]>
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> |
788f20eb | 23 | #include <linux/random.h> |
b765ead5 | 24 | #include <linux/iocontext.h> |
6f88a440 | 25 | #include <linux/capability.h> |
442a4f63 | 26 | #include <linux/ratelimit.h> |
59641015 | 27 | #include <linux/kthread.h> |
53b381b3 | 28 | #include <linux/raid/pq.h> |
803b2f54 | 29 | #include <linux/semaphore.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 YZ |
136 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
137 | struct btrfs_root *root, | |
138 | struct btrfs_device *device); | |
139 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 140 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 141 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 142 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
87ad58c5 | 143 | static void btrfs_close_one_device(struct btrfs_device *device); |
2b82032c | 144 | |
67a2c45e | 145 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 146 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
147 | struct list_head *btrfs_get_fs_uuids(void) |
148 | { | |
149 | return &fs_uuids; | |
150 | } | |
8a4b83cc | 151 | |
2208a378 ID |
152 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
153 | { | |
154 | struct btrfs_fs_devices *fs_devs; | |
155 | ||
78f2c9e6 | 156 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
157 | if (!fs_devs) |
158 | return ERR_PTR(-ENOMEM); | |
159 | ||
160 | mutex_init(&fs_devs->device_list_mutex); | |
161 | ||
162 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 163 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
164 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
165 | INIT_LIST_HEAD(&fs_devs->list); | |
166 | ||
167 | return fs_devs; | |
168 | } | |
169 | ||
170 | /** | |
171 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
172 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
173 | * generated. | |
174 | * | |
175 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
176 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
177 | * can be destroyed with kfree() right away. | |
178 | */ | |
179 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
180 | { | |
181 | struct btrfs_fs_devices *fs_devs; | |
182 | ||
183 | fs_devs = __alloc_fs_devices(); | |
184 | if (IS_ERR(fs_devs)) | |
185 | return fs_devs; | |
186 | ||
187 | if (fsid) | |
188 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
189 | else | |
190 | generate_random_uuid(fs_devs->fsid); | |
191 | ||
192 | return fs_devs; | |
193 | } | |
194 | ||
e4404d6e YZ |
195 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
196 | { | |
197 | struct btrfs_device *device; | |
198 | WARN_ON(fs_devices->opened); | |
199 | while (!list_empty(&fs_devices->devices)) { | |
200 | device = list_entry(fs_devices->devices.next, | |
201 | struct btrfs_device, dev_list); | |
202 | list_del(&device->dev_list); | |
606686ee | 203 | rcu_string_free(device->name); |
e4404d6e YZ |
204 | kfree(device); |
205 | } | |
206 | kfree(fs_devices); | |
207 | } | |
208 | ||
b8b8ff59 LC |
209 | static void btrfs_kobject_uevent(struct block_device *bdev, |
210 | enum kobject_action action) | |
211 | { | |
212 | int ret; | |
213 | ||
214 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
215 | if (ret) | |
efe120a0 | 216 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
217 | action, |
218 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
219 | &disk_to_dev(bdev->bd_disk)->kobj); | |
220 | } | |
221 | ||
143bede5 | 222 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
223 | { |
224 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 225 | |
2b82032c YZ |
226 | while (!list_empty(&fs_uuids)) { |
227 | fs_devices = list_entry(fs_uuids.next, | |
228 | struct btrfs_fs_devices, list); | |
229 | list_del(&fs_devices->list); | |
e4404d6e | 230 | free_fs_devices(fs_devices); |
8a4b83cc | 231 | } |
8a4b83cc CM |
232 | } |
233 | ||
12bd2fc0 ID |
234 | static struct btrfs_device *__alloc_device(void) |
235 | { | |
236 | struct btrfs_device *dev; | |
237 | ||
78f2c9e6 | 238 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
239 | if (!dev) |
240 | return ERR_PTR(-ENOMEM); | |
241 | ||
242 | INIT_LIST_HEAD(&dev->dev_list); | |
243 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 244 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
245 | |
246 | spin_lock_init(&dev->io_lock); | |
247 | ||
248 | spin_lock_init(&dev->reada_lock); | |
249 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 250 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 251 | btrfs_device_data_ordered_init(dev); |
d0164adc MG |
252 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
253 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
12bd2fc0 ID |
254 | |
255 | return dev; | |
256 | } | |
257 | ||
a1b32a59 CM |
258 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
259 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
260 | { |
261 | struct btrfs_device *dev; | |
8a4b83cc | 262 | |
c6e30871 | 263 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 264 | if (dev->devid == devid && |
8f18cf13 | 265 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 266 | return dev; |
a443755f | 267 | } |
8a4b83cc CM |
268 | } |
269 | return NULL; | |
270 | } | |
271 | ||
a1b32a59 | 272 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 273 | { |
8a4b83cc CM |
274 | struct btrfs_fs_devices *fs_devices; |
275 | ||
c6e30871 | 276 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
277 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
278 | return fs_devices; | |
279 | } | |
280 | return NULL; | |
281 | } | |
282 | ||
beaf8ab3 SB |
283 | static int |
284 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
285 | int flush, struct block_device **bdev, | |
286 | struct buffer_head **bh) | |
287 | { | |
288 | int ret; | |
289 | ||
290 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
291 | ||
292 | if (IS_ERR(*bdev)) { | |
293 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
294 | goto error; |
295 | } | |
296 | ||
297 | if (flush) | |
298 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
299 | ret = set_blocksize(*bdev, 4096); | |
300 | if (ret) { | |
301 | blkdev_put(*bdev, flags); | |
302 | goto error; | |
303 | } | |
304 | invalidate_bdev(*bdev); | |
305 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
306 | if (IS_ERR(*bh)) { |
307 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
308 | blkdev_put(*bdev, flags); |
309 | goto error; | |
310 | } | |
311 | ||
312 | return 0; | |
313 | ||
314 | error: | |
315 | *bdev = NULL; | |
316 | *bh = NULL; | |
317 | return ret; | |
318 | } | |
319 | ||
ffbd517d CM |
320 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
321 | struct bio *head, struct bio *tail) | |
322 | { | |
323 | ||
324 | struct bio *old_head; | |
325 | ||
326 | old_head = pending_bios->head; | |
327 | pending_bios->head = head; | |
328 | if (pending_bios->tail) | |
329 | tail->bi_next = old_head; | |
330 | else | |
331 | pending_bios->tail = tail; | |
332 | } | |
333 | ||
8b712842 CM |
334 | /* |
335 | * we try to collect pending bios for a device so we don't get a large | |
336 | * number of procs sending bios down to the same device. This greatly | |
337 | * improves the schedulers ability to collect and merge the bios. | |
338 | * | |
339 | * But, it also turns into a long list of bios to process and that is sure | |
340 | * to eventually make the worker thread block. The solution here is to | |
341 | * make some progress and then put this work struct back at the end of | |
342 | * the list if the block device is congested. This way, multiple devices | |
343 | * can make progress from a single worker thread. | |
344 | */ | |
143bede5 | 345 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
346 | { |
347 | struct bio *pending; | |
348 | struct backing_dev_info *bdi; | |
b64a2851 | 349 | struct btrfs_fs_info *fs_info; |
ffbd517d | 350 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
351 | struct bio *tail; |
352 | struct bio *cur; | |
353 | int again = 0; | |
ffbd517d | 354 | unsigned long num_run; |
d644d8a1 | 355 | unsigned long batch_run = 0; |
b64a2851 | 356 | unsigned long limit; |
b765ead5 | 357 | unsigned long last_waited = 0; |
d84275c9 | 358 | int force_reg = 0; |
0e588859 | 359 | int sync_pending = 0; |
211588ad CM |
360 | struct blk_plug plug; |
361 | ||
362 | /* | |
363 | * this function runs all the bios we've collected for | |
364 | * a particular device. We don't want to wander off to | |
365 | * another device without first sending all of these down. | |
366 | * So, setup a plug here and finish it off before we return | |
367 | */ | |
368 | blk_start_plug(&plug); | |
8b712842 | 369 | |
bedf762b | 370 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
371 | fs_info = device->dev_root->fs_info; |
372 | limit = btrfs_async_submit_limit(fs_info); | |
373 | limit = limit * 2 / 3; | |
374 | ||
8b712842 CM |
375 | loop: |
376 | spin_lock(&device->io_lock); | |
377 | ||
a6837051 | 378 | loop_lock: |
d84275c9 | 379 | num_run = 0; |
ffbd517d | 380 | |
8b712842 CM |
381 | /* take all the bios off the list at once and process them |
382 | * later on (without the lock held). But, remember the | |
383 | * tail and other pointers so the bios can be properly reinserted | |
384 | * into the list if we hit congestion | |
385 | */ | |
d84275c9 | 386 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 387 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
388 | force_reg = 1; |
389 | } else { | |
ffbd517d | 390 | pending_bios = &device->pending_bios; |
d84275c9 CM |
391 | force_reg = 0; |
392 | } | |
ffbd517d CM |
393 | |
394 | pending = pending_bios->head; | |
395 | tail = pending_bios->tail; | |
8b712842 | 396 | WARN_ON(pending && !tail); |
8b712842 CM |
397 | |
398 | /* | |
399 | * if pending was null this time around, no bios need processing | |
400 | * at all and we can stop. Otherwise it'll loop back up again | |
401 | * and do an additional check so no bios are missed. | |
402 | * | |
403 | * device->running_pending is used to synchronize with the | |
404 | * schedule_bio code. | |
405 | */ | |
ffbd517d CM |
406 | if (device->pending_sync_bios.head == NULL && |
407 | device->pending_bios.head == NULL) { | |
8b712842 CM |
408 | again = 0; |
409 | device->running_pending = 0; | |
ffbd517d CM |
410 | } else { |
411 | again = 1; | |
412 | device->running_pending = 1; | |
8b712842 | 413 | } |
ffbd517d CM |
414 | |
415 | pending_bios->head = NULL; | |
416 | pending_bios->tail = NULL; | |
417 | ||
8b712842 CM |
418 | spin_unlock(&device->io_lock); |
419 | ||
d397712b | 420 | while (pending) { |
ffbd517d CM |
421 | |
422 | rmb(); | |
d84275c9 CM |
423 | /* we want to work on both lists, but do more bios on the |
424 | * sync list than the regular list | |
425 | */ | |
426 | if ((num_run > 32 && | |
427 | pending_bios != &device->pending_sync_bios && | |
428 | device->pending_sync_bios.head) || | |
429 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
430 | device->pending_bios.head)) { | |
ffbd517d CM |
431 | spin_lock(&device->io_lock); |
432 | requeue_list(pending_bios, pending, tail); | |
433 | goto loop_lock; | |
434 | } | |
435 | ||
8b712842 CM |
436 | cur = pending; |
437 | pending = pending->bi_next; | |
438 | cur->bi_next = NULL; | |
b64a2851 | 439 | |
ee863954 DS |
440 | /* |
441 | * atomic_dec_return implies a barrier for waitqueue_active | |
442 | */ | |
66657b31 | 443 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
444 | waitqueue_active(&fs_info->async_submit_wait)) |
445 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 446 | |
dac56212 | 447 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 448 | |
2ab1ba68 CM |
449 | /* |
450 | * if we're doing the sync list, record that our | |
451 | * plug has some sync requests on it | |
452 | * | |
453 | * If we're doing the regular list and there are | |
454 | * sync requests sitting around, unplug before | |
455 | * we add more | |
456 | */ | |
457 | if (pending_bios == &device->pending_sync_bios) { | |
458 | sync_pending = 1; | |
459 | } else if (sync_pending) { | |
460 | blk_finish_plug(&plug); | |
461 | blk_start_plug(&plug); | |
462 | sync_pending = 0; | |
463 | } | |
464 | ||
21adbd5c | 465 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
466 | num_run++; |
467 | batch_run++; | |
853d8ec4 DS |
468 | |
469 | cond_resched(); | |
8b712842 CM |
470 | |
471 | /* | |
472 | * we made progress, there is more work to do and the bdi | |
473 | * is now congested. Back off and let other work structs | |
474 | * run instead | |
475 | */ | |
57fd5a5f | 476 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 477 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 478 | struct io_context *ioc; |
8b712842 | 479 | |
b765ead5 CM |
480 | ioc = current->io_context; |
481 | ||
482 | /* | |
483 | * the main goal here is that we don't want to | |
484 | * block if we're going to be able to submit | |
485 | * more requests without blocking. | |
486 | * | |
487 | * This code does two great things, it pokes into | |
488 | * the elevator code from a filesystem _and_ | |
489 | * it makes assumptions about how batching works. | |
490 | */ | |
491 | if (ioc && ioc->nr_batch_requests > 0 && | |
492 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
493 | (last_waited == 0 || | |
494 | ioc->last_waited == last_waited)) { | |
495 | /* | |
496 | * we want to go through our batch of | |
497 | * requests and stop. So, we copy out | |
498 | * the ioc->last_waited time and test | |
499 | * against it before looping | |
500 | */ | |
501 | last_waited = ioc->last_waited; | |
853d8ec4 | 502 | cond_resched(); |
b765ead5 CM |
503 | continue; |
504 | } | |
8b712842 | 505 | spin_lock(&device->io_lock); |
ffbd517d | 506 | requeue_list(pending_bios, pending, tail); |
a6837051 | 507 | device->running_pending = 1; |
8b712842 CM |
508 | |
509 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
510 | btrfs_queue_work(fs_info->submit_workers, |
511 | &device->work); | |
8b712842 CM |
512 | goto done; |
513 | } | |
d85c8a6f CM |
514 | /* unplug every 64 requests just for good measure */ |
515 | if (batch_run % 64 == 0) { | |
516 | blk_finish_plug(&plug); | |
517 | blk_start_plug(&plug); | |
518 | sync_pending = 0; | |
519 | } | |
8b712842 | 520 | } |
ffbd517d | 521 | |
51684082 CM |
522 | cond_resched(); |
523 | if (again) | |
524 | goto loop; | |
525 | ||
526 | spin_lock(&device->io_lock); | |
527 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
528 | goto loop_lock; | |
529 | spin_unlock(&device->io_lock); | |
530 | ||
8b712842 | 531 | done: |
211588ad | 532 | blk_finish_plug(&plug); |
8b712842 CM |
533 | } |
534 | ||
b2950863 | 535 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
536 | { |
537 | struct btrfs_device *device; | |
538 | ||
539 | device = container_of(work, struct btrfs_device, work); | |
540 | run_scheduled_bios(device); | |
541 | } | |
542 | ||
4fde46f0 AJ |
543 | |
544 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
545 | { | |
546 | struct btrfs_fs_devices *fs_devs; | |
547 | struct btrfs_device *dev; | |
548 | ||
549 | if (!cur_dev->name) | |
550 | return; | |
551 | ||
552 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
553 | int del = 1; | |
554 | ||
555 | if (fs_devs->opened) | |
556 | continue; | |
557 | if (fs_devs->seeding) | |
558 | continue; | |
559 | ||
560 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
561 | ||
562 | if (dev == cur_dev) | |
563 | continue; | |
564 | if (!dev->name) | |
565 | continue; | |
566 | ||
567 | /* | |
568 | * Todo: This won't be enough. What if the same device | |
569 | * comes back (with new uuid and) with its mapper path? | |
570 | * But for now, this does help as mostly an admin will | |
571 | * either use mapper or non mapper path throughout. | |
572 | */ | |
573 | rcu_read_lock(); | |
574 | del = strcmp(rcu_str_deref(dev->name), | |
575 | rcu_str_deref(cur_dev->name)); | |
576 | rcu_read_unlock(); | |
577 | if (!del) | |
578 | break; | |
579 | } | |
580 | ||
581 | if (!del) { | |
582 | /* delete the stale device */ | |
583 | if (fs_devs->num_devices == 1) { | |
584 | btrfs_sysfs_remove_fsid(fs_devs); | |
585 | list_del(&fs_devs->list); | |
586 | free_fs_devices(fs_devs); | |
587 | } else { | |
588 | fs_devs->num_devices--; | |
589 | list_del(&dev->dev_list); | |
590 | rcu_string_free(dev->name); | |
591 | kfree(dev); | |
592 | } | |
593 | break; | |
594 | } | |
595 | } | |
596 | } | |
597 | ||
60999ca4 DS |
598 | /* |
599 | * Add new device to list of registered devices | |
600 | * | |
601 | * Returns: | |
602 | * 1 - first time device is seen | |
603 | * 0 - device already known | |
604 | * < 0 - error | |
605 | */ | |
a1b32a59 | 606 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
607 | struct btrfs_super_block *disk_super, |
608 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
609 | { | |
610 | struct btrfs_device *device; | |
611 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 612 | struct rcu_string *name; |
60999ca4 | 613 | int ret = 0; |
8a4b83cc CM |
614 | u64 found_transid = btrfs_super_generation(disk_super); |
615 | ||
616 | fs_devices = find_fsid(disk_super->fsid); | |
617 | if (!fs_devices) { | |
2208a378 ID |
618 | fs_devices = alloc_fs_devices(disk_super->fsid); |
619 | if (IS_ERR(fs_devices)) | |
620 | return PTR_ERR(fs_devices); | |
621 | ||
8a4b83cc | 622 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 623 | |
8a4b83cc CM |
624 | device = NULL; |
625 | } else { | |
a443755f CM |
626 | device = __find_device(&fs_devices->devices, devid, |
627 | disk_super->dev_item.uuid); | |
8a4b83cc | 628 | } |
443f24fe | 629 | |
8a4b83cc | 630 | if (!device) { |
2b82032c YZ |
631 | if (fs_devices->opened) |
632 | return -EBUSY; | |
633 | ||
12bd2fc0 ID |
634 | device = btrfs_alloc_device(NULL, &devid, |
635 | disk_super->dev_item.uuid); | |
636 | if (IS_ERR(device)) { | |
8a4b83cc | 637 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 638 | return PTR_ERR(device); |
8a4b83cc | 639 | } |
606686ee JB |
640 | |
641 | name = rcu_string_strdup(path, GFP_NOFS); | |
642 | if (!name) { | |
8a4b83cc CM |
643 | kfree(device); |
644 | return -ENOMEM; | |
645 | } | |
606686ee | 646 | rcu_assign_pointer(device->name, name); |
90519d66 | 647 | |
e5e9a520 | 648 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 649 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 650 | fs_devices->num_devices++; |
e5e9a520 CM |
651 | mutex_unlock(&fs_devices->device_list_mutex); |
652 | ||
60999ca4 | 653 | ret = 1; |
2b82032c | 654 | device->fs_devices = fs_devices; |
606686ee | 655 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
656 | /* |
657 | * When FS is already mounted. | |
658 | * 1. If you are here and if the device->name is NULL that | |
659 | * means this device was missing at time of FS mount. | |
660 | * 2. If you are here and if the device->name is different | |
661 | * from 'path' that means either | |
662 | * a. The same device disappeared and reappeared with | |
663 | * different name. or | |
664 | * b. The missing-disk-which-was-replaced, has | |
665 | * reappeared now. | |
666 | * | |
667 | * We must allow 1 and 2a above. But 2b would be a spurious | |
668 | * and unintentional. | |
669 | * | |
670 | * Further in case of 1 and 2a above, the disk at 'path' | |
671 | * would have missed some transaction when it was away and | |
672 | * in case of 2a the stale bdev has to be updated as well. | |
673 | * 2b must not be allowed at all time. | |
674 | */ | |
675 | ||
676 | /* | |
0f23ae74 CM |
677 | * For now, we do allow update to btrfs_fs_device through the |
678 | * btrfs dev scan cli after FS has been mounted. We're still | |
679 | * tracking a problem where systems fail mount by subvolume id | |
680 | * when we reject replacement on a mounted FS. | |
b96de000 | 681 | */ |
0f23ae74 | 682 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
683 | /* |
684 | * That is if the FS is _not_ mounted and if you | |
685 | * are here, that means there is more than one | |
686 | * disk with same uuid and devid.We keep the one | |
687 | * with larger generation number or the last-in if | |
688 | * generation are equal. | |
689 | */ | |
0f23ae74 | 690 | return -EEXIST; |
77bdae4d | 691 | } |
b96de000 | 692 | |
606686ee | 693 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
694 | if (!name) |
695 | return -ENOMEM; | |
606686ee JB |
696 | rcu_string_free(device->name); |
697 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
698 | if (device->missing) { |
699 | fs_devices->missing_devices--; | |
700 | device->missing = 0; | |
701 | } | |
8a4b83cc CM |
702 | } |
703 | ||
77bdae4d AJ |
704 | /* |
705 | * Unmount does not free the btrfs_device struct but would zero | |
706 | * generation along with most of the other members. So just update | |
707 | * it back. We need it to pick the disk with largest generation | |
708 | * (as above). | |
709 | */ | |
710 | if (!fs_devices->opened) | |
711 | device->generation = found_transid; | |
712 | ||
4fde46f0 AJ |
713 | /* |
714 | * if there is new btrfs on an already registered device, | |
715 | * then remove the stale device entry. | |
716 | */ | |
02feae3c AJ |
717 | if (ret > 0) |
718 | btrfs_free_stale_device(device); | |
4fde46f0 | 719 | |
8a4b83cc | 720 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
721 | |
722 | return ret; | |
8a4b83cc CM |
723 | } |
724 | ||
e4404d6e YZ |
725 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
726 | { | |
727 | struct btrfs_fs_devices *fs_devices; | |
728 | struct btrfs_device *device; | |
729 | struct btrfs_device *orig_dev; | |
730 | ||
2208a378 ID |
731 | fs_devices = alloc_fs_devices(orig->fsid); |
732 | if (IS_ERR(fs_devices)) | |
733 | return fs_devices; | |
e4404d6e | 734 | |
adbbb863 | 735 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 736 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 737 | |
46224705 | 738 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 739 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
740 | struct rcu_string *name; |
741 | ||
12bd2fc0 ID |
742 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
743 | orig_dev->uuid); | |
744 | if (IS_ERR(device)) | |
e4404d6e YZ |
745 | goto error; |
746 | ||
606686ee JB |
747 | /* |
748 | * This is ok to do without rcu read locked because we hold the | |
749 | * uuid mutex so nothing we touch in here is going to disappear. | |
750 | */ | |
e755f780 | 751 | if (orig_dev->name) { |
78f2c9e6 DS |
752 | name = rcu_string_strdup(orig_dev->name->str, |
753 | GFP_KERNEL); | |
e755f780 AJ |
754 | if (!name) { |
755 | kfree(device); | |
756 | goto error; | |
757 | } | |
758 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 759 | } |
e4404d6e | 760 | |
e4404d6e YZ |
761 | list_add(&device->dev_list, &fs_devices->devices); |
762 | device->fs_devices = fs_devices; | |
763 | fs_devices->num_devices++; | |
764 | } | |
adbbb863 | 765 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
766 | return fs_devices; |
767 | error: | |
adbbb863 | 768 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
769 | free_fs_devices(fs_devices); |
770 | return ERR_PTR(-ENOMEM); | |
771 | } | |
772 | ||
9eaed21e | 773 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 774 | { |
c6e30871 | 775 | struct btrfs_device *device, *next; |
443f24fe | 776 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 777 | |
dfe25020 CM |
778 | mutex_lock(&uuid_mutex); |
779 | again: | |
46224705 | 780 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 781 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 782 | if (device->in_fs_metadata) { |
63a212ab | 783 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
784 | (!latest_dev || |
785 | device->generation > latest_dev->generation)) { | |
786 | latest_dev = device; | |
a6b0d5c8 | 787 | } |
2b82032c | 788 | continue; |
a6b0d5c8 | 789 | } |
2b82032c | 790 | |
8dabb742 SB |
791 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
792 | /* | |
793 | * In the first step, keep the device which has | |
794 | * the correct fsid and the devid that is used | |
795 | * for the dev_replace procedure. | |
796 | * In the second step, the dev_replace state is | |
797 | * read from the device tree and it is known | |
798 | * whether the procedure is really active or | |
799 | * not, which means whether this device is | |
800 | * used or whether it should be removed. | |
801 | */ | |
802 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
803 | continue; | |
804 | } | |
805 | } | |
2b82032c | 806 | if (device->bdev) { |
d4d77629 | 807 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
808 | device->bdev = NULL; |
809 | fs_devices->open_devices--; | |
810 | } | |
811 | if (device->writeable) { | |
812 | list_del_init(&device->dev_alloc_list); | |
813 | device->writeable = 0; | |
8dabb742 SB |
814 | if (!device->is_tgtdev_for_dev_replace) |
815 | fs_devices->rw_devices--; | |
2b82032c | 816 | } |
e4404d6e YZ |
817 | list_del_init(&device->dev_list); |
818 | fs_devices->num_devices--; | |
606686ee | 819 | rcu_string_free(device->name); |
e4404d6e | 820 | kfree(device); |
dfe25020 | 821 | } |
2b82032c YZ |
822 | |
823 | if (fs_devices->seed) { | |
824 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
825 | goto again; |
826 | } | |
827 | ||
443f24fe | 828 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 829 | |
dfe25020 | 830 | mutex_unlock(&uuid_mutex); |
dfe25020 | 831 | } |
a0af469b | 832 | |
1f78160c XG |
833 | static void __free_device(struct work_struct *work) |
834 | { | |
835 | struct btrfs_device *device; | |
836 | ||
837 | device = container_of(work, struct btrfs_device, rcu_work); | |
838 | ||
839 | if (device->bdev) | |
840 | blkdev_put(device->bdev, device->mode); | |
841 | ||
606686ee | 842 | rcu_string_free(device->name); |
1f78160c XG |
843 | kfree(device); |
844 | } | |
845 | ||
846 | static void free_device(struct rcu_head *head) | |
847 | { | |
848 | struct btrfs_device *device; | |
849 | ||
850 | device = container_of(head, struct btrfs_device, rcu); | |
851 | ||
852 | INIT_WORK(&device->rcu_work, __free_device); | |
853 | schedule_work(&device->rcu_work); | |
854 | } | |
855 | ||
2b82032c | 856 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 857 | { |
2037a093 | 858 | struct btrfs_device *device, *tmp; |
e4404d6e | 859 | |
2b82032c YZ |
860 | if (--fs_devices->opened > 0) |
861 | return 0; | |
8a4b83cc | 862 | |
c9513edb | 863 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 864 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
f190aa47 | 865 | btrfs_close_one_device(device); |
8a4b83cc | 866 | } |
c9513edb XG |
867 | mutex_unlock(&fs_devices->device_list_mutex); |
868 | ||
e4404d6e YZ |
869 | WARN_ON(fs_devices->open_devices); |
870 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
871 | fs_devices->opened = 0; |
872 | fs_devices->seeding = 0; | |
2b82032c | 873 | |
8a4b83cc CM |
874 | return 0; |
875 | } | |
876 | ||
2b82032c YZ |
877 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
878 | { | |
e4404d6e | 879 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
880 | int ret; |
881 | ||
882 | mutex_lock(&uuid_mutex); | |
883 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
884 | if (!fs_devices->opened) { |
885 | seed_devices = fs_devices->seed; | |
886 | fs_devices->seed = NULL; | |
887 | } | |
2b82032c | 888 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
889 | |
890 | while (seed_devices) { | |
891 | fs_devices = seed_devices; | |
892 | seed_devices = fs_devices->seed; | |
893 | __btrfs_close_devices(fs_devices); | |
894 | free_fs_devices(fs_devices); | |
895 | } | |
bc178622 ES |
896 | /* |
897 | * Wait for rcu kworkers under __btrfs_close_devices | |
898 | * to finish all blkdev_puts so device is really | |
899 | * free when umount is done. | |
900 | */ | |
901 | rcu_barrier(); | |
2b82032c YZ |
902 | return ret; |
903 | } | |
904 | ||
e4404d6e YZ |
905 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
906 | fmode_t flags, void *holder) | |
8a4b83cc | 907 | { |
d5e2003c | 908 | struct request_queue *q; |
8a4b83cc CM |
909 | struct block_device *bdev; |
910 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 911 | struct btrfs_device *device; |
443f24fe | 912 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
913 | struct buffer_head *bh; |
914 | struct btrfs_super_block *disk_super; | |
a0af469b | 915 | u64 devid; |
2b82032c | 916 | int seeding = 1; |
a0af469b | 917 | int ret = 0; |
8a4b83cc | 918 | |
d4d77629 TH |
919 | flags |= FMODE_EXCL; |
920 | ||
c6e30871 | 921 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
922 | if (device->bdev) |
923 | continue; | |
dfe25020 CM |
924 | if (!device->name) |
925 | continue; | |
926 | ||
f63e0cca ES |
927 | /* Just open everything we can; ignore failures here */ |
928 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
929 | &bdev, &bh)) | |
beaf8ab3 | 930 | continue; |
a0af469b CM |
931 | |
932 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 933 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
934 | if (devid != device->devid) |
935 | goto error_brelse; | |
936 | ||
2b82032c YZ |
937 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
938 | BTRFS_UUID_SIZE)) | |
939 | goto error_brelse; | |
940 | ||
941 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
942 | if (!latest_dev || |
943 | device->generation > latest_dev->generation) | |
944 | latest_dev = device; | |
a0af469b | 945 | |
2b82032c YZ |
946 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
947 | device->writeable = 0; | |
948 | } else { | |
949 | device->writeable = !bdev_read_only(bdev); | |
950 | seeding = 0; | |
951 | } | |
952 | ||
d5e2003c | 953 | q = bdev_get_queue(bdev); |
90180da4 | 954 | if (blk_queue_discard(q)) |
d5e2003c | 955 | device->can_discard = 1; |
d5e2003c | 956 | |
8a4b83cc | 957 | device->bdev = bdev; |
dfe25020 | 958 | device->in_fs_metadata = 0; |
15916de8 CM |
959 | device->mode = flags; |
960 | ||
c289811c CM |
961 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
962 | fs_devices->rotating = 1; | |
963 | ||
a0af469b | 964 | fs_devices->open_devices++; |
55e50e45 ID |
965 | if (device->writeable && |
966 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
967 | fs_devices->rw_devices++; |
968 | list_add(&device->dev_alloc_list, | |
969 | &fs_devices->alloc_list); | |
970 | } | |
4f6c9328 | 971 | brelse(bh); |
a0af469b | 972 | continue; |
a061fc8d | 973 | |
a0af469b CM |
974 | error_brelse: |
975 | brelse(bh); | |
d4d77629 | 976 | blkdev_put(bdev, flags); |
a0af469b | 977 | continue; |
8a4b83cc | 978 | } |
a0af469b | 979 | if (fs_devices->open_devices == 0) { |
20bcd649 | 980 | ret = -EINVAL; |
a0af469b CM |
981 | goto out; |
982 | } | |
2b82032c YZ |
983 | fs_devices->seeding = seeding; |
984 | fs_devices->opened = 1; | |
443f24fe | 985 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 986 | fs_devices->total_rw_bytes = 0; |
a0af469b | 987 | out: |
2b82032c YZ |
988 | return ret; |
989 | } | |
990 | ||
991 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 992 | fmode_t flags, void *holder) |
2b82032c YZ |
993 | { |
994 | int ret; | |
995 | ||
996 | mutex_lock(&uuid_mutex); | |
997 | if (fs_devices->opened) { | |
e4404d6e YZ |
998 | fs_devices->opened++; |
999 | ret = 0; | |
2b82032c | 1000 | } else { |
15916de8 | 1001 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1002 | } |
8a4b83cc | 1003 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1004 | return ret; |
1005 | } | |
1006 | ||
6cf86a00 AJ |
1007 | void btrfs_release_disk_super(struct page *page) |
1008 | { | |
1009 | kunmap(page); | |
1010 | put_page(page); | |
1011 | } | |
1012 | ||
1013 | int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, | |
1014 | struct page **page, struct btrfs_super_block **disk_super) | |
1015 | { | |
1016 | void *p; | |
1017 | pgoff_t index; | |
1018 | ||
1019 | /* make sure our super fits in the device */ | |
