 | return (flags & (flags - 1)) == 0; | |
3740 | } | |
3741 | ||
837d5b6e ID |
3742 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3743 | { | |
a7e99c69 ID |
3744 | /* cancel requested || normal exit path */ |
3745 | return atomic_read(&fs_info->balance_cancel_req) || | |
3746 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3747 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3748 | } |
3749 | ||
c9e9f97b ID |
3750 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3751 | { | |
0940ebf6 ID |
3752 | int ret; |
3753 | ||
c9e9f97b | 3754 | unset_balance_control(fs_info); |
6bccf3ab | 3755 | ret = del_balance_item(fs_info); |
0f788c58 | 3756 | if (ret) |
34d97007 | 3757 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f | 3758 | |
171938e5 | 3759 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
c9e9f97b ID |
3760 | } |
3761 | ||
bdcd3c97 AM |
3762 | /* Non-zero return value signifies invalidity */ |
3763 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3764 | u64 allowed) | |
3765 | { | |
3766 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3767 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3768 | (bctl_arg->target & ~allowed))); | |
3769 | } | |
3770 | ||
c9e9f97b ID |
3771 | /* |
3772 | * Should be called with both balance and volume mutexes held | |
3773 | */ | |
3774 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3775 | struct btrfs_ioctl_balance_args *bargs) | |
3776 | { | |
3777 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
14506127 | 3778 | u64 meta_target, data_target; |
f43ffb60 | 3779 | u64 allowed; |
e4837f8f | 3780 | int mixed = 0; |
c9e9f97b | 3781 | int ret; |
8dabb742 | 3782 | u64 num_devices; |
de98ced9 | 3783 | unsigned seq; |
c9e9f97b | 3784 | |
837d5b6e | 3785 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3786 | atomic_read(&fs_info->balance_pause_req) || |
3787 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3788 | ret = -EINVAL; |
3789 | goto out; | |
3790 | } | |
3791 | ||
e4837f8f ID |
3792 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3793 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3794 | mixed = 1; | |
3795 | ||
f43ffb60 ID |
3796 | /* |
3797 | * In case of mixed groups both data and meta should be picked, | |
3798 | * and identical options should be given for both of them. | |
3799 | */ | |
e4837f8f ID |
3800 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3801 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3802 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3803 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3804 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3805 | btrfs_err(fs_info, |
3806 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3807 | ret = -EINVAL; |
3808 | goto out; | |
3809 | } | |
3810 | } | |
3811 | ||
8dabb742 | 3812 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3813 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3814 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3815 | BUG_ON(num_devices < 1); | |
3816 | num_devices--; | |
3817 | } | |
73beece9 | 3818 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
88be159c AH |
3819 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3820 | if (num_devices > 1) | |
e4d8ec0f | 3821 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3822 | if (num_devices > 2) |
3823 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3824 | if (num_devices > 3) | |
3825 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3826 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3827 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3828 | btrfs_err(fs_info, |
3829 | "unable to start balance with target data profile %llu", | |
3830 | bctl->data.target); | |
e4d8ec0f ID |
3831 | ret = -EINVAL; |
3832 | goto out; | |
3833 | } | |
bdcd3c97 | 3834 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3835 | btrfs_err(fs_info, |
5d163e0e JM |
3836 | "unable to start balance with target metadata profile %llu", |
3837 | bctl->meta.target); | |
e4d8ec0f ID |
3838 | ret = -EINVAL; |
3839 | goto out; | |
3840 | } | |
bdcd3c97 | 3841 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3842 | btrfs_err(fs_info, |
5d163e0e JM |
3843 | "unable to start balance with target system profile %llu", |
3844 | bctl->sys.target); | |
e4d8ec0f ID |
3845 | ret = -EINVAL; |
3846 | goto out; | |
3847 | } | |
3848 | ||
e4d8ec0f ID |
3849 | /* allow to reduce meta or sys integrity only if force set */ |
3850 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3851 | BTRFS_BLOCK_GROUP_RAID10 | |
3852 | BTRFS_BLOCK_GROUP_RAID5 | | |
3853 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3854 | do { |
3855 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3856 | ||
3857 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3858 | (fs_info->avail_system_alloc_bits & allowed) && | |
3859 | !(bctl->sys.target & allowed)) || | |
3860 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3861 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3862 | !(bctl->meta.target & allowed))) { | |
3863 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3864 | btrfs_info(fs_info, |
3865 | "force reducing metadata integrity"); | |
de98ced9 | 3866 | } else { |
5d163e0e JM |
3867 | btrfs_err(fs_info, |
3868 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3869 | ret = -EINVAL; |
3870 | goto out; | |
3871 | } | |
e4d8ec0f | 3872 | } |
de98ced9 | 3873 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3874 | |
14506127 AB |
3875 | /* if we're not converting, the target field is uninitialized */ |
3876 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3877 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3878 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3879 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3880 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3881 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 3882 | btrfs_warn(fs_info, |
5d163e0e | 3883 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
14506127 | 3884 | meta_target, data_target); |
ee592d07 ST |
3885 | } |
3886 | ||
5af3e8cc | 3887 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
943c6e99 ZL |
3888 | fs_info->num_tolerated_disk_barrier_failures = min( |
3889 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3890 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3891 | bctl->sys.target)); | |
5af3e8cc SB |
3892 | } |
3893 | ||
6bccf3ab | 3894 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3895 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3896 | goto out; |
3897 | ||
59641015 ID |
3898 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3899 | BUG_ON(ret == -EEXIST); | |
3900 | set_balance_control(bctl); | |
3901 | } else { | |
3902 | BUG_ON(ret != -EEXIST); | |
3903 | spin_lock(&fs_info->balance_lock); | |
3904 | update_balance_args(bctl); | |
3905 | spin_unlock(&fs_info->balance_lock); | |
3906 | } | |
c9e9f97b | 3907 | |
837d5b6e | 3908 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3909 | mutex_unlock(&fs_info->balance_mutex); |
3910 | ||
3911 | ret = __btrfs_balance(fs_info); | |
3912 | ||
3913 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3914 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3915 | |
bf023ecf ID |
3916 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3917 | fs_info->num_tolerated_disk_barrier_failures = | |
3918 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3919 | } | |
3920 | ||
c9e9f97b ID |
3921 | if (bargs) { |
3922 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3923 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3924 | } |
3925 | ||
3a01aa7a ID |
3926 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3927 | balance_need_close(fs_info)) { | |
3928 | __cancel_balance(fs_info); | |
3929 | } | |
3930 | ||
837d5b6e | 3931 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3932 | |
3933 | return ret; | |
3934 | out: | |
59641015 ID |
3935 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3936 | __cancel_balance(fs_info); | |
ed0fb78f | 3937 | else { |
59641015 | 3938 | kfree(bctl); |
171938e5 | 3939 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
ed0fb78f | 3940 | } |
59641015 ID |
3941 | return ret; |
3942 | } | |
3943 | ||
3944 | static int balance_kthread(void *data) | |
3945 | { | |
2b6ba629 | 3946 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3947 | int ret = 0; |
59641015 ID |
3948 | |
3949 | mutex_lock(&fs_info->volume_mutex); | |
3950 | mutex_lock(&fs_info->balance_mutex); | |
3951 | ||
2b6ba629 | 3952 | if (fs_info->balance_ctl) { |
efe120a0 | 3953 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3954 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3955 | } |
59641015 ID |
3956 | |
3957 | mutex_unlock(&fs_info->balance_mutex); | |
3958 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3959 | |
59641015 ID |
3960 | return ret; |
3961 | } | |
3962 | ||
2b6ba629 ID |
3963 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3964 | { | |
3965 | struct task_struct *tsk; | |
3966 | ||
3967 | spin_lock(&fs_info->balance_lock); | |
3968 | if (!fs_info->balance_ctl) { | |
3969 | spin_unlock(&fs_info->balance_lock); | |
3970 | return 0; | |
3971 | } | |
3972 | spin_unlock(&fs_info->balance_lock); | |
3973 | ||
3cdde224 | 3974 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 3975 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3976 | return 0; |
3977 | } | |
3978 | ||
3979 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3980 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3981 | } |
3982 | ||
68310a5e | 3983 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3984 | { |
59641015 ID |
3985 | struct btrfs_balance_control *bctl; |
3986 | struct btrfs_balance_item *item; | |
3987 | struct btrfs_disk_balance_args disk_bargs; | |
3988 | struct btrfs_path *path; | |
3989 | struct extent_buffer *leaf; | |
3990 | struct btrfs_key key; | |
3991 | int ret; | |
3992 | ||
3993 | path = btrfs_alloc_path(); | |
3994 | if (!path) | |
3995 | return -ENOMEM; | |
3996 | ||
59641015 | 3997 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3998 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
3999 | key.offset = 0; |
4000 | ||
68310a5e | 4001 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4002 | if (ret < 0) |
68310a5e | 4003 | goto out; |
59641015 ID |
4004 | if (ret > 0) { /* ret = -ENOENT; */ |
4005 | ret = 0; | |
68310a5e ID |
4006 | goto out; |
4007 | } | |
4008 | ||
4009 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4010 | if (!bctl) { | |
4011 | ret = -ENOMEM; | |
4012 | goto out; | |
59641015 ID |
4013 | } |
4014 | ||
4015 | leaf = path->nodes[0]; | |
4016 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4017 | ||
68310a5e ID |
4018 | bctl->fs_info = fs_info; |
4019 | bctl->flags = btrfs_balance_flags(leaf, item); | |
4020 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4021 | |
4022 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4023 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4024 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4025 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4026 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4027 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4028 | ||
171938e5 | 4029 | WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)); |
ed0fb78f | 4030 | |
68310a5e ID |
4031 | mutex_lock(&fs_info->volume_mutex); |
4032 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 4033 | |
68310a5e ID |
4034 | set_balance_control(bctl); |
4035 | ||
4036 | mutex_unlock(&fs_info->balance_mutex); | |
4037 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4038 | out: |
4039 | btrfs_free_path(path); | |
ec44a35c CM |
4040 | return ret; |
4041 | } | |
4042 | ||
837d5b6e ID |
4043 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4044 | { | |
4045 | int ret = 0; | |
4046 | ||
4047 | mutex_lock(&fs_info->balance_mutex); | |
4048 | if (!fs_info->balance_ctl) { | |
4049 | mutex_unlock(&fs_info->balance_mutex); | |
4050 | return -ENOTCONN; | |
4051 | } | |
4052 | ||
4053 | if (atomic_read(&fs_info->balance_running)) { | |
4054 | atomic_inc(&fs_info->balance_pause_req); | |
4055 | mutex_unlock(&fs_info->balance_mutex); | |
4056 | ||
4057 | wait_event(fs_info->balance_wait_q, | |
4058 | atomic_read(&fs_info->balance_running) == 0); | |
4059 | ||
4060 | mutex_lock(&fs_info->balance_mutex); | |
4061 | /* we are good with balance_ctl ripped off from under us */ | |
4062 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4063 | atomic_dec(&fs_info->balance_pause_req); | |
4064 | } else { | |
4065 | ret = -ENOTCONN; | |
4066 | } | |
4067 | ||
4068 | mutex_unlock(&fs_info->balance_mutex); | |
4069 | return ret; | |
4070 | } | |
4071 | ||
a7e99c69 ID |
4072 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4073 | { | |
e649e587 ID |
4074 | if (fs_info->sb->s_flags & MS_RDONLY) |
4075 | return -EROFS; | |
4076 | ||
a7e99c69 ID |
4077 | mutex_lock(&fs_info->balance_mutex); |
4078 | if (!fs_info->balance_ctl) { | |
4079 | mutex_unlock(&fs_info->balance_mutex); | |
4080 | return -ENOTCONN; | |
4081 | } | |
4082 | ||
4083 | atomic_inc(&fs_info->balance_cancel_req); | |
4084 | /* | |
4085 | * if we are running just wait and return, balance item is | |
4086 | * deleted in btrfs_balance in this case | |
4087 | */ | |
4088 | if (atomic_read(&fs_info->balance_running)) { | |
4089 | mutex_unlock(&fs_info->balance_mutex); | |
4090 | wait_event(fs_info->balance_wait_q, | |
4091 | atomic_read(&fs_info->balance_running) == 0); | |
4092 | mutex_lock(&fs_info->balance_mutex); | |
4093 | } else { | |
4094 | /* __cancel_balance needs volume_mutex */ | |
4095 | mutex_unlock(&fs_info->balance_mutex); | |
4096 | mutex_lock(&fs_info->volume_mutex); | |
4097 | mutex_lock(&fs_info->balance_mutex); | |
4098 | ||
4099 | if (fs_info->balance_ctl) | |
4100 | __cancel_balance(fs_info); | |
4101 | ||
4102 | mutex_unlock(&fs_info->volume_mutex); | |
4103 | } | |
4104 | ||
4105 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4106 | atomic_dec(&fs_info->balance_cancel_req); | |
4107 | mutex_unlock(&fs_info->balance_mutex); | |
4108 | return 0; | |
4109 | } | |
4110 | ||
803b2f54 SB |
4111 | static int btrfs_uuid_scan_kthread(void *data) |
4112 | { | |
4113 | struct btrfs_fs_info *fs_info = data; | |
4114 | struct btrfs_root *root = fs_info->tree_root; | |
4115 | struct btrfs_key key; | |
4116 | struct btrfs_key max_key; | |
4117 | struct btrfs_path *path = NULL; | |
4118 | int ret = 0; | |
4119 | struct extent_buffer *eb; | |
4120 | int slot; | |
4121 | struct btrfs_root_item root_item; | |
4122 | u32 item_size; | |
f45388f3 | 4123 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4124 | |
4125 | path = btrfs_alloc_path(); | |
4126 | if (!path) { | |
4127 | ret = -ENOMEM; | |
4128 | goto out; | |
4129 | } | |
4130 | ||
4131 | key.objectid = 0; | |
4132 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4133 | key.offset = 0; | |
4134 | ||
4135 | max_key.objectid = (u64)-1; | |
4136 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
4137 | max_key.offset = (u64)-1; | |
4138 | ||
803b2f54 | 4139 | while (1) { |
6174d3cb | 4140 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4141 | if (ret) { |
4142 | if (ret > 0) | |
4143 | ret = 0; | |
4144 | break; | |
4145 | } | |
4146 | ||
4147 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4148 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4149 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4150 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4151 | goto skip; | |
4152 | ||
4153 | eb = path->nodes[0]; | |
4154 | slot = path->slots[0]; | |
4155 | item_size = btrfs_item_size_nr(eb, slot); | |
4156 | if (item_size < sizeof(root_item)) | |
4157 | goto skip; | |
4158 | ||
803b2f54 SB |
4159 | read_extent_buffer(eb, &root_item, |
4160 | btrfs_item_ptr_offset(eb, slot), | |
4161 | (int)sizeof(root_item)); | |
4162 | if (btrfs_root_refs(&root_item) == 0) | |
4163 | goto skip; | |
f45388f3 FDBM |
4164 | |
4165 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4166 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4167 | if (trans) | |
4168 | goto update_tree; | |
4169 | ||
4170 | btrfs_release_path(path); | |
803b2f54 SB |
4171 | /* |
4172 | * 1 - subvol uuid item | |
4173 | * 1 - received_subvol uuid item | |
4174 | */ | |
4175 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4176 | if (IS_ERR(trans)) { | |
4177 | ret = PTR_ERR(trans); | |
4178 | break; | |
4179 | } | |
f45388f3 FDBM |
4180 | continue; |
4181 | } else { | |
4182 | goto skip; | |
4183 | } | |
4184 | update_tree: | |
4185 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
6bccf3ab | 4186 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4187 | root_item.uuid, |
4188 | BTRFS_UUID_KEY_SUBVOL, | |
4189 | key.objectid); | |
4190 | if (ret < 0) { | |
efe120a0 | 4191 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4192 | ret); |
803b2f54 SB |
4193 | break; |
4194 | } | |
4195 | } | |
4196 | ||
4197 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
6bccf3ab | 4198 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4199 | root_item.received_uuid, |
4200 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4201 | key.objectid); | |
4202 | if (ret < 0) { | |
efe120a0 | 4203 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4204 | ret); |
803b2f54 SB |
4205 | break; |
4206 | } | |
4207 | } | |
4208 | ||
f45388f3 | 4209 | skip: |
803b2f54 | 4210 | if (trans) { |
3a45bb20 | 4211 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4212 | trans = NULL; |
803b2f54 SB |
4213 | if (ret) |
4214 | break; | |
4215 | } | |
4216 | ||
803b2f54 SB |
4217 | btrfs_release_path(path); |
4218 | if (key.offset < (u64)-1) { | |
4219 | key.offset++; | |
4220 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4221 | key.offset = 0; | |
4222 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4223 | } else if (key.objectid < (u64)-1) { | |
4224 | key.offset = 0; | |
4225 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4226 | key.objectid++; | |
4227 | } else { | |
4228 | break; | |
4229 | } | |
4230 | cond_resched(); | |
4231 | } | |
4232 | ||
4233 | out: | |
4234 | btrfs_free_path(path); | |
f45388f3 | 4235 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4236 | btrfs_end_transaction(trans); |
803b2f54 | 4237 | if (ret) |
efe120a0 | 4238 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4239 | else |
afcdd129 | 4240 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4241 | up(&fs_info->uuid_tree_rescan_sem); |
4242 | return 0; | |
4243 | } | |
4244 | ||
70f80175 SB |
4245 | /* |
4246 | * Callback for btrfs_uuid_tree_iterate(). | |
4247 | * returns: | |
4248 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4249 | * < 0 if an error occurred. |
70f80175 SB |
4250 | * > 0 if the check failed, which means the caller shall remove the entry. |
4251 | */ | |
4252 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4253 | u8 *uuid, u8 type, u64 subid) | |
4254 | { | |
4255 | struct btrfs_key key; | |
4256 | int ret = 0; | |
4257 | struct btrfs_root *subvol_root; | |
4258 | ||
4259 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4260 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4261 | goto out; | |
4262 | ||
4263 | key.objectid = subid; | |
4264 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4265 | key.offset = (u64)-1; | |
4266 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4267 | if (IS_ERR(subvol_root)) { | |
4268 | ret = PTR_ERR(subvol_root); | |
4269 | if (ret == -ENOENT) | |
4270 | ret = 1; | |
4271 | goto out; | |
4272 | } | |
4273 | ||
4274 | switch (type) { | |
4275 | case BTRFS_UUID_KEY_SUBVOL: | |
4276 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4277 | ret = 1; | |
4278 | break; | |
4279 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4280 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4281 | BTRFS_UUID_SIZE)) | |
4282 | ret = 1; | |
4283 | break; | |
4284 | } | |
4285 | ||
4286 | out: | |
4287 | return ret; | |
4288 | } | |
4289 | ||
4290 | static int btrfs_uuid_rescan_kthread(void *data) | |
4291 | { | |
4292 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4293 | int ret; | |
4294 | ||
4295 | /* | |
4296 | * 1st step is to iterate through the existing UUID tree and | |
4297 | * to delete all entries that contain outdated data. | |
4298 | * 2nd step is to add all missing entries to the UUID tree. | |
4299 | */ | |
4300 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4301 | if (ret < 0) { | |
efe120a0 | 4302 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4303 | up(&fs_info->uuid_tree_rescan_sem); |
4304 | return ret; | |
4305 | } | |
4306 | return btrfs_uuid_scan_kthread(data); | |
4307 | } | |
4308 | ||
f7a81ea4 SB |
4309 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4310 | { | |
4311 | struct btrfs_trans_handle *trans; | |
4312 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4313 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4314 | struct task_struct *task; |
4315 | int ret; | |
f7a81ea4 SB |
4316 | |
4317 | /* | |
4318 | * 1 - root node | |
4319 | * 1 - root item | |
4320 | */ | |
4321 | trans = btrfs_start_transaction(tree_root, 2); | |
4322 | if (IS_ERR(trans)) | |
4323 | return PTR_ERR(trans); | |
4324 | ||
4325 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4326 | BTRFS_UUID_TREE_OBJECTID); | |
4327 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4328 | ret = PTR_ERR(uuid_root); |
66642832 | 4329 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4330 | btrfs_end_transaction(trans); |
6d13f549 | 4331 | return ret; |
f7a81ea4 SB |
4332 | } |
4333 | ||
4334 | fs_info->uuid_root = uuid_root; | |
4335 | ||
3a45bb20 | 4336 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4337 | if (ret) |
4338 | return ret; | |
4339 | ||
4340 | down(&fs_info->uuid_tree_rescan_sem); | |
4341 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4342 | if (IS_ERR(task)) { | |
70f80175 | 4343 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4344 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4345 | up(&fs_info->uuid_tree_rescan_sem); |
4346 | return PTR_ERR(task); | |
4347 | } | |
4348 | ||
4349 | return 0; | |
f7a81ea4 | 4350 | } |
803b2f54 | 4351 | |
70f80175 SB |
4352 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4353 | { | |
4354 | struct task_struct *task; | |
4355 | ||
4356 | down(&fs_info->uuid_tree_rescan_sem); | |
4357 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4358 | if (IS_ERR(task)) { | |
4359 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4360 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4361 | up(&fs_info->uuid_tree_rescan_sem); |
4362 | return PTR_ERR(task); | |
4363 | } | |
4364 | ||
4365 | return 0; | |
4366 | } | |
4367 | ||
8f18cf13 CM |
4368 | /* |
4369 | * shrinking a device means finding all of the device extents past | |
4370 | * the new size, and then following the back refs to the chunks. | |
4371 | * The chunk relocation code actually frees the device extent | |
4372 | */ | |
4373 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4374 | { | |
0b246afa JM |
4375 | struct btrfs_fs_info *fs_info = device->fs_info; |
4376 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4377 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4378 | struct btrfs_dev_extent *dev_extent = NULL; |
4379 | struct btrfs_path *path; | |
4380 | u64 length; | |
8f18cf13 CM |
4381 | u64 chunk_offset; |
4382 | int ret; | |
4383 | int slot; | |
ba1bf481 JB |
4384 | int failed = 0; |
4385 | bool retried = false; | |
53e489bc | 4386 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4387 | struct extent_buffer *l; |
4388 | struct btrfs_key key; | |
0b246afa | 4389 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4390 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4391 | u64 old_size = btrfs_device_get_total_bytes(device); |
4392 | u64 diff = old_size - new_size; | |
8f18cf13 | 4393 | |
63a212ab SB |
4394 | if (device->is_tgtdev_for_dev_replace) |
4395 | return -EINVAL; | |
4396 | ||
8f18cf13 CM |
4397 | path = btrfs_alloc_path(); |
4398 | if (!path) | |
4399 | return -ENOMEM; | |
4400 | ||
e4058b54 | 4401 | path->reada = READA_FORWARD; |
8f18cf13 | 4402 | |
34441361 | 4403 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4404 | |
7cc8e58d | 4405 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4406 | if (device->writeable) { |
2b82032c | 4407 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4408 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4409 | } |
34441361 | 4410 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4411 | |
ba1bf481 | 4412 | again: |
8f18cf13 CM |
4413 | key.objectid = device->devid; |
4414 | key.offset = (u64)-1; | |
4415 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4416 | ||
213e64da | 4417 | do { |
0b246afa | 4418 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4419 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4420 | if (ret < 0) { |
0b246afa | 4421 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4422 | goto done; |
67c5e7d4 | 4423 | } |
8f18cf13 CM |
4424 | |
4425 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4426 | if (ret) |
0b246afa | 4427 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4428 | if (ret < 0) |
4429 | goto done; | |
4430 | if (ret) { | |
4431 | ret = 0; | |
b3b4aa74 | 4432 | btrfs_release_path(path); |
bf1fb512 | 4433 | break; |
8f18cf13 CM |
4434 | } |
4435 | ||
4436 | l = path->nodes[0]; | |
4437 | slot = path->slots[0]; | |
4438 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4439 | ||
ba1bf481 | 4440 | if (key.objectid != device->devid) { |
0b246afa | 4441 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4442 | btrfs_release_path(path); |
bf1fb512 | 4443 | break; |
ba1bf481 | 4444 | } |
8f18cf13 CM |
4445 | |
4446 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4447 | length = btrfs_dev_extent_length(l, dev_extent); | |
4448 | ||
ba1bf481 | 4449 | if (key.offset + length <= new_size) { |
0b246afa | 4450 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4451 | btrfs_release_path(path); |
d6397bae | 4452 | break; |
ba1bf481 | 4453 | } |
8f18cf13 | 4454 | |
8f18cf13 | 4455 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4456 | btrfs_release_path(path); |
8f18cf13 | 4457 | |
0b246afa JM |
4458 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4459 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4460 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4461 | goto done; |
ba1bf481 JB |
4462 | if (ret == -ENOSPC) |
4463 | failed++; | |
213e64da | 4464 | } while (key.offset-- > 0); |
ba1bf481 JB |
4465 | |
4466 | if (failed && !retried) { | |
4467 | failed = 0; | |
4468 | retried = true; | |
4469 | goto again; | |
4470 | } else if (failed && retried) { | |
4471 | ret = -ENOSPC; | |
ba1bf481 | 4472 | goto done; |
8f18cf13 CM |
4473 | } |
4474 | ||
d6397bae | 4475 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4476 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4477 | if (IS_ERR(trans)) { |
4478 | ret = PTR_ERR(trans); | |
4479 | goto done; | |
4480 | } | |
4481 | ||
34441361 | 4482 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4483 | |
4484 | /* | |
4485 | * We checked in the above loop all device extents that were already in | |
4486 | * the device tree. However before we have updated the device's | |
4487 | * total_bytes to the new size, we might have had chunk allocations that | |
4488 | * have not complete yet (new block groups attached to transaction | |
4489 | * handles), and therefore their device extents were not yet in the | |
4490 | * device tree and we missed them in the loop above. So if we have any | |
4491 | * pending chunk using a device extent that overlaps the device range | |
4492 | * that we can not use anymore, commit the current transaction and | |
4493 | * repeat the search on the device tree - this way we guarantee we will | |
4494 | * not have chunks using device extents that end beyond 'new_size'. | |
4495 | */ | |
4496 | if (!checked_pending_chunks) { | |
4497 | u64 start = new_size; | |
4498 | u64 len = old_size - new_size; | |
4499 | ||
499f377f JM |
4500 | if (contains_pending_extent(trans->transaction, device, |
4501 | &start, len)) { | |
34441361 | 4502 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4503 | checked_pending_chunks = true; |
4504 | failed = 0; | |
4505 | retried = false; | |
3a45bb20 | 4506 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4507 | if (ret) |
4508 | goto done; | |
4509 | goto again; | |
4510 | } | |
4511 | } | |
4512 | ||
7cc8e58d | 4513 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4514 | if (list_empty(&device->resized_list)) |
4515 | list_add_tail(&device->resized_list, | |
0b246afa | 4516 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4517 | |
d6397bae CB |
4518 | WARN_ON(diff > old_total); |
4519 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
34441361 | 4520 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4521 | |
4522 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4523 | ret = btrfs_update_device(trans, device); | |
3a45bb20 | 4524 | btrfs_end_transaction(trans); |
8f18cf13 CM |
4525 | done: |
4526 | btrfs_free_path(path); | |
53e489bc | 4527 | if (ret) { |
34441361 | 4528 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4529 | btrfs_device_set_total_bytes(device, old_size); |
4530 | if (device->writeable) | |
4531 | device->fs_devices->total_rw_bytes += diff; | |
a5ed45f8 | 4532 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4533 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4534 | } |
8f18cf13 CM |
4535 | return ret; |
4536 | } | |
4537 | ||
2ff7e61e | 4538 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4539 | struct btrfs_key *key, |
4540 | struct btrfs_chunk *chunk, int item_size) | |
4541 | { | |
0b246afa | 4542 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4543 | struct btrfs_disk_key disk_key; |
4544 | u32 array_size; | |
4545 | u8 *ptr; | |
4546 | ||
34441361 | 4547 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4548 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4549 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4550 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4551 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4552 | return -EFBIG; |
fe48a5c0 | 4553 | } |
0b86a832 CM |
4554 | |
4555 | ptr = super_copy->sys_chunk_array + array_size; | |
4556 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4557 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4558 | ptr += sizeof(disk_key); | |
4559 | memcpy(ptr, chunk, item_size); | |
4560 | item_size += sizeof(disk_key); | |
4561 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4562 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4563 | |
0b86a832 CM |
4564 | return 0; |
4565 | } | |
4566 | ||
73c5de00 AJ |
4567 | /* |
4568 | * sort the devices in descending order by max_avail, total_avail | |
4569 | */ | |
4570 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4571 | { |
73c5de00 AJ |
4572 | const struct btrfs_device_info *di_a = a; |
4573 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4574 | |
73c5de00 | 4575 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4576 | return -1; |
73c5de00 | 4577 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4578 | return 1; |
73c5de00 AJ |
4579 | if (di_a->total_avail > di_b->total_avail) |
4580 | return -1; | |
4581 | if (di_a->total_avail < di_b->total_avail) | |
4582 | return 1; | |
4583 | return 0; | |
b2117a39 | 4584 | } |
0b86a832 | 4585 | |
53b381b3 DW |
4586 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4587 | { | |
4588 | /* TODO allow them to set a preferred stripe size */ | |
ee22184b | 4589 | return SZ_64K; |
53b381b3 DW |
4590 | } |
4591 | ||
4592 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4593 | { | |
ffe2d203 | 4594 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4595 | return; |
4596 | ||
ceda0864 | 4597 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4598 | } |
4599 | ||
da17066c | 4600 | #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \ |
23f8f9b7 GH |
4601 | - sizeof(struct btrfs_chunk)) \ |
4602 | / sizeof(struct btrfs_stripe) + 1) | |
4603 | ||
4604 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4605 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4606 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4607 | / sizeof(struct btrfs_stripe) + 1) | |
4608 | ||
73c5de00 | 4609 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4610 | u64 start, u64 type) |
b2117a39 | 4611 | { |
2ff7e61e | 4612 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 AJ |
4613 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
4614 | struct list_head *cur; | |
4615 | struct map_lookup *map = NULL; | |
4616 | struct extent_map_tree *em_tree; | |
4617 | struct extent_map *em; | |
4618 | struct btrfs_device_info *devices_info = NULL; | |
4619 | u64 total_avail; | |
4620 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4621 | int data_stripes; /* number of stripes that count for |
4622 | block group size */ | |
73c5de00 AJ |
4623 | int sub_stripes; /* sub_stripes info for map */ |
4624 | int dev_stripes; /* stripes per dev */ | |