1020 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1021 | return 1; | |
1022 | ||
1023 | /* make sure our super fits in the page */ | |
1024 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1025 | return 1; | |
1026 | ||
1027 | /* make sure our super doesn't straddle pages on disk */ | |
1028 | index = bytenr >> PAGE_SHIFT; | |
1029 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1030 | return 1; | |
1031 | ||
1032 | /* pull in the page with our super */ | |
1033 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1034 | index, GFP_KERNEL); | |
1035 | ||
1036 | if (IS_ERR_OR_NULL(*page)) | |
1037 | return 1; | |
1038 | ||
1039 | p = kmap(*page); | |
1040 | ||
1041 | /* align our pointer to the offset of the super block */ | |
1042 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1043 | ||
1044 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1045 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1046 | btrfs_release_disk_super(*page); | |
1047 | return 1; | |
1048 | } | |
1049 | ||
1050 | if ((*disk_super)->label[0] && | |
1051 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1052 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1053 | ||
1054 | return 0; | |
1055 | } | |
1056 | ||
6f60cbd3 DS |
1057 | /* |
1058 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1059 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1060 | * is read via pagecache | |
1061 | */ | |
97288f2c | 1062 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1063 | struct btrfs_fs_devices **fs_devices_ret) |
1064 | { | |
1065 | struct btrfs_super_block *disk_super; | |
1066 | struct block_device *bdev; | |
6f60cbd3 | 1067 | struct page *page; |
6f60cbd3 | 1068 | int ret = -EINVAL; |
8a4b83cc | 1069 | u64 devid; |
f2984462 | 1070 | u64 transid; |
02db0844 | 1071 | u64 total_devices; |
6f60cbd3 | 1072 | u64 bytenr; |
8a4b83cc | 1073 | |
6f60cbd3 DS |
1074 | /* |
1075 | * we would like to check all the supers, but that would make | |
1076 | * a btrfs mount succeed after a mkfs from a different FS. | |
1077 | * So, we need to add a special mount option to scan for | |
1078 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1079 | */ | |
1080 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1081 | flags |= FMODE_EXCL; |
10f6327b | 1082 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1083 | |
1084 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1085 | if (IS_ERR(bdev)) { |
1086 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1087 | goto error; |
6f60cbd3 DS |
1088 | } |
1089 | ||
6cf86a00 | 1090 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) |
6f60cbd3 DS |
1091 | goto error_bdev_put; |
1092 | ||
a343832f | 1093 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1094 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1095 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1096 | |
8a4b83cc | 1097 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1098 | if (ret > 0) { |
1099 | if (disk_super->label[0]) { | |
60999ca4 DS |
1100 | printk(KERN_INFO "BTRFS: device label %s ", disk_super->label); |
1101 | } else { | |
1102 | printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid); | |
1103 | } | |
1104 | ||
1105 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); | |
1106 | ret = 0; | |
1107 | } | |
02db0844 JB |
1108 | if (!ret && fs_devices_ret) |
1109 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 | 1110 | |
6cf86a00 | 1111 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1112 | |
1113 | error_bdev_put: | |
d4d77629 | 1114 | blkdev_put(bdev, flags); |
8a4b83cc | 1115 | error: |
beaf8ab3 | 1116 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1117 | return ret; |
1118 | } | |
0b86a832 | 1119 | |
6d07bcec MX |
1120 | /* helper to account the used device space in the range */ |
1121 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1122 | u64 end, u64 *length) | |
1123 | { | |
1124 | struct btrfs_key key; | |
1125 | struct btrfs_root *root = device->dev_root; | |
1126 | struct btrfs_dev_extent *dev_extent; | |
1127 | struct btrfs_path *path; | |
1128 | u64 extent_end; | |
1129 | int ret; | |
1130 | int slot; | |
1131 | struct extent_buffer *l; | |
1132 | ||
1133 | *length = 0; | |
1134 | ||
63a212ab | 1135 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1136 | return 0; |
1137 | ||
1138 | path = btrfs_alloc_path(); | |
1139 | if (!path) | |
1140 | return -ENOMEM; | |
e4058b54 | 1141 | path->reada = READA_FORWARD; |
6d07bcec MX |
1142 | |
1143 | key.objectid = device->devid; | |
1144 | key.offset = start; | |
1145 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1146 | ||
1147 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1148 | if (ret < 0) | |
1149 | goto out; | |
1150 | if (ret > 0) { | |
1151 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1152 | if (ret < 0) | |
1153 | goto out; | |
1154 | } | |
1155 | ||
1156 | while (1) { | |
1157 | l = path->nodes[0]; | |
1158 | slot = path->slots[0]; | |
1159 | if (slot >= btrfs_header_nritems(l)) { | |
1160 | ret = btrfs_next_leaf(root, path); | |
1161 | if (ret == 0) | |
1162 | continue; | |
1163 | if (ret < 0) | |
1164 | goto out; | |
1165 | ||
1166 | break; | |
1167 | } | |
1168 | btrfs_item_key_to_cpu(l, &key, slot); | |
1169 | ||
1170 | if (key.objectid < device->devid) | |
1171 | goto next; | |
1172 | ||
1173 | if (key.objectid > device->devid) | |
1174 | break; | |
1175 | ||
962a298f | 1176 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1177 | goto next; |
1178 | ||
1179 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1180 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1181 | dev_extent); | |
1182 | if (key.offset <= start && extent_end > end) { | |
1183 | *length = end - start + 1; | |
1184 | break; | |
1185 | } else if (key.offset <= start && extent_end > start) | |
1186 | *length += extent_end - start; | |
1187 | else if (key.offset > start && extent_end <= end) | |
1188 | *length += extent_end - key.offset; | |
1189 | else if (key.offset > start && key.offset <= end) { | |
1190 | *length += end - key.offset + 1; | |
1191 | break; | |
1192 | } else if (key.offset > end) | |
1193 | break; | |
1194 | ||
1195 | next: | |
1196 | path->slots[0]++; | |
1197 | } | |
1198 | ret = 0; | |
1199 | out: | |
1200 | btrfs_free_path(path); | |
1201 | return ret; | |
1202 | } | |
1203 | ||
499f377f | 1204 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1205 | struct btrfs_device *device, |
1206 | u64 *start, u64 len) | |
1207 | { | |
499f377f | 1208 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; |
6df9a95e | 1209 | struct extent_map *em; |
499f377f | 1210 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1211 | int ret = 0; |
1b984508 | 1212 | u64 physical_start = *start; |
6df9a95e | 1213 | |
499f377f JM |
1214 | if (transaction) |
1215 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1216 | again: |
1217 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1218 | struct map_lookup *map; |
1219 | int i; | |
1220 | ||
95617d69 | 1221 | map = em->map_lookup; |
6df9a95e | 1222 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1223 | u64 end; |
1224 | ||
6df9a95e JB |
1225 | if (map->stripes[i].dev != device) |
1226 | continue; | |
1b984508 | 1227 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1228 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1229 | physical_start) |
6df9a95e | 1230 | continue; |
c152b63e FM |
1231 | /* |
1232 | * Make sure that while processing the pinned list we do | |
1233 | * not override our *start with a lower value, because | |
1234 | * we can have pinned chunks that fall within this | |
1235 | * device hole and that have lower physical addresses | |
1236 | * than the pending chunks we processed before. If we | |
1237 | * do not take this special care we can end up getting | |
1238 | * 2 pending chunks that start at the same physical | |
1239 | * device offsets because the end offset of a pinned | |
1240 | * chunk can be equal to the start offset of some | |
1241 | * pending chunk. | |
1242 | */ | |
1243 | end = map->stripes[i].physical + em->orig_block_len; | |
1244 | if (end > *start) { | |
1245 | *start = end; | |
1246 | ret = 1; | |
1247 | } | |
6df9a95e JB |
1248 | } |
1249 | } | |
499f377f JM |
1250 | if (search_list != &fs_info->pinned_chunks) { |
1251 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1252 | goto again; |
1253 | } | |
6df9a95e JB |
1254 | |
1255 | return ret; | |
1256 | } | |
1257 | ||
1258 | ||
0b86a832 | 1259 | /* |
499f377f JM |
1260 | * find_free_dev_extent_start - find free space in the specified device |
1261 | * @device: the device which we search the free space in | |
1262 | * @num_bytes: the size of the free space that we need | |
1263 | * @search_start: the position from which to begin the search | |
1264 | * @start: store the start of the free space. | |
1265 | * @len: the size of the free space. that we find, or the size | |
1266 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1267 | * |
0b86a832 CM |
1268 | * this uses a pretty simple search, the expectation is that it is |
1269 | * called very infrequently and that a given device has a small number | |
1270 | * of extents | |
7bfc837d MX |
1271 | * |
1272 | * @start is used to store the start of the free space if we find. But if we | |
1273 | * don't find suitable free space, it will be used to store the start position | |
1274 | * of the max free space. | |
1275 | * | |
1276 | * @len is used to store the size of the free space that we find. | |
1277 | * But if we don't find suitable free space, it is used to store the size of | |
1278 | * the max free space. | |
0b86a832 | 1279 | */ |
499f377f JM |
1280 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1281 | struct btrfs_device *device, u64 num_bytes, | |
1282 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 CM |
1283 | { |
1284 | struct btrfs_key key; | |
1285 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1286 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1287 | struct btrfs_path *path; |
7bfc837d MX |
1288 | u64 hole_size; |
1289 | u64 max_hole_start; | |
1290 | u64 max_hole_size; | |
1291 | u64 extent_end; | |
0b86a832 CM |
1292 | u64 search_end = device->total_bytes; |
1293 | int ret; | |
7bfc837d | 1294 | int slot; |
0b86a832 | 1295 | struct extent_buffer *l; |
8cdc7c5b FM |
1296 | u64 min_search_start; |
1297 | ||
1298 | /* | |
1299 | * We don't want to overwrite the superblock on the drive nor any area | |
1300 | * used by the boot loader (grub for example), so we make sure to start | |
1301 | * at an offset of at least 1MB. | |
1302 | */ | |
1303 | min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024); | |
1304 | search_start = max(search_start, min_search_start); | |
0b86a832 | 1305 | |
6df9a95e JB |
1306 | path = btrfs_alloc_path(); |
1307 | if (!path) | |
1308 | return -ENOMEM; | |
f2ab7618 | 1309 | |
7bfc837d MX |
1310 | max_hole_start = search_start; |
1311 | max_hole_size = 0; | |
1312 | ||
f2ab7618 | 1313 | again: |
63a212ab | 1314 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1315 | ret = -ENOSPC; |
6df9a95e | 1316 | goto out; |
7bfc837d MX |
1317 | } |
1318 | ||
e4058b54 | 1319 | path->reada = READA_FORWARD; |
6df9a95e JB |
1320 | path->search_commit_root = 1; |
1321 | path->skip_locking = 1; | |
7bfc837d | 1322 | |
0b86a832 CM |
1323 | key.objectid = device->devid; |
1324 | key.offset = search_start; | |
1325 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1326 | |
125ccb0a | 1327 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1328 | if (ret < 0) |
7bfc837d | 1329 | goto out; |
1fcbac58 YZ |
1330 | if (ret > 0) { |
1331 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1332 | if (ret < 0) | |
7bfc837d | 1333 | goto out; |
1fcbac58 | 1334 | } |
7bfc837d | 1335 | |
0b86a832 CM |
1336 | while (1) { |
1337 | l = path->nodes[0]; | |
1338 | slot = path->slots[0]; | |
1339 | if (slot >= btrfs_header_nritems(l)) { | |
1340 | ret = btrfs_next_leaf(root, path); | |
1341 | if (ret == 0) | |
1342 | continue; | |
1343 | if (ret < 0) | |
7bfc837d MX |
1344 | goto out; |
1345 | ||
1346 | break; | |
0b86a832 CM |
1347 | } |
1348 | btrfs_item_key_to_cpu(l, &key, slot); | |
1349 | ||
1350 | if (key.objectid < device->devid) | |
1351 | goto next; | |
1352 | ||
1353 | if (key.objectid > device->devid) | |
7bfc837d | 1354 | break; |
0b86a832 | 1355 | |
962a298f | 1356 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1357 | goto next; |
9779b72f | 1358 | |
7bfc837d MX |
1359 | if (key.offset > search_start) { |
1360 | hole_size = key.offset - search_start; | |
9779b72f | 1361 | |
6df9a95e JB |
1362 | /* |
1363 | * Have to check before we set max_hole_start, otherwise | |
1364 | * we could end up sending back this offset anyway. | |
1365 | */ | |
499f377f | 1366 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1367 | &search_start, |
1b984508 FL |
1368 | hole_size)) { |
1369 | if (key.offset >= search_start) { | |
1370 | hole_size = key.offset - search_start; | |
1371 | } else { | |
1372 | WARN_ON_ONCE(1); | |
1373 | hole_size = 0; | |
1374 | } | |
1375 | } | |
6df9a95e | 1376 | |
7bfc837d MX |
1377 | if (hole_size > max_hole_size) { |
1378 | max_hole_start = search_start; | |
1379 | max_hole_size = hole_size; | |
1380 | } | |
9779b72f | 1381 | |
7bfc837d MX |
1382 | /* |
1383 | * If this free space is greater than which we need, | |
1384 | * it must be the max free space that we have found | |
1385 | * until now, so max_hole_start must point to the start | |
1386 | * of this free space and the length of this free space | |
1387 | * is stored in max_hole_size. Thus, we return | |
1388 | * max_hole_start and max_hole_size and go back to the | |
1389 | * caller. | |
1390 | */ | |
1391 | if (hole_size >= num_bytes) { | |
1392 | ret = 0; | |
1393 | goto out; | |
0b86a832 CM |
1394 | } |
1395 | } | |
0b86a832 | 1396 | |
0b86a832 | 1397 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1398 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1399 | dev_extent); | |
1400 | if (extent_end > search_start) | |
1401 | search_start = extent_end; | |
0b86a832 CM |
1402 | next: |
1403 | path->slots[0]++; | |
1404 | cond_resched(); | |
1405 | } | |
0b86a832 | 1406 | |
38c01b96 | 1407 | /* |
1408 | * At this point, search_start should be the end of | |
1409 | * allocated dev extents, and when shrinking the device, | |
1410 | * search_end may be smaller than search_start. | |
1411 | */ | |
f2ab7618 | 1412 | if (search_end > search_start) { |
38c01b96 | 1413 | hole_size = search_end - search_start; |
1414 | ||
499f377f | 1415 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1416 | hole_size)) { |
1417 | btrfs_release_path(path); | |
1418 | goto again; | |
1419 | } | |
0b86a832 | 1420 | |
f2ab7618 ZL |
1421 | if (hole_size > max_hole_size) { |
1422 | max_hole_start = search_start; | |
1423 | max_hole_size = hole_size; | |
1424 | } | |
6df9a95e JB |
1425 | } |
1426 | ||
7bfc837d | 1427 | /* See above. */ |
f2ab7618 | 1428 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1429 | ret = -ENOSPC; |
1430 | else | |
1431 | ret = 0; | |
1432 | ||
1433 | out: | |
2b82032c | 1434 | btrfs_free_path(path); |
7bfc837d | 1435 | *start = max_hole_start; |
b2117a39 | 1436 | if (len) |
7bfc837d | 1437 | *len = max_hole_size; |
0b86a832 CM |
1438 | return ret; |
1439 | } | |
1440 | ||
499f377f JM |
1441 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1442 | struct btrfs_device *device, u64 num_bytes, | |
1443 | u64 *start, u64 *len) | |
1444 | { | |
499f377f | 1445 | /* FIXME use last free of some kind */ |
499f377f | 1446 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1447 | num_bytes, 0, start, len); |
499f377f JM |
1448 | } |
1449 | ||
b2950863 | 1450 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1451 | struct btrfs_device *device, |
2196d6e8 | 1452 | u64 start, u64 *dev_extent_len) |
8f18cf13 CM |
1453 | { |
1454 | int ret; | |
1455 | struct btrfs_path *path; | |
1456 | struct btrfs_root *root = device->dev_root; | |
1457 | struct btrfs_key key; | |
a061fc8d CM |
1458 | struct btrfs_key found_key; |
1459 | struct extent_buffer *leaf = NULL; | |
1460 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1461 | |
1462 | path = btrfs_alloc_path(); | |
1463 | if (!path) | |
1464 | return -ENOMEM; | |
1465 | ||
1466 | key.objectid = device->devid; | |
1467 | key.offset = start; | |
1468 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1469 | again: |
8f18cf13 | 1470 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1471 | if (ret > 0) { |
1472 | ret = btrfs_previous_item(root, path, key.objectid, | |
1473 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1474 | if (ret) |
1475 | goto out; | |
a061fc8d CM |
1476 | leaf = path->nodes[0]; |
1477 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1478 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1479 | struct btrfs_dev_extent); | |
1480 | BUG_ON(found_key.offset > start || found_key.offset + | |
1481 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1482 | key = found_key; |
1483 | btrfs_release_path(path); | |
1484 | goto again; | |
a061fc8d CM |
1485 | } else if (ret == 0) { |
1486 | leaf = path->nodes[0]; | |
1487 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1488 | struct btrfs_dev_extent); | |
79787eaa | 1489 | } else { |
a4553fef | 1490 | btrfs_std_error(root->fs_info, ret, "Slot search failed"); |
79787eaa | 1491 | goto out; |
a061fc8d | 1492 | } |
8f18cf13 | 1493 | |
2196d6e8 MX |
1494 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1495 | ||
8f18cf13 | 1496 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1497 | if (ret) { |
a4553fef | 1498 | btrfs_std_error(root->fs_info, ret, |
79787eaa | 1499 | "Failed to remove dev extent item"); |
13212b54 | 1500 | } else { |
3204d33c | 1501 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1502 | } |
b0b802d7 | 1503 | out: |
8f18cf13 CM |
1504 | btrfs_free_path(path); |
1505 | return ret; | |
1506 | } | |
1507 | ||
48a3b636 ES |
1508 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1509 | struct btrfs_device *device, | |
1510 | u64 chunk_tree, u64 chunk_objectid, | |
1511 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1512 | { |
1513 | int ret; | |
1514 | struct btrfs_path *path; | |
1515 | struct btrfs_root *root = device->dev_root; | |
1516 | struct btrfs_dev_extent *extent; | |
1517 | struct extent_buffer *leaf; | |
1518 | struct btrfs_key key; | |
1519 | ||
dfe25020 | 1520 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1521 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1522 | path = btrfs_alloc_path(); |
1523 | if (!path) | |
1524 | return -ENOMEM; | |
1525 | ||
0b86a832 | 1526 | key.objectid = device->devid; |
2b82032c | 1527 | key.offset = start; |
0b86a832 CM |
1528 | key.type = BTRFS_DEV_EXTENT_KEY; |
1529 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1530 | sizeof(*extent)); | |
2cdcecbc MF |
1531 | if (ret) |
1532 | goto out; | |
0b86a832 CM |
1533 | |
1534 | leaf = path->nodes[0]; | |
1535 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1536 | struct btrfs_dev_extent); | |
e17cade2 CM |
1537 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1538 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1539 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1540 | ||
1541 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1542 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1543 | |
0b86a832 CM |
1544 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1545 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1546 | out: |
0b86a832 CM |
1547 | btrfs_free_path(path); |
1548 | return ret; | |
1549 | } | |
1550 | ||
6df9a95e | 1551 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1552 | { |
6df9a95e JB |
1553 | struct extent_map_tree *em_tree; |
1554 | struct extent_map *em; | |
1555 | struct rb_node *n; | |
1556 | u64 ret = 0; | |
0b86a832 | 1557 | |
6df9a95e JB |
1558 | em_tree = &fs_info->mapping_tree.map_tree; |
1559 | read_lock(&em_tree->lock); | |
1560 | n = rb_last(&em_tree->map); | |
1561 | if (n) { | |
1562 | em = rb_entry(n, struct extent_map, rb_node); | |
1563 | ret = em->start + em->len; | |
0b86a832 | 1564 | } |
6df9a95e JB |
1565 | read_unlock(&em_tree->lock); |
1566 | ||
0b86a832 CM |
1567 | return ret; |
1568 | } | |
1569 | ||
53f10659 ID |
1570 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1571 | u64 *devid_ret) | |
0b86a832 CM |
1572 | { |
1573 | int ret; | |
1574 | struct btrfs_key key; | |
1575 | struct btrfs_key found_key; | |
2b82032c YZ |
1576 | struct btrfs_path *path; |
1577 | ||
2b82032c YZ |
1578 | path = btrfs_alloc_path(); |
1579 | if (!path) | |
1580 | return -ENOMEM; | |
0b86a832 CM |
1581 | |
1582 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1583 | key.type = BTRFS_DEV_ITEM_KEY; | |
1584 | key.offset = (u64)-1; | |
1585 | ||
53f10659 | 1586 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1587 | if (ret < 0) |
1588 | goto error; | |
1589 | ||
79787eaa | 1590 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1591 | |
53f10659 ID |
1592 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1593 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1594 | BTRFS_DEV_ITEM_KEY); |
1595 | if (ret) { | |
53f10659 | 1596 | *devid_ret = 1; |
0b86a832 CM |
1597 | } else { |
1598 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1599 | path->slots[0]); | |
53f10659 | 1600 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1601 | } |
1602 | ret = 0; | |
1603 | error: | |
2b82032c | 1604 | btrfs_free_path(path); |
0b86a832 CM |
1605 | return ret; |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * the device information is stored in the chunk root | |
1610 | * the btrfs_device struct should be fully filled in | |
1611 | */ | |
48a3b636 ES |
1612 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1613 | struct btrfs_root *root, | |
1614 | struct btrfs_device *device) | |
0b86a832 CM |
1615 | { |
1616 | int ret; | |
1617 | struct btrfs_path *path; | |
1618 | struct btrfs_dev_item *dev_item; | |
1619 | struct extent_buffer *leaf; | |
1620 | struct btrfs_key key; | |
1621 | unsigned long ptr; | |
0b86a832 CM |
1622 | |
1623 | root = root->fs_info->chunk_root; | |
1624 | ||
1625 | path = btrfs_alloc_path(); | |
1626 | if (!path) | |
1627 | return -ENOMEM; | |
1628 | ||
0b86a832 CM |
1629 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1630 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1631 | key.offset = device->devid; |
0b86a832 CM |
1632 | |
1633 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1634 | sizeof(*dev_item)); |
0b86a832 CM |
1635 | if (ret) |
1636 | goto out; | |
1637 | ||
1638 | leaf = path->nodes[0]; | |
1639 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1640 | ||
1641 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1642 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1643 | btrfs_set_device_type(leaf, dev_item, device->type); |
1644 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1645 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1646 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1647 | btrfs_set_device_total_bytes(leaf, dev_item, |
1648 | btrfs_device_get_disk_total_bytes(device)); | |
1649 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1650 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1651 | btrfs_set_device_group(leaf, dev_item, 0); |
1652 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1653 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1654 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1655 | |
410ba3a2 | 1656 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1657 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1658 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1659 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1660 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1661 | |
2b82032c | 1662 | ret = 0; |
0b86a832 CM |
1663 | out: |
1664 | btrfs_free_path(path); | |
1665 | return ret; | |
1666 | } | |
8f18cf13 | 1667 | |
5a1972bd QW |
1668 | /* |
1669 | * Function to update ctime/mtime for a given device path. | |
1670 | * Mainly used for ctime/mtime based probe like libblkid. | |
1671 | */ | |
1672 | static void update_dev_time(char *path_name) | |
1673 | { | |
1674 | struct file *filp; | |
1675 | ||
1676 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1677 | if (IS_ERR(filp)) |
5a1972bd QW |
1678 | return; |
1679 | file_update_time(filp); | |
1680 | filp_close(filp, NULL); | |
5a1972bd QW |
1681 | } |
1682 | ||
a061fc8d CM |
1683 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1684 | struct btrfs_device *device) | |
1685 | { | |
1686 | int ret; | |
1687 | struct btrfs_path *path; | |
a061fc8d | 1688 | struct btrfs_key key; |
a061fc8d CM |
1689 | struct btrfs_trans_handle *trans; |
1690 | ||
1691 | root = root->fs_info->chunk_root; | |
1692 | ||
1693 | path = btrfs_alloc_path(); | |
1694 | if (!path) | |
1695 | return -ENOMEM; | |
1696 | ||
a22285a6 | 1697 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1698 | if (IS_ERR(trans)) { |
1699 | btrfs_free_path(path); | |
1700 | return PTR_ERR(trans); | |
1701 | } | |
a061fc8d CM |
1702 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1703 | key.type = BTRFS_DEV_ITEM_KEY; | |
1704 | key.offset = device->devid; | |
1705 | ||
1706 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1707 | if (ret < 0) | |
1708 | goto out; | |
1709 | ||
1710 | if (ret > 0) { | |
1711 | ret = -ENOENT; | |
1712 | goto out; | |
1713 | } | |
1714 | ||
1715 | ret = btrfs_del_item(trans, root, path); | |
1716 | if (ret) | |
1717 | goto out; | |
a061fc8d CM |
1718 | out: |
1719 | btrfs_free_path(path); | |
1720 | btrfs_commit_transaction(trans, root); | |
1721 | return ret; | |
1722 | } | |
1723 | ||
3cc31a0d DS |
1724 | /* |
1725 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1726 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1727 | * replace. | |
1728 | */ | |
1729 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1730 | u64 num_devices) | |
a061fc8d | 1731 | { |
a061fc8d | 1732 | u64 all_avail; |
de98ced9 | 1733 | unsigned seq; |
418775a2 | 1734 | int i; |
a061fc8d | 1735 | |
de98ced9 | 1736 | do { |
bd45ffbc | 1737 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1738 | |
bd45ffbc AJ |
1739 | all_avail = fs_info->avail_data_alloc_bits | |
1740 | fs_info->avail_system_alloc_bits | | |
1741 | fs_info->avail_metadata_alloc_bits; | |
1742 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1743 | |
418775a2 DS |
1744 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1745 | if (!(all_avail & btrfs_raid_group[i])) | |
1746 | continue; | |
a061fc8d | 1747 | |
418775a2 DS |
1748 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1749 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1750 | |
418775a2 DS |
1751 | if (ret) |
1752 | return ret; | |
1753 | } | |
53b381b3 DW |
1754 | } |
1755 | ||
bd45ffbc | 1756 | return 0; |
f1fa7f26 AJ |
1757 | } |
1758 | ||
6b526ed7 | 1759 | int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid) |
f1fa7f26 AJ |
1760 | { |
1761 | struct btrfs_device *device; | |
1762 | struct btrfs_device *next_device; | |
f1fa7f26 | 1763 | struct btrfs_fs_devices *cur_devices; |
f1fa7f26 | 1764 | u64 num_devices; |
f1fa7f26 AJ |
1765 | int ret = 0; |
1766 | bool clear_super = false; | |
42b67427 | 1767 | char *dev_name = NULL; |
f1fa7f26 AJ |
1768 | |
1769 | mutex_lock(&uuid_mutex); | |
1770 | ||
3cc31a0d DS |
1771 | num_devices = root->fs_info->fs_devices->num_devices; |
1772 | btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0); | |
1773 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1774 | WARN_ON(num_devices < 1); | |
1775 | num_devices--; | |
1776 | } | |
1777 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0); | |
1778 | ||
1779 | ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1); | |
f1fa7f26 AJ |
1780 | if (ret) |
1781 | goto out; | |
1782 | ||
5c5c0df0 | 1783 | ret = btrfs_find_device_by_devspec(root, devid, device_path, |
24fc572f AJ |
1784 | &device); |
1785 | if (ret) | |
1786 | goto out; | |
dfe25020 | 1787 | |
63a212ab | 1788 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1789 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1790 | goto out; |
63a212ab SB |
1791 | } |
1792 | ||
2b82032c | 1793 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1794 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1795 | goto out; |
2b82032c YZ |
1796 | } |
1797 | ||
1798 | if (device->writeable) { | |
0c1daee0 | 1799 | lock_chunks(root); |
2b82032c | 1800 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1801 | device->fs_devices->rw_devices--; |
0c1daee0 | 1802 | unlock_chunks(root); |
42b67427 AJ |
1803 | dev_name = kstrdup(device->name->str, GFP_KERNEL); |
1804 | if (!dev_name) { | |
1805 | ret = -ENOMEM; | |
1806 | goto error_undo; | |
1807 | } | |
1f78160c | 1808 | clear_super = true; |
dfe25020 | 1809 | } |
a061fc8d | 1810 | |
d7901554 | 1811 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1812 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1813 | mutex_lock(&uuid_mutex); |
a061fc8d | 1814 | if (ret) |
9b3517e9 | 1815 | goto error_undo; |
a061fc8d | 1816 | |
63a212ab SB |
1817 | /* |
1818 | * TODO: the superblock still includes this device in its num_devices | |
1819 | * counter although write_all_supers() is not locked out. This | |
1820 | * could give a filesystem state which requires a degraded mount. | |
1821 | */ | |
a061fc8d CM |
1822 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1823 | if (ret) | |
9b3517e9 | 1824 | goto error_undo; |
a061fc8d | 1825 | |
2b82032c | 1826 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1827 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1828 | |
1829 | /* | |
1830 | * the device list mutex makes sure that we don't change | |
1831 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1832 | * the device supers. Whoever is writing all supers, should |
1833 | * lock the device list mutex before getting the number of | |
1834 | * devices in the super block (super_copy). Conversely, | |
1835 | * whoever updates the number of devices in the super block | |
1836 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1837 | */ |
1f78160c XG |
1838 | |
1839 | cur_devices = device->fs_devices; | |
e5e9a520 | 1840 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1841 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1842 | |
e4404d6e | 1843 | device->fs_devices->num_devices--; |
02db0844 | 1844 | device->fs_devices->total_devices--; |
2b82032c | 1845 | |
cd02dca5 | 1846 | if (device->missing) |
3a7d55c8 | 1847 | device->fs_devices->missing_devices--; |
cd02dca5 | 1848 | |
2b82032c YZ |
1849 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1850 | struct btrfs_device, dev_list); | |
8ed01abe AJ |
1851 | if (root->fs_info->sb->s_bdev && |
1852 | (root->fs_info->sb->s_bdev == device->bdev)) | |
2b82032c YZ |
1853 | root->fs_info->sb->s_bdev = next_device->bdev; |
1854 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1855 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1856 | ||
0bfaa9c5 | 1857 | if (device->bdev) { |
e4404d6e | 1858 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1859 | /* remove sysfs entry */ |
32576040 | 1860 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); |
0bfaa9c5 | 1861 | } |
99994cde | 1862 | |
1f78160c | 1863 | call_rcu(&device->rcu, free_device); |
e4404d6e | 1864 | |
6c41761f DS |
1865 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1866 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
d7306801 | 1867 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2b82032c | 1868 | |
1f78160c | 1869 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1870 | struct btrfs_fs_devices *fs_devices; |
1871 | fs_devices = root->fs_info->fs_devices; | |
1872 | while (fs_devices) { | |
8321cf25 RS |
1873 | if (fs_devices->seed == cur_devices) { |