4625 | int devs_max; /* max devs to use */ | |
4626 | int devs_min; /* min devs needed */ | |
4627 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4628 | int ncopies; /* how many copies to data has */ | |
4629 | int ret; | |
4630 | u64 max_stripe_size; | |
4631 | u64 max_chunk_size; | |
4632 | u64 stripe_size; | |
4633 | u64 num_bytes; | |
53b381b3 | 4634 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4635 | int ndevs; |
4636 | int i; | |
4637 | int j; | |
31e50229 | 4638 | int index; |
593060d7 | 4639 | |
0c460c0d | 4640 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4641 | |
73c5de00 AJ |
4642 | if (list_empty(&fs_devices->alloc_list)) |
4643 | return -ENOSPC; | |
b2117a39 | 4644 | |
31e50229 | 4645 | index = __get_raid_index(type); |
73c5de00 | 4646 | |
31e50229 LB |
4647 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4648 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4649 | devs_max = btrfs_raid_array[index].devs_max; | |
4650 | devs_min = btrfs_raid_array[index].devs_min; | |
4651 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4652 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4653 | |
9b3f68b9 | 4654 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4655 | max_stripe_size = SZ_1G; |
73c5de00 | 4656 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 GH |
4657 | if (!devs_max) |
4658 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4659 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4660 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4661 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4662 | max_stripe_size = SZ_1G; | |
1100373f | 4663 | else |
ee22184b | 4664 | max_stripe_size = SZ_256M; |
73c5de00 | 4665 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4666 | if (!devs_max) |
4667 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4668 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4669 | max_stripe_size = SZ_32M; |
73c5de00 | 4670 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4671 | if (!devs_max) |
4672 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4673 | } else { |
351fd353 | 4674 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4675 | type); |
4676 | BUG_ON(1); | |
9b3f68b9 CM |
4677 | } |
4678 | ||
2b82032c YZ |
4679 | /* we don't want a chunk larger than 10% of writeable space */ |
4680 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4681 | max_chunk_size); | |
9b3f68b9 | 4682 | |
31e818fe | 4683 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4684 | GFP_NOFS); |
4685 | if (!devices_info) | |
4686 | return -ENOMEM; | |
0cad8a11 | 4687 | |
73c5de00 | 4688 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4689 | |
9f680ce0 | 4690 | /* |
73c5de00 AJ |
4691 | * in the first pass through the devices list, we gather information |
4692 | * about the available holes on each device. | |
9f680ce0 | 4693 | */ |
73c5de00 AJ |
4694 | ndevs = 0; |
4695 | while (cur != &fs_devices->alloc_list) { | |
4696 | struct btrfs_device *device; | |
4697 | u64 max_avail; | |
4698 | u64 dev_offset; | |
b2117a39 | 4699 | |
73c5de00 | 4700 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4701 | |
73c5de00 | 4702 | cur = cur->next; |
b2117a39 | 4703 | |
73c5de00 | 4704 | if (!device->writeable) { |
31b1a2bd | 4705 | WARN(1, KERN_ERR |
efe120a0 | 4706 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4707 | continue; |
4708 | } | |
b2117a39 | 4709 | |
63a212ab SB |
4710 | if (!device->in_fs_metadata || |
4711 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4712 | continue; |
b2117a39 | 4713 | |
73c5de00 AJ |
4714 | if (device->total_bytes > device->bytes_used) |
4715 | total_avail = device->total_bytes - device->bytes_used; | |
4716 | else | |
4717 | total_avail = 0; | |
38c01b96 | 4718 | |
4719 | /* If there is no space on this device, skip it. */ | |
4720 | if (total_avail == 0) | |
4721 | continue; | |
b2117a39 | 4722 | |
6df9a95e | 4723 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4724 | max_stripe_size * dev_stripes, |
4725 | &dev_offset, &max_avail); | |
4726 | if (ret && ret != -ENOSPC) | |
4727 | goto error; | |
b2117a39 | 4728 | |
73c5de00 AJ |
4729 | if (ret == 0) |
4730 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4731 | |
73c5de00 AJ |
4732 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4733 | continue; | |
b2117a39 | 4734 | |
063d006f ES |
4735 | if (ndevs == fs_devices->rw_devices) { |
4736 | WARN(1, "%s: found more than %llu devices\n", | |
4737 | __func__, fs_devices->rw_devices); | |
4738 | break; | |
4739 | } | |
73c5de00 AJ |
4740 | devices_info[ndevs].dev_offset = dev_offset; |
4741 | devices_info[ndevs].max_avail = max_avail; | |
4742 | devices_info[ndevs].total_avail = total_avail; | |
4743 | devices_info[ndevs].dev = device; | |
4744 | ++ndevs; | |
4745 | } | |
b2117a39 | 4746 | |
73c5de00 AJ |
4747 | /* |
4748 | * now sort the devices by hole size / available space | |
4749 | */ | |
4750 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4751 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4752 | |
73c5de00 AJ |
4753 | /* round down to number of usable stripes */ |
4754 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4755 | |
73c5de00 AJ |
4756 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4757 | ret = -ENOSPC; | |
4758 | goto error; | |
b2117a39 | 4759 | } |
9f680ce0 | 4760 | |
73c5de00 AJ |
4761 | if (devs_max && ndevs > devs_max) |
4762 | ndevs = devs_max; | |
4763 | /* | |
4764 | * the primary goal is to maximize the number of stripes, so use as many | |
4765 | * devices as possible, even if the stripes are not maximum sized. | |
4766 | */ | |
4767 | stripe_size = devices_info[ndevs-1].max_avail; | |
4768 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4769 | |
53b381b3 DW |
4770 | /* |
4771 | * this will have to be fixed for RAID1 and RAID10 over | |
4772 | * more drives | |
4773 | */ | |
4774 | data_stripes = num_stripes / ncopies; | |
4775 | ||
53b381b3 DW |
4776 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4777 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
da17066c | 4778 | info->stripesize); |
53b381b3 DW |
4779 | data_stripes = num_stripes - 1; |
4780 | } | |
4781 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4782 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
da17066c | 4783 | info->stripesize); |
53b381b3 DW |
4784 | data_stripes = num_stripes - 2; |
4785 | } | |
86db2578 CM |
4786 | |
4787 | /* | |
4788 | * Use the number of data stripes to figure out how big this chunk | |
4789 | * is really going to be in terms of logical address space, | |
4790 | * and compare that answer with the max chunk size | |
4791 | */ | |
4792 | if (stripe_size * data_stripes > max_chunk_size) { | |
4793 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4794 | |
4795 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4796 | |
4797 | /* bump the answer up to a 16MB boundary */ | |
4798 | stripe_size = (stripe_size + mask) & ~mask; | |
4799 | ||
4800 | /* but don't go higher than the limits we found | |
4801 | * while searching for free extents | |
4802 | */ | |
4803 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4804 | stripe_size = devices_info[ndevs-1].max_avail; | |
4805 | } | |
4806 | ||
b8b93add | 4807 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4808 | |
4809 | /* align to BTRFS_STRIPE_LEN */ | |
42c61ab6 | 4810 | stripe_size = div64_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4811 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4812 | |
4813 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4814 | if (!map) { | |
4815 | ret = -ENOMEM; | |
4816 | goto error; | |
4817 | } | |
4818 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4819 | |
73c5de00 AJ |
4820 | for (i = 0; i < ndevs; ++i) { |
4821 | for (j = 0; j < dev_stripes; ++j) { | |
4822 | int s = i * dev_stripes + j; | |
4823 | map->stripes[s].dev = devices_info[i].dev; | |
4824 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4825 | j * stripe_size; | |
6324fbf3 | 4826 | } |
6324fbf3 | 4827 | } |
da17066c | 4828 | map->sector_size = info->sectorsize; |
53b381b3 DW |
4829 | map->stripe_len = raid_stripe_len; |
4830 | map->io_align = raid_stripe_len; | |
4831 | map->io_width = raid_stripe_len; | |
2b82032c | 4832 | map->type = type; |
2b82032c | 4833 | map->sub_stripes = sub_stripes; |
0b86a832 | 4834 | |
53b381b3 | 4835 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4836 | |
6bccf3ab | 4837 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4838 | |
172ddd60 | 4839 | em = alloc_extent_map(); |
2b82032c | 4840 | if (!em) { |
298a8f9c | 4841 | kfree(map); |
b2117a39 MX |
4842 | ret = -ENOMEM; |
4843 | goto error; | |
593060d7 | 4844 | } |
298a8f9c | 4845 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4846 | em->map_lookup = map; |
2b82032c | 4847 | em->start = start; |
73c5de00 | 4848 | em->len = num_bytes; |
2b82032c YZ |
4849 | em->block_start = 0; |
4850 | em->block_len = em->len; | |
6df9a95e | 4851 | em->orig_block_len = stripe_size; |
593060d7 | 4852 | |
0b246afa | 4853 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4854 | write_lock(&em_tree->lock); |
09a2a8f9 | 4855 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4856 | if (!ret) { |
4857 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
490b54d6 | 4858 | refcount_inc(&em->refs); |
6df9a95e | 4859 | } |
890871be | 4860 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4861 | if (ret) { |
4862 | free_extent_map(em); | |
1dd4602f | 4863 | goto error; |
0f5d42b2 | 4864 | } |
0b86a832 | 4865 | |
2ff7e61e | 4866 | ret = btrfs_make_block_group(trans, info, 0, type, |
04487488 JB |
4867 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, |
4868 | start, num_bytes); | |
6df9a95e JB |
4869 | if (ret) |
4870 | goto error_del_extent; | |
2b82032c | 4871 | |
7cc8e58d MX |
4872 | for (i = 0; i < map->num_stripes; i++) { |
4873 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4874 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4875 | } | |
43530c46 | 4876 | |
a5ed45f8 | 4877 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4878 | |
0f5d42b2 | 4879 | free_extent_map(em); |
0b246afa | 4880 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4881 | |
b2117a39 | 4882 | kfree(devices_info); |
2b82032c | 4883 | return 0; |
b2117a39 | 4884 | |
6df9a95e | 4885 | error_del_extent: |
0f5d42b2 JB |
4886 | write_lock(&em_tree->lock); |
4887 | remove_extent_mapping(em_tree, em); | |
4888 | write_unlock(&em_tree->lock); | |
4889 | ||
4890 | /* One for our allocation */ | |
4891 | free_extent_map(em); | |
4892 | /* One for the tree reference */ | |
4893 | free_extent_map(em); | |
495e64f4 FM |
4894 | /* One for the pending_chunks list reference */ |
4895 | free_extent_map(em); | |
b2117a39 | 4896 | error: |
b2117a39 MX |
4897 | kfree(devices_info); |
4898 | return ret; | |
2b82032c YZ |
4899 | } |
4900 | ||
6df9a95e | 4901 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
6bccf3ab | 4902 | struct btrfs_fs_info *fs_info, |
6df9a95e | 4903 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4904 | { |
6bccf3ab JM |
4905 | struct btrfs_root *extent_root = fs_info->extent_root; |
4906 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4907 | struct btrfs_key key; |
2b82032c YZ |
4908 | struct btrfs_device *device; |
4909 | struct btrfs_chunk *chunk; | |
4910 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4911 | struct extent_map *em; |
4912 | struct map_lookup *map; | |
4913 | size_t item_size; | |
4914 | u64 dev_offset; | |
4915 | u64 stripe_size; | |
4916 | int i = 0; | |
140e639f | 4917 | int ret = 0; |
2b82032c | 4918 | |
592d92ee LB |
4919 | em = get_chunk_map(fs_info, chunk_offset, chunk_size); |
4920 | if (IS_ERR(em)) | |
4921 | return PTR_ERR(em); | |
6df9a95e | 4922 | |
95617d69 | 4923 | map = em->map_lookup; |
6df9a95e JB |
4924 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4925 | stripe_size = em->orig_block_len; | |
4926 | ||
2b82032c | 4927 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4928 | if (!chunk) { |
4929 | ret = -ENOMEM; | |
4930 | goto out; | |
4931 | } | |
4932 | ||
50460e37 FM |
4933 | /* |
4934 | * Take the device list mutex to prevent races with the final phase of | |
4935 | * a device replace operation that replaces the device object associated | |
4936 | * with the map's stripes, because the device object's id can change | |
4937 | * at any time during that final phase of the device replace operation | |
4938 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4939 | */ | |
0b246afa | 4940 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
4941 | for (i = 0; i < map->num_stripes; i++) { |
4942 | device = map->stripes[i].dev; | |
4943 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4944 | |
0b86a832 | 4945 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4946 | if (ret) |
50460e37 | 4947 | break; |
6df9a95e JB |
4948 | ret = btrfs_alloc_dev_extent(trans, device, |
4949 | chunk_root->root_key.objectid, | |
4950 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4951 | chunk_offset, dev_offset, | |
4952 | stripe_size); | |
4953 | if (ret) | |
50460e37 FM |
4954 | break; |
4955 | } | |
4956 | if (ret) { | |
0b246afa | 4957 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 4958 | goto out; |
2b82032c YZ |
4959 | } |
4960 | ||
2b82032c | 4961 | stripe = &chunk->stripe; |
6df9a95e JB |
4962 | for (i = 0; i < map->num_stripes; i++) { |
4963 | device = map->stripes[i].dev; | |
4964 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4965 | |
e17cade2 CM |
4966 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4967 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4968 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4969 | stripe++; |
0b86a832 | 4970 | } |
0b246afa | 4971 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4972 | |
2b82032c | 4973 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4974 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4975 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4976 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4977 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4978 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4979 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 4980 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 4981 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4982 | |
2b82032c YZ |
4983 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4984 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4985 | key.offset = chunk_offset; | |
0b86a832 | 4986 | |
2b82032c | 4987 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4988 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4989 | /* | |
4990 | * TODO: Cleanup of inserted chunk root in case of | |
4991 | * failure. | |
4992 | */ | |
2ff7e61e | 4993 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 4994 | } |
1abe9b8a | 4995 | |
6df9a95e | 4996 | out: |
0b86a832 | 4997 | kfree(chunk); |
6df9a95e | 4998 | free_extent_map(em); |
4ed1d16e | 4999 | return ret; |
2b82032c | 5000 | } |
0b86a832 | 5001 | |
2b82032c YZ |
5002 | /* |
5003 | * Chunk allocation falls into two parts. The first part does works | |
5004 | * that make the new allocated chunk useable, but not do any operation | |
5005 | * that modifies the chunk tree. The second part does the works that | |
5006 | * require modifying the chunk tree. This division is important for the | |
5007 | * bootstrap process of adding storage to a seed btrfs. | |
5008 | */ | |
5009 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 5010 | struct btrfs_fs_info *fs_info, u64 type) |
2b82032c YZ |
5011 | { |
5012 | u64 chunk_offset; | |
2b82032c | 5013 | |
0b246afa JM |
5014 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
5015 | chunk_offset = find_next_chunk(fs_info); | |
72b468c8 | 5016 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5017 | } |
5018 | ||
d397712b | 5019 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5020 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5021 | { |
5022 | u64 chunk_offset; | |
5023 | u64 sys_chunk_offset; | |
2b82032c | 5024 | u64 alloc_profile; |
2b82032c YZ |
5025 | int ret; |
5026 | ||
6df9a95e | 5027 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5028 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5029 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5030 | if (ret) |
5031 | return ret; | |
2b82032c | 5032 | |
0b246afa | 5033 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5034 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5035 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5036 | return ret; |
2b82032c YZ |
5037 | } |
5038 | ||
d20983b4 MX |
5039 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5040 | { | |
5041 | int max_errors; | |
5042 | ||
5043 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5044 | BTRFS_BLOCK_GROUP_RAID10 | | |
5045 | BTRFS_BLOCK_GROUP_RAID5 | | |
5046 | BTRFS_BLOCK_GROUP_DUP)) { | |
5047 | max_errors = 1; | |
5048 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5049 | max_errors = 2; | |
5050 | } else { | |
5051 | max_errors = 0; | |
005d6427 | 5052 | } |
2b82032c | 5053 | |
d20983b4 | 5054 | return max_errors; |
2b82032c YZ |
5055 | } |
5056 | ||
2ff7e61e | 5057 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5058 | { |
5059 | struct extent_map *em; | |
5060 | struct map_lookup *map; | |
2b82032c | 5061 | int readonly = 0; |
d20983b4 | 5062 | int miss_ndevs = 0; |
2b82032c YZ |
5063 | int i; |
5064 | ||
592d92ee LB |
5065 | em = get_chunk_map(fs_info, chunk_offset, 1); |
5066 | if (IS_ERR(em)) | |
2b82032c YZ |
5067 | return 1; |
5068 | ||
95617d69 | 5069 | map = em->map_lookup; |
2b82032c | 5070 | for (i = 0; i < map->num_stripes; i++) { |
d20983b4 MX |
5071 | if (map->stripes[i].dev->missing) { |
5072 | miss_ndevs++; | |
5073 | continue; | |
5074 | } | |
5075 | ||
2b82032c YZ |
5076 | if (!map->stripes[i].dev->writeable) { |
5077 | readonly = 1; | |
d20983b4 | 5078 | goto end; |
2b82032c YZ |
5079 | } |
5080 | } | |
d20983b4 MX |
5081 | |
5082 | /* | |
5083 | * If the number of missing devices is larger than max errors, | |
5084 | * we can not write the data into that chunk successfully, so | |
5085 | * set it readonly. | |
5086 | */ | |
5087 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5088 | readonly = 1; | |
5089 | end: | |
0b86a832 | 5090 | free_extent_map(em); |
2b82032c | 5091 | return readonly; |
0b86a832 CM |
5092 | } |
5093 | ||
5094 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5095 | { | |
a8067e02 | 5096 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5097 | } |
5098 | ||
5099 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5100 | { | |
5101 | struct extent_map *em; | |
5102 | ||
d397712b | 5103 | while (1) { |
890871be | 5104 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5105 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5106 | if (em) | |
5107 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5108 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5109 | if (!em) |
5110 | break; | |
0b86a832 CM |
5111 | /* once for us */ |
5112 | free_extent_map(em); | |
5113 | /* once for the tree */ | |
5114 | free_extent_map(em); | |
5115 | } | |
5116 | } | |
5117 | ||
5d964051 | 5118 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5119 | { |
5120 | struct extent_map *em; | |
5121 | struct map_lookup *map; | |
f188591e CM |
5122 | int ret; |
5123 | ||
592d92ee LB |
5124 | em = get_chunk_map(fs_info, logical, len); |
5125 | if (IS_ERR(em)) | |
5126 | /* | |
5127 | * We could return errors for these cases, but that could get | |
5128 | * ugly and we'd probably do the same thing which is just not do | |
5129 | * anything else and exit, so return 1 so the callers don't try | |
5130 | * to use other copies. | |
5131 | */ | |
fb7669b5 | 5132 | return 1; |
fb7669b5 | 5133 | |
95617d69 | 5134 | map = em->map_lookup; |
f188591e CM |
5135 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5136 | ret = map->num_stripes; | |
321aecc6 CM |
5137 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5138 | ret = map->sub_stripes; | |
53b381b3 DW |
5139 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5140 | ret = 2; | |
5141 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5142 | ret = 3; | |
f188591e CM |
5143 | else |
5144 | ret = 1; | |
5145 | free_extent_map(em); | |
ad6d620e | 5146 | |
73beece9 | 5147 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
6fad823f LB |
5148 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5149 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5150 | ret++; |
73beece9 | 5151 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5152 | |
f188591e CM |
5153 | return ret; |
5154 | } | |
5155 | ||
2ff7e61e | 5156 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5157 | struct btrfs_mapping_tree *map_tree, |
5158 | u64 logical) | |
5159 | { | |
5160 | struct extent_map *em; | |
5161 | struct map_lookup *map; | |
0b246afa | 5162 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5163 | |
592d92ee | 5164 | em = get_chunk_map(fs_info, logical, len); |
539b50d2 | 5165 | WARN_ON(IS_ERR(em)); |
53b381b3 | 5166 | |
95617d69 | 5167 | map = em->map_lookup; |
ffe2d203 | 5168 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 5169 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
5170 | free_extent_map(em); |
5171 | return len; | |
5172 | } | |
5173 | ||
592d92ee | 5174 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5175 | u64 logical, u64 len, int mirror_num) |
5176 | { | |
5177 | struct extent_map *em; | |
5178 | struct map_lookup *map; | |
53b381b3 DW |
5179 | int ret = 0; |
5180 | ||
592d92ee | 5181 | em = get_chunk_map(fs_info, logical, len); |
539b50d2 | 5182 | WARN_ON(IS_ERR(em)); |
53b381b3 | 5183 | |
95617d69 | 5184 | map = em->map_lookup; |
ffe2d203 | 5185 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
5186 | ret = 1; |
5187 | free_extent_map(em); | |
5188 | return ret; | |
5189 | } | |
5190 | ||
30d9861f SB |
5191 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5192 | struct map_lookup *map, int first, int num, | |
5193 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5194 | { |
5195 | int i; | |
30d9861f SB |
5196 | int tolerance; |
5197 | struct btrfs_device *srcdev; | |
5198 | ||
5199 | if (dev_replace_is_ongoing && | |
5200 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5201 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5202 | srcdev = fs_info->dev_replace.srcdev; | |
5203 | else | |
5204 | srcdev = NULL; | |
5205 | ||
5206 | /* | |
5207 | * try to avoid the drive that is the source drive for a | |
5208 | * dev-replace procedure, only choose it if no other non-missing | |
5209 | * mirror is available | |
5210 | */ | |
5211 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5212 | if (map->stripes[optimal].dev->bdev && | |
5213 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5214 | return optimal; | |
5215 | for (i = first; i < first + num; i++) { | |
5216 | if (map->stripes[i].dev->bdev && | |
5217 | (tolerance || map->stripes[i].dev != srcdev)) | |
5218 | return i; | |
5219 | } | |
dfe25020 | 5220 | } |
30d9861f | 5221 | |
dfe25020 CM |
5222 | /* we couldn't find one that doesn't fail. Just return something |
5223 | * and the io error handling code will clean up eventually | |
5224 | */ | |
5225 | return optimal; | |
5226 | } | |
5227 | ||
53b381b3 DW |
5228 | static inline int parity_smaller(u64 a, u64 b) |
5229 | { | |
5230 | return a > b; | |
5231 | } | |
5232 | ||
5233 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5234 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5235 | { |
5236 | struct btrfs_bio_stripe s; | |
5237 | int i; | |
5238 | u64 l; | |
5239 | int again = 1; | |
5240 | ||
5241 | while (again) { | |
5242 | again = 0; | |
cc7539ed | 5243 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5244 | if (parity_smaller(bbio->raid_map[i], |
5245 | bbio->raid_map[i+1])) { | |
53b381b3 | 5246 | s = bbio->stripes[i]; |
8e5cfb55 | 5247 | l = bbio->raid_map[i]; |
53b381b3 | 5248 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5249 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5250 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5251 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5252 | |
53b381b3 DW |
5253 | again = 1; |
5254 | } | |
5255 | } | |
5256 | } | |
5257 | } | |
5258 | ||
6e9606d2 ZL |
5259 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5260 | { | |
5261 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5262 | /* the size of the btrfs_bio */ |
6e9606d2 | 5263 | sizeof(struct btrfs_bio) + |
e57cf21e | 5264 | /* plus the variable array for the stripes */ |
6e9606d2 | 5265 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5266 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5267 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5268 | /* |
5269 | * plus the raid_map, which includes both the tgt dev | |
5270 | * and the stripes | |
5271 | */ | |
5272 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5273 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5274 | |
5275 | atomic_set(&bbio->error, 0); | |
140475ae | 5276 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5277 | |
5278 | return bbio; | |
5279 | } | |
5280 | ||
5281 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5282 | { | |
140475ae ER |
5283 | WARN_ON(!refcount_read(&bbio->refs)); |
5284 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5285 | } |
5286 | ||
5287 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5288 | { | |
5289 | if (!bbio) | |
5290 | return; | |
140475ae | 5291 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5292 | kfree(bbio); |
5293 | } | |
5294 | ||
0b3d4cd3 LB |
5295 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5296 | /* | |
5297 | * Please note that, discard won't be sent to target device of device | |
5298 | * replace. | |
5299 | */ | |
5300 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5301 | u64 logical, u64 length, | |
5302 | struct btrfs_bio **bbio_ret) | |
5303 | { | |
5304 | struct extent_map *em; | |
5305 | struct map_lookup *map; | |
5306 | struct btrfs_bio *bbio; | |
5307 | u64 offset; | |
5308 | u64 stripe_nr; | |
5309 | u64 stripe_nr_end; | |
5310 | u64 stripe_end_offset; | |
5311 | u64 stripe_cnt; | |
5312 | u64 stripe_len; | |
5313 | u64 stripe_offset; | |
5314 | u64 num_stripes; | |
5315 | u32 stripe_index; | |
5316 | u32 factor = 0; | |
5317 | u32 sub_stripes = 0; | |
5318 | u64 stripes_per_dev = 0; | |
5319 | u32 remaining_stripes = 0; | |
5320 | u32 last_stripe = 0; | |
5321 | int ret = 0; | |
5322 | int i; | |
5323 | ||
5324 | /* discard always return a bbio */ | |
5325 | ASSERT(bbio_ret); | |
5326 | ||
5327 | em = get_chunk_map(fs_info, logical, length); | |
5328 | if (IS_ERR(em)) | |
5329 | return PTR_ERR(em); | |
5330 | ||
5331 | map = em->map_lookup; | |
5332 | /* we don't discard raid56 yet */ | |
5333 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5334 | ret = -EOPNOTSUPP; | |
5335 | goto out; | |
5336 | } | |
5337 | ||
5338 | offset = logical - em->start; | |
5339 | length = min_t(u64, em->len - offset, length); | |
5340 | ||
5341 | stripe_len = map->stripe_len; | |
5342 | /* | |
5343 | * stripe_nr counts the total number of stripes we have to stride | |
5344 | * to get to this block | |
5345 | */ | |
5346 | stripe_nr = div64_u64(offset, stripe_len); | |
5347 | ||
5348 | /* stripe_offset is the offset of this block in its stripe */ | |
5349 | stripe_offset = offset - stripe_nr * stripe_len; | |
5350 | ||
5351 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5352 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5353 | stripe_cnt = stripe_nr_end - stripe_nr; |
5354 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5355 | (offset + length); | |
5356 | /* | |
5357 | * after this, stripe_nr is the number of stripes on this | |
5358 | * device we have to walk to find the data, and stripe_index is | |
5359 | * the number of our device in the stripe array | |
5360 | */ | |
5361 | num_stripes = 1; | |
5362 | stripe_index = 0; | |
5363 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5364 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5365 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5366 | sub_stripes = 1; | |
5367 | else | |
5368 | sub_stripes = map->sub_stripes; | |
5369 | ||
5370 | factor = map->num_stripes / sub_stripes; | |
5371 | num_stripes = min_t(u64, map->num_stripes, | |
5372 | sub_stripes * stripe_cnt); | |
5373 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5374 | stripe_index *= sub_stripes; | |
5375 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5376 | &remaining_stripes); | |
5377 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5378 | last_stripe *= sub_stripes; | |
5379 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5380 | BTRFS_BLOCK_GROUP_DUP)) { | |
5381 | num_stripes = map->num_stripes; | |
5382 | } else { | |
5383 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5384 | &stripe_index); | |
5385 | } | |
5386 | ||
5387 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5388 | if (!bbio) { | |
5389 | ret = -ENOMEM; | |
5390 | goto out; | |
5391 | } | |
5392 | ||
5393 | for (i = 0; i < num_stripes; i++) { | |
5394 | bbio->stripes[i].physical = | |
5395 | map->stripes[stripe_index].physical + | |
5396 | stripe_offset + stripe_nr * map->stripe_len; | |
5397 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5398 | ||
5399 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5400 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5401 | bbio->stripes[i].length = stripes_per_dev * | |
5402 | map->stripe_len; | |
5403 | ||
5404 | if (i / sub_stripes < remaining_stripes) | |
5405 | bbio->stripes[i].length += | |
5406 | map->stripe_len; | |
5407 | ||
5408 | /* | |
5409 | * Special for the first stripe and | |
5410 | * the last stripe: | |
5411 | * | |
5412 | * |-------|...|-------| | |
5413 | * |----------| | |
5414 | * off end_off | |
5415 | */ | |
5416 | if (i < sub_stripes) | |
5417 | bbio->stripes[i].length -= | |
5418 | stripe_offset; | |
5419 | ||
5420 | if (stripe_index >= last_stripe && | |
5421 | stripe_index <= (last_stripe + | |
5422 | sub_stripes - 1)) | |
5423 | bbio->stripes[i].length -= | |
5424 | stripe_end_offset; | |
5425 | ||
5426 | if (i == sub_stripes - 1) | |
5427 | stripe_offset = 0; | |
5428 | } else { | |
5429 | bbio->stripes[i].length = length; | |
5430 | } | |
5431 | ||
5432 | stripe_index++; | |
5433 | if (stripe_index == map->num_stripes) { | |
5434 | stripe_index = 0; | |
5435 | stripe_nr++; | |
5436 | } | |
5437 | } | |
5438 | ||
5439 | *bbio_ret = bbio; | |
5440 | bbio->map_type = map->type; | |
5441 | bbio->num_stripes = num_stripes; | |
5442 | out: | |
5443 | free_extent_map(em); | |
5444 | return ret; | |
5445 | } | |
5446 | ||
5ab56090 LB |
5447 | /* |
5448 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5449 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5450 | * left cursor can also be read from the target drive. | |
5451 | * | |
5452 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5453 | * array of stripes. | |
5454 | * For READ, it also needs to be supported using the same mirror number. | |
5455 | * | |
5456 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5457 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5458 | * case. | |
5459 | */ | |
5460 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5461 | u64 logical, u64 length, | |
5462 | u64 srcdev_devid, int *mirror_num, | |
5463 | u64 *physical) | |
5464 | { | |
5465 | struct btrfs_bio *bbio = NULL; | |
5466 | int num_stripes; | |
5467 | int index_srcdev = 0; | |
5468 | int found = 0; | |
5469 | u64 physical_of_found = 0; | |
5470 | int i; | |
5471 | int ret = 0; | |
5472 | ||
5473 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5474 | logical, &length, &bbio, 0, 0); | |
5475 | if (ret) { | |
5476 | ASSERT(bbio == NULL); | |
5477 | return ret; | |
5478 | } | |
5479 | ||
5480 | num_stripes = bbio->num_stripes; | |
5481 | if (*mirror_num > num_stripes) { | |
5482 | /* | |
5483 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5484 | * that means that the requested area is not left of the left | |
5485 | * cursor | |
5486 | */ | |
5487 | btrfs_put_bbio(bbio); | |
5488 | return -EIO; | |
5489 | } | |
5490 | ||
5491 | /* | |
5492 | * process the rest of the function using the mirror_num of the source | |
5493 | * drive. Therefore look it up first. At the end, patch the device | |
5494 | * pointer to the one of the target drive. | |
5495 | */ | |
5496 | for (i = 0; i < num_stripes; i++) { | |
5497 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5498 | continue; | |
5499 | ||
5500 | /* | |
5501 | * In case of DUP, in order to keep it simple, only add the | |
5502 | * mirror with the lowest physical address | |
5503 | */ | |
5504 | if (found && | |
5505 | physical_of_found <= bbio->stripes[i].physical) | |
5506 | continue; | |
5507 | ||
5508 | index_srcdev = i; | |
5509 | found = 1; | |
5510 | physical_of_found = bbio->stripes[i].physical; | |
5511 | } | |
5512 | ||
5513 | btrfs_put_bbio(bbio); | |
5514 | ||
5515 | ASSERT(found); | |
5516 | if (!found) | |
5517 | return -EIO; | |
5518 | ||