1874 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1875 | break; |
8321cf25 | 1876 | } |
e4404d6e | 1877 | fs_devices = fs_devices->seed; |
2b82032c | 1878 | } |
1f78160c | 1879 | cur_devices->seed = NULL; |
1f78160c | 1880 | __btrfs_close_devices(cur_devices); |
1f78160c | 1881 | free_fs_devices(cur_devices); |
2b82032c YZ |
1882 | } |
1883 | ||
5af3e8cc SB |
1884 | root->fs_info->num_tolerated_disk_barrier_failures = |
1885 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1886 | ||
2b82032c YZ |
1887 | /* |
1888 | * at this point, the device is zero sized. We want to | |
1889 | * remove it from the devices list and zero out the old super | |
1890 | */ | |
24fc572f | 1891 | if (clear_super) { |
42b67427 | 1892 | struct block_device *bdev; |
4d90d28b | 1893 | |
42b67427 AJ |
1894 | bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL, |
1895 | root->fs_info->bdev_holder); | |
1896 | if (!IS_ERR(bdev)) { | |
1897 | btrfs_scratch_superblocks(bdev, dev_name); | |
24fc572f AJ |
1898 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
1899 | } | |
5a1972bd QW |
1900 | } |
1901 | ||
a061fc8d | 1902 | out: |
42b67427 AJ |
1903 | kfree(dev_name); |
1904 | ||
a061fc8d | 1905 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1906 | return ret; |
24fc572f | 1907 | |
9b3517e9 ID |
1908 | error_undo: |
1909 | if (device->writeable) { | |
0c1daee0 | 1910 | lock_chunks(root); |
9b3517e9 ID |
1911 | list_add(&device->dev_alloc_list, |
1912 | &root->fs_info->fs_devices->alloc_list); | |
c3929c36 | 1913 | device->fs_devices->rw_devices++; |
0c1daee0 | 1914 | unlock_chunks(root); |
9b3517e9 | 1915 | } |
24fc572f | 1916 | goto out; |
a061fc8d CM |
1917 | } |
1918 | ||
084b6e7c QW |
1919 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1920 | struct btrfs_device *srcdev) | |
e93c89c1 | 1921 | { |
d51908ce AJ |
1922 | struct btrfs_fs_devices *fs_devices; |
1923 | ||
e93c89c1 | 1924 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 1925 | |
25e8e911 AJ |
1926 | /* |
1927 | * in case of fs with no seed, srcdev->fs_devices will point | |
1928 | * to fs_devices of fs_info. However when the dev being replaced is | |
1929 | * a seed dev it will point to the seed's local fs_devices. In short | |
1930 | * srcdev will have its correct fs_devices in both the cases. | |
1931 | */ | |
1932 | fs_devices = srcdev->fs_devices; | |
d51908ce | 1933 | |
e93c89c1 SB |
1934 | list_del_rcu(&srcdev->dev_list); |
1935 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 1936 | fs_devices->num_devices--; |
82372bc8 | 1937 | if (srcdev->missing) |
d51908ce | 1938 | fs_devices->missing_devices--; |
e93c89c1 | 1939 | |
82372bc8 MX |
1940 | if (srcdev->writeable) { |
1941 | fs_devices->rw_devices--; | |
1942 | /* zero out the old super if it is writable */ | |
31388ab2 | 1943 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); |
1357272f ID |
1944 | } |
1945 | ||
82372bc8 | 1946 | if (srcdev->bdev) |
d51908ce | 1947 | fs_devices->open_devices--; |
084b6e7c QW |
1948 | } |
1949 | ||
1950 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
1951 | struct btrfs_device *srcdev) | |
1952 | { | |
1953 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 SB |
1954 | |
1955 | call_rcu(&srcdev->rcu, free_device); | |
94d5f0c2 AJ |
1956 | |
1957 | /* | |
1958 | * unless fs_devices is seed fs, num_devices shouldn't go | |
1959 | * zero | |
1960 | */ | |
1961 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
1962 | ||
1963 | /* if this is no devs we rather delete the fs_devices */ | |
1964 | if (!fs_devices->num_devices) { | |
1965 | struct btrfs_fs_devices *tmp_fs_devices; | |
1966 | ||
1967 | tmp_fs_devices = fs_info->fs_devices; | |
1968 | while (tmp_fs_devices) { | |
1969 | if (tmp_fs_devices->seed == fs_devices) { | |
1970 | tmp_fs_devices->seed = fs_devices->seed; | |
1971 | break; | |
1972 | } | |
1973 | tmp_fs_devices = tmp_fs_devices->seed; | |
1974 | } | |
1975 | fs_devices->seed = NULL; | |
8bef8401 AJ |
1976 | __btrfs_close_devices(fs_devices); |
1977 | free_fs_devices(fs_devices); | |
94d5f0c2 | 1978 | } |
e93c89c1 SB |
1979 | } |
1980 | ||
1981 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
1982 | struct btrfs_device *tgtdev) | |
1983 | { | |
1984 | struct btrfs_device *next_device; | |
1985 | ||
67a2c45e | 1986 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
1987 | WARN_ON(!tgtdev); |
1988 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 1989 | |
32576040 | 1990 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 1991 | |
e93c89c1 | 1992 | if (tgtdev->bdev) { |
31388ab2 | 1993 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); |
e93c89c1 SB |
1994 | fs_info->fs_devices->open_devices--; |
1995 | } | |
1996 | fs_info->fs_devices->num_devices--; | |
e93c89c1 SB |
1997 | |
1998 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
1999 | struct btrfs_device, dev_list); | |
8ed01abe AJ |
2000 | if (fs_info->sb->s_bdev && |
2001 | (tgtdev->bdev == fs_info->sb->s_bdev)) | |
e93c89c1 SB |
2002 | fs_info->sb->s_bdev = next_device->bdev; |
2003 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
2004 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2005 | list_del_rcu(&tgtdev->dev_list); | |
2006 | ||
2007 | call_rcu(&tgtdev->rcu, free_device); | |
2008 | ||
2009 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2010 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2011 | } |
2012 | ||
48a3b636 ES |
2013 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
2014 | struct btrfs_device **device) | |
7ba15b7d SB |
2015 | { |
2016 | int ret = 0; | |
2017 | struct btrfs_super_block *disk_super; | |
2018 | u64 devid; | |
2019 | u8 *dev_uuid; | |
2020 | struct block_device *bdev; | |
2021 | struct buffer_head *bh; | |
2022 | ||
2023 | *device = NULL; | |
2024 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
2025 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
2026 | if (ret) | |
2027 | return ret; | |
2028 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2029 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2030 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 2031 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
2032 | disk_super->fsid); |
2033 | brelse(bh); | |
2034 | if (!*device) | |
2035 | ret = -ENOENT; | |
2036 | blkdev_put(bdev, FMODE_READ); | |
2037 | return ret; | |
2038 | } | |
2039 | ||
2040 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
2041 | char *device_path, | |
2042 | struct btrfs_device **device) | |
2043 | { | |
2044 | *device = NULL; | |
2045 | if (strcmp(device_path, "missing") == 0) { | |
2046 | struct list_head *devices; | |
2047 | struct btrfs_device *tmp; | |
2048 | ||
2049 | devices = &root->fs_info->fs_devices->devices; | |
2050 | /* | |
2051 | * It is safe to read the devices since the volume_mutex | |
2052 | * is held by the caller. | |
2053 | */ | |
2054 | list_for_each_entry(tmp, devices, dev_list) { | |
2055 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2056 | *device = tmp; | |
2057 | break; | |
2058 | } | |
2059 | } | |
2060 | ||
d74a6259 AJ |
2061 | if (!*device) |
2062 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2063 | |
2064 | return 0; | |
2065 | } else { | |
2066 | return btrfs_find_device_by_path(root, device_path, device); | |
2067 | } | |
2068 | } | |
2069 | ||
5c5c0df0 DS |
2070 | /* |
2071 | * Lookup a device given by device id, or the path if the id is 0. | |
2072 | */ | |
2073 | int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid, | |
2074 | char *devpath, | |
24e0474b AJ |
2075 | struct btrfs_device **device) |
2076 | { | |
2077 | int ret; | |
2078 | ||
5c5c0df0 | 2079 | if (devid) { |
24e0474b | 2080 | ret = 0; |
5c5c0df0 | 2081 | *device = btrfs_find_device(root->fs_info, devid, NULL, |
24e0474b AJ |
2082 | NULL); |
2083 | if (!*device) | |
2084 | ret = -ENOENT; | |
2085 | } else { | |
5c5c0df0 | 2086 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2087 | return -EINVAL; |
2088 | ||
5c5c0df0 | 2089 | ret = btrfs_find_device_missing_or_by_path(root, devpath, |
24e0474b AJ |
2090 | device); |
2091 | } | |
2092 | return ret; | |
2093 | } | |
2094 | ||
2b82032c YZ |
2095 | /* |
2096 | * does all the dirty work required for changing file system's UUID. | |
2097 | */ | |
125ccb0a | 2098 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
2099 | { |
2100 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
2101 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 2102 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 2103 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
2104 | struct btrfs_device *device; |
2105 | u64 super_flags; | |
2106 | ||
2107 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2108 | if (!fs_devices->seeding) |
2b82032c YZ |
2109 | return -EINVAL; |
2110 | ||
2208a378 ID |
2111 | seed_devices = __alloc_fs_devices(); |
2112 | if (IS_ERR(seed_devices)) | |
2113 | return PTR_ERR(seed_devices); | |
2b82032c | 2114 | |
e4404d6e YZ |
2115 | old_devices = clone_fs_devices(fs_devices); |
2116 | if (IS_ERR(old_devices)) { | |
2117 | kfree(seed_devices); | |
2118 | return PTR_ERR(old_devices); | |
2b82032c | 2119 | } |
e4404d6e | 2120 | |
2b82032c YZ |
2121 | list_add(&old_devices->list, &fs_uuids); |
2122 | ||
e4404d6e YZ |
2123 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2124 | seed_devices->opened = 1; | |
2125 | INIT_LIST_HEAD(&seed_devices->devices); | |
2126 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2127 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
2128 | |
2129 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
2130 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2131 | synchronize_rcu); | |
2196d6e8 MX |
2132 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2133 | device->fs_devices = seed_devices; | |
c9513edb | 2134 | |
2196d6e8 | 2135 | lock_chunks(root); |
e4404d6e | 2136 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
2196d6e8 | 2137 | unlock_chunks(root); |
e4404d6e | 2138 | |
2b82032c YZ |
2139 | fs_devices->seeding = 0; |
2140 | fs_devices->num_devices = 0; | |
2141 | fs_devices->open_devices = 0; | |
69611ac8 | 2142 | fs_devices->missing_devices = 0; |
69611ac8 | 2143 | fs_devices->rotating = 0; |
e4404d6e | 2144 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2145 | |
2146 | generate_random_uuid(fs_devices->fsid); | |
2147 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2148 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
f7171750 FDBM |
2149 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2150 | ||
2b82032c YZ |
2151 | super_flags = btrfs_super_flags(disk_super) & |
2152 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2153 | btrfs_set_super_flags(disk_super, super_flags); | |
2154 | ||
2155 | return 0; | |
2156 | } | |
2157 | ||
2158 | /* | |
2159 | * strore the expected generation for seed devices in device items. | |
2160 | */ | |
2161 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2162 | struct btrfs_root *root) | |
2163 | { | |
2164 | struct btrfs_path *path; | |
2165 | struct extent_buffer *leaf; | |
2166 | struct btrfs_dev_item *dev_item; | |
2167 | struct btrfs_device *device; | |
2168 | struct btrfs_key key; | |
2169 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2170 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2171 | u64 devid; | |
2172 | int ret; | |
2173 | ||
2174 | path = btrfs_alloc_path(); | |
2175 | if (!path) | |
2176 | return -ENOMEM; | |
2177 | ||
2178 | root = root->fs_info->chunk_root; | |
2179 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2180 | key.offset = 0; | |
2181 | key.type = BTRFS_DEV_ITEM_KEY; | |
2182 | ||
2183 | while (1) { | |
2184 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2185 | if (ret < 0) | |
2186 | goto error; | |
2187 | ||
2188 | leaf = path->nodes[0]; | |
2189 | next_slot: | |
2190 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2191 | ret = btrfs_next_leaf(root, path); | |
2192 | if (ret > 0) | |
2193 | break; | |
2194 | if (ret < 0) | |
2195 | goto error; | |
2196 | leaf = path->nodes[0]; | |
2197 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2198 | btrfs_release_path(path); |
2b82032c YZ |
2199 | continue; |
2200 | } | |
2201 | ||
2202 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2203 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2204 | key.type != BTRFS_DEV_ITEM_KEY) | |
2205 | break; | |
2206 | ||
2207 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2208 | struct btrfs_dev_item); | |
2209 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2210 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2211 | BTRFS_UUID_SIZE); |
1473b24e | 2212 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2213 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
2214 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2215 | fs_uuid); | |
79787eaa | 2216 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2217 | |
2218 | if (device->fs_devices->seeding) { | |
2219 | btrfs_set_device_generation(leaf, dev_item, | |
2220 | device->generation); | |
2221 | btrfs_mark_buffer_dirty(leaf); | |
2222 | } | |
2223 | ||
2224 | path->slots[0]++; | |
2225 | goto next_slot; | |
2226 | } | |
2227 | ret = 0; | |
2228 | error: | |
2229 | btrfs_free_path(path); | |
2230 | return ret; | |
2231 | } | |
2232 | ||
788f20eb CM |
2233 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
2234 | { | |
d5e2003c | 2235 | struct request_queue *q; |
788f20eb CM |
2236 | struct btrfs_trans_handle *trans; |
2237 | struct btrfs_device *device; | |
2238 | struct block_device *bdev; | |
788f20eb | 2239 | struct list_head *devices; |
2b82032c | 2240 | struct super_block *sb = root->fs_info->sb; |
606686ee | 2241 | struct rcu_string *name; |
3c1dbdf5 | 2242 | u64 tmp; |
2b82032c | 2243 | int seeding_dev = 0; |
788f20eb CM |
2244 | int ret = 0; |
2245 | ||
2b82032c | 2246 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 2247 | return -EROFS; |
788f20eb | 2248 | |
a5d16333 | 2249 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 2250 | root->fs_info->bdev_holder); |
7f59203a JB |
2251 | if (IS_ERR(bdev)) |
2252 | return PTR_ERR(bdev); | |
a2135011 | 2253 | |
2b82032c YZ |
2254 | if (root->fs_info->fs_devices->seeding) { |
2255 | seeding_dev = 1; | |
2256 | down_write(&sb->s_umount); | |
2257 | mutex_lock(&uuid_mutex); | |
2258 | } | |
2259 | ||
8c8bee1d | 2260 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2261 | |
788f20eb | 2262 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
2263 | |
2264 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 2265 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2266 | if (device->bdev == bdev) { |
2267 | ret = -EEXIST; | |
d25628bd LB |
2268 | mutex_unlock( |
2269 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2270 | goto error; |
788f20eb CM |
2271 | } |
2272 | } | |
d25628bd | 2273 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2274 | |
12bd2fc0 ID |
2275 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
2276 | if (IS_ERR(device)) { | |
788f20eb | 2277 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2278 | ret = PTR_ERR(device); |
2b82032c | 2279 | goto error; |
788f20eb CM |
2280 | } |
2281 | ||
78f2c9e6 | 2282 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2283 | if (!name) { |
788f20eb | 2284 | kfree(device); |
2b82032c YZ |
2285 | ret = -ENOMEM; |
2286 | goto error; | |
788f20eb | 2287 | } |
606686ee | 2288 | rcu_assign_pointer(device->name, name); |
2b82032c | 2289 | |
a22285a6 | 2290 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2291 | if (IS_ERR(trans)) { |
606686ee | 2292 | rcu_string_free(device->name); |
98d5dc13 TI |
2293 | kfree(device); |
2294 | ret = PTR_ERR(trans); | |
2295 | goto error; | |
2296 | } | |
2297 | ||
d5e2003c JB |
2298 | q = bdev_get_queue(bdev); |
2299 | if (blk_queue_discard(q)) | |
2300 | device->can_discard = 1; | |
2b82032c | 2301 | device->writeable = 1; |
2b82032c | 2302 | device->generation = trans->transid; |
788f20eb CM |
2303 | device->io_width = root->sectorsize; |
2304 | device->io_align = root->sectorsize; | |
2305 | device->sector_size = root->sectorsize; | |
2306 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2307 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2308 | device->commit_total_bytes = device->total_bytes; |
788f20eb CM |
2309 | device->dev_root = root->fs_info->dev_root; |
2310 | device->bdev = bdev; | |
dfe25020 | 2311 | device->in_fs_metadata = 1; |
63a212ab | 2312 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2313 | device->mode = FMODE_EXCL; |
27087f37 | 2314 | device->dev_stats_valid = 1; |
2b82032c | 2315 | set_blocksize(device->bdev, 4096); |
788f20eb | 2316 | |
2b82032c YZ |
2317 | if (seeding_dev) { |
2318 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2319 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2320 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2321 | } |
788f20eb | 2322 | |
2b82032c | 2323 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2324 | |
e5e9a520 | 2325 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
2196d6e8 | 2326 | lock_chunks(root); |
1f78160c | 2327 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2328 | list_add(&device->dev_alloc_list, |
2329 | &root->fs_info->fs_devices->alloc_list); | |
2330 | root->fs_info->fs_devices->num_devices++; | |
2331 | root->fs_info->fs_devices->open_devices++; | |
2332 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2333 | root->fs_info->fs_devices->total_devices++; |
2b82032c | 2334 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2335 | |
2bf64758 JB |
2336 | spin_lock(&root->fs_info->free_chunk_lock); |
2337 | root->fs_info->free_chunk_space += device->total_bytes; | |
2338 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2339 | ||
c289811c CM |
2340 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2341 | root->fs_info->fs_devices->rotating = 1; | |
2342 | ||
3c1dbdf5 | 2343 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); |
6c41761f | 2344 | btrfs_set_super_total_bytes(root->fs_info->super_copy, |
3c1dbdf5 | 2345 | tmp + device->total_bytes); |
788f20eb | 2346 | |
3c1dbdf5 | 2347 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); |
6c41761f | 2348 | btrfs_set_super_num_devices(root->fs_info->super_copy, |
3c1dbdf5 | 2349 | tmp + 1); |
0d39376a AJ |
2350 | |
2351 | /* add sysfs device entry */ | |
e3bd6973 | 2352 | btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device); |
0d39376a | 2353 | |
2196d6e8 MX |
2354 | /* |
2355 | * we've got more storage, clear any full flags on the space | |
2356 | * infos | |
2357 | */ | |
2358 | btrfs_clear_space_info_full(root->fs_info); | |
2359 | ||
2360 | unlock_chunks(root); | |
e5e9a520 | 2361 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2362 | |
2b82032c | 2363 | if (seeding_dev) { |
2196d6e8 | 2364 | lock_chunks(root); |
2b82032c | 2365 | ret = init_first_rw_device(trans, root, device); |
2196d6e8 | 2366 | unlock_chunks(root); |
005d6427 DS |
2367 | if (ret) { |
2368 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2369 | goto error_trans; |
005d6427 | 2370 | } |
2196d6e8 MX |
2371 | } |
2372 | ||
2373 | ret = btrfs_add_device(trans, root, device); | |
2374 | if (ret) { | |
2375 | btrfs_abort_transaction(trans, root, ret); | |
2376 | goto error_trans; | |
2377 | } | |
2378 | ||
2379 | if (seeding_dev) { | |
2380 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2381 | ||
2b82032c | 2382 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2383 | if (ret) { |
2384 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2385 | goto error_trans; |
005d6427 | 2386 | } |
b2373f25 AJ |
2387 | |
2388 | /* Sprouting would change fsid of the mounted root, | |
2389 | * so rename the fsid on the sysfs | |
2390 | */ | |
2391 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2392 | root->fs_info->fsid); | |
c1b7e474 | 2393 | if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj, |
2e7910d6 | 2394 | fsid_buf)) |
f14d104d DS |
2395 | btrfs_warn(root->fs_info, |
2396 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2397 | } |
2398 | ||
5af3e8cc SB |
2399 | root->fs_info->num_tolerated_disk_barrier_failures = |
2400 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2401 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2402 | |
2b82032c YZ |
2403 | if (seeding_dev) { |
2404 | mutex_unlock(&uuid_mutex); | |
2405 | up_write(&sb->s_umount); | |
788f20eb | 2406 | |
79787eaa JM |
2407 | if (ret) /* transaction commit */ |
2408 | return ret; | |
2409 | ||
2b82032c | 2410 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa | 2411 | if (ret < 0) |
a4553fef | 2412 | btrfs_std_error(root->fs_info, ret, |
79787eaa JM |
2413 | "Failed to relocate sys chunks after " |
2414 | "device initialization. This can be fixed " | |
2415 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2416 | trans = btrfs_attach_transaction(root); |
2417 | if (IS_ERR(trans)) { | |
2418 | if (PTR_ERR(trans) == -ENOENT) | |
2419 | return 0; | |
2420 | return PTR_ERR(trans); | |
2421 | } | |
2422 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2423 | } |
c9e9f97b | 2424 | |
5a1972bd QW |
2425 | /* Update ctime/mtime for libblkid */ |
2426 | update_dev_time(device_path); | |
2b82032c | 2427 | return ret; |
79787eaa JM |
2428 | |
2429 | error_trans: | |
79787eaa | 2430 | btrfs_end_transaction(trans, root); |
606686ee | 2431 | rcu_string_free(device->name); |
32576040 | 2432 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); |
79787eaa | 2433 | kfree(device); |
2b82032c | 2434 | error: |
e525fd89 | 2435 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2436 | if (seeding_dev) { |
2437 | mutex_unlock(&uuid_mutex); | |
2438 | up_write(&sb->s_umount); | |
2439 | } | |
c9e9f97b | 2440 | return ret; |
788f20eb CM |
2441 | } |
2442 | ||
e93c89c1 | 2443 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2444 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2445 | struct btrfs_device **device_out) |
2446 | { | |
2447 | struct request_queue *q; | |
2448 | struct btrfs_device *device; | |
2449 | struct block_device *bdev; | |
2450 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2451 | struct list_head *devices; | |
2452 | struct rcu_string *name; | |
12bd2fc0 | 2453 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2454 | int ret = 0; |
2455 | ||
2456 | *device_out = NULL; | |
1c43366d MX |
2457 | if (fs_info->fs_devices->seeding) { |
2458 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2459 | return -EINVAL; |
1c43366d | 2460 | } |
e93c89c1 SB |
2461 | |
2462 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2463 | fs_info->bdev_holder); | |
1c43366d MX |
2464 | if (IS_ERR(bdev)) { |
2465 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2466 | return PTR_ERR(bdev); |
1c43366d | 2467 | } |
e93c89c1 SB |
2468 | |
2469 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2470 | ||
2471 | devices = &fs_info->fs_devices->devices; | |
2472 | list_for_each_entry(device, devices, dev_list) { | |
2473 | if (device->bdev == bdev) { | |
1c43366d | 2474 | btrfs_err(fs_info, "target device is in the filesystem!"); |
e93c89c1 SB |
2475 | ret = -EEXIST; |
2476 | goto error; | |
2477 | } | |
2478 | } | |
2479 | ||
1c43366d | 2480 | |
7cc8e58d MX |
2481 | if (i_size_read(bdev->bd_inode) < |
2482 | btrfs_device_get_total_bytes(srcdev)) { | |
1c43366d MX |
2483 | btrfs_err(fs_info, "target device is smaller than source device!"); |
2484 | ret = -EINVAL; | |
2485 | goto error; | |
2486 | } | |
2487 | ||
2488 | ||
12bd2fc0 ID |
2489 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2490 | if (IS_ERR(device)) { | |
2491 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2492 | goto error; |
2493 | } | |
2494 | ||
2495 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2496 | if (!name) { | |
2497 | kfree(device); | |
2498 | ret = -ENOMEM; | |
2499 | goto error; | |
2500 | } | |
2501 | rcu_assign_pointer(device->name, name); | |
2502 | ||
2503 | q = bdev_get_queue(bdev); | |
2504 | if (blk_queue_discard(q)) | |
2505 | device->can_discard = 1; | |
2506 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2507 | device->writeable = 1; | |
e93c89c1 SB |
2508 | device->generation = 0; |
2509 | device->io_width = root->sectorsize; | |
2510 | device->io_align = root->sectorsize; | |
2511 | device->sector_size = root->sectorsize; | |
7cc8e58d MX |
2512 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2513 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2514 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2515 | ASSERT(list_empty(&srcdev->resized_list)); |
2516 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2517 | device->commit_bytes_used = device->bytes_used; |
e93c89c1 SB |
2518 | device->dev_root = fs_info->dev_root; |
2519 | device->bdev = bdev; | |
2520 | device->in_fs_metadata = 1; | |
2521 | device->is_tgtdev_for_dev_replace = 1; | |
2522 | device->mode = FMODE_EXCL; | |
27087f37 | 2523 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2524 | set_blocksize(device->bdev, 4096); |
2525 | device->fs_devices = fs_info->fs_devices; | |
2526 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2527 | fs_info->fs_devices->num_devices++; | |
2528 | fs_info->fs_devices->open_devices++; | |
e93c89c1 SB |
2529 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2530 | ||
2531 | *device_out = device; | |
2532 | return ret; | |
2533 | ||
2534 | error: | |
2535 | blkdev_put(bdev, FMODE_EXCL); | |
2536 | return ret; | |
2537 | } | |
2538 | ||
2539 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2540 | struct btrfs_device *tgtdev) | |
2541 | { | |
2542 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2543 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2544 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2545 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2546 | tgtdev->dev_root = fs_info->dev_root; | |
2547 | tgtdev->in_fs_metadata = 1; | |
2548 | } | |
2549 | ||
d397712b CM |
2550 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2551 | struct btrfs_device *device) | |
0b86a832 CM |
2552 | { |
2553 | int ret; | |
2554 | struct btrfs_path *path; | |
2555 | struct btrfs_root *root; | |
2556 | struct btrfs_dev_item *dev_item; | |
2557 | struct extent_buffer *leaf; | |
2558 | struct btrfs_key key; | |
2559 | ||
2560 | root = device->dev_root->fs_info->chunk_root; | |
2561 | ||
2562 | path = btrfs_alloc_path(); | |
2563 | if (!path) | |
2564 | return -ENOMEM; | |
2565 | ||
2566 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2567 | key.type = BTRFS_DEV_ITEM_KEY; | |
2568 | key.offset = device->devid; | |
2569 | ||
2570 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2571 | if (ret < 0) | |
2572 | goto out; | |
2573 | ||
2574 | if (ret > 0) { | |
2575 | ret = -ENOENT; | |
2576 | goto out; | |
2577 | } | |
2578 | ||
2579 | leaf = path->nodes[0]; | |
2580 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2581 | ||
2582 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2583 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2584 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2585 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2586 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2587 | btrfs_set_device_total_bytes(leaf, dev_item, |
2588 | btrfs_device_get_disk_total_bytes(device)); | |
2589 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2590 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2591 | btrfs_mark_buffer_dirty(leaf); |
2592 | ||
2593 | out: | |
2594 | btrfs_free_path(path); | |
2595 | return ret; | |
2596 | } | |
2597 | ||
2196d6e8 | 2598 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2599 | struct btrfs_device *device, u64 new_size) |
2600 | { | |
2601 | struct btrfs_super_block *super_copy = | |
6c41761f | 2602 | device->dev_root->fs_info->super_copy; |
935e5cc9 | 2603 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2604 | u64 old_total; |
2605 | u64 diff; | |
8f18cf13 | 2606 | |
2b82032c YZ |
2607 | if (!device->writeable) |
2608 | return -EACCES; | |
2196d6e8 MX |
2609 | |
2610 | lock_chunks(device->dev_root); | |
2611 | old_total = btrfs_super_total_bytes(super_copy); | |
2612 | diff = new_size - device->total_bytes; | |
2613 | ||
63a212ab | 2614 | if (new_size <= device->total_bytes || |
2196d6e8 MX |
2615 | device->is_tgtdev_for_dev_replace) { |
2616 | unlock_chunks(device->dev_root); | |
2b82032c | 2617 | return -EINVAL; |
2196d6e8 | 2618 | } |
2b82032c | 2619 | |
935e5cc9 | 2620 | fs_devices = device->dev_root->fs_info->fs_devices; |
2b82032c | 2621 | |
8f18cf13 | 2622 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2623 | device->fs_devices->total_rw_bytes += diff; |
2624 | ||
7cc8e58d MX |
2625 | btrfs_device_set_total_bytes(device, new_size); |
2626 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4184ea7f | 2627 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
935e5cc9 MX |
2628 | if (list_empty(&device->resized_list)) |
2629 | list_add_tail(&device->resized_list, | |
2630 | &fs_devices->resized_devices); | |
2196d6e8 | 2631 | unlock_chunks(device->dev_root); |
4184ea7f | 2632 | |
8f18cf13 CM |
2633 | return btrfs_update_device(trans, device); |
2634 | } | |
2635 | ||
2636 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
a688a04a | 2637 | struct btrfs_root *root, u64 chunk_objectid, |
8f18cf13 CM |
2638 | u64 chunk_offset) |
2639 | { | |
2640 | int ret; | |
2641 | struct btrfs_path *path; | |
2642 | struct btrfs_key key; | |
2643 | ||
2644 | root = root->fs_info->chunk_root; | |
2645 | path = btrfs_alloc_path(); | |
2646 | if (!path) | |
2647 | return -ENOMEM; | |
2648 | ||
2649 | key.objectid = chunk_objectid; | |
2650 | key.offset = chunk_offset; | |
2651 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2652 | ||
2653 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2654 | if (ret < 0) |
2655 | goto out; | |
2656 | else if (ret > 0) { /* Logic error or corruption */ | |
a4553fef | 2657 | btrfs_std_error(root->fs_info, -ENOENT, |
79787eaa JM |
2658 | "Failed lookup while freeing chunk."); |
2659 | ret = -ENOENT; | |
2660 | goto out; | |
2661 | } | |
8f18cf13 CM |
2662 | |
2663 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2664 | if (ret < 0) |
a4553fef | 2665 | btrfs_std_error(root->fs_info, ret, |
79787eaa JM |
2666 | "Failed to delete chunk item."); |
2667 | out: | |
8f18cf13 | 2668 | btrfs_free_path(path); |
65a246c5 | 2669 | return ret; |
8f18cf13 CM |
2670 | } |
2671 | ||
b2950863 | 2672 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2673 | chunk_offset) |
2674 | { | |
6c41761f | 2675 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2676 | struct btrfs_disk_key *disk_key; |
2677 | struct btrfs_chunk *chunk; | |
2678 | u8 *ptr; | |
2679 | int ret = 0; | |
2680 | u32 num_stripes; | |
2681 | u32 array_size; | |
2682 | u32 len = 0; | |
2683 | u32 cur; | |
2684 | struct btrfs_key key; | |
2685 | ||
2196d6e8 | 2686 | lock_chunks(root); |
8f18cf13 CM |
2687 | array_size = btrfs_super_sys_array_size(super_copy); |
2688 | ||
2689 | ptr = super_copy->sys_chunk_array; | |
2690 | cur = 0; | |
2691 | ||
2692 | while (cur < array_size) { | |
2693 | disk_key = (struct btrfs_disk_key *)ptr; | |
2694 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2695 | ||
2696 | len = sizeof(*disk_key); | |
2697 | ||
2698 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2699 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2700 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2701 | len += btrfs_chunk_item_size(num_stripes); | |
2702 | } else { | |
2703 | ret = -EIO; | |
2704 | break; | |
2705 | } | |
2706 | if (key.objectid == chunk_objectid && | |
2707 | key.offset == chunk_offset) { | |
2708 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2709 | array_size -= len; | |
2710 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2711 | } else { | |
2712 | ptr += len; | |
2713 | cur += len; | |
2714 | } | |
2715 | } | |
2196d6e8 | 2716 | unlock_chunks(root); |
8f18cf13 CM |
2717 | return ret; |
2718 | } | |
2719 | ||
47ab2a6c JB |
2720 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
2721 | struct btrfs_root *root, u64 chunk_offset) | |
8f18cf13 CM |
2722 | { |
2723 | struct extent_map_tree *em_tree; | |
8f18cf13 | 2724 | struct extent_map *em; |
47ab2a6c | 2725 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
8f18cf13 | 2726 | struct map_lookup *map; |
2196d6e8 | 2727 | u64 dev_extent_len = 0; |
47ab2a6c | 2728 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2729 | int i, ret = 0; |
8f18cf13 | 2730 | |
47ab2a6c | 2731 | /* Just in case */ |
8f18cf13 | 2732 | root = root->fs_info->chunk_root; |
8f18cf13 CM |
2733 | em_tree = &root->fs_info->mapping_tree.map_tree; |
2734 | ||
890871be | 2735 | read_lock(&em_tree->lock); |
8f18cf13 | 2736 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2737 | read_unlock(&em_tree->lock); |
8f18cf13 | 2738 | |
47ab2a6c JB |
2739 | if (!em || em->start > chunk_offset || |
2740 | em->start + em->len < chunk_offset) { | |
2741 | /* | |
2742 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2743 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2744 | * do anything we still error out. |
2745 | */ | |
2746 | ASSERT(0); | |
2747 | if (em) | |
2748 | free_extent_map(em); | |
2749 | return -EINVAL; | |
2750 | } | |
95617d69 | 2751 | map = em->map_lookup; |
39c2d7fa | 2752 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 2753 | check_system_chunk(trans, extent_root, map->type); |
39c2d7fa | 2754 | unlock_chunks(root->fs_info->chunk_root); |
8f18cf13 CM |
2755 | |
2756 | for (i = 0; i < map->num_stripes; i++) { | |
47ab2a6c | 2757 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2758 | ret = btrfs_free_dev_extent(trans, device, |
2759 | map->stripes[i].physical, | |
2760 | &dev_extent_len); | |
47ab2a6c JB |
2761 | if (ret) { |
2762 | btrfs_abort_transaction(trans, root, ret); | |
2763 | goto out; | |
2764 | } | |
a061fc8d | 2765 | |
2196d6e8 MX |
2766 | if (device->bytes_used > 0) { |
2767 | lock_chunks(root); | |
2768 | btrfs_device_set_bytes_used(device, | |
2769 | device->bytes_used - dev_extent_len); | |
2770 | spin_lock(&root->fs_info->free_chunk_lock); | |
2771 | root->fs_info->free_chunk_space += dev_extent_len; | |
2772 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2773 | btrfs_clear_space_info_full(root->fs_info); | |
2774 | unlock_chunks(root); | |
2775 | } | |
a061fc8d | 2776 | |
dfe25020 CM |
2777 | if (map->stripes[i].dev) { |
2778 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c JB |
2779 | if (ret) { |
2780 | btrfs_abort_transaction(trans, root, ret); | |
2781 | goto out; | |
2782 | } | |
dfe25020 | 2783 | } |
8f18cf13 | 2784 | } |
a688a04a | 2785 | ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset); |
47ab2a6c JB |
2786 | if (ret) { |
2787 | btrfs_abort_transaction(trans, root, ret); | |
2788 | goto out; | |
2789 | } | |
8f18cf13 | 2790 | |
1abe9b8a | 2791 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2792 | ||
8f18cf13 CM |
2793 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2794 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
47ab2a6c JB |
2795 | if (ret) { |
2796 | btrfs_abort_transaction(trans, root, ret); | |
2797 | goto out; | |
2798 | } | |
8f18cf13 CM |
2799 | } |
2800 | ||
04216820 | 2801 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em); |
47ab2a6c JB |
2802 | if (ret) { |
2803 | btrfs_abort_transaction(trans, extent_root, ret); | |
2804 | goto out; | |
2805 | } | |
2b82032c | 2806 | |
47ab2a6c | 2807 | out: |
2b82032c YZ |
2808 | /* once for us */ |
2809 | free_extent_map(em); | |
47ab2a6c JB |
2810 | return ret; |
2811 | } | |
2b82032c | 2812 | |
dc2ee4e2 | 2813 | static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset) |
47ab2a6c JB |
2814 | { |
2815 | struct btrfs_root *extent_root; | |
2816 | struct btrfs_trans_handle *trans; | |
2817 | int ret; | |
2b82032c | 2818 | |
47ab2a6c JB |
2819 | root = root->fs_info->chunk_root; |
2820 | extent_root = root->fs_info->extent_root; | |
2821 | ||
67c5e7d4 FM |
2822 | /* |
2823 | * Prevent races with automatic removal of unused block groups. | |
2824 | * After we relocate and before we remove the chunk with offset | |
2825 | * chunk_offset, automatic removal of the block group can kick in, | |
2826 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2827 | * | |
2828 | * Make sure to acquire this mutex before doing a tree search (dev | |
2829 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2830 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2831 | * we release the path used to search the chunk/dev tree and before | |
2832 | * the current task acquires this mutex and calls us. | |
2833 | */ | |
2834 | ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex)); | |
2835 | ||
47ab2a6c JB |
2836 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2837 | if (ret) | |
2838 | return -ENOSPC; | |
2839 | ||
2840 | /* step one, relocate all the extents inside this chunk */ | |
55e3a601 | 2841 | btrfs_scrub_pause(root); |
47ab2a6c | 2842 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
55e3a601 | 2843 | btrfs_scrub_continue(root); |
47ab2a6c JB |
2844 | if (ret) |
2845 | return ret; | |
2846 | ||
7fd01182 FM |
2847 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2848 | chunk_offset); | |
47ab2a6c JB |
2849 | if (IS_ERR(trans)) { |
2850 | ret = PTR_ERR(trans); | |
a4553fef | 2851 | btrfs_std_error(root->fs_info, ret, NULL); |
47ab2a6c JB |
2852 | return ret; |
2853 | } | |
2854 | ||
2855 | /* | |
2856 | * step two, delete the device extents and the | |
2857 | * chunk tree entries | |
2858 | */ | |
2859 | ret = btrfs_remove_chunk(trans, root, chunk_offset); | |
2b82032c | 2860 | btrfs_end_transaction(trans, root); |
47ab2a6c | 2861 | return ret; |
2b82032c YZ |
2862 | } |
2863 | ||
2864 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2865 | { | |
2866 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2867 | struct btrfs_path *path; | |
2868 | struct extent_buffer *leaf; | |
2869 | struct btrfs_chunk *chunk; | |
2870 | struct btrfs_key key; | |
2871 | struct btrfs_key found_key; | |
2b82032c | 2872 | u64 chunk_type; |
ba1bf481 JB |
2873 | bool retried = false; |
2874 | int failed = 0; | |
2b82032c YZ |
2875 | int ret; |
2876 | ||
2877 | path = btrfs_alloc_path(); | |
2878 | if (!path) | |
2879 | return -ENOMEM; | |
2880 | ||
ba1bf481 | 2881 | again: |
2b82032c YZ |
2882 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2883 | key.offset = (u64)-1; | |
2884 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2885 | ||
2886 | while (1) { | |
67c5e7d4 | 2887 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
2b82032c | 2888 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
2889 | if (ret < 0) { |
2890 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c | 2891 | goto error; |
67c5e7d4 | 2892 | } |
79787eaa | 2893 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2894 | |
2895 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2896 | key.type); | |
67c5e7d4 FM |
2897 | if (ret) |
2898 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c YZ |
2899 | if (ret < 0) |
2900 | goto error; | |
2901 | if (ret > 0) | |
2902 | break; | |
1a40e23b | 2903 | |
2b82032c YZ |
2904 | leaf = path->nodes[0]; |
2905 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2906 | |
2b82032c YZ |
2907 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2908 | struct btrfs_chunk); | |
2909 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2910 | btrfs_release_path(path); |
8f18cf13 | 2911 | |
2b82032c | 2912 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
a688a04a | 2913 | ret = btrfs_relocate_chunk(chunk_root, |
2b82032c | 2914 | found_key.offset); |
ba1bf481 JB |
2915 | if (ret == -ENOSPC) |
2916 | failed++; | |
14586651 HS |
2917 | else |
2918 | BUG_ON(ret); | |
2b82032c | 2919 | } |
67c5e7d4 | 2920 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 2921 | |
2b82032c YZ |
2922 | if (found_key.offset == 0) |
2923 | break; | |
2924 | key.offset = found_key.offset - 1; | |
2925 | } | |
2926 | ret = 0; | |
ba1bf481 JB |
2927 | if (failed && !retried) { |
2928 | failed = 0; | |
2929 | retried = true; | |
2930 | goto again; | |
fae7f21c | 2931 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2932 | ret = -ENOSPC; |
2933 | } | |
2b82032c YZ |
2934 | error: |
2935 | btrfs_free_path(path); | |
2936 | return ret; | |
8f18cf13 CM |
2937 | } |
2938 | ||
0940ebf6 ID |
2939 | static int insert_balance_item(struct btrfs_root *root, |
2940 | struct btrfs_balance_control *bctl) | |
2941 | { | |
2942 | struct btrfs_trans_handle *trans; | |
2943 | struct btrfs_balance_item *item; | |
2944 | struct btrfs_disk_balance_args disk_bargs; | |
2945 | struct btrfs_path *path; | |
2946 | struct extent_buffer *leaf; | |
2947 | struct btrfs_key key; | |
2948 | int ret, err; | |
2949 | ||
2950 | path = btrfs_alloc_path(); | |
2951 | if (!path) | |
2952 | return -ENOMEM; | |
2953 | ||
2954 | trans = btrfs_start_transaction(root, 0); | |
2955 | if (IS_ERR(trans)) { | |
2956 | btrfs_free_path(path); | |
2957 | return PTR_ERR(trans); | |
2958 | } | |
2959 | ||
2960 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 2961 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
2962 | key.offset = 0; |
2963 | ||
2964 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2965 | sizeof(*item)); | |
2966 | if (ret) | |
2967 | goto out; | |
2968 | ||
2969 | leaf = path->nodes[0]; | |
2970 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2971 | ||
2972 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2973 | ||
2974 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2975 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2976 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2977 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2978 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2979 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2980 | ||
2981 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2982 | ||
2983 | btrfs_mark_buffer_dirty(leaf); | |
2984 | out: | |
2985 | btrfs_free_path(path); | |
2986 | err = btrfs_commit_transaction(trans, root); | |
2987 | if (err && !ret) | |
2988 | ret = err; | |
2989 | return ret; | |
2990 | } | |
2991 | ||
2992 | static int del_balance_item(struct btrfs_root *root) | |
2993 | { | |
2994 | struct btrfs_trans_handle *trans; | |
2995 | struct btrfs_path *path; | |
2996 | struct btrfs_key key; | |
2997 | int ret, err; | |
2998 | ||
2999 | path = btrfs_alloc_path(); | |
3000 | if (!path) | |
3001 | return -ENOMEM; | |
3002 | ||
3003 | trans = btrfs_start_transaction(root, 0); | |
3004 | if (IS_ERR(trans)) { | |
3005 | btrfs_free_path(path); | |
3006 | return PTR_ERR(trans); | |
3007 | } | |
3008 | ||
3009 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3010 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3011 | key.offset = 0; |
3012 | ||
3013 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3014 | if (ret < 0) | |
3015 | goto out; | |
3016 | if (ret > 0) { | |
3017 | ret = -ENOENT; | |
3018 | goto out; | |
3019 | } | |
3020 | ||
3021 | ret = btrfs_del_item(trans, root, path); | |
3022 | out: | |
3023 | btrfs_free_path(path); | |
3024 | err = btrfs_commit_transaction(trans, root); | |
3025 | if (err && !ret) | |
3026 | ret = err; | |
3027 | return ret; | |
3028 | } | |
3029 | ||
59641015 ID |
3030 | /* |
3031 | * This is a heuristic used to reduce the number of chunks balanced on | |
3032 | * resume after balance was interrupted. | |
3033 | */ | |
3034 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3035 | { | |
3036 | /* | |
3037 | * Turn on soft mode for chunk types that were being converted. | |
3038 | */ | |
3039 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3040 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3041 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3042 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3043 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3044 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3045 | ||
3046 | /* | |
3047 | * Turn on usage filter if is not already used. The idea is | |
3048 | * that chunks that we have already balanced should be | |
3049 | * reasonably full. Don't do it for chunks that are being | |
3050 | * converted - that will keep us from relocating unconverted | |
3051 | * (albeit full) chunks. | |
3052 | */ | |
3053 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3054 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3055 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3056 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3057 | bctl->data.usage = 90; | |
3058 | } | |
3059 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3060 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3061 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3062 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3063 | bctl->sys.usage = 90; | |
3064 | } | |
3065 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3066 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3067 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3068 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3069 | bctl->meta.usage = 90; | |
3070 | } | |
3071 | } | |
3072 | ||
c9e9f97b ID |
3073 | /* |
3074 | * Should be called with both balance and volume mutexes held to | |
3075 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3076 | * restriper. Same goes for unset_balance_control. | |
3077 | */ | |
3078 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3079 | { | |
3080 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3081 | ||
3082 | BUG_ON(fs_info->balance_ctl); | |
3083 | ||
3084 | spin_lock(&fs_info->balance_lock); | |
3085 | fs_info->balance_ctl = bctl; | |
3086 | spin_unlock(&fs_info->balance_lock); | |
3087 | } | |
3088 | ||
3089 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3090 | { | |
3091 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3092 | ||
3093 | BUG_ON(!fs_info->balance_ctl); | |
3094 | ||
3095 | spin_lock(&fs_info->balance_lock); | |
3096 | fs_info->balance_ctl = NULL; | |
3097 | spin_unlock(&fs_info->balance_lock); | |
3098 | ||
3099 | kfree(bctl); | |
3100 | } | |
3101 | ||
ed25e9b2 ID |
3102 | /* |
3103 | * Balance filters. Return 1 if chunk should be filtered out | |
3104 | * (should not be balanced). | |
3105 | */ | |
899c81ea | 3106 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3107 | struct btrfs_balance_args *bargs) |
3108 | { | |
899c81ea ID |
3109 | chunk_type = chunk_to_extended(chunk_type) & |
3110 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3111 | |
899c81ea | 3112 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3113 | return 0; |
3114 | ||
3115 | return 1; | |
3116 | } | |
3117 | ||
dba72cb3 | 3118 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3119 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3120 | { |
3121 | struct btrfs_block_group_cache *cache; | |
3122 | u64 chunk_used; | |
3123 | u64 user_thresh_min; | |
3124 | u64 user_thresh_max; | |
3125 | int ret = 1; | |
3126 | ||
3127 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3128 | chunk_used = btrfs_block_group_used(&cache->item); | |
3129 | ||
3130 | if (bargs->usage_min == 0) | |
3131 | user_thresh_min = 0; | |
3132 | else | |
3133 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3134 | bargs->usage_min); | |
3135 | ||
3136 | if (bargs->usage_max == 0) | |
3137 | user_thresh_max = 1; | |
3138 | else if (bargs->usage_max > 100) | |
3139 | user_thresh_max = cache->key.offset; | |
3140 | else | |
3141 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3142 | bargs->usage_max); | |
3143 | ||
3144 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3145 | ret = 0; | |
3146 | ||
3147 | btrfs_put_block_group(cache); | |
3148 | return ret; | |
3149 | } | |
3150 | ||
dba72cb3 | 3151 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3152 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3153 | { |
3154 | struct btrfs_block_group_cache *cache; | |
3155 | u64 chunk_used, user_thresh; | |
3156 | int ret = 1; | |
3157 | ||
3158 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3159 | chunk_used = btrfs_block_group_used(&cache->item); | |
3160 | ||
bc309467 | 3161 | if (bargs->usage_min == 0) |
3e39cea6 | 3162 | user_thresh = 1; |
a105bb88 ID |
3163 | else if (bargs->usage > 100) |
3164 | user_thresh = cache->key.offset; | |
3165 | else | |
3166 | user_thresh = div_factor_fine(cache->key.offset, | |
3167 | bargs->usage); | |
3168 | ||
5ce5b3c0 ID |
3169 | if (chunk_used < user_thresh) |
3170 | ret = 0; | |
3171 | ||
3172 | btrfs_put_block_group(cache); | |
3173 | return ret; | |
3174 | } | |
3175 | ||
409d404b ID |
3176 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3177 | struct btrfs_chunk *chunk, | |
3178 | struct btrfs_balance_args *bargs) | |
3179 | { | |
3180 | struct btrfs_stripe *stripe; | |
3181 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3182 | int i; | |
3183 | ||
3184 | for (i = 0; i < num_stripes; i++) { | |
3185 | stripe = btrfs_stripe_nr(chunk, i); | |
3186 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3187 | return 0; | |
3188 | } | |
3189 | ||
3190 | return 1; | |
3191 | } | |
3192 | ||
94e60d5a ID |
3193 | /* [pstart, pend) */ |
3194 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3195 | struct btrfs_chunk *chunk, | |
3196 | u64 chunk_offset, | |
3197 | struct btrfs_balance_args *bargs) | |
3198 | { | |
3199 | struct btrfs_stripe *stripe; | |
3200 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3201 | u64 stripe_offset; | |
3202 | u64 stripe_length; | |
3203 | int factor; | |
3204 | int i; | |
3205 | ||
3206 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3207 | return 0; | |
3208 | ||
3209 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3210 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3211 | factor = num_stripes / 2; | |
3212 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3213 | factor = num_stripes - 1; | |
3214 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3215 | factor = num_stripes - 2; | |
3216 | } else { | |
3217 | factor = num_stripes; | |
3218 | } | |
94e60d5a ID |
3219 | |
3220 | for (i = 0; i < num_stripes; i++) { | |
3221 | stripe = btrfs_stripe_nr(chunk, i); | |
3222 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3223 | continue; | |
3224 | ||
3225 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3226 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3227 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3228 | |
3229 | if (stripe_offset < bargs->pend && | |
3230 | stripe_offset + stripe_length > bargs->pstart) | |
3231 | return 0; | |
3232 | } | |
3233 | ||
3234 | return 1; | |
3235 | } | |
3236 | ||
ea67176a ID |
3237 | /* [vstart, vend) */ |
3238 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3239 | struct btrfs_chunk *chunk, | |
3240 | u64 chunk_offset, | |
3241 | struct btrfs_balance_args *bargs) | |
3242 | { | |
3243 | if (chunk_offset < bargs->vend && | |
3244 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3245 | /* at least part of the chunk is inside this vrange */ | |
3246 | return 0; | |
3247 | ||
3248 | return 1; | |
3249 | } | |
3250 | ||
dee32d0a GAP |
3251 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3252 | struct btrfs_chunk *chunk, | |
3253 | struct btrfs_balance_args *bargs) | |
3254 | { | |
3255 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3256 | ||
3257 | if (bargs->stripes_min <= num_stripes | |
3258 | && num_stripes <= bargs->stripes_max) | |
3259 | return 0; | |
3260 | ||
3261 | return 1; | |
3262 | } | |
3263 | ||
899c81ea | 3264 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3265 | struct btrfs_balance_args *bargs) |
3266 | { | |
3267 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3268 | return 0; | |
3269 | ||
899c81ea ID |
3270 | chunk_type = chunk_to_extended(chunk_type) & |
3271 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3272 | |
899c81ea | 3273 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3274 | return 1; |
3275 | ||
3276 | return 0; | |
3277 | } | |
3278 | ||
f43ffb60 ID |
3279 | static int should_balance_chunk(struct btrfs_root *root, |
3280 | struct extent_buffer *leaf, | |
3281 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3282 | { | |
3283 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3284 | struct btrfs_balance_args *bargs = NULL; | |
3285 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3286 | ||
3287 | /* type filter */ | |
3288 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3289 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3290 | return 0; | |
3291 | } | |
3292 | ||
3293 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3294 | bargs = &bctl->data; | |
3295 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3296 | bargs = &bctl->sys; | |
3297 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3298 | bargs = &bctl->meta; | |
3299 | ||
ed25e9b2 ID |
3300 | /* profiles filter */ |
3301 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3302 | chunk_profiles_filter(chunk_type, bargs)) { | |
3303 | return 0; | |
5ce5b3c0 ID |
3304 | } |
3305 | ||
3306 | /* usage filter */ | |
3307 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3308 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3309 | return 0; | |
bc309467 DS |
3310 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
3311 | chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3312 | return 0; | |
409d404b ID |
3313 | } |
3314 | ||
3315 | /* devid filter */ | |
3316 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3317 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3318 | return 0; | |
94e60d5a ID |
3319 | } |
3320 | ||
3321 | /* drange filter, makes sense only with devid filter */ | |
3322 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3323 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3324 | return 0; | |
ea67176a ID |
3325 | } |
3326 | ||
3327 | /* vrange filter */ | |
3328 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3329 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3330 | return 0; | |
ed25e9b2 ID |
3331 | } |
3332 | ||
dee32d0a GAP |
3333 | /* stripes filter */ |
3334 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3335 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3336 | return 0; | |
3337 | } | |
3338 | ||
cfa4c961 ID |
3339 | /* soft profile changing mode */ |
3340 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3341 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3342 | return 0; | |
3343 | } | |
3344 | ||
7d824b6f DS |
3345 | /* |
3346 | * limited by count, must be the last filter | |
3347 | */ | |
3348 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3349 | if (bargs->limit == 0) | |
3350 | return 0; | |
3351 | else | |
3352 | bargs->limit--; | |
12907fc7 DS |
3353 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3354 | /* | |
3355 | * Same logic as the 'limit' filter; the minimum cannot be | |
3356 | * determined here because we do not have the global informatoin | |
3357 | * about the count of all chunks that satisfy the filters. | |
3358 | */ | |
3359 | if (bargs->limit_max == 0) | |
3360 | return 0; | |
3361 | else | |
3362 | bargs->limit_max--; | |
7d824b6f DS |
3363 | } |
3364 | ||
f43ffb60 ID |
3365 | return 1; |
3366 | } | |
3367 | ||
c9e9f97b | 3368 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3369 | { |
19a39dce | 3370 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3371 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3372 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3373 | struct list_head *devices; | |
ec44a35c CM |
3374 | struct btrfs_device *device; |
3375 | u64 old_size; | |
3376 | u64 size_to_free; | |
12907fc7 | 3377 | u64 chunk_type; |
f43ffb60 | 3378 | struct btrfs_chunk *chunk; |
ec44a35c CM |
3379 | struct btrfs_path *path; |
3380 | struct btrfs_key key; | |
ec44a35c | 3381 | struct btrfs_key found_key; |
c9e9f97b | 3382 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3383 | struct extent_buffer *leaf; |
3384 | int slot; | |
c9e9f97b ID |
3385 | int ret; |
3386 | int enospc_errors = 0; | |
19a39dce | 3387 | bool counting = true; |
12907fc7 | 3388 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3389 | u64 limit_data = bctl->data.limit; |
3390 | u64 limit_meta = bctl->meta.limit; | |
3391 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3392 | u32 count_data = 0; |
3393 | u32 count_meta = 0; | |
3394 | u32 count_sys = 0; | |
2c9fe835 | 3395 | int chunk_reserved = 0; |
ec44a35c | 3396 | |
ec44a35c | 3397 | /* step one make some room on all the devices */ |
c9e9f97b | 3398 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3399 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3400 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3401 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3402 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
2b82032c | 3403 | if (!device->writeable || |
7cc8e58d MX |
3404 | btrfs_device_get_total_bytes(device) - |
3405 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3406 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3407 | continue; |
3408 | ||
3409 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3410 | if (ret == -ENOSPC) |
3411 | break; | |
ec44a35c CM |
3412 | BUG_ON(ret); |
3413 | ||
a22285a6 | 3414 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 3415 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
3416 | |
3417 | ret = btrfs_grow_device(trans, device, old_size); | |
3418 | BUG_ON(ret); | |
3419 | ||
3420 | btrfs_end_transaction(trans, dev_root); | |
3421 | } | |
3422 | ||
3423 | /* step two, relocate all the chunks */ | |
3424 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3425 | if (!path) { |
3426 | ret = -ENOMEM; | |
3427 | goto error; | |
3428 | } | |
19a39dce ID |
3429 | |
3430 | /* zero out stat counters */ | |
3431 | spin_lock(&fs_info->balance_lock); | |
3432 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3433 | spin_unlock(&fs_info->balance_lock); | |
3434 | again: | |
7d824b6f | 3435 | if (!counting) { |
12907fc7 DS |
3436 | /* |
3437 | * The single value limit and min/max limits use the same bytes | |
3438 | * in the | |
3439 | */ | |
7d824b6f DS |
3440 | bctl->data.limit = limit_data; |
3441 | bctl->meta.limit = limit_meta; | |
3442 | bctl->sys.limit = limit_sys; | |
3443 | } | |
ec44a35c CM |
3444 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3445 | key.offset = (u64)-1; | |
3446 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3447 | ||
d397712b | 3448 | while (1) { |
19a39dce | 3449 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3450 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3451 | ret = -ECANCELED; |
3452 | goto error; | |
3453 | } | |
3454 | ||
67c5e7d4 | 3455 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3456 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3457 | if (ret < 0) { |
3458 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3459 | goto error; |
67c5e7d4 | 3460 | } |
ec44a35c CM |
3461 | |
3462 | /* | |
3463 | * this shouldn't happen, it means the last relocate | |
3464 | * failed | |
3465 | */ | |
3466 | if (ret == 0) | |
c9e9f97b | 3467 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3468 | |
3469 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3470 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3471 | if (ret) { |
67c5e7d4 | 3472 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3473 | ret = 0; |
ec44a35c | 3474 | break; |
c9e9f97b | 3475 | } |
7d9eb12c | 3476 | |
f43ffb60 ID |
3477 | leaf = path->nodes[0]; |
3478 | slot = path->slots[0]; | |
3479 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3480 | |
67c5e7d4 FM |
3481 | if (found_key.objectid != key.objectid) { |
3482 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3483 | break; |
67c5e7d4 | 3484 | } |
7d9eb12c | 3485 | |
f43ffb60 | 3486 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3487 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3488 | |
19a39dce ID |
3489 | if (!counting) { |
3490 | spin_lock(&fs_info->balance_lock); | |
3491 | bctl->stat.considered++; | |
3492 | spin_unlock(&fs_info->balance_lock); | |
3493 | } | |
3494 | ||
f43ffb60 ID |
3495 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3496 | found_key.offset); | |
2c9fe835 | 3497 | |
b3b4aa74 | 3498 | btrfs_release_path(path); |
67c5e7d4 FM |
3499 | if (!ret) { |
3500 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3501 | goto loop; |
67c5e7d4 | 3502 | } |
f43ffb60 | 3503 | |
19a39dce | 3504 | if (counting) { |
67c5e7d4 | 3505 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3506 | spin_lock(&fs_info->balance_lock); |
3507 | bctl->stat.expected++; | |
3508 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3509 | |
3510 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3511 | count_data++; | |
3512 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3513 | count_sys++; | |
3514 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3515 | count_meta++; | |
3516 | ||
3517 | goto loop; | |
3518 | } | |
3519 | ||
3520 | /* | |
3521 | * Apply limit_min filter, no need to check if the LIMITS | |
3522 | * filter is used, limit_min is 0 by default | |
3523 | */ | |
3524 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3525 | count_data < bctl->data.limit_min) | |
3526 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3527 | count_meta < bctl->meta.limit_min) | |
3528 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3529 | count_sys < bctl->sys.limit_min)) { | |
3530 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3531 | goto loop; |
3532 | } | |
3533 | ||
2c9fe835 ZL |
3534 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) { |
3535 | trans = btrfs_start_transaction(chunk_root, 0); | |
3536 | if (IS_ERR(trans)) { | |
3537 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3538 | ret = PTR_ERR(trans); | |
3539 | goto error; | |
3540 | } | |
3541 | ||
3542 | ret = btrfs_force_chunk_alloc(trans, chunk_root, | |
3543 | BTRFS_BLOCK_GROUP_DATA); | |
8a7d656f | 3544 | btrfs_end_transaction(trans, chunk_root); |
2c9fe835 ZL |
3545 | if (ret < 0) { |
3546 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3547 | goto error; | |
3548 | } | |
2c9fe835 ZL |
3549 | chunk_reserved = 1; |
3550 | } | |
3551 | ||
ec44a35c | 3552 | ret = btrfs_relocate_chunk(chunk_root, |
ec44a35c | 3553 | found_key.offset); |
67c5e7d4 | 3554 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3555 | if (ret && ret != -ENOSPC) |
3556 | goto error; | |
19a39dce | 3557 | if (ret == -ENOSPC) { |
c9e9f97b | 3558 | enospc_errors++; |
19a39dce ID |
3559 | } else { |
3560 | spin_lock(&fs_info->balance_lock); | |
3561 | bctl->stat.completed++; | |
3562 | spin_unlock(&fs_info->balance_lock); | |
3563 | } | |
f43ffb60 | 3564 | loop: |
795a3321 ID |
3565 | if (found_key.offset == 0) |
3566 | break; | |
ba1bf481 | 3567 | key.offset = found_key.offset - 1; |
ec44a35c | 3568 | } |
c9e9f97b | 3569 | |
19a39dce ID |
3570 | if (counting) { |
3571 | btrfs_release_path(path); | |
3572 | counting = false; | |
3573 | goto again; | |
3574 | } | |
ec44a35c CM |
3575 | error: |
3576 | btrfs_free_path(path); | |
c9e9f97b | 3577 | if (enospc_errors) { |
efe120a0 | 3578 | btrfs_info(fs_info, "%d enospc errors during balance", |
c9e9f97b ID |
3579 | enospc_errors); |
3580 | if (!ret) | |
3581 | ret = -ENOSPC; | |
3582 | } | |
3583 | ||
ec44a35c CM |
3584 | return ret; |
3585 | } | |
3586 | ||
0c460c0d ID |
3587 | /** |
3588 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3589 | * @flags: profile to validate | |
3590 | * @extended: if true @flags is treated as an extended profile | |
3591 | */ | |
3592 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3593 | { | |
3594 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3595 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3596 | ||