5519 | *mirror_num = index_srcdev + 1; | |
5520 | *physical = physical_of_found; | |
5521 | return ret; | |
5522 | } | |
5523 | ||
73c0f228 LB |
5524 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5525 | struct btrfs_bio **bbio_ret, | |
5526 | struct btrfs_dev_replace *dev_replace, | |
5527 | int *num_stripes_ret, int *max_errors_ret) | |
5528 | { | |
5529 | struct btrfs_bio *bbio = *bbio_ret; | |
5530 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5531 | int tgtdev_indexes = 0; | |
5532 | int num_stripes = *num_stripes_ret; | |
5533 | int max_errors = *max_errors_ret; | |
5534 | int i; | |
5535 | ||
5536 | if (op == BTRFS_MAP_WRITE) { | |
5537 | int index_where_to_add; | |
5538 | ||
5539 | /* | |
5540 | * duplicate the write operations while the dev replace | |
5541 | * procedure is running. Since the copying of the old disk to | |
5542 | * the new disk takes place at run time while the filesystem is | |
5543 | * mounted writable, the regular write operations to the old | |
5544 | * disk have to be duplicated to go to the new disk as well. | |
5545 | * | |
5546 | * Note that device->missing is handled by the caller, and that | |
5547 | * the write to the old disk is already set up in the stripes | |
5548 | * array. | |
5549 | */ | |
5550 | index_where_to_add = num_stripes; | |
5551 | for (i = 0; i < num_stripes; i++) { | |
5552 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5553 | /* write to new disk, too */ | |
5554 | struct btrfs_bio_stripe *new = | |
5555 | bbio->stripes + index_where_to_add; | |
5556 | struct btrfs_bio_stripe *old = | |
5557 | bbio->stripes + i; | |
5558 | ||
5559 | new->physical = old->physical; | |
5560 | new->length = old->length; | |
5561 | new->dev = dev_replace->tgtdev; | |
5562 | bbio->tgtdev_map[i] = index_where_to_add; | |
5563 | index_where_to_add++; | |
5564 | max_errors++; | |
5565 | tgtdev_indexes++; | |
5566 | } | |
5567 | } | |
5568 | num_stripes = index_where_to_add; | |
5569 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5570 | int index_srcdev = 0; | |
5571 | int found = 0; | |
5572 | u64 physical_of_found = 0; | |
5573 | ||
5574 | /* | |
5575 | * During the dev-replace procedure, the target drive can also | |
5576 | * be used to read data in case it is needed to repair a corrupt | |
5577 | * block elsewhere. This is possible if the requested area is | |
5578 | * left of the left cursor. In this area, the target drive is a | |
5579 | * full copy of the source drive. | |
5580 | */ | |
5581 | for (i = 0; i < num_stripes; i++) { | |
5582 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5583 | /* | |
5584 | * In case of DUP, in order to keep it simple, | |
5585 | * only add the mirror with the lowest physical | |
5586 | * address | |
5587 | */ | |
5588 | if (found && | |
5589 | physical_of_found <= | |
5590 | bbio->stripes[i].physical) | |
5591 | continue; | |
5592 | index_srcdev = i; | |
5593 | found = 1; | |
5594 | physical_of_found = bbio->stripes[i].physical; | |
5595 | } | |
5596 | } | |
5597 | if (found) { | |
5598 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5599 | bbio->stripes + num_stripes; | |
5600 | ||
5601 | tgtdev_stripe->physical = physical_of_found; | |
5602 | tgtdev_stripe->length = | |
5603 | bbio->stripes[index_srcdev].length; | |
5604 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5605 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5606 | ||
5607 | tgtdev_indexes++; | |
5608 | num_stripes++; | |
5609 | } | |
5610 | } | |
5611 | ||
5612 | *num_stripes_ret = num_stripes; | |
5613 | *max_errors_ret = max_errors; | |
5614 | bbio->num_tgtdevs = tgtdev_indexes; | |
5615 | *bbio_ret = bbio; | |
5616 | } | |
5617 | ||
2b19a1fe LB |
5618 | static bool need_full_stripe(enum btrfs_map_op op) |
5619 | { | |
5620 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5621 | } | |
5622 | ||
cf8cddd3 CH |
5623 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5624 | enum btrfs_map_op op, | |
f2d8d74d | 5625 | u64 logical, u64 *length, |
a1d3c478 | 5626 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5627 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5628 | { |
5629 | struct extent_map *em; | |
5630 | struct map_lookup *map; | |
0b86a832 | 5631 | u64 offset; |
593060d7 CM |
5632 | u64 stripe_offset; |
5633 | u64 stripe_nr; | |
53b381b3 | 5634 | u64 stripe_len; |
9d644a62 | 5635 | u32 stripe_index; |
cea9e445 | 5636 | int i; |
de11cc12 | 5637 | int ret = 0; |
f2d8d74d | 5638 | int num_stripes; |
a236aed1 | 5639 | int max_errors = 0; |
2c8cdd6e | 5640 | int tgtdev_indexes = 0; |
a1d3c478 | 5641 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5642 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5643 | int dev_replace_is_ongoing = 0; | |
5644 | int num_alloc_stripes; | |
ad6d620e SB |
5645 | int patch_the_first_stripe_for_dev_replace = 0; |
5646 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5647 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5648 | |
0b3d4cd3 LB |
5649 | if (op == BTRFS_MAP_DISCARD) |
5650 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5651 | *length, bbio_ret); | |
5652 | ||
592d92ee LB |
5653 | em = get_chunk_map(fs_info, logical, *length); |
5654 | if (IS_ERR(em)) | |
5655 | return PTR_ERR(em); | |
0b86a832 | 5656 | |
95617d69 | 5657 | map = em->map_lookup; |
0b86a832 | 5658 | offset = logical - em->start; |
593060d7 | 5659 | |
53b381b3 | 5660 | stripe_len = map->stripe_len; |
593060d7 CM |
5661 | stripe_nr = offset; |
5662 | /* | |
5663 | * stripe_nr counts the total number of stripes we have to stride | |
5664 | * to get to this block | |
5665 | */ | |
47c5713f | 5666 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5667 | |
53b381b3 | 5668 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5669 | if (offset < stripe_offset) { |
5d163e0e JM |
5670 | btrfs_crit(fs_info, |
5671 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5672 | stripe_offset, offset, em->start, logical, |
5673 | stripe_len); | |
5674 | free_extent_map(em); | |
5675 | return -EINVAL; | |
5676 | } | |
593060d7 CM |
5677 | |
5678 | /* stripe_offset is the offset of this block in its stripe*/ | |
5679 | stripe_offset = offset - stripe_offset; | |
5680 | ||
53b381b3 | 5681 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5682 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5683 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5684 | raid56_full_stripe_start = offset; | |
5685 | ||
5686 | /* allow a write of a full stripe, but make sure we don't | |
5687 | * allow straddling of stripes | |
5688 | */ | |
47c5713f DS |
5689 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5690 | full_stripe_len); | |
53b381b3 DW |
5691 | raid56_full_stripe_start *= full_stripe_len; |
5692 | } | |
5693 | ||
0b3d4cd3 | 5694 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5695 | u64 max_len; |
5696 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5697 | For other RAID types and for RAID[56] reads, just allow a single | |
5698 | stripe (on a single disk). */ | |
ffe2d203 | 5699 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5700 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5701 | max_len = stripe_len * nr_data_stripes(map) - |
5702 | (offset - raid56_full_stripe_start); | |
5703 | } else { | |
5704 | /* we limit the length of each bio to what fits in a stripe */ | |
5705 | max_len = stripe_len - stripe_offset; | |
5706 | } | |
5707 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5708 | } else { |
5709 | *length = em->len - offset; | |
5710 | } | |
f2d8d74d | 5711 | |
53b381b3 DW |
5712 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5713 | it cares about is the length */ | |
a1d3c478 | 5714 | if (!bbio_ret) |
cea9e445 CM |
5715 | goto out; |
5716 | ||
73beece9 | 5717 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5718 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5719 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5720 | btrfs_dev_replace_unlock(dev_replace, 0); |
5721 | else | |
5722 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5723 | |
ad6d620e | 5724 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5725 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5726 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5727 | dev_replace->srcdev->devid, | |
5728 | &mirror_num, | |
5729 | &physical_to_patch_in_first_stripe); | |
5730 | if (ret) | |
ad6d620e | 5731 | goto out; |
5ab56090 LB |
5732 | else |
5733 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5734 | } else if (mirror_num > map->num_stripes) { |
5735 | mirror_num = 0; | |
5736 | } | |
5737 | ||
f2d8d74d | 5738 | num_stripes = 1; |
cea9e445 | 5739 | stripe_index = 0; |
fce3bb9a | 5740 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5741 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5742 | &stripe_index); | |
0b3d4cd3 | 5743 | if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS) |
28e1cc7d | 5744 | mirror_num = 1; |
fce3bb9a | 5745 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
0b3d4cd3 | 5746 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
f2d8d74d | 5747 | num_stripes = map->num_stripes; |
2fff734f | 5748 | else if (mirror_num) |
f188591e | 5749 | stripe_index = mirror_num - 1; |
dfe25020 | 5750 | else { |
30d9861f | 5751 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5752 | map->num_stripes, |
30d9861f SB |
5753 | current->pid % map->num_stripes, |
5754 | dev_replace_is_ongoing); | |
a1d3c478 | 5755 | mirror_num = stripe_index + 1; |
dfe25020 | 5756 | } |
2fff734f | 5757 | |
611f0e00 | 5758 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
0b3d4cd3 | 5759 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) { |
f2d8d74d | 5760 | num_stripes = map->num_stripes; |
a1d3c478 | 5761 | } else if (mirror_num) { |
f188591e | 5762 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5763 | } else { |
5764 | mirror_num = 1; | |
5765 | } | |
2fff734f | 5766 | |
321aecc6 | 5767 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5768 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5769 | |
47c5713f | 5770 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5771 | stripe_index *= map->sub_stripes; |
5772 | ||
cf8cddd3 | 5773 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
f2d8d74d | 5774 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5775 | else if (mirror_num) |
5776 | stripe_index += mirror_num - 1; | |
dfe25020 | 5777 | else { |
3e74317a | 5778 | int old_stripe_index = stripe_index; |
30d9861f SB |
5779 | stripe_index = find_live_mirror(fs_info, map, |
5780 | stripe_index, | |
dfe25020 | 5781 | map->sub_stripes, stripe_index + |
30d9861f SB |
5782 | current->pid % map->sub_stripes, |
5783 | dev_replace_is_ongoing); | |
3e74317a | 5784 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5785 | } |
53b381b3 | 5786 | |
ffe2d203 | 5787 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5788 | if (need_raid_map && |
cf8cddd3 | 5789 | (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS || |
af8e2d1d | 5790 | mirror_num > 1)) { |
53b381b3 | 5791 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5792 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5793 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5794 | |
5795 | /* RAID[56] write or recovery. Return all stripes */ | |
5796 | num_stripes = map->num_stripes; | |
5797 | max_errors = nr_parity_stripes(map); | |
5798 | ||
53b381b3 DW |
5799 | *length = map->stripe_len; |
5800 | stripe_index = 0; | |
5801 | stripe_offset = 0; | |
5802 | } else { | |
5803 | /* | |
5804 | * Mirror #0 or #1 means the original data block. | |
5805 | * Mirror #2 is RAID5 parity block. | |
5806 | * Mirror #3 is RAID6 Q block. | |
5807 | */ | |
47c5713f DS |
5808 | stripe_nr = div_u64_rem(stripe_nr, |
5809 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5810 | if (mirror_num > 1) |
5811 | stripe_index = nr_data_stripes(map) + | |
5812 | mirror_num - 2; | |
5813 | ||
5814 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5815 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5816 | &stripe_index); | |
0b3d4cd3 LB |
5817 | if ((op != BTRFS_MAP_WRITE && |
5818 | op != BTRFS_MAP_GET_READ_MIRRORS) && | |
5819 | mirror_num <= 1) | |
28e1cc7d | 5820 | mirror_num = 1; |
53b381b3 | 5821 | } |
8790d502 CM |
5822 | } else { |
5823 | /* | |
47c5713f DS |
5824 | * after this, stripe_nr is the number of stripes on this |
5825 | * device we have to walk to find the data, and stripe_index is | |
5826 | * the number of our device in the stripe array | |
8790d502 | 5827 | */ |
47c5713f DS |
5828 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5829 | &stripe_index); | |
a1d3c478 | 5830 | mirror_num = stripe_index + 1; |
8790d502 | 5831 | } |
e042d1ec | 5832 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5833 | btrfs_crit(fs_info, |
5834 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5835 | stripe_index, map->num_stripes); |
5836 | ret = -EINVAL; | |
5837 | goto out; | |
5838 | } | |
cea9e445 | 5839 | |
472262f3 | 5840 | num_alloc_stripes = num_stripes; |
6fad823f | 5841 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5842 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5843 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5844 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5845 | num_alloc_stripes++; |
2c8cdd6e | 5846 | tgtdev_indexes = num_stripes; |
ad6d620e | 5847 | } |
2c8cdd6e | 5848 | |
6e9606d2 | 5849 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5850 | if (!bbio) { |
5851 | ret = -ENOMEM; | |
5852 | goto out; | |
5853 | } | |
6fad823f | 5854 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5855 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5856 | |
8e5cfb55 | 5857 | /* build raid_map */ |
2b19a1fe LB |
5858 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5859 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5860 | u64 tmp; |
9d644a62 | 5861 | unsigned rot; |
8e5cfb55 ZL |
5862 | |
5863 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5864 | sizeof(struct btrfs_bio_stripe) * | |
5865 | num_alloc_stripes + | |
5866 | sizeof(int) * tgtdev_indexes); | |
5867 | ||
5868 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5869 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5870 | |
5871 | /* Fill in the logical address of each stripe */ | |
5872 | tmp = stripe_nr * nr_data_stripes(map); | |
5873 | for (i = 0; i < nr_data_stripes(map); i++) | |
5874 | bbio->raid_map[(i+rot) % num_stripes] = | |
5875 | em->start + (tmp + i) * map->stripe_len; | |
5876 | ||
5877 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5878 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5879 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5880 | RAID6_Q_STRIPE; | |
5881 | } | |
5882 | ||
b89203f7 | 5883 | |
0b3d4cd3 LB |
5884 | for (i = 0; i < num_stripes; i++) { |
5885 | bbio->stripes[i].physical = | |
5886 | map->stripes[stripe_index].physical + | |
5887 | stripe_offset + | |
5888 | stripe_nr * map->stripe_len; | |
5889 | bbio->stripes[i].dev = | |
5890 | map->stripes[stripe_index].dev; | |
5891 | stripe_index++; | |
593060d7 | 5892 | } |
de11cc12 | 5893 | |
2b19a1fe | 5894 | if (need_full_stripe(op)) |
d20983b4 | 5895 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5896 | |
8e5cfb55 ZL |
5897 | if (bbio->raid_map) |
5898 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5899 | |
73c0f228 | 5900 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5901 | need_full_stripe(op)) { |
73c0f228 LB |
5902 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5903 | &max_errors); | |
472262f3 SB |
5904 | } |
5905 | ||
de11cc12 | 5906 | *bbio_ret = bbio; |
10f11900 | 5907 | bbio->map_type = map->type; |
de11cc12 LZ |
5908 | bbio->num_stripes = num_stripes; |
5909 | bbio->max_errors = max_errors; | |
5910 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5911 | |
5912 | /* | |
5913 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5914 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5915 | * available as a mirror | |
5916 | */ | |
5917 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5918 | WARN_ON(num_stripes > 1); | |
5919 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5920 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5921 | bbio->mirror_num = map->num_stripes + 1; | |
5922 | } | |
cea9e445 | 5923 | out: |
73beece9 LB |
5924 | if (dev_replace_is_ongoing) { |
5925 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5926 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5927 | } | |
0b86a832 | 5928 | free_extent_map(em); |
de11cc12 | 5929 | return ret; |
0b86a832 CM |
5930 | } |
5931 | ||
cf8cddd3 | 5932 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 5933 | u64 logical, u64 *length, |
a1d3c478 | 5934 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5935 | { |
b3d3fa51 | 5936 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5937 | mirror_num, 0); |
f2d8d74d CM |
5938 | } |
5939 | ||
af8e2d1d | 5940 | /* For Scrub/replace */ |
cf8cddd3 | 5941 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 5942 | u64 logical, u64 *length, |
825ad4c9 | 5943 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 5944 | { |
825ad4c9 | 5945 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
5946 | } |
5947 | ||
ab8d0fc4 | 5948 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
5949 | u64 chunk_start, u64 physical, u64 devid, |
5950 | u64 **logical, int *naddrs, int *stripe_len) | |
5951 | { | |
a512bbf8 YZ |
5952 | struct extent_map *em; |
5953 | struct map_lookup *map; | |
5954 | u64 *buf; | |
5955 | u64 bytenr; | |
5956 | u64 length; | |
5957 | u64 stripe_nr; | |
53b381b3 | 5958 | u64 rmap_len; |
a512bbf8 YZ |
5959 | int i, j, nr = 0; |
5960 | ||
592d92ee LB |
5961 | em = get_chunk_map(fs_info, chunk_start, 1); |
5962 | if (IS_ERR(em)) | |
835d974f | 5963 | return -EIO; |
835d974f | 5964 | |
95617d69 | 5965 | map = em->map_lookup; |
a512bbf8 | 5966 | length = em->len; |
53b381b3 DW |
5967 | rmap_len = map->stripe_len; |
5968 | ||
a512bbf8 | 5969 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5970 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5971 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5972 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5973 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5974 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5975 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5976 | } | |
a512bbf8 | 5977 | |
31e818fe | 5978 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5979 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5980 | |
5981 | for (i = 0; i < map->num_stripes; i++) { | |
5982 | if (devid && map->stripes[i].dev->devid != devid) | |
5983 | continue; | |
5984 | if (map->stripes[i].physical > physical || | |
5985 | map->stripes[i].physical + length <= physical) | |
5986 | continue; | |
5987 | ||
5988 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 5989 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5990 | |
5991 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5992 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5993 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5994 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5995 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5996 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5997 | * Alternatively, just use rmap_len below instead of | |
5998 | * map->stripe_len */ | |
5999 | ||
6000 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6001 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6002 | for (j = 0; j < nr; j++) { |
6003 | if (buf[j] == bytenr) | |
6004 | break; | |
6005 | } | |
934d375b CM |
6006 | if (j == nr) { |
6007 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6008 | buf[nr++] = bytenr; |
934d375b | 6009 | } |
a512bbf8 YZ |
6010 | } |
6011 | ||
a512bbf8 YZ |
6012 | *logical = buf; |
6013 | *naddrs = nr; | |
53b381b3 | 6014 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6015 | |
6016 | free_extent_map(em); | |
6017 | return 0; | |
f2d8d74d CM |
6018 | } |
6019 | ||
4246a0b6 | 6020 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6021 | { |
326e1dbb MS |
6022 | bio->bi_private = bbio->private; |
6023 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6024 | bio_endio(bio); |
326e1dbb | 6025 | |
6e9606d2 | 6026 | btrfs_put_bbio(bbio); |
8408c716 MX |
6027 | } |
6028 | ||
4246a0b6 | 6029 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6030 | { |
9be3395b | 6031 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6032 | int is_orig_bio = 0; |
8790d502 | 6033 | |
4246a0b6 | 6034 | if (bio->bi_error) { |
a1d3c478 | 6035 | atomic_inc(&bbio->error); |
4246a0b6 | 6036 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { |
442a4f63 | 6037 | unsigned int stripe_index = |
9be3395b | 6038 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6039 | struct btrfs_device *dev; |
442a4f63 SB |
6040 | |
6041 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6042 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6043 | if (dev->bdev) { |
37226b21 | 6044 | if (bio_op(bio) == REQ_OP_WRITE) |
597a60fa SB |
6045 | btrfs_dev_stat_inc(dev, |
6046 | BTRFS_DEV_STAT_WRITE_ERRS); | |
6047 | else | |
6048 | btrfs_dev_stat_inc(dev, | |
6049 | BTRFS_DEV_STAT_READ_ERRS); | |
70fd7614 | 6050 | if (bio->bi_opf & REQ_PREFLUSH) |
597a60fa SB |
6051 | btrfs_dev_stat_inc(dev, |
6052 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
6053 | btrfs_dev_stat_print_on_error(dev); | |
6054 | } | |
442a4f63 SB |
6055 | } |
6056 | } | |
8790d502 | 6057 | |
a1d3c478 | 6058 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6059 | is_orig_bio = 1; |
6060 | ||
c404e0dc MX |
6061 | btrfs_bio_counter_dec(bbio->fs_info); |
6062 | ||
a1d3c478 | 6063 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6064 | if (!is_orig_bio) { |
6065 | bio_put(bio); | |
a1d3c478 | 6066 | bio = bbio->orig_bio; |
7d2b4daa | 6067 | } |
c7b22bb1 | 6068 | |
9be3395b | 6069 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6070 | /* only send an error to the higher layers if it is |
53b381b3 | 6071 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6072 | */ |
a1d3c478 | 6073 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4246a0b6 | 6074 | bio->bi_error = -EIO; |
5dbc8fca | 6075 | } else { |
1259ab75 CM |
6076 | /* |
6077 | * this bio is actually up to date, we didn't | |
6078 | * go over the max number of errors | |
6079 | */ | |
4246a0b6 | 6080 | bio->bi_error = 0; |
1259ab75 | 6081 | } |
c55f1396 | 6082 | |
4246a0b6 | 6083 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6084 | } else if (!is_orig_bio) { |
8790d502 CM |
6085 | bio_put(bio); |
6086 | } | |
8790d502 CM |
6087 | } |
6088 | ||
8b712842 CM |
6089 | /* |
6090 | * see run_scheduled_bios for a description of why bios are collected for | |
6091 | * async submit. | |
6092 | * | |
6093 | * This will add one bio to the pending list for a device and make sure | |
6094 | * the work struct is scheduled. | |
6095 | */ | |
2ff7e61e | 6096 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6097 | struct bio *bio) |
8b712842 | 6098 | { |
0b246afa | 6099 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6100 | int should_queue = 1; |
ffbd517d | 6101 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6102 | |
53b381b3 | 6103 | if (device->missing || !device->bdev) { |
4246a0b6 | 6104 | bio_io_error(bio); |
53b381b3 DW |
6105 | return; |
6106 | } | |
6107 | ||
8b712842 | 6108 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6109 | if (bio_op(bio) == REQ_OP_READ) { |
492bb6de | 6110 | bio_get(bio); |
4e49ea4a | 6111 | btrfsic_submit_bio(bio); |
492bb6de | 6112 | bio_put(bio); |
143bede5 | 6113 | return; |
8b712842 CM |
6114 | } |
6115 | ||
6116 | /* | |
0986fe9e | 6117 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
6118 | * higher layers. Otherwise, the async bio makes it appear we have |
6119 | * made progress against dirty pages when we've really just put it | |
6120 | * on a queue for later | |
6121 | */ | |
0b246afa | 6122 | atomic_inc(&fs_info->nr_async_bios); |
492bb6de | 6123 | WARN_ON(bio->bi_next); |
8b712842 | 6124 | bio->bi_next = NULL; |
8b712842 CM |
6125 | |
6126 | spin_lock(&device->io_lock); | |
67f055c7 | 6127 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6128 | pending_bios = &device->pending_sync_bios; |
6129 | else | |
6130 | pending_bios = &device->pending_bios; | |
8b712842 | 6131 | |
ffbd517d CM |
6132 | if (pending_bios->tail) |
6133 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6134 | |
ffbd517d CM |
6135 | pending_bios->tail = bio; |
6136 | if (!pending_bios->head) | |
6137 | pending_bios->head = bio; | |
8b712842 CM |
6138 | if (device->running_pending) |
6139 | should_queue = 0; | |
6140 | ||
6141 | spin_unlock(&device->io_lock); | |
6142 | ||
6143 | if (should_queue) | |
0b246afa | 6144 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6145 | } |
6146 | ||
2ff7e61e JM |
6147 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6148 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6149 | { |
6150 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6151 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6152 | |
6153 | bio->bi_private = bbio; | |
9be3395b | 6154 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6155 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6156 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6157 | #ifdef DEBUG |
6158 | { | |
6159 | struct rcu_string *name; | |
6160 | ||
6161 | rcu_read_lock(); | |
6162 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6163 | btrfs_debug(fs_info, |
6164 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6165 | bio_op(bio), bio->bi_opf, | |
6166 | (u64)bio->bi_iter.bi_sector, | |
6167 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6168 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6169 | rcu_read_unlock(); |
6170 | } | |
6171 | #endif | |
6172 | bio->bi_bdev = dev->bdev; | |
c404e0dc | 6173 | |
2ff7e61e | 6174 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6175 | |
de1ee92a | 6176 | if (async) |
2ff7e61e | 6177 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6178 | else |
4e49ea4a | 6179 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6180 | } |
6181 | ||
de1ee92a JB |
6182 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6183 | { | |
6184 | atomic_inc(&bbio->error); | |
6185 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6186 | /* Should be the original bio. */ |
8408c716 MX |
6187 | WARN_ON(bio != bbio->orig_bio); |
6188 | ||
9be3395b | 6189 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6190 | bio->bi_iter.bi_sector = logical >> 9; |
4246a0b6 CH |
6191 | bio->bi_error = -EIO; |
6192 | btrfs_end_bbio(bbio, bio); | |
de1ee92a JB |
6193 | } |
6194 | } | |
6195 | ||
2ff7e61e | 6196 | int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
8b712842 | 6197 | int mirror_num, int async_submit) |
0b86a832 | 6198 | { |
0b86a832 | 6199 | struct btrfs_device *dev; |
8790d502 | 6200 | struct bio *first_bio = bio; |
4f024f37 | 6201 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6202 | u64 length = 0; |
6203 | u64 map_length; | |
0b86a832 | 6204 | int ret; |
08da757d ZL |
6205 | int dev_nr; |
6206 | int total_devs; | |
a1d3c478 | 6207 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6208 | |
4f024f37 | 6209 | length = bio->bi_iter.bi_size; |
0b86a832 | 6210 | map_length = length; |
cea9e445 | 6211 | |
0b246afa JM |
6212 | btrfs_bio_counter_inc_blocked(fs_info); |
6213 | ret = __btrfs_map_block(fs_info, bio_op(bio), logical, | |
37226b21 | 6214 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6215 | if (ret) { |
0b246afa | 6216 | btrfs_bio_counter_dec(fs_info); |
79787eaa | 6217 | return ret; |
c404e0dc | 6218 | } |
cea9e445 | 6219 | |
a1d3c478 | 6220 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6221 | bbio->orig_bio = first_bio; |
6222 | bbio->private = first_bio->bi_private; | |
6223 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6224 | bbio->fs_info = fs_info; |
53b381b3 DW |
6225 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6226 | ||
ad1ba2a0 | 6227 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6228 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6229 | /* In this case, map_length has been set to the length of |
6230 | a single stripe; not the whole write */ | |
37226b21 | 6231 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6232 | ret = raid56_parity_write(fs_info, bio, bbio, |
6233 | map_length); | |
53b381b3 | 6234 | } else { |
2ff7e61e JM |
6235 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6236 | map_length, mirror_num, 1); | |
53b381b3 | 6237 | } |
4245215d | 6238 | |
0b246afa | 6239 | btrfs_bio_counter_dec(fs_info); |
c404e0dc | 6240 | return ret; |
53b381b3 DW |
6241 | } |
6242 | ||
cea9e445 | 6243 | if (map_length < length) { |
0b246afa | 6244 | btrfs_crit(fs_info, |
5d163e0e JM |
6245 | "mapping failed logical %llu bio len %llu len %llu", |
6246 | logical, length, map_length); | |
cea9e445 CM |
6247 | BUG(); |
6248 | } | |
a1d3c478 | 6249 | |
08da757d | 6250 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6251 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6252 | if (!dev || !dev->bdev || |
a967efb3 | 6253 | (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) { |
de1ee92a | 6254 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6255 | continue; |
6256 | } | |
6257 | ||
3aa8e074 | 6258 | if (dev_nr < total_devs - 1) |
9be3395b | 6259 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
3aa8e074 | 6260 | else |
a1d3c478 | 6261 | bio = first_bio; |
de1ee92a | 6262 | |
2ff7e61e JM |
6263 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6264 | dev_nr, async_submit); | |
8790d502 | 6265 | } |
0b246afa | 6266 | btrfs_bio_counter_dec(fs_info); |
0b86a832 CM |
6267 | return 0; |
6268 | } | |
6269 | ||
aa1b8cd4 | 6270 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6271 | u8 *uuid, u8 *fsid) |
0b86a832 | 6272 | { |
2b82032c YZ |
6273 | struct btrfs_device *device; |
6274 | struct btrfs_fs_devices *cur_devices; | |
6275 | ||
aa1b8cd4 | 6276 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6277 | while (cur_devices) { |
6278 | if (!fsid || | |
6279 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6280 | device = __find_device(&cur_devices->devices, | |
6281 | devid, uuid); | |
6282 | if (device) | |
6283 | return device; | |
6284 | } | |
6285 | cur_devices = cur_devices->seed; | |
6286 | } | |
6287 | return NULL; | |
0b86a832 CM |
6288 | } |
6289 | ||
2ff7e61e | 6290 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6291 | u64 devid, u8 *dev_uuid) |
6292 | { | |
6293 | struct btrfs_device *device; | |
dfe25020 | 6294 | |
12bd2fc0 ID |
6295 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6296 | if (IS_ERR(device)) | |
7cbd8a83 | 6297 | return NULL; |
12bd2fc0 ID |
6298 | |
6299 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6300 | device->fs_devices = fs_devices; |
dfe25020 | 6301 | fs_devices->num_devices++; |
12bd2fc0 ID |
6302 | |
6303 | device->missing = 1; | |
cd02dca5 | 6304 | fs_devices->missing_devices++; |
12bd2fc0 | 6305 | |
dfe25020 CM |
6306 | return device; |
6307 | } | |
6308 | ||
12bd2fc0 ID |
6309 | /** |
6310 | * btrfs_alloc_device - allocate struct btrfs_device | |
6311 | * @fs_info: used only for generating a new devid, can be NULL if | |
6312 | * devid is provided (i.e. @devid != NULL). | |
6313 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6314 | * is generated. | |
6315 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6316 | * is generated. | |
6317 | * | |
6318 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6319 | * on error. Returned struct is not linked onto any lists and can be | |
6320 | * destroyed with kfree() right away. | |
6321 | */ | |
6322 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6323 | const u64 *devid, | |
6324 | const u8 *uuid) | |
6325 | { | |
6326 | struct btrfs_device *dev; | |
6327 | u64 tmp; | |
6328 | ||
fae7f21c | 6329 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6330 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6331 | |
6332 | dev = __alloc_device(); | |
6333 | if (IS_ERR(dev)) | |
6334 | return dev; | |
6335 | ||
6336 | if (devid) | |
6337 | tmp = *devid; | |
6338 | else { | |
6339 | int ret; | |
6340 | ||
6341 | ret = find_next_devid(fs_info, &tmp); | |
6342 | if (ret) { | |
6343 | kfree(dev); | |
6344 | return ERR_PTR(ret); | |
6345 | } | |
6346 | } | |
6347 | dev->devid = tmp; | |
6348 | ||
6349 | if (uuid) | |
6350 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6351 | else | |
6352 | generate_random_uuid(dev->uuid); | |
6353 | ||
9e0af237 LB |
6354 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6355 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6356 | |
6357 | return dev; | |
6358 | } | |
6359 | ||
e06cd3dd | 6360 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6361 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6362 | struct extent_buffer *leaf, |
6363 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6364 | { |
0b86a832 | 6365 | u64 length; |
f04b772b | 6366 | u64 stripe_len; |
e06cd3dd LB |
6367 | u16 num_stripes; |
6368 | u16 sub_stripes; | |
6369 | u64 type; | |
0b86a832 | 6370 | |