3597 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3598 | ||
3599 | /* 1) check that all other bits are zeroed */ | |
3600 | if (flags & ~mask) | |
3601 | return 0; | |
3602 | ||
3603 | /* 2) see if profile is reduced */ | |
3604 | if (flags == 0) | |
3605 | return !extended; /* "0" is valid for usual profiles */ | |
3606 | ||
3607 | /* true if exactly one bit set */ | |
3608 | return (flags & (flags - 1)) == 0; | |
3609 | } | |
3610 | ||
837d5b6e ID |
3611 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3612 | { | |
a7e99c69 ID |
3613 | /* cancel requested || normal exit path */ |
3614 | return atomic_read(&fs_info->balance_cancel_req) || | |
3615 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3616 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3617 | } |
3618 | ||
c9e9f97b ID |
3619 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3620 | { | |
0940ebf6 ID |
3621 | int ret; |
3622 | ||
c9e9f97b | 3623 | unset_balance_control(fs_info); |
0940ebf6 | 3624 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 | 3625 | if (ret) |
a4553fef | 3626 | btrfs_std_error(fs_info, ret, NULL); |
ed0fb78f ID |
3627 | |
3628 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3629 | } |
3630 | ||
bdcd3c97 AM |
3631 | /* Non-zero return value signifies invalidity */ |
3632 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3633 | u64 allowed) | |
3634 | { | |
3635 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3636 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3637 | (bctl_arg->target & ~allowed))); | |
3638 | } | |
3639 | ||
c9e9f97b ID |
3640 | /* |
3641 | * Should be called with both balance and volume mutexes held | |
3642 | */ | |
3643 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3644 | struct btrfs_ioctl_balance_args *bargs) | |
3645 | { | |
3646 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3647 | u64 allowed; |
e4837f8f | 3648 | int mixed = 0; |
c9e9f97b | 3649 | int ret; |
8dabb742 | 3650 | u64 num_devices; |
de98ced9 | 3651 | unsigned seq; |
c9e9f97b | 3652 | |
837d5b6e | 3653 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3654 | atomic_read(&fs_info->balance_pause_req) || |
3655 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3656 | ret = -EINVAL; |
3657 | goto out; | |
3658 | } | |
3659 | ||
e4837f8f ID |
3660 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3661 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3662 | mixed = 1; | |
3663 | ||
f43ffb60 ID |
3664 | /* |
3665 | * In case of mixed groups both data and meta should be picked, | |
3666 | * and identical options should be given for both of them. | |
3667 | */ | |
e4837f8f ID |
3668 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3669 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3670 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3671 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3672 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
efe120a0 FH |
3673 | btrfs_err(fs_info, "with mixed groups data and " |
3674 | "metadata balance options must be the same"); | |
f43ffb60 ID |
3675 | ret = -EINVAL; |
3676 | goto out; | |
3677 | } | |
3678 | } | |
3679 | ||
8dabb742 | 3680 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3681 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3682 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3683 | BUG_ON(num_devices < 1); | |
3684 | num_devices--; | |
3685 | } | |
73beece9 | 3686 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
e4d8ec0f | 3687 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3688 | if (num_devices == 1) |
e4d8ec0f | 3689 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8250dabe | 3690 | else if (num_devices > 1) |
e4d8ec0f | 3691 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3692 | if (num_devices > 2) |
3693 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3694 | if (num_devices > 3) | |
3695 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3696 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3697 | if (validate_convert_profile(&bctl->data, allowed)) { |
efe120a0 FH |
3698 | btrfs_err(fs_info, "unable to start balance with target " |
3699 | "data profile %llu", | |
c1c9ff7c | 3700 | bctl->data.target); |
e4d8ec0f ID |
3701 | ret = -EINVAL; |
3702 | goto out; | |
3703 | } | |
bdcd3c97 | 3704 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 FH |
3705 | btrfs_err(fs_info, |
3706 | "unable to start balance with target metadata profile %llu", | |
c1c9ff7c | 3707 | bctl->meta.target); |
e4d8ec0f ID |
3708 | ret = -EINVAL; |
3709 | goto out; | |
3710 | } | |
bdcd3c97 | 3711 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 FH |
3712 | btrfs_err(fs_info, |
3713 | "unable to start balance with target system profile %llu", | |
c1c9ff7c | 3714 | bctl->sys.target); |
e4d8ec0f ID |
3715 | ret = -EINVAL; |
3716 | goto out; | |
3717 | } | |
3718 | ||
e4d8ec0f ID |
3719 | /* allow to reduce meta or sys integrity only if force set */ |
3720 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3721 | BTRFS_BLOCK_GROUP_RAID10 | |
3722 | BTRFS_BLOCK_GROUP_RAID5 | | |
3723 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3724 | do { |
3725 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3726 | ||
3727 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3728 | (fs_info->avail_system_alloc_bits & allowed) && | |
3729 | !(bctl->sys.target & allowed)) || | |
3730 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3731 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3732 | !(bctl->meta.target & allowed))) { | |
3733 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
efe120a0 | 3734 | btrfs_info(fs_info, "force reducing metadata integrity"); |
de98ced9 | 3735 | } else { |
efe120a0 FH |
3736 | btrfs_err(fs_info, "balance will reduce metadata " |
3737 | "integrity, use force if you want this"); | |
de98ced9 MX |
3738 | ret = -EINVAL; |
3739 | goto out; | |
3740 | } | |
e4d8ec0f | 3741 | } |
de98ced9 | 3742 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3743 | |
ee592d07 ST |
3744 | if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) < |
3745 | btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) { | |
3746 | btrfs_warn(fs_info, | |
fedc0045 | 3747 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
ee592d07 ST |
3748 | bctl->meta.target, bctl->data.target); |
3749 | } | |
3750 | ||
5af3e8cc | 3751 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
943c6e99 ZL |
3752 | fs_info->num_tolerated_disk_barrier_failures = min( |
3753 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3754 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3755 | bctl->sys.target)); | |
5af3e8cc SB |
3756 | } |
3757 | ||
0940ebf6 | 3758 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3759 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3760 | goto out; |
3761 | ||
59641015 ID |
3762 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3763 | BUG_ON(ret == -EEXIST); | |
3764 | set_balance_control(bctl); | |
3765 | } else { | |
3766 | BUG_ON(ret != -EEXIST); | |
3767 | spin_lock(&fs_info->balance_lock); | |
3768 | update_balance_args(bctl); | |
3769 | spin_unlock(&fs_info->balance_lock); | |
3770 | } | |
c9e9f97b | 3771 | |
837d5b6e | 3772 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3773 | mutex_unlock(&fs_info->balance_mutex); |
3774 | ||
3775 | ret = __btrfs_balance(fs_info); | |
3776 | ||
3777 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3778 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3779 | |
bf023ecf ID |
3780 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3781 | fs_info->num_tolerated_disk_barrier_failures = | |
3782 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3783 | } | |
3784 | ||
c9e9f97b ID |
3785 | if (bargs) { |
3786 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3787 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3788 | } |
3789 | ||
3a01aa7a ID |
3790 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3791 | balance_need_close(fs_info)) { | |
3792 | __cancel_balance(fs_info); | |
3793 | } | |
3794 | ||
837d5b6e | 3795 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3796 | |
3797 | return ret; | |
3798 | out: | |
59641015 ID |
3799 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3800 | __cancel_balance(fs_info); | |
ed0fb78f | 3801 | else { |
59641015 | 3802 | kfree(bctl); |
ed0fb78f ID |
3803 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3804 | } | |
59641015 ID |
3805 | return ret; |
3806 | } | |
3807 | ||
3808 | static int balance_kthread(void *data) | |
3809 | { | |
2b6ba629 | 3810 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3811 | int ret = 0; |
59641015 ID |
3812 | |
3813 | mutex_lock(&fs_info->volume_mutex); | |
3814 | mutex_lock(&fs_info->balance_mutex); | |
3815 | ||
2b6ba629 | 3816 | if (fs_info->balance_ctl) { |
efe120a0 | 3817 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3818 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3819 | } |
59641015 ID |
3820 | |
3821 | mutex_unlock(&fs_info->balance_mutex); | |
3822 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3823 | |
59641015 ID |
3824 | return ret; |
3825 | } | |
3826 | ||
2b6ba629 ID |
3827 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3828 | { | |
3829 | struct task_struct *tsk; | |
3830 | ||
3831 | spin_lock(&fs_info->balance_lock); | |
3832 | if (!fs_info->balance_ctl) { | |
3833 | spin_unlock(&fs_info->balance_lock); | |
3834 | return 0; | |
3835 | } | |
3836 | spin_unlock(&fs_info->balance_lock); | |
3837 | ||
3838 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
efe120a0 | 3839 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3840 | return 0; |
3841 | } | |
3842 | ||
3843 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3844 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3845 | } |
3846 | ||
68310a5e | 3847 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3848 | { |
59641015 ID |
3849 | struct btrfs_balance_control *bctl; |
3850 | struct btrfs_balance_item *item; | |
3851 | struct btrfs_disk_balance_args disk_bargs; | |
3852 | struct btrfs_path *path; | |
3853 | struct extent_buffer *leaf; | |
3854 | struct btrfs_key key; | |
3855 | int ret; | |
3856 | ||
3857 | path = btrfs_alloc_path(); | |
3858 | if (!path) | |
3859 | return -ENOMEM; | |
3860 | ||
59641015 | 3861 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3862 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
3863 | key.offset = 0; |
3864 | ||
68310a5e | 3865 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3866 | if (ret < 0) |
68310a5e | 3867 | goto out; |
59641015 ID |
3868 | if (ret > 0) { /* ret = -ENOENT; */ |
3869 | ret = 0; | |
68310a5e ID |
3870 | goto out; |
3871 | } | |
3872 | ||
3873 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3874 | if (!bctl) { | |
3875 | ret = -ENOMEM; | |
3876 | goto out; | |
59641015 ID |
3877 | } |
3878 | ||
3879 | leaf = path->nodes[0]; | |
3880 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3881 | ||
68310a5e ID |
3882 | bctl->fs_info = fs_info; |
3883 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3884 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3885 | |
3886 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3887 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3888 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3889 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3890 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3891 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3892 | ||
ed0fb78f ID |
3893 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3894 | ||
68310a5e ID |
3895 | mutex_lock(&fs_info->volume_mutex); |
3896 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3897 | |
68310a5e ID |
3898 | set_balance_control(bctl); |
3899 | ||
3900 | mutex_unlock(&fs_info->balance_mutex); | |
3901 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3902 | out: |
3903 | btrfs_free_path(path); | |
ec44a35c CM |
3904 | return ret; |
3905 | } | |
3906 | ||
837d5b6e ID |
3907 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3908 | { | |
3909 | int ret = 0; | |
3910 | ||
3911 | mutex_lock(&fs_info->balance_mutex); | |
3912 | if (!fs_info->balance_ctl) { | |
3913 | mutex_unlock(&fs_info->balance_mutex); | |
3914 | return -ENOTCONN; | |
3915 | } | |
3916 | ||
3917 | if (atomic_read(&fs_info->balance_running)) { | |
3918 | atomic_inc(&fs_info->balance_pause_req); | |
3919 | mutex_unlock(&fs_info->balance_mutex); | |
3920 | ||
3921 | wait_event(fs_info->balance_wait_q, | |
3922 | atomic_read(&fs_info->balance_running) == 0); | |
3923 | ||
3924 | mutex_lock(&fs_info->balance_mutex); | |
3925 | /* we are good with balance_ctl ripped off from under us */ | |
3926 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3927 | atomic_dec(&fs_info->balance_pause_req); | |
3928 | } else { | |
3929 | ret = -ENOTCONN; | |
3930 | } | |
3931 | ||
3932 | mutex_unlock(&fs_info->balance_mutex); | |
3933 | return ret; | |
3934 | } | |
3935 | ||
a7e99c69 ID |
3936 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3937 | { | |
e649e587 ID |
3938 | if (fs_info->sb->s_flags & MS_RDONLY) |
3939 | return -EROFS; | |
3940 | ||
a7e99c69 ID |
3941 | mutex_lock(&fs_info->balance_mutex); |
3942 | if (!fs_info->balance_ctl) { | |
3943 | mutex_unlock(&fs_info->balance_mutex); | |
3944 | return -ENOTCONN; | |
3945 | } | |
3946 | ||
3947 | atomic_inc(&fs_info->balance_cancel_req); | |
3948 | /* | |
3949 | * if we are running just wait and return, balance item is | |
3950 | * deleted in btrfs_balance in this case | |
3951 | */ | |
3952 | if (atomic_read(&fs_info->balance_running)) { | |
3953 | mutex_unlock(&fs_info->balance_mutex); | |
3954 | wait_event(fs_info->balance_wait_q, | |
3955 | atomic_read(&fs_info->balance_running) == 0); | |
3956 | mutex_lock(&fs_info->balance_mutex); | |
3957 | } else { | |
3958 | /* __cancel_balance needs volume_mutex */ | |
3959 | mutex_unlock(&fs_info->balance_mutex); | |
3960 | mutex_lock(&fs_info->volume_mutex); | |
3961 | mutex_lock(&fs_info->balance_mutex); | |
3962 | ||
3963 | if (fs_info->balance_ctl) | |
3964 | __cancel_balance(fs_info); | |
3965 | ||
3966 | mutex_unlock(&fs_info->volume_mutex); | |
3967 | } | |
3968 | ||
3969 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3970 | atomic_dec(&fs_info->balance_cancel_req); | |
3971 | mutex_unlock(&fs_info->balance_mutex); | |
3972 | return 0; | |
3973 | } | |
3974 | ||
803b2f54 SB |
3975 | static int btrfs_uuid_scan_kthread(void *data) |
3976 | { | |
3977 | struct btrfs_fs_info *fs_info = data; | |
3978 | struct btrfs_root *root = fs_info->tree_root; | |
3979 | struct btrfs_key key; | |
3980 | struct btrfs_key max_key; | |
3981 | struct btrfs_path *path = NULL; | |
3982 | int ret = 0; | |
3983 | struct extent_buffer *eb; | |
3984 | int slot; | |
3985 | struct btrfs_root_item root_item; | |
3986 | u32 item_size; | |
f45388f3 | 3987 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
3988 | |
3989 | path = btrfs_alloc_path(); | |
3990 | if (!path) { | |
3991 | ret = -ENOMEM; | |
3992 | goto out; | |
3993 | } | |
3994 | ||
3995 | key.objectid = 0; | |
3996 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3997 | key.offset = 0; | |
3998 | ||
3999 | max_key.objectid = (u64)-1; | |
4000 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
4001 | max_key.offset = (u64)-1; | |
4002 | ||
803b2f54 | 4003 | while (1) { |
6174d3cb | 4004 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4005 | if (ret) { |
4006 | if (ret > 0) | |
4007 | ret = 0; | |
4008 | break; | |
4009 | } | |
4010 | ||
4011 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4012 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4013 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4014 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4015 | goto skip; | |
4016 | ||
4017 | eb = path->nodes[0]; | |
4018 | slot = path->slots[0]; | |
4019 | item_size = btrfs_item_size_nr(eb, slot); | |
4020 | if (item_size < sizeof(root_item)) | |
4021 | goto skip; | |
4022 | ||
803b2f54 SB |
4023 | read_extent_buffer(eb, &root_item, |
4024 | btrfs_item_ptr_offset(eb, slot), | |
4025 | (int)sizeof(root_item)); | |
4026 | if (btrfs_root_refs(&root_item) == 0) | |
4027 | goto skip; | |
f45388f3 FDBM |
4028 | |
4029 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4030 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4031 | if (trans) | |
4032 | goto update_tree; | |
4033 | ||
4034 | btrfs_release_path(path); | |
803b2f54 SB |
4035 | /* |
4036 | * 1 - subvol uuid item | |
4037 | * 1 - received_subvol uuid item | |
4038 | */ | |
4039 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4040 | if (IS_ERR(trans)) { | |
4041 | ret = PTR_ERR(trans); | |
4042 | break; | |
4043 | } | |
f45388f3 FDBM |
4044 | continue; |
4045 | } else { | |
4046 | goto skip; | |
4047 | } | |
4048 | update_tree: | |
4049 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
803b2f54 SB |
4050 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
4051 | root_item.uuid, | |
4052 | BTRFS_UUID_KEY_SUBVOL, | |
4053 | key.objectid); | |
4054 | if (ret < 0) { | |
efe120a0 | 4055 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4056 | ret); |
803b2f54 SB |
4057 | break; |
4058 | } | |
4059 | } | |
4060 | ||
4061 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
803b2f54 SB |
4062 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
4063 | root_item.received_uuid, | |
4064 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4065 | key.objectid); | |
4066 | if (ret < 0) { | |
efe120a0 | 4067 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4068 | ret); |
803b2f54 SB |
4069 | break; |
4070 | } | |
4071 | } | |
4072 | ||
f45388f3 | 4073 | skip: |
803b2f54 SB |
4074 | if (trans) { |
4075 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 4076 | trans = NULL; |
803b2f54 SB |
4077 | if (ret) |
4078 | break; | |
4079 | } | |
4080 | ||
803b2f54 SB |
4081 | btrfs_release_path(path); |
4082 | if (key.offset < (u64)-1) { | |
4083 | key.offset++; | |
4084 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4085 | key.offset = 0; | |
4086 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4087 | } else if (key.objectid < (u64)-1) { | |
4088 | key.offset = 0; | |
4089 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4090 | key.objectid++; | |
4091 | } else { | |
4092 | break; | |
4093 | } | |
4094 | cond_resched(); | |
4095 | } | |
4096 | ||
4097 | out: | |
4098 | btrfs_free_path(path); | |
f45388f3 FDBM |
4099 | if (trans && !IS_ERR(trans)) |
4100 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 4101 | if (ret) |
efe120a0 | 4102 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 SB |
4103 | else |
4104 | fs_info->update_uuid_tree_gen = 1; | |
803b2f54 SB |
4105 | up(&fs_info->uuid_tree_rescan_sem); |
4106 | return 0; | |
4107 | } | |
4108 | ||
70f80175 SB |
4109 | /* |
4110 | * Callback for btrfs_uuid_tree_iterate(). | |
4111 | * returns: | |
4112 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4113 | * < 0 if an error occurred. |
70f80175 SB |
4114 | * > 0 if the check failed, which means the caller shall remove the entry. |
4115 | */ | |
4116 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4117 | u8 *uuid, u8 type, u64 subid) | |
4118 | { | |
4119 | struct btrfs_key key; | |
4120 | int ret = 0; | |
4121 | struct btrfs_root *subvol_root; | |
4122 | ||
4123 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4124 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4125 | goto out; | |
4126 | ||
4127 | key.objectid = subid; | |
4128 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4129 | key.offset = (u64)-1; | |
4130 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4131 | if (IS_ERR(subvol_root)) { | |
4132 | ret = PTR_ERR(subvol_root); | |
4133 | if (ret == -ENOENT) | |
4134 | ret = 1; | |
4135 | goto out; | |
4136 | } | |
4137 | ||
4138 | switch (type) { | |
4139 | case BTRFS_UUID_KEY_SUBVOL: | |
4140 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4141 | ret = 1; | |
4142 | break; | |
4143 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4144 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4145 | BTRFS_UUID_SIZE)) | |
4146 | ret = 1; | |
4147 | break; | |
4148 | } | |
4149 | ||
4150 | out: | |
4151 | return ret; | |
4152 | } | |
4153 | ||
4154 | static int btrfs_uuid_rescan_kthread(void *data) | |
4155 | { | |
4156 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4157 | int ret; | |
4158 | ||
4159 | /* | |
4160 | * 1st step is to iterate through the existing UUID tree and | |
4161 | * to delete all entries that contain outdated data. | |
4162 | * 2nd step is to add all missing entries to the UUID tree. | |
4163 | */ | |
4164 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4165 | if (ret < 0) { | |
efe120a0 | 4166 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4167 | up(&fs_info->uuid_tree_rescan_sem); |
4168 | return ret; | |
4169 | } | |
4170 | return btrfs_uuid_scan_kthread(data); | |
4171 | } | |
4172 | ||
f7a81ea4 SB |
4173 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4174 | { | |
4175 | struct btrfs_trans_handle *trans; | |
4176 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4177 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4178 | struct task_struct *task; |
4179 | int ret; | |
f7a81ea4 SB |
4180 | |
4181 | /* | |
4182 | * 1 - root node | |
4183 | * 1 - root item | |
4184 | */ | |
4185 | trans = btrfs_start_transaction(tree_root, 2); | |
4186 | if (IS_ERR(trans)) | |
4187 | return PTR_ERR(trans); | |
4188 | ||
4189 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4190 | BTRFS_UUID_TREE_OBJECTID); | |
4191 | if (IS_ERR(uuid_root)) { | |
6d13f549 DS |
4192 | ret = PTR_ERR(uuid_root); |
4193 | btrfs_abort_transaction(trans, tree_root, ret); | |
4194 | return ret; | |
f7a81ea4 SB |
4195 | } |
4196 | ||
4197 | fs_info->uuid_root = uuid_root; | |
4198 | ||
803b2f54 SB |
4199 | ret = btrfs_commit_transaction(trans, tree_root); |
4200 | if (ret) | |
4201 | return ret; | |
4202 | ||
4203 | down(&fs_info->uuid_tree_rescan_sem); | |
4204 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4205 | if (IS_ERR(task)) { | |
70f80175 | 4206 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4207 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4208 | up(&fs_info->uuid_tree_rescan_sem); |
4209 | return PTR_ERR(task); | |
4210 | } | |
4211 | ||
4212 | return 0; | |
f7a81ea4 | 4213 | } |
803b2f54 | 4214 | |
70f80175 SB |
4215 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4216 | { | |
4217 | struct task_struct *task; | |
4218 | ||
4219 | down(&fs_info->uuid_tree_rescan_sem); | |
4220 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4221 | if (IS_ERR(task)) { | |
4222 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4223 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4224 | up(&fs_info->uuid_tree_rescan_sem); |
4225 | return PTR_ERR(task); | |
4226 | } | |
4227 | ||
4228 | return 0; | |
4229 | } | |
4230 | ||
8f18cf13 CM |
4231 | /* |
4232 | * shrinking a device means finding all of the device extents past | |
4233 | * the new size, and then following the back refs to the chunks. | |
4234 | * The chunk relocation code actually frees the device extent | |
4235 | */ | |
4236 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4237 | { | |
4238 | struct btrfs_trans_handle *trans; | |
4239 | struct btrfs_root *root = device->dev_root; | |
4240 | struct btrfs_dev_extent *dev_extent = NULL; | |
4241 | struct btrfs_path *path; | |
4242 | u64 length; | |
8f18cf13 CM |
4243 | u64 chunk_offset; |
4244 | int ret; | |
4245 | int slot; | |
ba1bf481 JB |
4246 | int failed = 0; |
4247 | bool retried = false; | |
53e489bc | 4248 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4249 | struct extent_buffer *l; |
4250 | struct btrfs_key key; | |
6c41761f | 4251 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 4252 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4253 | u64 old_size = btrfs_device_get_total_bytes(device); |
4254 | u64 diff = old_size - new_size; | |
8f18cf13 | 4255 | |
63a212ab SB |
4256 | if (device->is_tgtdev_for_dev_replace) |
4257 | return -EINVAL; | |
4258 | ||
8f18cf13 CM |
4259 | path = btrfs_alloc_path(); |
4260 | if (!path) | |
4261 | return -ENOMEM; | |
4262 | ||
e4058b54 | 4263 | path->reada = READA_FORWARD; |
8f18cf13 | 4264 | |
7d9eb12c CM |
4265 | lock_chunks(root); |
4266 | ||
7cc8e58d | 4267 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4268 | if (device->writeable) { |
2b82032c | 4269 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
4270 | spin_lock(&root->fs_info->free_chunk_lock); |
4271 | root->fs_info->free_chunk_space -= diff; | |
4272 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4273 | } | |
7d9eb12c | 4274 | unlock_chunks(root); |
8f18cf13 | 4275 | |
ba1bf481 | 4276 | again: |
8f18cf13 CM |
4277 | key.objectid = device->devid; |
4278 | key.offset = (u64)-1; | |
4279 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4280 | ||
213e64da | 4281 | do { |
67c5e7d4 | 4282 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4283 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 FM |
4284 | if (ret < 0) { |
4285 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 | 4286 | goto done; |
67c5e7d4 | 4287 | } |
8f18cf13 CM |
4288 | |
4289 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 FM |
4290 | if (ret) |
4291 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 CM |
4292 | if (ret < 0) |
4293 | goto done; | |
4294 | if (ret) { | |
4295 | ret = 0; | |
b3b4aa74 | 4296 | btrfs_release_path(path); |
bf1fb512 | 4297 | break; |
8f18cf13 CM |
4298 | } |
4299 | ||
4300 | l = path->nodes[0]; | |
4301 | slot = path->slots[0]; | |
4302 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4303 | ||
ba1bf481 | 4304 | if (key.objectid != device->devid) { |
67c5e7d4 | 4305 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4306 | btrfs_release_path(path); |
bf1fb512 | 4307 | break; |
ba1bf481 | 4308 | } |
8f18cf13 CM |
4309 | |
4310 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4311 | length = btrfs_dev_extent_length(l, dev_extent); | |
4312 | ||
ba1bf481 | 4313 | if (key.offset + length <= new_size) { |
67c5e7d4 | 4314 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4315 | btrfs_release_path(path); |
d6397bae | 4316 | break; |
ba1bf481 | 4317 | } |
8f18cf13 | 4318 | |
8f18cf13 | 4319 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4320 | btrfs_release_path(path); |
8f18cf13 | 4321 | |
dc2ee4e2 | 4322 | ret = btrfs_relocate_chunk(root, chunk_offset); |
67c5e7d4 | 4323 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
ba1bf481 | 4324 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4325 | goto done; |
ba1bf481 JB |
4326 | if (ret == -ENOSPC) |
4327 | failed++; | |
213e64da | 4328 | } while (key.offset-- > 0); |
ba1bf481 JB |
4329 | |
4330 | if (failed && !retried) { | |
4331 | failed = 0; | |
4332 | retried = true; | |
4333 | goto again; | |
4334 | } else if (failed && retried) { | |
4335 | ret = -ENOSPC; | |
ba1bf481 | 4336 | goto done; |
8f18cf13 CM |
4337 | } |
4338 | ||
d6397bae | 4339 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4340 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4341 | if (IS_ERR(trans)) { |
4342 | ret = PTR_ERR(trans); | |
4343 | goto done; | |
4344 | } | |
4345 | ||
d6397bae | 4346 | lock_chunks(root); |
53e489bc FM |
4347 | |
4348 | /* | |
4349 | * We checked in the above loop all device extents that were already in | |
4350 | * the device tree. However before we have updated the device's | |
4351 | * total_bytes to the new size, we might have had chunk allocations that | |
4352 | * have not complete yet (new block groups attached to transaction | |
4353 | * handles), and therefore their device extents were not yet in the | |
4354 | * device tree and we missed them in the loop above. So if we have any | |
4355 | * pending chunk using a device extent that overlaps the device range | |
4356 | * that we can not use anymore, commit the current transaction and | |
4357 | * repeat the search on the device tree - this way we guarantee we will | |
4358 | * not have chunks using device extents that end beyond 'new_size'. | |
4359 | */ | |
4360 | if (!checked_pending_chunks) { | |
4361 | u64 start = new_size; | |
4362 | u64 len = old_size - new_size; | |
4363 | ||
499f377f JM |
4364 | if (contains_pending_extent(trans->transaction, device, |
4365 | &start, len)) { | |
53e489bc FM |
4366 | unlock_chunks(root); |
4367 | checked_pending_chunks = true; | |
4368 | failed = 0; | |
4369 | retried = false; | |
4370 | ret = btrfs_commit_transaction(trans, root); | |
4371 | if (ret) | |
4372 | goto done; | |
4373 | goto again; | |
4374 | } | |
4375 | } | |
4376 | ||
7cc8e58d | 4377 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4378 | if (list_empty(&device->resized_list)) |
4379 | list_add_tail(&device->resized_list, | |
4380 | &root->fs_info->fs_devices->resized_devices); | |
d6397bae | 4381 | |
d6397bae CB |
4382 | WARN_ON(diff > old_total); |
4383 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4384 | unlock_chunks(root); | |
2196d6e8 MX |
4385 | |
4386 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4387 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4388 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4389 | done: |
4390 | btrfs_free_path(path); | |
53e489bc FM |
4391 | if (ret) { |
4392 | lock_chunks(root); | |
4393 | btrfs_device_set_total_bytes(device, old_size); | |
4394 | if (device->writeable) | |
4395 | device->fs_devices->total_rw_bytes += diff; | |
4396 | spin_lock(&root->fs_info->free_chunk_lock); | |
4397 | root->fs_info->free_chunk_space += diff; | |
4398 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4399 | unlock_chunks(root); | |
4400 | } | |
8f18cf13 CM |
4401 | return ret; |
4402 | } | |
4403 | ||
125ccb0a | 4404 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4405 | struct btrfs_key *key, |
4406 | struct btrfs_chunk *chunk, int item_size) | |
4407 | { | |
6c41761f | 4408 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
4409 | struct btrfs_disk_key disk_key; |
4410 | u32 array_size; | |
4411 | u8 *ptr; | |
4412 | ||
fe48a5c0 | 4413 | lock_chunks(root); |
0b86a832 | 4414 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4415 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 MX |
4416 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
4417 | unlock_chunks(root); | |
0b86a832 | 4418 | return -EFBIG; |
fe48a5c0 | 4419 | } |
0b86a832 CM |
4420 | |
4421 | ptr = super_copy->sys_chunk_array + array_size; | |
4422 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4423 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4424 | ptr += sizeof(disk_key); | |
4425 | memcpy(ptr, chunk, item_size); | |
4426 | item_size += sizeof(disk_key); | |
4427 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 MX |
4428 | unlock_chunks(root); |
4429 | ||
0b86a832 CM |
4430 | return 0; |
4431 | } | |
4432 | ||
73c5de00 AJ |
4433 | /* |
4434 | * sort the devices in descending order by max_avail, total_avail | |
4435 | */ | |
4436 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4437 | { |
73c5de00 AJ |
4438 | const struct btrfs_device_info *di_a = a; |
4439 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4440 | |
73c5de00 | 4441 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4442 | return -1; |
73c5de00 | 4443 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4444 | return 1; |
73c5de00 AJ |
4445 | if (di_a->total_avail > di_b->total_avail) |
4446 | return -1; | |
4447 | if (di_a->total_avail < di_b->total_avail) | |
4448 | return 1; | |
4449 | return 0; | |
b2117a39 | 4450 | } |
0b86a832 | 4451 | |
53b381b3 DW |
4452 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4453 | { | |
4454 | /* TODO allow them to set a preferred stripe size */ | |
ee22184b | 4455 | return SZ_64K; |
53b381b3 DW |
4456 | } |
4457 | ||