e17cade2 | 6371 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6372 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6373 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6374 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6375 | type = btrfs_chunk_type(leaf, chunk); | |
6376 | ||
f04b772b | 6377 | if (!num_stripes) { |
0b246afa | 6378 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6379 | num_stripes); |
6380 | return -EIO; | |
6381 | } | |
0b246afa JM |
6382 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6383 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6384 | return -EIO; |
6385 | } | |
0b246afa JM |
6386 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6387 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6388 | btrfs_chunk_sector_size(leaf, chunk)); |
6389 | return -EIO; | |
6390 | } | |
0b246afa JM |
6391 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6392 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6393 | return -EIO; |
6394 | } | |
3d8da678 | 6395 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6396 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6397 | stripe_len); |
6398 | return -EIO; | |
6399 | } | |
6400 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6401 | type) { |
0b246afa | 6402 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6403 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6404 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6405 | btrfs_chunk_type(leaf, chunk)); | |
6406 | return -EIO; | |
6407 | } | |
e06cd3dd LB |
6408 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6409 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6410 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6411 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6412 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6413 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6414 | num_stripes != 1)) { | |
0b246afa | 6415 | btrfs_err(fs_info, |
e06cd3dd LB |
6416 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6417 | num_stripes, sub_stripes, | |
6418 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6419 | return -EIO; | |
6420 | } | |
6421 | ||
6422 | return 0; | |
6423 | } | |
6424 | ||
2ff7e61e | 6425 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6426 | struct extent_buffer *leaf, |
6427 | struct btrfs_chunk *chunk) | |
6428 | { | |
0b246afa | 6429 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6430 | struct map_lookup *map; |
6431 | struct extent_map *em; | |
6432 | u64 logical; | |
6433 | u64 length; | |
6434 | u64 stripe_len; | |
6435 | u64 devid; | |
6436 | u8 uuid[BTRFS_UUID_SIZE]; | |
6437 | int num_stripes; | |
6438 | int ret; | |
6439 | int i; | |
6440 | ||
6441 | logical = key->offset; | |
6442 | length = btrfs_chunk_length(leaf, chunk); | |
6443 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6444 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
6445 | ||
2ff7e61e | 6446 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6447 | if (ret) |
6448 | return ret; | |
a061fc8d | 6449 | |
890871be | 6450 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6451 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6452 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6453 | |
6454 | /* already mapped? */ | |
6455 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6456 | free_extent_map(em); | |
0b86a832 CM |
6457 | return 0; |
6458 | } else if (em) { | |
6459 | free_extent_map(em); | |
6460 | } | |
0b86a832 | 6461 | |
172ddd60 | 6462 | em = alloc_extent_map(); |
0b86a832 CM |
6463 | if (!em) |
6464 | return -ENOMEM; | |
593060d7 | 6465 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6466 | if (!map) { |
6467 | free_extent_map(em); | |
6468 | return -ENOMEM; | |
6469 | } | |
6470 | ||
298a8f9c | 6471 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6472 | em->map_lookup = map; |
0b86a832 CM |
6473 | em->start = logical; |
6474 | em->len = length; | |
70c8a91c | 6475 | em->orig_start = 0; |
0b86a832 | 6476 | em->block_start = 0; |
c8b97818 | 6477 | em->block_len = em->len; |
0b86a832 | 6478 | |
593060d7 CM |
6479 | map->num_stripes = num_stripes; |
6480 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6481 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6482 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6483 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6484 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6485 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6486 | for (i = 0; i < num_stripes; i++) { |
6487 | map->stripes[i].physical = | |
6488 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6489 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6490 | read_extent_buffer(leaf, uuid, (unsigned long) |
6491 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6492 | BTRFS_UUID_SIZE); | |
0b246afa | 6493 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6494 | uuid, NULL); |
3cdde224 | 6495 | if (!map->stripes[i].dev && |
0b246afa | 6496 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 CM |
6497 | free_extent_map(em); |
6498 | return -EIO; | |
6499 | } | |
dfe25020 CM |
6500 | if (!map->stripes[i].dev) { |
6501 | map->stripes[i].dev = | |
2ff7e61e JM |
6502 | add_missing_dev(fs_info->fs_devices, devid, |
6503 | uuid); | |
dfe25020 | 6504 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6505 | free_extent_map(em); |
6506 | return -EIO; | |
6507 | } | |
0b246afa | 6508 | btrfs_warn(fs_info, "devid %llu uuid %pU is missing", |
5d163e0e | 6509 | devid, uuid); |
dfe25020 CM |
6510 | } |
6511 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6512 | } |
6513 | ||
890871be | 6514 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6515 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6516 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6517 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6518 | free_extent_map(em); |
6519 | ||
6520 | return 0; | |
6521 | } | |
6522 | ||
143bede5 | 6523 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6524 | struct btrfs_dev_item *dev_item, |
6525 | struct btrfs_device *device) | |
6526 | { | |
6527 | unsigned long ptr; | |
0b86a832 CM |
6528 | |
6529 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6530 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6531 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6532 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6533 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6534 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6535 | device->type = btrfs_device_type(leaf, dev_item); |
6536 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6537 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6538 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6539 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6540 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6541 | |
410ba3a2 | 6542 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6543 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6544 | } |
6545 | ||
2ff7e61e | 6546 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6547 | u8 *fsid) |
2b82032c YZ |
6548 | { |
6549 | struct btrfs_fs_devices *fs_devices; | |
6550 | int ret; | |
6551 | ||
b367e47f | 6552 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c | 6553 | |
0b246afa | 6554 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6555 | while (fs_devices) { |
5f375835 MX |
6556 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6557 | return fs_devices; | |
6558 | ||
2b82032c YZ |
6559 | fs_devices = fs_devices->seed; |
6560 | } | |
6561 | ||
6562 | fs_devices = find_fsid(fsid); | |
6563 | if (!fs_devices) { | |
0b246afa | 6564 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6565 | return ERR_PTR(-ENOENT); |
6566 | ||
6567 | fs_devices = alloc_fs_devices(fsid); | |
6568 | if (IS_ERR(fs_devices)) | |
6569 | return fs_devices; | |
6570 | ||
6571 | fs_devices->seeding = 1; | |
6572 | fs_devices->opened = 1; | |
6573 | return fs_devices; | |
2b82032c | 6574 | } |
e4404d6e YZ |
6575 | |
6576 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6577 | if (IS_ERR(fs_devices)) |
6578 | return fs_devices; | |
2b82032c | 6579 | |
97288f2c | 6580 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
0b246afa | 6581 | fs_info->bdev_holder); |
48d28232 JL |
6582 | if (ret) { |
6583 | free_fs_devices(fs_devices); | |
5f375835 | 6584 | fs_devices = ERR_PTR(ret); |
2b82032c | 6585 | goto out; |
48d28232 | 6586 | } |
2b82032c YZ |
6587 | |
6588 | if (!fs_devices->seeding) { | |
6589 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6590 | free_fs_devices(fs_devices); |
5f375835 | 6591 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6592 | goto out; |
6593 | } | |
6594 | ||
0b246afa JM |
6595 | fs_devices->seed = fs_info->fs_devices->seed; |
6596 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6597 | out: |
5f375835 | 6598 | return fs_devices; |
2b82032c YZ |
6599 | } |
6600 | ||
2ff7e61e | 6601 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6602 | struct extent_buffer *leaf, |
6603 | struct btrfs_dev_item *dev_item) | |
6604 | { | |
0b246afa | 6605 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6606 | struct btrfs_device *device; |
6607 | u64 devid; | |
6608 | int ret; | |
2b82032c | 6609 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6610 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6611 | ||
0b86a832 | 6612 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6613 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6614 | BTRFS_UUID_SIZE); |
1473b24e | 6615 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6616 | BTRFS_UUID_SIZE); |
6617 | ||
0b246afa | 6618 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) { |
2ff7e61e | 6619 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6620 | if (IS_ERR(fs_devices)) |
6621 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6622 | } |
6623 | ||
0b246afa | 6624 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6625 | if (!device) { |
0b246afa | 6626 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
2b82032c YZ |
6627 | return -EIO; |
6628 | ||
2ff7e61e | 6629 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
5f375835 MX |
6630 | if (!device) |
6631 | return -ENOMEM; | |
0b246afa | 6632 | btrfs_warn(fs_info, "devid %llu uuid %pU missing", |
33b97e43 | 6633 | devid, dev_uuid); |
5f375835 | 6634 | } else { |
0b246afa | 6635 | if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6636 | return -EIO; |
6637 | ||
6638 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6639 | /* |
6640 | * this happens when a device that was properly setup | |
6641 | * in the device info lists suddenly goes bad. | |
6642 | * device->bdev is NULL, and so we have to set | |
6643 | * device->missing to one here | |
6644 | */ | |
5f375835 | 6645 | device->fs_devices->missing_devices++; |
cd02dca5 | 6646 | device->missing = 1; |
2b82032c | 6647 | } |
5f375835 MX |
6648 | |
6649 | /* Move the device to its own fs_devices */ | |
6650 | if (device->fs_devices != fs_devices) { | |
6651 | ASSERT(device->missing); | |
6652 | ||
6653 | list_move(&device->dev_list, &fs_devices->devices); | |
6654 | device->fs_devices->num_devices--; | |
6655 | fs_devices->num_devices++; | |
6656 | ||
6657 | device->fs_devices->missing_devices--; | |
6658 | fs_devices->missing_devices++; | |
6659 | ||
6660 | device->fs_devices = fs_devices; | |
6661 | } | |
2b82032c YZ |
6662 | } |
6663 | ||
0b246afa | 6664 | if (device->fs_devices != fs_info->fs_devices) { |
2b82032c YZ |
6665 | BUG_ON(device->writeable); |
6666 | if (device->generation != | |
6667 | btrfs_device_generation(leaf, dev_item)) | |
6668 | return -EINVAL; | |
6324fbf3 | 6669 | } |
0b86a832 CM |
6670 | |
6671 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6672 | device->in_fs_metadata = 1; |
63a212ab | 6673 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6674 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6675 | atomic64_add(device->total_bytes - device->bytes_used, |
6676 | &fs_info->free_chunk_space); | |
2bf64758 | 6677 | } |
0b86a832 | 6678 | ret = 0; |
0b86a832 CM |
6679 | return ret; |
6680 | } | |
6681 | ||
6bccf3ab | 6682 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6683 | { |
6bccf3ab | 6684 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6685 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6686 | struct extent_buffer *sb; |
0b86a832 | 6687 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6688 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6689 | u8 *array_ptr; |
6690 | unsigned long sb_array_offset; | |
84eed90f | 6691 | int ret = 0; |
0b86a832 CM |
6692 | u32 num_stripes; |
6693 | u32 array_size; | |
6694 | u32 len = 0; | |
1ffb22cf | 6695 | u32 cur_offset; |
e06cd3dd | 6696 | u64 type; |
84eed90f | 6697 | struct btrfs_key key; |
0b86a832 | 6698 | |
0b246afa | 6699 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6700 | /* |
6701 | * This will create extent buffer of nodesize, superblock size is | |
6702 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6703 | * overallocate but we can keep it as-is, only the first page is used. | |
6704 | */ | |
2ff7e61e | 6705 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6706 | if (IS_ERR(sb)) |
6707 | return PTR_ERR(sb); | |
4db8c528 | 6708 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6709 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6710 | /* |
01327610 | 6711 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6712 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6713 | * pages up-to-date when the page is larger: extent does not cover the |
6714 | * whole page and consequently check_page_uptodate does not find all | |
6715 | * the page's extents up-to-date (the hole beyond sb), | |
6716 | * write_extent_buffer then triggers a WARN_ON. | |
6717 | * | |
6718 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6719 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6720 | * to silence the warning eg. on PowerPC 64. | |
6721 | */ | |
09cbfeaf | 6722 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6723 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6724 | |
a061fc8d | 6725 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6726 | array_size = btrfs_super_sys_array_size(super_copy); |
6727 | ||
1ffb22cf DS |
6728 | array_ptr = super_copy->sys_chunk_array; |
6729 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6730 | cur_offset = 0; | |
0b86a832 | 6731 | |
1ffb22cf DS |
6732 | while (cur_offset < array_size) { |
6733 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6734 | len = sizeof(*disk_key); |
6735 | if (cur_offset + len > array_size) | |
6736 | goto out_short_read; | |
6737 | ||
0b86a832 CM |
6738 | btrfs_disk_key_to_cpu(&key, disk_key); |
6739 | ||
1ffb22cf DS |
6740 | array_ptr += len; |
6741 | sb_array_offset += len; | |
6742 | cur_offset += len; | |
0b86a832 | 6743 | |
0d81ba5d | 6744 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6745 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6746 | /* |
6747 | * At least one btrfs_chunk with one stripe must be | |
6748 | * present, exact stripe count check comes afterwards | |
6749 | */ | |
6750 | len = btrfs_chunk_item_size(1); | |
6751 | if (cur_offset + len > array_size) | |
6752 | goto out_short_read; | |
6753 | ||
6754 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6755 | if (!num_stripes) { |
ab8d0fc4 JM |
6756 | btrfs_err(fs_info, |
6757 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6758 | num_stripes, cur_offset); |
6759 | ret = -EIO; | |
6760 | break; | |
6761 | } | |
6762 | ||
e06cd3dd LB |
6763 | type = btrfs_chunk_type(sb, chunk); |
6764 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6765 | btrfs_err(fs_info, |
e06cd3dd LB |
6766 | "invalid chunk type %llu in sys_array at offset %u", |
6767 | type, cur_offset); | |
6768 | ret = -EIO; | |
6769 | break; | |
6770 | } | |
6771 | ||
e3540eab DS |
6772 | len = btrfs_chunk_item_size(num_stripes); |
6773 | if (cur_offset + len > array_size) | |
6774 | goto out_short_read; | |
6775 | ||
2ff7e61e | 6776 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6777 | if (ret) |
6778 | break; | |
0b86a832 | 6779 | } else { |
ab8d0fc4 JM |
6780 | btrfs_err(fs_info, |
6781 | "unexpected item type %u in sys_array at offset %u", | |
6782 | (u32)key.type, cur_offset); | |
84eed90f CM |
6783 | ret = -EIO; |
6784 | break; | |
0b86a832 | 6785 | } |
1ffb22cf DS |
6786 | array_ptr += len; |
6787 | sb_array_offset += len; | |
6788 | cur_offset += len; | |
0b86a832 | 6789 | } |
d865177a | 6790 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6791 | free_extent_buffer_stale(sb); |
84eed90f | 6792 | return ret; |
e3540eab DS |
6793 | |
6794 | out_short_read: | |
ab8d0fc4 | 6795 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6796 | len, cur_offset); |
d865177a | 6797 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6798 | free_extent_buffer_stale(sb); |
e3540eab | 6799 | return -EIO; |
0b86a832 CM |
6800 | } |
6801 | ||
5b4aacef | 6802 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6803 | { |
5b4aacef | 6804 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6805 | struct btrfs_path *path; |
6806 | struct extent_buffer *leaf; | |
6807 | struct btrfs_key key; | |
6808 | struct btrfs_key found_key; | |
6809 | int ret; | |
6810 | int slot; | |
99e3ecfc | 6811 | u64 total_dev = 0; |
0b86a832 | 6812 | |
0b86a832 CM |
6813 | path = btrfs_alloc_path(); |
6814 | if (!path) | |
6815 | return -ENOMEM; | |
6816 | ||
b367e47f | 6817 | mutex_lock(&uuid_mutex); |
34441361 | 6818 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6819 | |
395927a9 FDBM |
6820 | /* |
6821 | * Read all device items, and then all the chunk items. All | |
6822 | * device items are found before any chunk item (their object id | |
6823 | * is smaller than the lowest possible object id for a chunk | |
6824 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6825 | */ |
6826 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6827 | key.offset = 0; | |
6828 | key.type = 0; | |
0b86a832 | 6829 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6830 | if (ret < 0) |
6831 | goto error; | |
d397712b | 6832 | while (1) { |
0b86a832 CM |
6833 | leaf = path->nodes[0]; |
6834 | slot = path->slots[0]; | |
6835 | if (slot >= btrfs_header_nritems(leaf)) { | |
6836 | ret = btrfs_next_leaf(root, path); | |
6837 | if (ret == 0) | |
6838 | continue; | |
6839 | if (ret < 0) | |
6840 | goto error; | |
6841 | break; | |
6842 | } | |
6843 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6844 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6845 | struct btrfs_dev_item *dev_item; | |
6846 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6847 | struct btrfs_dev_item); |
2ff7e61e | 6848 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
6849 | if (ret) |
6850 | goto error; | |
99e3ecfc | 6851 | total_dev++; |
0b86a832 CM |
6852 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6853 | struct btrfs_chunk *chunk; | |
6854 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 6855 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
6856 | if (ret) |
6857 | goto error; | |
0b86a832 CM |
6858 | } |
6859 | path->slots[0]++; | |
6860 | } | |
99e3ecfc LB |
6861 | |
6862 | /* | |
6863 | * After loading chunk tree, we've got all device information, | |
6864 | * do another round of validation checks. | |
6865 | */ | |
0b246afa JM |
6866 | if (total_dev != fs_info->fs_devices->total_devices) { |
6867 | btrfs_err(fs_info, | |
99e3ecfc | 6868 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 6869 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
6870 | total_dev); |
6871 | ret = -EINVAL; | |
6872 | goto error; | |
6873 | } | |
0b246afa JM |
6874 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
6875 | fs_info->fs_devices->total_rw_bytes) { | |
6876 | btrfs_err(fs_info, | |
99e3ecfc | 6877 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
6878 | btrfs_super_total_bytes(fs_info->super_copy), |
6879 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
6880 | ret = -EINVAL; |
6881 | goto error; | |
6882 | } | |
0b86a832 CM |
6883 | ret = 0; |
6884 | error: | |
34441361 | 6885 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
6886 | mutex_unlock(&uuid_mutex); |
6887 | ||
2b82032c | 6888 | btrfs_free_path(path); |
0b86a832 CM |
6889 | return ret; |
6890 | } | |
442a4f63 | 6891 | |
cb517eab MX |
6892 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6893 | { | |
6894 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6895 | struct btrfs_device *device; | |
6896 | ||
29cc83f6 LB |
6897 | while (fs_devices) { |
6898 | mutex_lock(&fs_devices->device_list_mutex); | |
6899 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 6900 | device->fs_info = fs_info; |
29cc83f6 LB |
6901 | mutex_unlock(&fs_devices->device_list_mutex); |
6902 | ||
6903 | fs_devices = fs_devices->seed; | |
6904 | } | |
cb517eab MX |
6905 | } |
6906 | ||
733f4fbb SB |
6907 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6908 | { | |
6909 | int i; | |
6910 | ||
6911 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6912 | btrfs_dev_stat_reset(dev, i); | |
6913 | } | |
6914 | ||
6915 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6916 | { | |
6917 | struct btrfs_key key; | |
6918 | struct btrfs_key found_key; | |
6919 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6920 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6921 | struct extent_buffer *eb; | |
6922 | int slot; | |
6923 | int ret = 0; | |
6924 | struct btrfs_device *device; | |
6925 | struct btrfs_path *path = NULL; | |
6926 | int i; | |
6927 | ||
6928 | path = btrfs_alloc_path(); | |
6929 | if (!path) { | |
6930 | ret = -ENOMEM; | |
6931 | goto out; | |
6932 | } | |
6933 | ||
6934 | mutex_lock(&fs_devices->device_list_mutex); | |
6935 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6936 | int item_size; | |
6937 | struct btrfs_dev_stats_item *ptr; | |
6938 | ||
242e2956 DS |
6939 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6940 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6941 | key.offset = device->devid; |
6942 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6943 | if (ret) { | |
733f4fbb SB |
6944 | __btrfs_reset_dev_stats(device); |
6945 | device->dev_stats_valid = 1; | |
6946 | btrfs_release_path(path); | |
6947 | continue; | |
6948 | } | |
6949 | slot = path->slots[0]; | |
6950 | eb = path->nodes[0]; | |
6951 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6952 | item_size = btrfs_item_size_nr(eb, slot); | |
6953 | ||
6954 | ptr = btrfs_item_ptr(eb, slot, | |
6955 | struct btrfs_dev_stats_item); | |
6956 | ||
6957 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6958 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6959 | btrfs_dev_stat_set(device, i, | |
6960 | btrfs_dev_stats_value(eb, ptr, i)); | |
6961 | else | |
6962 | btrfs_dev_stat_reset(device, i); | |
6963 | } | |
6964 | ||
6965 | device->dev_stats_valid = 1; | |
6966 | btrfs_dev_stat_print_on_load(device); | |
6967 | btrfs_release_path(path); | |
6968 | } | |
6969 | mutex_unlock(&fs_devices->device_list_mutex); | |
6970 | ||
6971 | out: | |
6972 | btrfs_free_path(path); | |
6973 | return ret < 0 ? ret : 0; | |
6974 | } | |
6975 | ||
6976 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6bccf3ab | 6977 | struct btrfs_fs_info *fs_info, |
733f4fbb SB |
6978 | struct btrfs_device *device) |
6979 | { | |
6bccf3ab | 6980 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
6981 | struct btrfs_path *path; |
6982 | struct btrfs_key key; | |
6983 | struct extent_buffer *eb; | |
6984 | struct btrfs_dev_stats_item *ptr; | |
6985 | int ret; | |
6986 | int i; | |
6987 | ||
242e2956 DS |
6988 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6989 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6990 | key.offset = device->devid; |
6991 | ||
6992 | path = btrfs_alloc_path(); | |
fa252992 DS |
6993 | if (!path) |
6994 | return -ENOMEM; | |
733f4fbb SB |
6995 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
6996 | if (ret < 0) { | |
0b246afa | 6997 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 6998 | "error %d while searching for dev_stats item for device %s", |
606686ee | 6999 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7000 | goto out; |
7001 | } | |
7002 | ||
7003 | if (ret == 0 && | |
7004 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7005 | /* need to delete old one and insert a new one */ | |
7006 | ret = btrfs_del_item(trans, dev_root, path); | |
7007 | if (ret != 0) { | |
0b246afa | 7008 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7009 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7010 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7011 | goto out; |
7012 | } | |
7013 | ret = 1; | |
7014 | } | |
7015 | ||
7016 | if (ret == 1) { | |
7017 | /* need to insert a new item */ | |
7018 | btrfs_release_path(path); | |
7019 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7020 | &key, sizeof(*ptr)); | |
7021 | if (ret < 0) { | |
0b246afa | 7022 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7023 | "insert dev_stats item for device %s failed %d", |
7024 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7025 | goto out; |
7026 | } | |
7027 | } | |
7028 | ||
7029 | eb = path->nodes[0]; | |
7030 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7031 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7032 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7033 | btrfs_dev_stat_read(device, i)); | |
7034 | btrfs_mark_buffer_dirty(eb); | |
7035 | ||
7036 | out: | |
7037 | btrfs_free_path(path); | |
7038 | return ret; | |
7039 | } | |
7040 | ||
7041 | /* | |
7042 | * called from commit_transaction. Writes all changed device stats to disk. | |
7043 | */ | |
7044 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7045 | struct btrfs_fs_info *fs_info) | |
7046 | { | |
733f4fbb SB |
7047 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7048 | struct btrfs_device *device; | |
addc3fa7 | 7049 | int stats_cnt; |
733f4fbb SB |
7050 | int ret = 0; |
7051 | ||
7052 | mutex_lock(&fs_devices->device_list_mutex); | |
7053 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 7054 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
7055 | continue; |
7056 | ||
addc3fa7 | 7057 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
6bccf3ab | 7058 | ret = update_dev_stat_item(trans, fs_info, device); |
733f4fbb | 7059 | if (!ret) |
addc3fa7 | 7060 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7061 | } |
7062 | mutex_unlock(&fs_devices->device_list_mutex); | |
7063 | ||
7064 | return ret; | |
7065 | } | |
7066 | ||
442a4f63 SB |
7067 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7068 | { | |
7069 | btrfs_dev_stat_inc(dev, index); | |
7070 | btrfs_dev_stat_print_on_error(dev); | |
7071 | } | |
7072 | ||
48a3b636 | 7073 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7074 | { |
733f4fbb SB |
7075 | if (!dev->dev_stats_valid) |
7076 | return; | |
fb456252 | 7077 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7078 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7079 | rcu_str_deref(dev->name), |
442a4f63 SB |
7080 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7081 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7082 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7083 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7084 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7085 | } |
c11d2c23 | 7086 | |
733f4fbb SB |
7087 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7088 | { | |
a98cdb85 SB |
7089 | int i; |
7090 | ||
7091 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7092 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7093 | break; | |
7094 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7095 | return; /* all values == 0, suppress message */ | |
7096 | ||
fb456252 | 7097 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7098 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7099 | rcu_str_deref(dev->name), |
733f4fbb SB |
7100 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7101 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7102 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7103 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7104 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7105 | } | |
7106 | ||
2ff7e61e | 7107 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7108 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7109 | { |
7110 | struct btrfs_device *dev; | |
0b246afa | 7111 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7112 | int i; |
7113 | ||
7114 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7115 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7116 | mutex_unlock(&fs_devices->device_list_mutex); |
7117 | ||
7118 | if (!dev) { | |
0b246afa | 7119 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7120 | return -ENODEV; |
733f4fbb | 7121 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7122 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7123 | return -ENODEV; |
b27f7c0c | 7124 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7125 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7126 | if (stats->nr_items > i) | |
7127 | stats->values[i] = | |
7128 | btrfs_dev_stat_read_and_reset(dev, i); | |
7129 | else | |
7130 | btrfs_dev_stat_reset(dev, i); | |
7131 | } | |
7132 | } else { | |
7133 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7134 | if (stats->nr_items > i) | |
7135 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7136 | } | |
7137 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7138 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7139 | return 0; | |
7140 | } | |
a8a6dab7 | 7141 | |
da353f6b | 7142 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7143 | { |
7144 | struct buffer_head *bh; | |
7145 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7146 | int copy_num; |
a8a6dab7 | 7147 | |
12b1c263 AJ |
7148 | if (!bdev) |
7149 | return; | |
a8a6dab7 | 7150 | |
12b1c263 AJ |
7151 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7152 | copy_num++) { | |
a8a6dab7 | 7153 | |
12b1c263 AJ |
7154 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7155 | continue; | |
7156 | ||
7157 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7158 | ||
7159 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7160 | set_buffer_dirty(bh); | |
7161 | sync_dirty_buffer(bh); | |
7162 | brelse(bh); | |
7163 | } | |
7164 | ||
7165 | /* Notify udev that device has changed */ | |
7166 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7167 | ||
7168 | /* Update ctime/mtime for device path for libblkid */ | |
7169 | update_dev_time(device_path); | |
a8a6dab7 | 7170 | } |
935e5cc9 MX |
7171 | |
7172 | /* | |
7173 | * Update the size of all devices, which is used for writing out the | |
7174 | * super blocks. | |
7175 | */ | |
7176 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7177 | { | |
7178 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7179 | struct btrfs_device *curr, *next; | |
7180 | ||
7181 | if (list_empty(&fs_devices->resized_devices)) | |
7182 | return; | |
7183 | ||
7184 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7185 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7186 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7187 | resized_list) { | |
7188 | list_del_init(&curr->resized_list); | |
7189 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7190 | } | |
34441361 | 7191 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7192 | mutex_unlock(&fs_devices->device_list_mutex); |
7193 | } | |
ce7213c7 MX |
7194 | |
7195 | /* Must be invoked during the transaction commit */ | |
2ff7e61e | 7196 | void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info, |
ce7213c7 MX |
7197 | struct btrfs_transaction *transaction) |
7198 | { | |
7199 | struct extent_map *em; | |
7200 | struct map_lookup *map; | |
7201 | struct btrfs_device *dev; | |
7202 | int i; | |
7203 | ||
7204 | if (list_empty(&transaction->pending_chunks)) | |
7205 | return; | |
7206 | ||
7207 | /* In order to kick the device replace finish process */ | |
34441361 | 7208 | mutex_lock(&fs_info->chunk_mutex); |
ce7213c7 | 7209 | list_for_each_entry(em, &transaction->pending_chunks, list) { |
95617d69 | 7210 | map = em->map_lookup; |
ce7213c7 MX |
7211 | |
7212 | for (i = 0; i < map->num_stripes; i++) { | |
7213 | dev = map->stripes[i].dev; | |
7214 | dev->commit_bytes_used = dev->bytes_used; | |
7215 | } | |
7216 | } | |
34441361 | 7217 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7218 | } |
5a13f430 AJ |
7219 | |
7220 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7221 | { | |
7222 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7223 | while (fs_devices) { | |
7224 | fs_devices->fs_info = fs_info; | |
7225 | fs_devices = fs_devices->seed; | |
7226 | } | |
7227 | } | |
7228 | ||
7229 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7230 | { | |
7231 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7232 | while (fs_devices) { | |
7233 | fs_devices->fs_info = NULL; | |
7234 | fs_devices = fs_devices->seed; | |
7235 | } | |
7236 | } |