4458 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4459 | { | |
ffe2d203 | 4460 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4461 | return; |
4462 | ||
ceda0864 | 4463 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4464 | } |
4465 | ||
23f8f9b7 GH |
4466 | #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \ |
4467 | - sizeof(struct btrfs_item) \ | |
4468 | - sizeof(struct btrfs_chunk)) \ | |
4469 | / sizeof(struct btrfs_stripe) + 1) | |
4470 | ||
4471 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4472 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4473 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4474 | / sizeof(struct btrfs_stripe) + 1) | |
4475 | ||
73c5de00 | 4476 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4477 | struct btrfs_root *extent_root, u64 start, |
4478 | u64 type) | |
b2117a39 | 4479 | { |
73c5de00 AJ |
4480 | struct btrfs_fs_info *info = extent_root->fs_info; |
4481 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4482 | struct list_head *cur; | |
4483 | struct map_lookup *map = NULL; | |
4484 | struct extent_map_tree *em_tree; | |
4485 | struct extent_map *em; | |
4486 | struct btrfs_device_info *devices_info = NULL; | |
4487 | u64 total_avail; | |
4488 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4489 | int data_stripes; /* number of stripes that count for |
4490 | block group size */ | |
73c5de00 AJ |
4491 | int sub_stripes; /* sub_stripes info for map */ |
4492 | int dev_stripes; /* stripes per dev */ | |
4493 | int devs_max; /* max devs to use */ | |
4494 | int devs_min; /* min devs needed */ | |
4495 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4496 | int ncopies; /* how many copies to data has */ | |
4497 | int ret; | |
4498 | u64 max_stripe_size; | |
4499 | u64 max_chunk_size; | |
4500 | u64 stripe_size; | |
4501 | u64 num_bytes; | |
53b381b3 | 4502 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4503 | int ndevs; |
4504 | int i; | |
4505 | int j; | |
31e50229 | 4506 | int index; |
593060d7 | 4507 | |
0c460c0d | 4508 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4509 | |
73c5de00 AJ |
4510 | if (list_empty(&fs_devices->alloc_list)) |
4511 | return -ENOSPC; | |
b2117a39 | 4512 | |
31e50229 | 4513 | index = __get_raid_index(type); |
73c5de00 | 4514 | |
31e50229 LB |
4515 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4516 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4517 | devs_max = btrfs_raid_array[index].devs_max; | |
4518 | devs_min = btrfs_raid_array[index].devs_min; | |
4519 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4520 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4521 | |
9b3f68b9 | 4522 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4523 | max_stripe_size = SZ_1G; |
73c5de00 | 4524 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 GH |
4525 | if (!devs_max) |
4526 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4527 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4528 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4529 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4530 | max_stripe_size = SZ_1G; | |
1100373f | 4531 | else |
ee22184b | 4532 | max_stripe_size = SZ_256M; |
73c5de00 | 4533 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4534 | if (!devs_max) |
4535 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4536 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4537 | max_stripe_size = SZ_32M; |
73c5de00 | 4538 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4539 | if (!devs_max) |
4540 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4541 | } else { |
351fd353 | 4542 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4543 | type); |
4544 | BUG_ON(1); | |
9b3f68b9 CM |
4545 | } |
4546 | ||
2b82032c YZ |
4547 | /* we don't want a chunk larger than 10% of writeable space */ |
4548 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4549 | max_chunk_size); | |
9b3f68b9 | 4550 | |
31e818fe | 4551 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4552 | GFP_NOFS); |
4553 | if (!devices_info) | |
4554 | return -ENOMEM; | |
0cad8a11 | 4555 | |
73c5de00 | 4556 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4557 | |
9f680ce0 | 4558 | /* |
73c5de00 AJ |
4559 | * in the first pass through the devices list, we gather information |
4560 | * about the available holes on each device. | |
9f680ce0 | 4561 | */ |
73c5de00 AJ |
4562 | ndevs = 0; |
4563 | while (cur != &fs_devices->alloc_list) { | |
4564 | struct btrfs_device *device; | |
4565 | u64 max_avail; | |
4566 | u64 dev_offset; | |
b2117a39 | 4567 | |
73c5de00 | 4568 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4569 | |
73c5de00 | 4570 | cur = cur->next; |
b2117a39 | 4571 | |
73c5de00 | 4572 | if (!device->writeable) { |
31b1a2bd | 4573 | WARN(1, KERN_ERR |
efe120a0 | 4574 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4575 | continue; |
4576 | } | |
b2117a39 | 4577 | |
63a212ab SB |
4578 | if (!device->in_fs_metadata || |
4579 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4580 | continue; |
b2117a39 | 4581 | |
73c5de00 AJ |
4582 | if (device->total_bytes > device->bytes_used) |
4583 | total_avail = device->total_bytes - device->bytes_used; | |
4584 | else | |
4585 | total_avail = 0; | |
38c01b96 | 4586 | |
4587 | /* If there is no space on this device, skip it. */ | |
4588 | if (total_avail == 0) | |
4589 | continue; | |
b2117a39 | 4590 | |
6df9a95e | 4591 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4592 | max_stripe_size * dev_stripes, |
4593 | &dev_offset, &max_avail); | |
4594 | if (ret && ret != -ENOSPC) | |
4595 | goto error; | |
b2117a39 | 4596 | |
73c5de00 AJ |
4597 | if (ret == 0) |
4598 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4599 | |
73c5de00 AJ |
4600 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4601 | continue; | |
b2117a39 | 4602 | |
063d006f ES |
4603 | if (ndevs == fs_devices->rw_devices) { |
4604 | WARN(1, "%s: found more than %llu devices\n", | |
4605 | __func__, fs_devices->rw_devices); | |
4606 | break; | |
4607 | } | |
73c5de00 AJ |
4608 | devices_info[ndevs].dev_offset = dev_offset; |
4609 | devices_info[ndevs].max_avail = max_avail; | |
4610 | devices_info[ndevs].total_avail = total_avail; | |
4611 | devices_info[ndevs].dev = device; | |
4612 | ++ndevs; | |
4613 | } | |
b2117a39 | 4614 | |
73c5de00 AJ |
4615 | /* |
4616 | * now sort the devices by hole size / available space | |
4617 | */ | |
4618 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4619 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4620 | |
73c5de00 AJ |
4621 | /* round down to number of usable stripes */ |
4622 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4623 | |
73c5de00 AJ |
4624 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4625 | ret = -ENOSPC; | |
4626 | goto error; | |
b2117a39 | 4627 | } |
9f680ce0 | 4628 | |
73c5de00 AJ |
4629 | if (devs_max && ndevs > devs_max) |
4630 | ndevs = devs_max; | |
4631 | /* | |
4632 | * the primary goal is to maximize the number of stripes, so use as many | |
4633 | * devices as possible, even if the stripes are not maximum sized. | |
4634 | */ | |
4635 | stripe_size = devices_info[ndevs-1].max_avail; | |
4636 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4637 | |
53b381b3 DW |
4638 | /* |
4639 | * this will have to be fixed for RAID1 and RAID10 over | |
4640 | * more drives | |
4641 | */ | |
4642 | data_stripes = num_stripes / ncopies; | |
4643 | ||
53b381b3 DW |
4644 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4645 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4646 | btrfs_super_stripesize(info->super_copy)); | |
4647 | data_stripes = num_stripes - 1; | |
4648 | } | |
4649 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4650 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4651 | btrfs_super_stripesize(info->super_copy)); | |
4652 | data_stripes = num_stripes - 2; | |
4653 | } | |
86db2578 CM |
4654 | |
4655 | /* | |
4656 | * Use the number of data stripes to figure out how big this chunk | |
4657 | * is really going to be in terms of logical address space, | |
4658 | * and compare that answer with the max chunk size | |
4659 | */ | |
4660 | if (stripe_size * data_stripes > max_chunk_size) { | |
4661 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4662 | |
4663 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4664 | |
4665 | /* bump the answer up to a 16MB boundary */ | |
4666 | stripe_size = (stripe_size + mask) & ~mask; | |
4667 | ||
4668 | /* but don't go higher than the limits we found | |
4669 | * while searching for free extents | |
4670 | */ | |
4671 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4672 | stripe_size = devices_info[ndevs-1].max_avail; | |
4673 | } | |
4674 | ||
b8b93add | 4675 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4676 | |
4677 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4678 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4679 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4680 | |
4681 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4682 | if (!map) { | |
4683 | ret = -ENOMEM; | |
4684 | goto error; | |
4685 | } | |
4686 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4687 | |
73c5de00 AJ |
4688 | for (i = 0; i < ndevs; ++i) { |
4689 | for (j = 0; j < dev_stripes; ++j) { | |
4690 | int s = i * dev_stripes + j; | |
4691 | map->stripes[s].dev = devices_info[i].dev; | |
4692 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4693 | j * stripe_size; | |
6324fbf3 | 4694 | } |
6324fbf3 | 4695 | } |
2b82032c | 4696 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4697 | map->stripe_len = raid_stripe_len; |
4698 | map->io_align = raid_stripe_len; | |
4699 | map->io_width = raid_stripe_len; | |
2b82032c | 4700 | map->type = type; |
2b82032c | 4701 | map->sub_stripes = sub_stripes; |
0b86a832 | 4702 | |
53b381b3 | 4703 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4704 | |
73c5de00 | 4705 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4706 | |
172ddd60 | 4707 | em = alloc_extent_map(); |
2b82032c | 4708 | if (!em) { |
298a8f9c | 4709 | kfree(map); |
b2117a39 MX |
4710 | ret = -ENOMEM; |
4711 | goto error; | |
593060d7 | 4712 | } |
298a8f9c | 4713 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4714 | em->map_lookup = map; |
2b82032c | 4715 | em->start = start; |
73c5de00 | 4716 | em->len = num_bytes; |
2b82032c YZ |
4717 | em->block_start = 0; |
4718 | em->block_len = em->len; | |
6df9a95e | 4719 | em->orig_block_len = stripe_size; |
593060d7 | 4720 | |
2b82032c | 4721 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4722 | write_lock(&em_tree->lock); |
09a2a8f9 | 4723 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4724 | if (!ret) { |
4725 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4726 | atomic_inc(&em->refs); | |
4727 | } | |
890871be | 4728 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4729 | if (ret) { |
4730 | free_extent_map(em); | |
1dd4602f | 4731 | goto error; |
0f5d42b2 | 4732 | } |
0b86a832 | 4733 | |
04487488 JB |
4734 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4735 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4736 | start, num_bytes); | |
6df9a95e JB |
4737 | if (ret) |
4738 | goto error_del_extent; | |
2b82032c | 4739 | |
7cc8e58d MX |
4740 | for (i = 0; i < map->num_stripes; i++) { |
4741 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4742 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4743 | } | |
43530c46 | 4744 | |
1c116187 MX |
4745 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4746 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4747 | map->num_stripes); | |
4748 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4749 | ||
0f5d42b2 | 4750 | free_extent_map(em); |
53b381b3 DW |
4751 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4752 | ||
b2117a39 | 4753 | kfree(devices_info); |
2b82032c | 4754 | return 0; |
b2117a39 | 4755 | |
6df9a95e | 4756 | error_del_extent: |
0f5d42b2 JB |
4757 | write_lock(&em_tree->lock); |
4758 | remove_extent_mapping(em_tree, em); | |
4759 | write_unlock(&em_tree->lock); | |
4760 | ||
4761 | /* One for our allocation */ | |
4762 | free_extent_map(em); | |
4763 | /* One for the tree reference */ | |
4764 | free_extent_map(em); | |
495e64f4 FM |
4765 | /* One for the pending_chunks list reference */ |
4766 | free_extent_map(em); | |
b2117a39 | 4767 | error: |
b2117a39 MX |
4768 | kfree(devices_info); |
4769 | return ret; | |
2b82032c YZ |
4770 | } |
4771 | ||
6df9a95e | 4772 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4773 | struct btrfs_root *extent_root, |
6df9a95e | 4774 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4775 | { |
2b82032c YZ |
4776 | struct btrfs_key key; |
4777 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4778 | struct btrfs_device *device; | |
4779 | struct btrfs_chunk *chunk; | |
4780 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4781 | struct extent_map_tree *em_tree; |
4782 | struct extent_map *em; | |
4783 | struct map_lookup *map; | |
4784 | size_t item_size; | |
4785 | u64 dev_offset; | |
4786 | u64 stripe_size; | |
4787 | int i = 0; | |
140e639f | 4788 | int ret = 0; |
2b82032c | 4789 | |
6df9a95e JB |
4790 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4791 | read_lock(&em_tree->lock); | |
4792 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4793 | read_unlock(&em_tree->lock); | |
4794 | ||
4795 | if (!em) { | |
4796 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4797 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4798 | return -EINVAL; | |
4799 | } | |
4800 | ||
4801 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4802 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
351fd353 | 4803 | " %Lu-%Lu, found %Lu-%Lu", chunk_offset, |
6df9a95e JB |
4804 | chunk_size, em->start, em->len); |
4805 | free_extent_map(em); | |
4806 | return -EINVAL; | |
4807 | } | |
4808 | ||
95617d69 | 4809 | map = em->map_lookup; |
6df9a95e JB |
4810 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4811 | stripe_size = em->orig_block_len; | |
4812 | ||
2b82032c | 4813 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4814 | if (!chunk) { |
4815 | ret = -ENOMEM; | |
4816 | goto out; | |
4817 | } | |
4818 | ||
50460e37 FM |
4819 | /* |
4820 | * Take the device list mutex to prevent races with the final phase of | |
4821 | * a device replace operation that replaces the device object associated | |
4822 | * with the map's stripes, because the device object's id can change | |
4823 | * at any time during that final phase of the device replace operation | |
4824 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4825 | */ | |
4826 | mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
6df9a95e JB |
4827 | for (i = 0; i < map->num_stripes; i++) { |
4828 | device = map->stripes[i].dev; | |
4829 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4830 | |
0b86a832 | 4831 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4832 | if (ret) |
50460e37 | 4833 | break; |
6df9a95e JB |
4834 | ret = btrfs_alloc_dev_extent(trans, device, |
4835 | chunk_root->root_key.objectid, | |
4836 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4837 | chunk_offset, dev_offset, | |
4838 | stripe_size); | |
4839 | if (ret) | |
50460e37 FM |
4840 | break; |
4841 | } | |
4842 | if (ret) { | |
4843 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
4844 | goto out; | |
2b82032c YZ |
4845 | } |
4846 | ||
2b82032c | 4847 | stripe = &chunk->stripe; |
6df9a95e JB |
4848 | for (i = 0; i < map->num_stripes; i++) { |
4849 | device = map->stripes[i].dev; | |
4850 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4851 | |
e17cade2 CM |
4852 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4853 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4854 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4855 | stripe++; |
0b86a832 | 4856 | } |
50460e37 | 4857 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4858 | |
2b82032c | 4859 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4860 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4861 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4862 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4863 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4864 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4865 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4866 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4867 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4868 | |
2b82032c YZ |
4869 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4870 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4871 | key.offset = chunk_offset; | |
0b86a832 | 4872 | |
2b82032c | 4873 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4874 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4875 | /* | |
4876 | * TODO: Cleanup of inserted chunk root in case of | |
4877 | * failure. | |
4878 | */ | |
125ccb0a | 4879 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4880 | item_size); |
8f18cf13 | 4881 | } |
1abe9b8a | 4882 | |
6df9a95e | 4883 | out: |
0b86a832 | 4884 | kfree(chunk); |
6df9a95e | 4885 | free_extent_map(em); |
4ed1d16e | 4886 | return ret; |
2b82032c | 4887 | } |
0b86a832 | 4888 | |
2b82032c YZ |
4889 | /* |
4890 | * Chunk allocation falls into two parts. The first part does works | |
4891 | * that make the new allocated chunk useable, but not do any operation | |
4892 | * that modifies the chunk tree. The second part does the works that | |
4893 | * require modifying the chunk tree. This division is important for the | |
4894 | * bootstrap process of adding storage to a seed btrfs. | |
4895 | */ | |
4896 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4897 | struct btrfs_root *extent_root, u64 type) | |
4898 | { | |
4899 | u64 chunk_offset; | |
2b82032c | 4900 | |
a9629596 | 4901 | ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex)); |
6df9a95e JB |
4902 | chunk_offset = find_next_chunk(extent_root->fs_info); |
4903 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
4904 | } |
4905 | ||
d397712b | 4906 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4907 | struct btrfs_root *root, |
4908 | struct btrfs_device *device) | |
4909 | { | |
4910 | u64 chunk_offset; | |
4911 | u64 sys_chunk_offset; | |
2b82032c | 4912 | u64 alloc_profile; |
2b82032c YZ |
4913 | struct btrfs_fs_info *fs_info = root->fs_info; |
4914 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4915 | int ret; | |
4916 | ||
6df9a95e | 4917 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 4918 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
4919 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
4920 | alloc_profile); | |
79787eaa JM |
4921 | if (ret) |
4922 | return ret; | |
2b82032c | 4923 | |
6df9a95e | 4924 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 4925 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
4926 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
4927 | alloc_profile); | |
79787eaa | 4928 | return ret; |
2b82032c YZ |
4929 | } |
4930 | ||
d20983b4 MX |
4931 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
4932 | { | |
4933 | int max_errors; | |
4934 | ||
4935 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4936 | BTRFS_BLOCK_GROUP_RAID10 | | |
4937 | BTRFS_BLOCK_GROUP_RAID5 | | |
4938 | BTRFS_BLOCK_GROUP_DUP)) { | |
4939 | max_errors = 1; | |
4940 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4941 | max_errors = 2; | |
4942 | } else { | |
4943 | max_errors = 0; | |
005d6427 | 4944 | } |
2b82032c | 4945 | |
d20983b4 | 4946 | return max_errors; |
2b82032c YZ |
4947 | } |
4948 | ||
4949 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4950 | { | |
4951 | struct extent_map *em; | |
4952 | struct map_lookup *map; | |
4953 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4954 | int readonly = 0; | |
d20983b4 | 4955 | int miss_ndevs = 0; |
2b82032c YZ |
4956 | int i; |
4957 | ||
890871be | 4958 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4959 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4960 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4961 | if (!em) |
4962 | return 1; | |
4963 | ||
95617d69 | 4964 | map = em->map_lookup; |
2b82032c | 4965 | for (i = 0; i < map->num_stripes; i++) { |
d20983b4 MX |
4966 | if (map->stripes[i].dev->missing) { |
4967 | miss_ndevs++; | |
4968 | continue; | |
4969 | } | |
4970 | ||
2b82032c YZ |
4971 | if (!map->stripes[i].dev->writeable) { |
4972 | readonly = 1; | |
d20983b4 | 4973 | goto end; |
2b82032c YZ |
4974 | } |
4975 | } | |
d20983b4 MX |
4976 | |
4977 | /* | |
4978 | * If the number of missing devices is larger than max errors, | |
4979 | * we can not write the data into that chunk successfully, so | |
4980 | * set it readonly. | |
4981 | */ | |
4982 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
4983 | readonly = 1; | |
4984 | end: | |
0b86a832 | 4985 | free_extent_map(em); |
2b82032c | 4986 | return readonly; |
0b86a832 CM |
4987 | } |
4988 | ||
4989 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4990 | { | |
a8067e02 | 4991 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
4992 | } |
4993 | ||
4994 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4995 | { | |
4996 | struct extent_map *em; | |
4997 | ||
d397712b | 4998 | while (1) { |
890871be | 4999 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5000 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5001 | if (em) | |
5002 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5003 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5004 | if (!em) |
5005 | break; | |
0b86a832 CM |
5006 | /* once for us */ |
5007 | free_extent_map(em); | |
5008 | /* once for the tree */ | |
5009 | free_extent_map(em); | |
5010 | } | |
5011 | } | |
5012 | ||
5d964051 | 5013 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 5014 | { |
5d964051 | 5015 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
5016 | struct extent_map *em; |
5017 | struct map_lookup *map; | |
5018 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5019 | int ret; | |
5020 | ||
890871be | 5021 | read_lock(&em_tree->lock); |
f188591e | 5022 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 5023 | read_unlock(&em_tree->lock); |
f188591e | 5024 | |
fb7669b5 JB |
5025 | /* |
5026 | * We could return errors for these cases, but that could get ugly and | |
5027 | * we'd probably do the same thing which is just not do anything else | |
5028 | * and exit, so return 1 so the callers don't try to use other copies. | |
5029 | */ | |
5030 | if (!em) { | |
351fd353 | 5031 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
5032 | logical+len); |
5033 | return 1; | |
5034 | } | |
5035 | ||
5036 | if (em->start > logical || em->start + em->len < logical) { | |
ccf39f92 | 5037 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " |
351fd353 | 5038 | "%Lu-%Lu", logical, logical+len, em->start, |
fb7669b5 | 5039 | em->start + em->len); |
7d3d1744 | 5040 | free_extent_map(em); |
fb7669b5 JB |
5041 | return 1; |
5042 | } | |
5043 | ||
95617d69 | 5044 | map = em->map_lookup; |
f188591e CM |
5045 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5046 | ret = map->num_stripes; | |
321aecc6 CM |
5047 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5048 | ret = map->sub_stripes; | |
53b381b3 DW |
5049 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5050 | ret = 2; | |
5051 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5052 | ret = 3; | |
f188591e CM |
5053 | else |
5054 | ret = 1; | |
5055 | free_extent_map(em); | |
ad6d620e | 5056 | |
73beece9 | 5057 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
ad6d620e SB |
5058 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) |
5059 | ret++; | |
73beece9 | 5060 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5061 | |
f188591e CM |
5062 | return ret; |
5063 | } | |
5064 | ||
53b381b3 DW |
5065 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
5066 | struct btrfs_mapping_tree *map_tree, | |
5067 | u64 logical) | |
5068 | { | |
5069 | struct extent_map *em; | |
5070 | struct map_lookup *map; | |
5071 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5072 | unsigned long len = root->sectorsize; | |
5073 | ||
5074 | read_lock(&em_tree->lock); | |
5075 | em = lookup_extent_mapping(em_tree, logical, len); | |
5076 | read_unlock(&em_tree->lock); | |
5077 | BUG_ON(!em); | |
5078 | ||
5079 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
95617d69 | 5080 | map = em->map_lookup; |
ffe2d203 | 5081 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 5082 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
5083 | free_extent_map(em); |
5084 | return len; | |
5085 | } | |
5086 | ||
5087 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
5088 | u64 logical, u64 len, int mirror_num) | |
5089 | { | |
5090 | struct extent_map *em; | |
5091 | struct map_lookup *map; | |
5092 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5093 | int ret = 0; | |
5094 | ||
5095 | read_lock(&em_tree->lock); | |
5096 | em = lookup_extent_mapping(em_tree, logical, len); | |
5097 | read_unlock(&em_tree->lock); | |
5098 | BUG_ON(!em); | |
5099 | ||
5100 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
95617d69 | 5101 | map = em->map_lookup; |
ffe2d203 | 5102 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
5103 | ret = 1; |
5104 | free_extent_map(em); | |
5105 | return ret; | |
5106 | } | |
5107 | ||
30d9861f SB |
5108 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5109 | struct map_lookup *map, int first, int num, | |
5110 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5111 | { |
5112 | int i; | |
30d9861f SB |
5113 | int tolerance; |
5114 | struct btrfs_device *srcdev; | |
5115 | ||
5116 | if (dev_replace_is_ongoing && | |
5117 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5118 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5119 | srcdev = fs_info->dev_replace.srcdev; | |
5120 | else | |
5121 | srcdev = NULL; | |
5122 | ||
5123 | /* | |
5124 | * try to avoid the drive that is the source drive for a | |
5125 | * dev-replace procedure, only choose it if no other non-missing | |
5126 | * mirror is available | |
5127 | */ | |
5128 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5129 | if (map->stripes[optimal].dev->bdev && | |
5130 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5131 | return optimal; | |
5132 | for (i = first; i < first + num; i++) { | |
5133 | if (map->stripes[i].dev->bdev && | |
5134 | (tolerance || map->stripes[i].dev != srcdev)) | |
5135 | return i; | |
5136 | } | |
dfe25020 | 5137 | } |
30d9861f | 5138 | |
dfe25020 CM |
5139 | /* we couldn't find one that doesn't fail. Just return something |
5140 | * and the io error handling code will clean up eventually | |
5141 | */ | |
5142 | return optimal; | |
5143 | } | |
5144 | ||
53b381b3 DW |
5145 | static inline int parity_smaller(u64 a, u64 b) |
5146 | { | |
5147 | return a > b; | |
5148 | } | |
5149 | ||
5150 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5151 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5152 | { |
5153 | struct btrfs_bio_stripe s; | |
5154 | int i; | |
5155 | u64 l; | |
5156 | int again = 1; | |
5157 | ||
5158 | while (again) { | |
5159 | again = 0; | |
cc7539ed | 5160 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5161 | if (parity_smaller(bbio->raid_map[i], |
5162 | bbio->raid_map[i+1])) { | |
53b381b3 | 5163 | s = bbio->stripes[i]; |
8e5cfb55 | 5164 | l = bbio->raid_map[i]; |
53b381b3 | 5165 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5166 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5167 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5168 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5169 | |
53b381b3 DW |
5170 | again = 1; |
5171 | } | |
5172 | } | |
5173 | } | |
5174 | } | |
5175 | ||
6e9606d2 ZL |
5176 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5177 | { | |
5178 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5179 | /* the size of the btrfs_bio */ |
6e9606d2 | 5180 | sizeof(struct btrfs_bio) + |
e57cf21e | 5181 | /* plus the variable array for the stripes */ |
6e9606d2 | 5182 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5183 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5184 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5185 | /* |
5186 | * plus the raid_map, which includes both the tgt dev | |
5187 | * and the stripes | |
5188 | */ | |
5189 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5190 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5191 | |
5192 | atomic_set(&bbio->error, 0); | |
5193 | atomic_set(&bbio->refs, 1); | |
5194 | ||
5195 | return bbio; | |
5196 | } | |
5197 | ||
5198 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5199 | { | |
5200 | WARN_ON(!atomic_read(&bbio->refs)); | |
5201 | atomic_inc(&bbio->refs); | |
5202 | } | |
5203 | ||
5204 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5205 | { | |
5206 | if (!bbio) | |
5207 | return; | |
5208 | if (atomic_dec_and_test(&bbio->refs)) | |
5209 | kfree(bbio); | |
5210 | } | |
5211 | ||
3ec706c8 | 5212 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5213 | u64 logical, u64 *length, |
a1d3c478 | 5214 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5215 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5216 | { |
5217 | struct extent_map *em; | |
5218 | struct map_lookup *map; | |
3ec706c8 | 5219 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
5220 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5221 | u64 offset; | |
593060d7 | 5222 | u64 stripe_offset; |
fce3bb9a | 5223 | u64 stripe_end_offset; |
593060d7 | 5224 | u64 stripe_nr; |
fce3bb9a LD |
5225 | u64 stripe_nr_orig; |
5226 | u64 stripe_nr_end; | |
53b381b3 | 5227 | u64 stripe_len; |
9d644a62 | 5228 | u32 stripe_index; |
cea9e445 | 5229 | int i; |
de11cc12 | 5230 | int ret = 0; |
f2d8d74d | 5231 | int num_stripes; |
a236aed1 | 5232 | int max_errors = 0; |
2c8cdd6e | 5233 | int tgtdev_indexes = 0; |
a1d3c478 | 5234 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5235 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5236 | int dev_replace_is_ongoing = 0; | |
5237 | int num_alloc_stripes; | |
ad6d620e SB |
5238 | int patch_the_first_stripe_for_dev_replace = 0; |
5239 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5240 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5241 | |
890871be | 5242 | read_lock(&em_tree->lock); |
0b86a832 | 5243 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 5244 | read_unlock(&em_tree->lock); |
f2d8d74d | 5245 | |
3b951516 | 5246 | if (!em) { |
c2cf52eb | 5247 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 5248 | logical, *length); |
9bb91873 JB |
5249 | return -EINVAL; |
5250 | } | |
5251 | ||
5252 | if (em->start > logical || em->start + em->len < logical) { | |
5253 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
351fd353 | 5254 | "found %Lu-%Lu", logical, em->start, |
9bb91873 | 5255 | em->start + em->len); |
7d3d1744 | 5256 | free_extent_map(em); |
9bb91873 | 5257 | return -EINVAL; |
3b951516 | 5258 | } |
0b86a832 | 5259 | |
95617d69 | 5260 | map = em->map_lookup; |
0b86a832 | 5261 | offset = logical - em->start; |
593060d7 | 5262 | |
53b381b3 | 5263 | stripe_len = map->stripe_len; |
593060d7 CM |
5264 | stripe_nr = offset; |
5265 | /* | |
5266 | * stripe_nr counts the total number of stripes we have to stride | |
5267 | * to get to this block | |
5268 | */ | |
47c5713f | 5269 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5270 | |
53b381b3 | 5271 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
5272 | BUG_ON(offset < stripe_offset); |
5273 | ||
5274 | /* stripe_offset is the offset of this block in its stripe*/ | |
5275 | stripe_offset = offset - stripe_offset; | |
5276 | ||
53b381b3 | 5277 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5278 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5279 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5280 | raid56_full_stripe_start = offset; | |
5281 | ||
5282 | /* allow a write of a full stripe, but make sure we don't | |
5283 | * allow straddling of stripes | |
5284 | */ | |
47c5713f DS |
5285 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5286 | full_stripe_len); | |
53b381b3 DW |
5287 | raid56_full_stripe_start *= full_stripe_len; |
5288 | } | |
5289 | ||
5290 | if (rw & REQ_DISCARD) { | |
5291 | /* we don't discard raid56 yet */ | |
ffe2d203 | 5292 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5293 | ret = -EOPNOTSUPP; |
5294 | goto out; | |
5295 | } | |
fce3bb9a | 5296 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
5297 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
5298 | u64 max_len; | |
5299 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5300 | For other RAID types and for RAID[56] reads, just allow a single | |
5301 | stripe (on a single disk). */ | |
ffe2d203 | 5302 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
53b381b3 DW |
5303 | (rw & REQ_WRITE)) { |
5304 | max_len = stripe_len * nr_data_stripes(map) - | |
5305 | (offset - raid56_full_stripe_start); | |
5306 | } else { | |
5307 | /* we limit the length of each bio to what fits in a stripe */ | |
5308 | max_len = stripe_len - stripe_offset; | |
5309 | } | |
5310 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5311 | } else { |
5312 | *length = em->len - offset; | |
5313 | } | |
f2d8d74d | 5314 | |
53b381b3 DW |
5315 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5316 | it cares about is the length */ | |
a1d3c478 | 5317 | if (!bbio_ret) |
cea9e445 CM |
5318 | goto out; |
5319 | ||
73beece9 | 5320 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5321 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5322 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5323 | btrfs_dev_replace_unlock(dev_replace, 0); |
5324 | else | |
5325 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5326 | |
ad6d620e SB |
5327 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
5328 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
5329 | dev_replace->tgtdev != NULL) { | |
5330 | /* | |
5331 | * in dev-replace case, for repair case (that's the only | |
5332 | * case where the mirror is selected explicitly when | |
5333 | * calling btrfs_map_block), blocks left of the left cursor | |
5334 | * can also be read from the target drive. | |
5335 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5336 | * the last one to the array of stripes. For READ, it also | |
5337 | * needs to be supported using the same mirror number. | |
5338 | * If the requested block is not left of the left cursor, | |
5339 | * EIO is returned. This can happen because btrfs_num_copies() | |
5340 | * returns one more in the dev-replace case. | |
5341 | */ | |
5342 | u64 tmp_length = *length; | |
5343 | struct btrfs_bio *tmp_bbio = NULL; | |
5344 | int tmp_num_stripes; | |
5345 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5346 | int index_srcdev = 0; | |
5347 | int found = 0; | |
5348 | u64 physical_of_found = 0; | |
5349 | ||
5350 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
8e5cfb55 | 5351 | logical, &tmp_length, &tmp_bbio, 0, 0); |
ad6d620e SB |
5352 | if (ret) { |
5353 | WARN_ON(tmp_bbio != NULL); | |
5354 | goto out; | |
5355 | } | |
5356 | ||
5357 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5358 | if (mirror_num > tmp_num_stripes) { | |
5359 | /* | |
5360 | * REQ_GET_READ_MIRRORS does not contain this | |
5361 | * mirror, that means that the requested area | |
5362 | * is not left of the left cursor | |
5363 | */ | |
5364 | ret = -EIO; | |
6e9606d2 | 5365 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5366 | goto out; |
5367 | } | |
5368 | ||
5369 | /* | |
5370 | * process the rest of the function using the mirror_num | |
5371 | * of the source drive. Therefore look it up first. | |
5372 | * At the end, patch the device pointer to the one of the | |
5373 | * target drive. | |
5374 | */ | |
5375 | for (i = 0; i < tmp_num_stripes; i++) { | |
94a97dfe ZL |
5376 | if (tmp_bbio->stripes[i].dev->devid != srcdev_devid) |
5377 | continue; | |
5378 | ||
5379 | /* | |
5380 | * In case of DUP, in order to keep it simple, only add | |
5381 | * the mirror with the lowest physical address | |
5382 | */ | |
5383 | if (found && | |
5384 | physical_of_found <= tmp_bbio->stripes[i].physical) | |
5385 | continue; | |
5386 | ||
5387 | index_srcdev = i; | |
5388 | found = 1; | |
5389 | physical_of_found = tmp_bbio->stripes[i].physical; | |
ad6d620e SB |
5390 | } |
5391 | ||
94a97dfe ZL |
5392 | btrfs_put_bbio(tmp_bbio); |
5393 | ||
5394 | if (!found) { | |
ad6d620e SB |
5395 | WARN_ON(1); |
5396 | ret = -EIO; | |
ad6d620e SB |
5397 | goto out; |
5398 | } | |
5399 | ||
94a97dfe ZL |
5400 | mirror_num = index_srcdev + 1; |
5401 | patch_the_first_stripe_for_dev_replace = 1; | |
5402 | physical_to_patch_in_first_stripe = physical_of_found; | |
ad6d620e SB |
5403 | } else if (mirror_num > map->num_stripes) { |
5404 | mirror_num = 0; | |
5405 | } | |
5406 | ||
f2d8d74d | 5407 | num_stripes = 1; |
cea9e445 | 5408 | stripe_index = 0; |
fce3bb9a | 5409 | stripe_nr_orig = stripe_nr; |
fda2832f | 5410 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
b8b93add | 5411 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); |
fce3bb9a LD |
5412 | stripe_end_offset = stripe_nr_end * map->stripe_len - |
5413 | (offset + *length); | |
53b381b3 | 5414 | |
fce3bb9a LD |
5415 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5416 | if (rw & REQ_DISCARD) | |
5417 | num_stripes = min_t(u64, map->num_stripes, | |
5418 | stripe_nr_end - stripe_nr_orig); | |
47c5713f DS |
5419 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5420 | &stripe_index); | |
28e1cc7d MX |
5421 | if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))) |
5422 | mirror_num = 1; | |
fce3bb9a | 5423 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
29a8d9a0 | 5424 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5425 | num_stripes = map->num_stripes; |
2fff734f | 5426 | else if (mirror_num) |
f188591e | 5427 | stripe_index = mirror_num - 1; |
dfe25020 | 5428 | else { |
30d9861f | 5429 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5430 | map->num_stripes, |
30d9861f SB |
5431 | current->pid % map->num_stripes, |
5432 | dev_replace_is_ongoing); | |
a1d3c478 | 5433 | mirror_num = stripe_index + 1; |
dfe25020 | 5434 | } |
2fff734f | 5435 | |
611f0e00 | 5436 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 5437 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 5438 | num_stripes = map->num_stripes; |
a1d3c478 | 5439 | } else if (mirror_num) { |
f188591e | 5440 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5441 | } else { |
5442 | mirror_num = 1; | |
5443 | } | |
2fff734f | 5444 | |
321aecc6 | 5445 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5446 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5447 | |
47c5713f | 5448 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5449 | stripe_index *= map->sub_stripes; |
5450 | ||
29a8d9a0 | 5451 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5452 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
5453 | else if (rw & REQ_DISCARD) |
5454 | num_stripes = min_t(u64, map->sub_stripes * | |
5455 | (stripe_nr_end - stripe_nr_orig), | |
5456 | map->num_stripes); | |
321aecc6 CM |
5457 | else if (mirror_num) |
5458 | stripe_index += mirror_num - 1; | |
dfe25020 | 5459 | else { |
3e74317a | 5460 | int old_stripe_index = stripe_index; |
30d9861f SB |
5461 | stripe_index = find_live_mirror(fs_info, map, |
5462 | stripe_index, | |
dfe25020 | 5463 | map->sub_stripes, stripe_index + |
30d9861f SB |
5464 | current->pid % map->sub_stripes, |
5465 | dev_replace_is_ongoing); | |
3e74317a | 5466 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5467 | } |
53b381b3 | 5468 | |
ffe2d203 | 5469 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5470 | if (need_raid_map && |
af8e2d1d MX |
5471 | ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5472 | mirror_num > 1)) { | |
53b381b3 | 5473 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5474 | stripe_nr = div_u64(raid56_full_stripe_start, |
5475 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5476 | |
5477 | /* RAID[56] write or recovery. Return all stripes */ | |
5478 | num_stripes = map->num_stripes; | |
5479 | max_errors = nr_parity_stripes(map); | |
5480 | ||
53b381b3 DW |
5481 | *length = map->stripe_len; |
5482 | stripe_index = 0; | |
5483 | stripe_offset = 0; | |
5484 | } else { | |
5485 | /* | |
5486 | * Mirror #0 or #1 means the original data block. | |
5487 | * Mirror #2 is RAID5 parity block. | |
5488 | * Mirror #3 is RAID6 Q block. | |
5489 | */ | |
47c5713f DS |
5490 | stripe_nr = div_u64_rem(stripe_nr, |
5491 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5492 | if (mirror_num > 1) |
5493 | stripe_index = nr_data_stripes(map) + | |
5494 | mirror_num - 2; | |
5495 | ||
5496 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5497 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5498 | &stripe_index); | |
28e1cc7d MX |
5499 | if (!(rw & (REQ_WRITE | REQ_DISCARD | |
5500 | REQ_GET_READ_MIRRORS)) && mirror_num <= 1) | |
5501 | mirror_num = 1; | |
53b381b3 | 5502 | } |
8790d502 CM |
5503 | } else { |
5504 | /* | |
47c5713f DS |
5505 | * after this, stripe_nr is the number of stripes on this |
5506 | * device we have to walk to find the data, and stripe_index is | |
5507 | * the number of our device in the stripe array | |
8790d502 | 5508 | */ |
47c5713f DS |
5509 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5510 | &stripe_index); | |
a1d3c478 | 5511 | mirror_num = stripe_index + 1; |
8790d502 | 5512 | } |
593060d7 | 5513 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 5514 | |
472262f3 | 5515 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
5516 | if (dev_replace_is_ongoing) { |
5517 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
5518 | num_alloc_stripes <<= 1; | |
5519 | if (rw & REQ_GET_READ_MIRRORS) | |
5520 | num_alloc_stripes++; | |
2c8cdd6e | 5521 | tgtdev_indexes = num_stripes; |
ad6d620e | 5522 | } |
2c8cdd6e | 5523 | |
6e9606d2 | 5524 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5525 | if (!bbio) { |
5526 | ret = -ENOMEM; | |
5527 | goto out; | |
5528 | } | |
2c8cdd6e MX |
5529 | if (dev_replace_is_ongoing) |
5530 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
de11cc12 | 5531 | |
8e5cfb55 | 5532 | /* build raid_map */ |
ffe2d203 | 5533 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
8e5cfb55 ZL |
5534 | need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5535 | mirror_num > 1)) { | |
5536 | u64 tmp; | |
9d644a62 | 5537 | unsigned rot; |
8e5cfb55 ZL |
5538 | |
5539 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5540 | sizeof(struct btrfs_bio_stripe) * | |
5541 | num_alloc_stripes + | |
5542 | sizeof(int) * tgtdev_indexes); | |
5543 | ||
5544 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5545 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5546 | |
5547 | /* Fill in the logical address of each stripe */ | |
5548 | tmp = stripe_nr * nr_data_stripes(map); | |
5549 | for (i = 0; i < nr_data_stripes(map); i++) | |
5550 | bbio->raid_map[(i+rot) % num_stripes] = | |
5551 | em->start + (tmp + i) * map->stripe_len; | |
5552 | ||
5553 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5554 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5555 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5556 | RAID6_Q_STRIPE; | |
5557 | } | |
5558 | ||
fce3bb9a | 5559 | if (rw & REQ_DISCARD) { |
9d644a62 DS |
5560 | u32 factor = 0; |
5561 | u32 sub_stripes = 0; | |
ec9ef7a1 LZ |
5562 | u64 stripes_per_dev = 0; |
5563 | u32 remaining_stripes = 0; | |
b89203f7 | 5564 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5565 | |
5566 | if (map->type & | |
5567 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5568 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5569 | sub_stripes = 1; | |
5570 | else | |
5571 | sub_stripes = map->sub_stripes; | |
5572 | ||
5573 | factor = map->num_stripes / sub_stripes; | |
5574 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5575 | stripe_nr_orig, | |
5576 | factor, | |
5577 | &remaining_stripes); | |
b89203f7 LB |
5578 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5579 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5580 | } |
5581 | ||
fce3bb9a | 5582 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5583 | bbio->stripes[i].physical = |
f2d8d74d CM |
5584 | map->stripes[stripe_index].physical + |
5585 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5586 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5587 | |
ec9ef7a1 LZ |
5588 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5589 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5590 | bbio->stripes[i].length = stripes_per_dev * | |
5591 | map->stripe_len; | |
b89203f7 | 5592 | |
ec9ef7a1 LZ |
5593 | if (i / sub_stripes < remaining_stripes) |
5594 | bbio->stripes[i].length += | |
5595 | map->stripe_len; | |
b89203f7 LB |
5596 | |
5597 | /* | |
5598 | * Special for the first stripe and | |
5599 | * the last stripe: | |
5600 | * | |
5601 | * |-------|...|-------| | |
5602 | * |----------| | |
5603 | * off end_off | |
5604 | */ | |
ec9ef7a1 | 5605 | if (i < sub_stripes) |
a1d3c478 | 5606 | bbio->stripes[i].length -= |
fce3bb9a | 5607 | stripe_offset; |
b89203f7 LB |
5608 | |
5609 | if (stripe_index >= last_stripe && | |
5610 | stripe_index <= (last_stripe + | |
5611 | sub_stripes - 1)) | |
a1d3c478 | 5612 | bbio->stripes[i].length -= |
fce3bb9a | 5613 | stripe_end_offset; |
b89203f7 | 5614 | |
ec9ef7a1 LZ |
5615 | if (i == sub_stripes - 1) |
5616 | stripe_offset = 0; | |
fce3bb9a | 5617 | } else |
a1d3c478 | 5618 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5619 | |
5620 | stripe_index++; | |
5621 | if (stripe_index == map->num_stripes) { | |
5622 | /* This could only happen for RAID0/10 */ | |
5623 | stripe_index = 0; | |
5624 | stripe_nr++; | |
5625 | } | |
5626 | } | |
5627 | } else { | |
5628 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5629 | bbio->stripes[i].physical = |
212a17ab LT |
5630 | map->stripes[stripe_index].physical + |
5631 | stripe_offset + | |
5632 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5633 | bbio->stripes[i].dev = |
212a17ab | 5634 | map->stripes[stripe_index].dev; |
fce3bb9a | 5635 | stripe_index++; |
f2d8d74d | 5636 | } |
593060d7 | 5637 | } |
de11cc12 | 5638 | |
d20983b4 MX |
5639 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
5640 | max_errors = btrfs_chunk_max_errors(map); | |
de11cc12 | 5641 | |
8e5cfb55 ZL |
5642 | if (bbio->raid_map) |
5643 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5644 | |
2c8cdd6e | 5645 | tgtdev_indexes = 0; |
472262f3 SB |
5646 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
5647 | dev_replace->tgtdev != NULL) { | |
5648 | int index_where_to_add; | |
5649 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5650 | ||
5651 | /* | |
5652 | * duplicate the write operations while the dev replace | |
5653 | * procedure is running. Since the copying of the old disk | |
5654 | * to the new disk takes place at run time while the | |
5655 | * filesystem is mounted writable, the regular write | |
5656 | * operations to the old disk have to be duplicated to go | |
5657 | * to the new disk as well. | |
5658 | * Note that device->missing is handled by the caller, and | |
5659 | * that the write to the old disk is already set up in the | |
5660 | * stripes array. | |
5661 | */ | |
5662 | index_where_to_add = num_stripes; | |
5663 | for (i = 0; i < num_stripes; i++) { | |
5664 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5665 | /* write to new disk, too */ | |
5666 | struct btrfs_bio_stripe *new = | |
5667 | bbio->stripes + index_where_to_add; | |
5668 | struct btrfs_bio_stripe *old = | |
5669 | bbio->stripes + i; | |
5670 | ||
5671 | new->physical = old->physical; | |
5672 | new->length = old->length; | |
5673 | new->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5674 | bbio->tgtdev_map[i] = index_where_to_add; |
472262f3 SB |
5675 | index_where_to_add++; |
5676 | max_errors++; | |
2c8cdd6e | 5677 | tgtdev_indexes++; |
472262f3 SB |
5678 | } |
5679 | } | |
5680 | num_stripes = index_where_to_add; | |
ad6d620e SB |
5681 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
5682 | dev_replace->tgtdev != NULL) { | |
5683 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5684 | int index_srcdev = 0; | |
5685 | int found = 0; | |
5686 | u64 physical_of_found = 0; | |
5687 | ||
5688 | /* | |
5689 | * During the dev-replace procedure, the target drive can | |
5690 | * also be used to read data in case it is needed to repair | |
5691 | * a corrupt block elsewhere. This is possible if the | |
5692 | * requested area is left of the left cursor. In this area, | |
5693 | * the target drive is a full copy of the source drive. | |
5694 | */ | |
5695 | for (i = 0; i < num_stripes; i++) { | |
5696 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5697 | /* | |
5698 | * In case of DUP, in order to keep it | |
5699 | * simple, only add the mirror with the | |
5700 | * lowest physical address | |
5701 | */ | |
5702 | if (found && | |
5703 | physical_of_found <= | |
5704 | bbio->stripes[i].physical) | |
5705 | continue; | |
5706 | index_srcdev = i; | |
5707 | found = 1; | |
5708 | physical_of_found = bbio->stripes[i].physical; | |
5709 | } | |
5710 | } | |
5711 | if (found) { | |
258ece02 | 5712 | if (physical_of_found + map->stripe_len <= |
ad6d620e SB |
5713 | dev_replace->cursor_left) { |
5714 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5715 | bbio->stripes + num_stripes; | |
5716 | ||
5717 | tgtdev_stripe->physical = physical_of_found; | |
5718 | tgtdev_stripe->length = | |
5719 | bbio->stripes[index_srcdev].length; | |
5720 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5721 | bbio->tgtdev_map[index_srcdev] = num_stripes; |
ad6d620e | 5722 | |
2c8cdd6e | 5723 | tgtdev_indexes++; |
ad6d620e SB |
5724 | num_stripes++; |
5725 | } | |
5726 | } | |
472262f3 SB |
5727 | } |
5728 | ||
de11cc12 | 5729 | *bbio_ret = bbio; |
10f11900 | 5730 | bbio->map_type = map->type; |
de11cc12 LZ |
5731 | bbio->num_stripes = num_stripes; |
5732 | bbio->max_errors = max_errors; | |
5733 | bbio->mirror_num = mirror_num; | |
2c8cdd6e | 5734 | bbio->num_tgtdevs = tgtdev_indexes; |
ad6d620e SB |
5735 | |
5736 | /* | |
5737 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5738 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5739 | * available as a mirror | |
5740 | */ | |
5741 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5742 | WARN_ON(num_stripes > 1); | |
5743 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5744 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5745 | bbio->mirror_num = map->num_stripes + 1; | |
5746 | } | |
cea9e445 | 5747 | out: |
73beece9 LB |
5748 | if (dev_replace_is_ongoing) { |
5749 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5750 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5751 | } | |
0b86a832 | 5752 | free_extent_map(em); |
de11cc12 | 5753 | return ret; |
0b86a832 CM |
5754 | } |
5755 | ||
3ec706c8 | 5756 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5757 | u64 logical, u64 *length, |
a1d3c478 | 5758 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5759 | { |
3ec706c8 | 5760 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
8e5cfb55 | 5761 | mirror_num, 0); |
f2d8d74d CM |
5762 | } |
5763 | ||
af8e2d1d MX |
5764 | /* For Scrub/replace */ |
5765 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw, | |
5766 | u64 logical, u64 *length, | |
5767 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5768 | int need_raid_map) |
af8e2d1d MX |
5769 | { |
5770 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, | |
8e5cfb55 | 5771 | mirror_num, need_raid_map); |
af8e2d1d MX |
5772 | } |
5773 | ||
a512bbf8 YZ |
5774 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
5775 | u64 chunk_start, u64 physical, u64 devid, | |
5776 | u64 **logical, int *naddrs, int *stripe_len) | |
5777 | { | |
5778 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5779 | struct extent_map *em; | |
5780 | struct map_lookup *map; | |
5781 | u64 *buf; | |
5782 | u64 bytenr; | |
5783 | u64 length; | |
5784 | u64 stripe_nr; | |
53b381b3 | 5785 | u64 rmap_len; |
a512bbf8 YZ |
5786 | int i, j, nr = 0; |
5787 | ||
890871be | 5788 | read_lock(&em_tree->lock); |
a512bbf8 | 5789 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5790 | read_unlock(&em_tree->lock); |
a512bbf8 | 5791 | |
835d974f | 5792 | if (!em) { |
efe120a0 | 5793 | printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n", |
835d974f JB |
5794 | chunk_start); |
5795 | return -EIO; | |
5796 | } | |
5797 | ||
5798 | if (em->start != chunk_start) { | |
efe120a0 | 5799 | printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n", |
835d974f JB |
5800 | em->start, chunk_start); |
5801 | free_extent_map(em); | |
5802 | return -EIO; | |
5803 | } | |
95617d69 | 5804 | map = em->map_lookup; |
a512bbf8 YZ |
5805 | |
5806 | length = em->len; | |
53b381b3 DW |
5807 | rmap_len = map->stripe_len; |
5808 | ||
a512bbf8 | 5809 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5810 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5811 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5812 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5813 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5814 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5815 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5816 | } | |
a512bbf8 | 5817 | |
31e818fe | 5818 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5819 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5820 | |
5821 | for (i = 0; i < map->num_stripes; i++) { | |
5822 | if (devid && map->stripes[i].dev->devid != devid) | |
5823 | continue; | |
5824 | if (map->stripes[i].physical > physical || | |
5825 | map->stripes[i].physical + length <= physical) | |
5826 | continue; | |
5827 | ||
5828 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 5829 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5830 | |
5831 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5832 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5833 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5834 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5835 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5836 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5837 | * Alternatively, just use rmap_len below instead of | |
5838 | * map->stripe_len */ | |
5839 | ||
5840 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5841 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5842 | for (j = 0; j < nr; j++) { |
5843 | if (buf[j] == bytenr) | |
5844 | break; | |
5845 | } | |
934d375b CM |
5846 | if (j == nr) { |
5847 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5848 | buf[nr++] = bytenr; |
934d375b | 5849 | } |
a512bbf8 YZ |
5850 | } |
5851 | ||
a512bbf8 YZ |
5852 | *logical = buf; |
5853 | *naddrs = nr; | |
53b381b3 | 5854 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5855 | |
5856 | free_extent_map(em); | |
5857 | return 0; | |
f2d8d74d CM |
5858 | } |
5859 | ||
4246a0b6 | 5860 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 5861 | { |
326e1dbb MS |
5862 | bio->bi_private = bbio->private; |
5863 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 5864 | bio_endio(bio); |
326e1dbb | 5865 | |
6e9606d2 | 5866 | btrfs_put_bbio(bbio); |
8408c716 MX |
5867 | } |
5868 | ||
4246a0b6 | 5869 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 5870 | { |
9be3395b | 5871 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 5872 | int is_orig_bio = 0; |
8790d502 | 5873 | |
4246a0b6 | 5874 | if (bio->bi_error) { |
a1d3c478 | 5875 | atomic_inc(&bbio->error); |
4246a0b6 | 5876 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { |
442a4f63 | 5877 | unsigned int stripe_index = |
9be3395b | 5878 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 5879 | struct btrfs_device *dev; |
442a4f63 SB |
5880 | |
5881 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5882 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5883 | if (dev->bdev) { |
5884 | if (bio->bi_rw & WRITE) | |
5885 | btrfs_dev_stat_inc(dev, | |
5886 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5887 | else | |
5888 | btrfs_dev_stat_inc(dev, | |
5889 | BTRFS_DEV_STAT_READ_ERRS); | |
5890 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5891 | btrfs_dev_stat_inc(dev, | |
5892 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5893 | btrfs_dev_stat_print_on_error(dev); | |
5894 | } | |
442a4f63 SB |
5895 | } |
5896 | } | |
8790d502 | 5897 | |
a1d3c478 | 5898 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5899 | is_orig_bio = 1; |
5900 | ||
c404e0dc MX |
5901 | btrfs_bio_counter_dec(bbio->fs_info); |
5902 | ||
a1d3c478 | 5903 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5904 | if (!is_orig_bio) { |
5905 | bio_put(bio); | |
a1d3c478 | 5906 | bio = bbio->orig_bio; |
7d2b4daa | 5907 | } |
c7b22bb1 | 5908 | |
9be3395b | 5909 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 5910 | /* only send an error to the higher layers if it is |
53b381b3 | 5911 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5912 | */ |
a1d3c478 | 5913 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4246a0b6 | 5914 | bio->bi_error = -EIO; |
5dbc8fca | 5915 | } else { |
1259ab75 CM |
5916 | /* |
5917 | * this bio is actually up to date, we didn't | |
5918 | * go over the max number of errors | |
5919 | */ | |
4246a0b6 | 5920 | bio->bi_error = 0; |
1259ab75 | 5921 | } |
c55f1396 | 5922 | |
4246a0b6 | 5923 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 5924 | } else if (!is_orig_bio) { |
8790d502 CM |
5925 | bio_put(bio); |
5926 | } | |
8790d502 CM |
5927 | } |
5928 | ||
8b712842 CM |
5929 | /* |
5930 | * see run_scheduled_bios for a description of why bios are collected for | |
5931 | * async submit. | |
5932 | * | |
5933 | * This will add one bio to the pending list for a device and make sure | |
5934 | * the work struct is scheduled. | |
5935 | */ | |
48a3b636 ES |
5936 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
5937 | struct btrfs_device *device, | |
5938 | int rw, struct bio *bio) | |
8b712842 CM |
5939 | { |
5940 | int should_queue = 1; | |
ffbd517d | 5941 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5942 | |
53b381b3 | 5943 | if (device->missing || !device->bdev) { |
4246a0b6 | 5944 | bio_io_error(bio); |
53b381b3 DW |
5945 | return; |
5946 | } | |
5947 | ||
8b712842 | 5948 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5949 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5950 | bio_get(bio); |
21adbd5c | 5951 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5952 | bio_put(bio); |
143bede5 | 5953 | return; |
8b712842 CM |
5954 | } |
5955 | ||
5956 | /* | |
0986fe9e | 5957 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5958 | * higher layers. Otherwise, the async bio makes it appear we have |
5959 | * made progress against dirty pages when we've really just put it | |
5960 | * on a queue for later | |
5961 | */ | |
0986fe9e | 5962 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5963 | WARN_ON(bio->bi_next); |
8b712842 CM |
5964 | bio->bi_next = NULL; |
5965 | bio->bi_rw |= rw; | |
5966 | ||
5967 | spin_lock(&device->io_lock); | |
7b6d91da | 5968 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5969 | pending_bios = &device->pending_sync_bios; |
5970 | else | |
5971 | pending_bios = &device->pending_bios; | |
8b712842 | 5972 | |
ffbd517d CM |
5973 | if (pending_bios->tail) |
5974 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5975 | |
ffbd517d CM |
5976 | pending_bios->tail = bio; |
5977 | if (!pending_bios->head) | |
5978 | pending_bios->head = bio; | |
8b712842 CM |
5979 | if (device->running_pending) |
5980 | should_queue = 0; | |
5981 | ||
5982 | spin_unlock(&device->io_lock); | |
5983 | ||
5984 | if (should_queue) | |
a8c93d4e QW |
5985 | btrfs_queue_work(root->fs_info->submit_workers, |
5986 | &device->work); | |
8b712842 CM |
5987 | } |
5988 | ||
de1ee92a JB |
5989 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, |
5990 | struct bio *bio, u64 physical, int dev_nr, | |
5991 | int rw, int async) | |
5992 | { | |
5993 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5994 | ||
5995 | bio->bi_private = bbio; | |
9be3395b | 5996 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 5997 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 5998 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
5999 | #ifdef DEBUG |
6000 | { | |
6001 | struct rcu_string *name; | |
6002 | ||
6003 | rcu_read_lock(); | |
6004 | name = rcu_dereference(dev->name); | |
d1423248 | 6005 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a | 6006 | "(%s id %llu), size=%u\n", rw, |
1b6e4469 FF |
6007 | (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev, |
6008 | name->str, dev->devid, bio->bi_iter.bi_size); | |
de1ee92a JB |
6009 | rcu_read_unlock(); |
6010 | } | |
6011 | #endif | |
6012 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
6013 | |
6014 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
6015 | ||
de1ee92a | 6016 | if (async) |
53b381b3 | 6017 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
6018 | else |
6019 | btrfsic_submit_bio(rw, bio); | |
6020 | } | |
6021 | ||
de1ee92a JB |
6022 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6023 | { | |
6024 | atomic_inc(&bbio->error); | |
6025 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
8408c716 MX |
6026 | /* Shoud be the original bio. */ |
6027 | WARN_ON(bio != bbio->orig_bio); | |
6028 | ||
9be3395b | 6029 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6030 | bio->bi_iter.bi_sector = logical >> 9; |
4246a0b6 CH |
6031 | bio->bi_error = -EIO; |
6032 | btrfs_end_bbio(bbio, bio); | |
de1ee92a JB |
6033 | } |
6034 | } | |
6035 | ||
f188591e | 6036 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 6037 | int mirror_num, int async_submit) |
0b86a832 | 6038 | { |
0b86a832 | 6039 | struct btrfs_device *dev; |
8790d502 | 6040 | struct bio *first_bio = bio; |
4f024f37 | 6041 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6042 | u64 length = 0; |
6043 | u64 map_length; | |
0b86a832 | 6044 | int ret; |
08da757d ZL |
6045 | int dev_nr; |
6046 | int total_devs; | |
a1d3c478 | 6047 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6048 | |
4f024f37 | 6049 | length = bio->bi_iter.bi_size; |
0b86a832 | 6050 | map_length = length; |
cea9e445 | 6051 | |
c404e0dc | 6052 | btrfs_bio_counter_inc_blocked(root->fs_info); |
53b381b3 | 6053 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
8e5cfb55 | 6054 | mirror_num, 1); |
c404e0dc MX |
6055 | if (ret) { |
6056 | btrfs_bio_counter_dec(root->fs_info); | |
79787eaa | 6057 | return ret; |
c404e0dc | 6058 | } |
cea9e445 | 6059 | |
a1d3c478 | 6060 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6061 | bbio->orig_bio = first_bio; |
6062 | bbio->private = first_bio->bi_private; | |
6063 | bbio->end_io = first_bio->bi_end_io; | |
c404e0dc | 6064 | bbio->fs_info = root->fs_info; |
53b381b3 DW |
6065 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6066 | ||
ad1ba2a0 ZL |
6067 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
6068 | ((rw & WRITE) || (mirror_num > 1))) { | |
53b381b3 DW |
6069 | /* In this case, map_length has been set to the length of |
6070 | a single stripe; not the whole write */ | |
6071 | if (rw & WRITE) { | |
8e5cfb55 | 6072 | ret = raid56_parity_write(root, bio, bbio, map_length); |
53b381b3 | 6073 | } else { |
8e5cfb55 | 6074 | ret = raid56_parity_recover(root, bio, bbio, map_length, |
4245215d | 6075 | mirror_num, 1); |
53b381b3 | 6076 | } |
4245215d | 6077 | |
c404e0dc MX |
6078 | btrfs_bio_counter_dec(root->fs_info); |
6079 | return ret; | |
53b381b3 DW |
6080 | } |
6081 | ||
cea9e445 | 6082 | if (map_length < length) { |
c2cf52eb | 6083 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", |
c1c9ff7c | 6084 | logical, length, map_length); |
cea9e445 CM |
6085 | BUG(); |
6086 | } | |
a1d3c478 | 6087 | |
08da757d | 6088 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a JB |
6089 | dev = bbio->stripes[dev_nr].dev; |
6090 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
6091 | bbio_error(bbio, first_bio, logical); | |
de1ee92a JB |
6092 | continue; |
6093 | } | |
6094 | ||
a1d3c478 | 6095 | if (dev_nr < total_devs - 1) { |
9be3395b | 6096 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 6097 | BUG_ON(!bio); /* -ENOMEM */ |
326e1dbb | 6098 | } else |
a1d3c478 | 6099 | bio = first_bio; |
de1ee92a JB |
6100 | |
6101 | submit_stripe_bio(root, bbio, bio, | |
6102 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
6103 | async_submit); | |
8790d502 | 6104 | } |
c404e0dc | 6105 | btrfs_bio_counter_dec(root->fs_info); |
0b86a832 CM |
6106 | return 0; |
6107 | } | |
6108 | ||
aa1b8cd4 | 6109 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6110 | u8 *uuid, u8 *fsid) |
0b86a832 | 6111 | { |
2b82032c YZ |
6112 | struct btrfs_device *device; |
6113 | struct btrfs_fs_devices *cur_devices; | |
6114 | ||
aa1b8cd4 | 6115 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6116 | while (cur_devices) { |
6117 | if (!fsid || | |
6118 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6119 | device = __find_device(&cur_devices->devices, | |
6120 | devid, uuid); | |
6121 | if (device) | |
6122 | return device; | |
6123 | } | |
6124 | cur_devices = cur_devices->seed; | |
6125 | } | |
6126 | return NULL; | |
0b86a832 CM |
6127 | } |
6128 | ||
dfe25020 | 6129 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 6130 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6131 | u64 devid, u8 *dev_uuid) |
6132 | { | |
6133 | struct btrfs_device *device; | |
dfe25020 | 6134 | |
12bd2fc0 ID |
6135 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6136 | if (IS_ERR(device)) | |
7cbd8a83 | 6137 | return NULL; |
12bd2fc0 ID |
6138 | |
6139 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6140 | device->fs_devices = fs_devices; |
dfe25020 | 6141 | fs_devices->num_devices++; |
12bd2fc0 ID |
6142 | |
6143 | device->missing = 1; | |
cd02dca5 | 6144 | fs_devices->missing_devices++; |
12bd2fc0 | 6145 | |
dfe25020 CM |
6146 | return device; |
6147 | } | |
6148 | ||
12bd2fc0 ID |
6149 | /** |
6150 | * btrfs_alloc_device - allocate struct btrfs_device | |
6151 | * @fs_info: used only for generating a new devid, can be NULL if | |
6152 | * devid is provided (i.e. @devid != NULL). | |
6153 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6154 | * is generated. | |
6155 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6156 | * is generated. | |
6157 | * | |
6158 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6159 | * on error. Returned struct is not linked onto any lists and can be | |
6160 | * destroyed with kfree() right away. | |
6161 | */ | |
6162 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6163 | const u64 *devid, | |
6164 | const u8 *uuid) | |
6165 | { | |
6166 | struct btrfs_device *dev; | |
6167 | u64 tmp; | |
6168 | ||
fae7f21c | 6169 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6170 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6171 | |
6172 | dev = __alloc_device(); | |
6173 | if (IS_ERR(dev)) | |
6174 | return dev; | |
6175 | ||
6176 | if (devid) | |
6177 | tmp = *devid; | |
6178 | else { | |
6179 | int ret; | |
6180 | ||
6181 | ret = find_next_devid(fs_info, &tmp); | |
6182 | if (ret) { | |
6183 | kfree(dev); | |
6184 | return ERR_PTR(ret); | |
6185 | } | |
6186 | } | |
6187 | dev->devid = tmp; | |
6188 | ||
6189 | if (uuid) | |
6190 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6191 | else | |
6192 | generate_random_uuid(dev->uuid); | |
6193 | ||
9e0af237 LB |
6194 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6195 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6196 | |
6197 | return dev; | |
6198 | } | |
6199 | ||
0b86a832 CM |
6200 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
6201 | struct extent_buffer *leaf, | |
6202 | struct btrfs_chunk *chunk) | |
6203 | { | |
6204 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6205 | struct map_lookup *map; | |
6206 | struct extent_map *em; | |
6207 | u64 logical; | |
6208 | u64 length; | |
f04b772b | 6209 | u64 stripe_len; |
0b86a832 | 6210 | u64 devid; |
a443755f | 6211 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 6212 | int num_stripes; |
0b86a832 | 6213 | int ret; |
593060d7 | 6214 | int i; |
0b86a832 | 6215 | |
e17cade2 CM |
6216 | logical = key->offset; |
6217 | length = btrfs_chunk_length(leaf, chunk); | |
f04b772b QW |
6218 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6219 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
6220 | /* Validation check */ | |
6221 | if (!num_stripes) { | |
6222 | btrfs_err(root->fs_info, "invalid chunk num_stripes: %u", | |
6223 | num_stripes); | |
6224 | return -EIO; | |
6225 | } | |
6226 | if (!IS_ALIGNED(logical, root->sectorsize)) { | |
6227 | btrfs_err(root->fs_info, | |
6228 | "invalid chunk logical %llu", logical); | |
6229 | return -EIO; | |
6230 | } | |
6231 | if (!length || !IS_ALIGNED(length, root->sectorsize)) { | |
6232 | btrfs_err(root->fs_info, | |
6233 | "invalid chunk length %llu", length); | |
6234 | return -EIO; | |
6235 | } | |
6236 | if (!is_power_of_2(stripe_len)) { | |
6237 | btrfs_err(root->fs_info, "invalid chunk stripe length: %llu", | |
6238 | stripe_len); | |
6239 | return -EIO; | |
6240 | } | |
6241 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6242 | btrfs_chunk_type(leaf, chunk)) { | |
6243 | btrfs_err(root->fs_info, "unrecognized chunk type: %llu", | |
6244 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | | |
6245 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6246 | btrfs_chunk_type(leaf, chunk)); | |
6247 | return -EIO; | |
6248 | } | |
a061fc8d | 6249 | |
890871be | 6250 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6251 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6252 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6253 | |
6254 | /* already mapped? */ | |
6255 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6256 | free_extent_map(em); | |
0b86a832 CM |
6257 | return 0; |
6258 | } else if (em) { | |
6259 | free_extent_map(em); | |
6260 | } | |
0b86a832 | 6261 | |
172ddd60 | 6262 | em = alloc_extent_map(); |
0b86a832 CM |
6263 | if (!em) |
6264 | return -ENOMEM; | |
593060d7 | 6265 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6266 | if (!map) { |
6267 | free_extent_map(em); | |
6268 | return -ENOMEM; | |
6269 | } | |
6270 | ||
298a8f9c | 6271 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6272 | em->map_lookup = map; |
0b86a832 CM |
6273 | em->start = logical; |
6274 | em->len = length; | |
70c8a91c | 6275 | em->orig_start = 0; |
0b86a832 | 6276 | em->block_start = 0; |
c8b97818 | 6277 | em->block_len = em->len; |
0b86a832 | 6278 | |
593060d7 CM |
6279 | map->num_stripes = num_stripes; |
6280 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6281 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6282 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6283 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6284 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6285 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6286 | for (i = 0; i < num_stripes; i++) { |
6287 | map->stripes[i].physical = | |
6288 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6289 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6290 | read_extent_buffer(leaf, uuid, (unsigned long) |
6291 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6292 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
6293 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
6294 | uuid, NULL); | |
dfe25020 | 6295 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
6296 | free_extent_map(em); |
6297 | return -EIO; | |
6298 | } | |
dfe25020 CM |
6299 | if (!map->stripes[i].dev) { |
6300 | map->stripes[i].dev = | |
5f375835 MX |
6301 | add_missing_dev(root, root->fs_info->fs_devices, |
6302 | devid, uuid); | |
dfe25020 | 6303 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6304 | free_extent_map(em); |
6305 | return -EIO; | |
6306 | } | |
816fcebe AJ |
6307 | btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing", |
6308 | devid, uuid); | |
dfe25020 CM |
6309 | } |
6310 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6311 | } |
6312 | ||
890871be | 6313 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6314 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6315 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6316 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6317 | free_extent_map(em); |
6318 | ||
6319 | return 0; | |
6320 | } | |
6321 | ||
143bede5 | 6322 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6323 | struct btrfs_dev_item *dev_item, |
6324 | struct btrfs_device *device) | |
6325 | { | |
6326 | unsigned long ptr; | |
0b86a832 CM |
6327 | |
6328 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6329 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6330 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6331 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6332 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6333 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6334 | device->type = btrfs_device_type(leaf, dev_item); |
6335 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6336 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6337 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6338 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6339 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6340 | |
410ba3a2 | 6341 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6342 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6343 | } |
6344 | ||
5f375835 MX |
6345 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
6346 | u8 *fsid) | |
2b82032c YZ |
6347 | { |
6348 | struct btrfs_fs_devices *fs_devices; | |
6349 | int ret; | |
6350 | ||
b367e47f | 6351 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
6352 | |
6353 | fs_devices = root->fs_info->fs_devices->seed; | |
6354 | while (fs_devices) { | |
5f375835 MX |
6355 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6356 | return fs_devices; | |
6357 | ||
2b82032c YZ |
6358 | fs_devices = fs_devices->seed; |
6359 | } | |
6360 | ||
6361 | fs_devices = find_fsid(fsid); | |
6362 | if (!fs_devices) { | |
5f375835 MX |
6363 | if (!btrfs_test_opt(root, DEGRADED)) |
6364 | return ERR_PTR(-ENOENT); | |
6365 | ||
6366 | fs_devices = alloc_fs_devices(fsid); | |
6367 | if (IS_ERR(fs_devices)) | |
6368 | return fs_devices; | |
6369 | ||
6370 | fs_devices->seeding = 1; | |
6371 | fs_devices->opened = 1; | |
6372 | return fs_devices; | |
2b82032c | 6373 | } |
e4404d6e YZ |
6374 | |
6375 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6376 | if (IS_ERR(fs_devices)) |
6377 | return fs_devices; | |
2b82032c | 6378 | |
97288f2c | 6379 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 6380 | root->fs_info->bdev_holder); |
48d28232 JL |
6381 | if (ret) { |
6382 | free_fs_devices(fs_devices); | |
5f375835 | 6383 | fs_devices = ERR_PTR(ret); |
2b82032c | 6384 | goto out; |
48d28232 | 6385 | } |
2b82032c YZ |
6386 | |
6387 | if (!fs_devices->seeding) { | |
6388 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6389 | free_fs_devices(fs_devices); |
5f375835 | 6390 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6391 | goto out; |
6392 | } | |
6393 | ||
6394 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6395 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6396 | out: |
5f375835 | 6397 | return fs_devices; |
2b82032c YZ |
6398 | } |
6399 | ||
0d81ba5d | 6400 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6401 | struct extent_buffer *leaf, |
6402 | struct btrfs_dev_item *dev_item) | |
6403 | { | |
5f375835 | 6404 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
0b86a832 CM |
6405 | struct btrfs_device *device; |
6406 | u64 devid; | |
6407 | int ret; | |
2b82032c | 6408 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6409 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6410 | ||
0b86a832 | 6411 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6412 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6413 | BTRFS_UUID_SIZE); |
1473b24e | 6414 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6415 | BTRFS_UUID_SIZE); |
6416 | ||
6417 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5f375835 MX |
6418 | fs_devices = open_seed_devices(root, fs_uuid); |
6419 | if (IS_ERR(fs_devices)) | |
6420 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6421 | } |
6422 | ||
aa1b8cd4 | 6423 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6424 | if (!device) { |
e4404d6e | 6425 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
6426 | return -EIO; |
6427 | ||
5f375835 MX |
6428 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); |
6429 | if (!device) | |
6430 | return -ENOMEM; | |
33b97e43 AJ |
6431 | btrfs_warn(root->fs_info, "devid %llu uuid %pU missing", |
6432 | devid, dev_uuid); | |
5f375835 MX |
6433 | } else { |
6434 | if (!device->bdev && !btrfs_test_opt(root, DEGRADED)) | |
6435 | return -EIO; | |
6436 | ||
6437 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6438 | /* |
6439 | * this happens when a device that was properly setup | |
6440 | * in the device info lists suddenly goes bad. | |
6441 | * device->bdev is NULL, and so we have to set | |
6442 | * device->missing to one here | |
6443 | */ | |
5f375835 | 6444 | device->fs_devices->missing_devices++; |
cd02dca5 | 6445 | device->missing = 1; |
2b82032c | 6446 | } |
5f375835 MX |
6447 | |
6448 | /* Move the device to its own fs_devices */ | |
6449 | if (device->fs_devices != fs_devices) { | |
6450 | ASSERT(device->missing); | |
6451 | ||
6452 | list_move(&device->dev_list, &fs_devices->devices); | |
6453 | device->fs_devices->num_devices--; | |
6454 | fs_devices->num_devices++; | |
6455 | ||
6456 | device->fs_devices->missing_devices--; | |
6457 | fs_devices->missing_devices++; | |
6458 | ||
6459 | device->fs_devices = fs_devices; | |
6460 | } | |
2b82032c YZ |
6461 | } |
6462 | ||
6463 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6464 | BUG_ON(device->writeable); | |
6465 | if (device->generation != | |
6466 | btrfs_device_generation(leaf, dev_item)) | |
6467 | return -EINVAL; | |
6324fbf3 | 6468 | } |
0b86a832 CM |
6469 | |
6470 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6471 | device->in_fs_metadata = 1; |
63a212ab | 6472 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6473 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
6474 | spin_lock(&root->fs_info->free_chunk_lock); |
6475 | root->fs_info->free_chunk_space += device->total_bytes - | |
6476 | device->bytes_used; | |
6477 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6478 | } | |
0b86a832 | 6479 | ret = 0; |
0b86a832 CM |
6480 | return ret; |
6481 | } | |
6482 | ||
e4404d6e | 6483 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 6484 | { |
6c41761f | 6485 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 6486 | struct extent_buffer *sb; |
0b86a832 | 6487 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6488 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6489 | u8 *array_ptr; |
6490 | unsigned long sb_array_offset; | |
84eed90f | 6491 | int ret = 0; |
0b86a832 CM |
6492 | u32 num_stripes; |
6493 | u32 array_size; | |
6494 | u32 len = 0; | |
1ffb22cf | 6495 | u32 cur_offset; |
84eed90f | 6496 | struct btrfs_key key; |
0b86a832 | 6497 | |
a83fffb7 DS |
6498 | ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize); |
6499 | /* | |
6500 | * This will create extent buffer of nodesize, superblock size is | |
6501 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6502 | * overallocate but we can keep it as-is, only the first page is used. | |
6503 | */ | |
6504 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
a061fc8d CM |
6505 | if (!sb) |
6506 | return -ENOMEM; | |
4db8c528 | 6507 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6508 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
6509 | /* |
6510 | * The sb extent buffer is artifical and just used to read the system array. | |
4db8c528 | 6511 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6512 | * pages up-to-date when the page is larger: extent does not cover the |
6513 | * whole page and consequently check_page_uptodate does not find all | |
6514 | * the page's extents up-to-date (the hole beyond sb), | |
6515 | * write_extent_buffer then triggers a WARN_ON. | |
6516 | * | |
6517 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6518 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6519 | * to silence the warning eg. on PowerPC 64. | |
6520 | */ | |
09cbfeaf | 6521 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6522 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6523 | |
a061fc8d | 6524 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6525 | array_size = btrfs_super_sys_array_size(super_copy); |
6526 | ||
1ffb22cf DS |
6527 | array_ptr = super_copy->sys_chunk_array; |
6528 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6529 | cur_offset = 0; | |
0b86a832 | 6530 | |
1ffb22cf DS |
6531 | while (cur_offset < array_size) { |
6532 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6533 | len = sizeof(*disk_key); |
6534 | if (cur_offset + len > array_size) | |
6535 | goto out_short_read; | |
6536 | ||
0b86a832 CM |
6537 | btrfs_disk_key_to_cpu(&key, disk_key); |
6538 | ||
1ffb22cf DS |
6539 | array_ptr += len; |
6540 | sb_array_offset += len; | |
6541 | cur_offset += len; | |
0b86a832 | 6542 | |
0d81ba5d | 6543 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6544 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6545 | /* |
6546 | * At least one btrfs_chunk with one stripe must be | |
6547 | * present, exact stripe count check comes afterwards | |
6548 | */ | |
6549 | len = btrfs_chunk_item_size(1); | |
6550 | if (cur_offset + len > array_size) | |
6551 | goto out_short_read; | |
6552 | ||
6553 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 DS |
6554 | if (!num_stripes) { |
6555 | printk(KERN_ERR | |
6556 | "BTRFS: invalid number of stripes %u in sys_array at offset %u\n", | |
6557 | num_stripes, cur_offset); | |
6558 | ret = -EIO; | |
6559 | break; | |
6560 | } | |
6561 | ||
e3540eab DS |
6562 | len = btrfs_chunk_item_size(num_stripes); |
6563 | if (cur_offset + len > array_size) | |
6564 | goto out_short_read; | |
6565 | ||
0d81ba5d | 6566 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6567 | if (ret) |
6568 | break; | |
0b86a832 | 6569 | } else { |
93a3d467 DS |
6570 | printk(KERN_ERR |
6571 | "BTRFS: unexpected item type %u in sys_array at offset %u\n", | |
6572 | (u32)key.type, cur_offset); | |
84eed90f CM |
6573 | ret = -EIO; |
6574 | break; | |
0b86a832 | 6575 | } |
1ffb22cf DS |
6576 | array_ptr += len; |
6577 | sb_array_offset += len; | |
6578 | cur_offset += len; | |
0b86a832 | 6579 | } |
a061fc8d | 6580 | free_extent_buffer(sb); |
84eed90f | 6581 | return ret; |
e3540eab DS |
6582 | |
6583 | out_short_read: | |
6584 | printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n", | |
6585 | len, cur_offset); | |
6586 | free_extent_buffer(sb); | |
6587 | return -EIO; | |
0b86a832 CM |
6588 | } |
6589 | ||
6590 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6591 | { | |
6592 | struct btrfs_path *path; | |
6593 | struct extent_buffer *leaf; | |
6594 | struct btrfs_key key; | |
6595 | struct btrfs_key found_key; | |
6596 | int ret; | |
6597 | int slot; | |
6598 | ||
6599 | root = root->fs_info->chunk_root; | |
6600 | ||
6601 | path = btrfs_alloc_path(); | |
6602 | if (!path) | |
6603 | return -ENOMEM; | |
6604 | ||
b367e47f LZ |
6605 | mutex_lock(&uuid_mutex); |
6606 | lock_chunks(root); | |
6607 | ||
395927a9 FDBM |
6608 | /* |
6609 | * Read all device items, and then all the chunk items. All | |
6610 | * device items are found before any chunk item (their object id | |
6611 | * is smaller than the lowest possible object id for a chunk | |
6612 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6613 | */ |
6614 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6615 | key.offset = 0; | |
6616 | key.type = 0; | |
0b86a832 | 6617 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6618 | if (ret < 0) |
6619 | goto error; | |
d397712b | 6620 | while (1) { |
0b86a832 CM |
6621 | leaf = path->nodes[0]; |
6622 | slot = path->slots[0]; | |
6623 | if (slot >= btrfs_header_nritems(leaf)) { | |
6624 | ret = btrfs_next_leaf(root, path); | |
6625 | if (ret == 0) | |
6626 | continue; | |
6627 | if (ret < 0) | |
6628 | goto error; | |
6629 | break; | |
6630 | } | |
6631 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6632 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6633 | struct btrfs_dev_item *dev_item; | |
6634 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6635 | struct btrfs_dev_item); |
395927a9 FDBM |
6636 | ret = read_one_dev(root, leaf, dev_item); |
6637 | if (ret) | |
6638 | goto error; | |
0b86a832 CM |
6639 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6640 | struct btrfs_chunk *chunk; | |
6641 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6642 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6643 | if (ret) |
6644 | goto error; | |
0b86a832 CM |
6645 | } |
6646 | path->slots[0]++; | |
6647 | } | |
0b86a832 CM |
6648 | ret = 0; |
6649 | error: | |
b367e47f LZ |
6650 | unlock_chunks(root); |
6651 | mutex_unlock(&uuid_mutex); | |
6652 | ||
2b82032c | 6653 | btrfs_free_path(path); |
0b86a832 CM |
6654 | return ret; |
6655 | } | |
442a4f63 | 6656 | |
cb517eab MX |
6657 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6658 | { | |
6659 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6660 | struct btrfs_device *device; | |
6661 | ||
29cc83f6 LB |
6662 | while (fs_devices) { |
6663 | mutex_lock(&fs_devices->device_list_mutex); | |
6664 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6665 | device->dev_root = fs_info->dev_root; | |
6666 | mutex_unlock(&fs_devices->device_list_mutex); | |
6667 | ||
6668 | fs_devices = fs_devices->seed; | |
6669 | } | |
cb517eab MX |
6670 | } |
6671 | ||
733f4fbb SB |
6672 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6673 | { | |
6674 | int i; | |
6675 | ||
6676 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6677 | btrfs_dev_stat_reset(dev, i); | |
6678 | } | |
6679 | ||
6680 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6681 | { | |
6682 | struct btrfs_key key; | |
6683 | struct btrfs_key found_key; | |
6684 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6685 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6686 | struct extent_buffer *eb; | |
6687 | int slot; | |
6688 | int ret = 0; | |
6689 | struct btrfs_device *device; | |
6690 | struct btrfs_path *path = NULL; | |
6691 | int i; | |
6692 | ||
6693 | path = btrfs_alloc_path(); | |
6694 | if (!path) { | |
6695 | ret = -ENOMEM; | |
6696 | goto out; | |
6697 | } | |
6698 | ||
6699 | mutex_lock(&fs_devices->device_list_mutex); | |
6700 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6701 | int item_size; | |
6702 | struct btrfs_dev_stats_item *ptr; | |
6703 | ||
242e2956 DS |
6704 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6705 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6706 | key.offset = device->devid; |
6707 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6708 | if (ret) { | |
733f4fbb SB |
6709 | __btrfs_reset_dev_stats(device); |
6710 | device->dev_stats_valid = 1; | |
6711 | btrfs_release_path(path); | |
6712 | continue; | |
6713 | } | |
6714 | slot = path->slots[0]; | |
6715 | eb = path->nodes[0]; | |
6716 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6717 | item_size = btrfs_item_size_nr(eb, slot); | |
6718 | ||
6719 | ptr = btrfs_item_ptr(eb, slot, | |
6720 | struct btrfs_dev_stats_item); | |
6721 | ||
6722 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6723 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6724 | btrfs_dev_stat_set(device, i, | |
6725 | btrfs_dev_stats_value(eb, ptr, i)); | |
6726 | else | |
6727 | btrfs_dev_stat_reset(device, i); | |
6728 | } | |
6729 | ||
6730 | device->dev_stats_valid = 1; | |
6731 | btrfs_dev_stat_print_on_load(device); | |
6732 | btrfs_release_path(path); | |
6733 | } | |
6734 | mutex_unlock(&fs_devices->device_list_mutex); | |
6735 | ||
6736 | out: | |
6737 | btrfs_free_path(path); | |
6738 | return ret < 0 ? ret : 0; | |
6739 | } | |
6740 | ||
6741 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6742 | struct btrfs_root *dev_root, | |
6743 | struct btrfs_device *device) | |
6744 | { | |
6745 | struct btrfs_path *path; | |
6746 | struct btrfs_key key; | |
6747 | struct extent_buffer *eb; | |
6748 | struct btrfs_dev_stats_item *ptr; | |
6749 | int ret; | |
6750 | int i; | |
6751 | ||
242e2956 DS |
6752 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6753 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6754 | key.offset = device->devid; |
6755 | ||
6756 | path = btrfs_alloc_path(); | |
6757 | BUG_ON(!path); | |
6758 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6759 | if (ret < 0) { | |
ecaeb14b DS |
6760 | btrfs_warn_in_rcu(dev_root->fs_info, |
6761 | "error %d while searching for dev_stats item for device %s", | |
606686ee | 6762 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6763 | goto out; |
6764 | } | |
6765 | ||
6766 | if (ret == 0 && | |
6767 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6768 | /* need to delete old one and insert a new one */ | |
6769 | ret = btrfs_del_item(trans, dev_root, path); | |
6770 | if (ret != 0) { | |
ecaeb14b DS |
6771 | btrfs_warn_in_rcu(dev_root->fs_info, |
6772 | "delete too small dev_stats item for device %s failed %d", | |
606686ee | 6773 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6774 | goto out; |
6775 | } | |
6776 | ret = 1; | |
6777 | } | |
6778 | ||
6779 | if (ret == 1) { | |
6780 | /* need to insert a new item */ | |
6781 | btrfs_release_path(path); | |
6782 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6783 | &key, sizeof(*ptr)); | |
6784 | if (ret < 0) { | |
ecaeb14b DS |
6785 | btrfs_warn_in_rcu(dev_root->fs_info, |
6786 | "insert dev_stats item for device %s failed %d", | |
6787 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
6788 | goto out; |
6789 | } | |
6790 | } | |
6791 | ||
6792 | eb = path->nodes[0]; | |
6793 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6794 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6795 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6796 | btrfs_dev_stat_read(device, i)); | |
6797 | btrfs_mark_buffer_dirty(eb); | |
6798 | ||
6799 | out: | |
6800 | btrfs_free_path(path); | |
6801 | return ret; | |
6802 | } | |
6803 | ||
6804 | /* | |
6805 | * called from commit_transaction. Writes all changed device stats to disk. | |
6806 | */ | |
6807 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6808 | struct btrfs_fs_info *fs_info) | |
6809 | { | |
6810 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6811 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6812 | struct btrfs_device *device; | |
addc3fa7 | 6813 | int stats_cnt; |
733f4fbb SB |
6814 | int ret = 0; |
6815 | ||
6816 | mutex_lock(&fs_devices->device_list_mutex); | |
6817 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 6818 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
6819 | continue; |
6820 | ||
addc3fa7 | 6821 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
733f4fbb SB |
6822 | ret = update_dev_stat_item(trans, dev_root, device); |
6823 | if (!ret) | |
addc3fa7 | 6824 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
6825 | } |
6826 | mutex_unlock(&fs_devices->device_list_mutex); | |
6827 | ||
6828 | return ret; | |
6829 | } | |
6830 | ||
442a4f63 SB |
6831 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
6832 | { | |
6833 | btrfs_dev_stat_inc(dev, index); | |
6834 | btrfs_dev_stat_print_on_error(dev); | |
6835 | } | |
6836 | ||
48a3b636 | 6837 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 6838 | { |
733f4fbb SB |
6839 | if (!dev->dev_stats_valid) |
6840 | return; | |
b14af3b4 DS |
6841 | btrfs_err_rl_in_rcu(dev->dev_root->fs_info, |
6842 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
606686ee | 6843 | rcu_str_deref(dev->name), |
442a4f63 SB |
6844 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6845 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6846 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
6847 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
6848 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 6849 | } |
c11d2c23 | 6850 | |
733f4fbb SB |
6851 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
6852 | { | |
a98cdb85 SB |
6853 | int i; |
6854 | ||
6855 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6856 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6857 | break; | |
6858 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6859 | return; /* all values == 0, suppress message */ | |
6860 | ||
ecaeb14b DS |
6861 | btrfs_info_in_rcu(dev->dev_root->fs_info, |
6862 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
606686ee | 6863 | rcu_str_deref(dev->name), |
733f4fbb SB |
6864 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6865 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6866 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6867 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6868 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6869 | } | |
6870 | ||
c11d2c23 | 6871 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 6872 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
6873 | { |
6874 | struct btrfs_device *dev; | |
6875 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6876 | int i; | |
6877 | ||
6878 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 6879 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
6880 | mutex_unlock(&fs_devices->device_list_mutex); |
6881 | ||
6882 | if (!dev) { | |
efe120a0 | 6883 | btrfs_warn(root->fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 6884 | return -ENODEV; |
733f4fbb | 6885 | } else if (!dev->dev_stats_valid) { |
efe120a0 | 6886 | btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 6887 | return -ENODEV; |
b27f7c0c | 6888 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
6889 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
6890 | if (stats->nr_items > i) | |
6891 | stats->values[i] = | |
6892 | btrfs_dev_stat_read_and_reset(dev, i); | |
6893 | else | |
6894 | btrfs_dev_stat_reset(dev, i); | |
6895 | } | |
6896 | } else { | |
6897 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6898 | if (stats->nr_items > i) | |
6899 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6900 | } | |
6901 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6902 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6903 | return 0; | |
6904 | } | |
a8a6dab7 | 6905 | |
12b1c263 | 6906 | void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path) |
a8a6dab7 SB |
6907 | { |
6908 | struct buffer_head *bh; | |
6909 | struct btrfs_super_block *disk_super; | |
12b1c263 | 6910 | int copy_num; |
a8a6dab7 | 6911 | |
12b1c263 AJ |
6912 | if (!bdev) |
6913 | return; | |
a8a6dab7 | 6914 | |
12b1c263 AJ |
6915 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
6916 | copy_num++) { | |
a8a6dab7 | 6917 | |
12b1c263 AJ |
6918 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
6919 | continue; | |
6920 | ||
6921 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6922 | ||
6923 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6924 | set_buffer_dirty(bh); | |
6925 | sync_dirty_buffer(bh); | |
6926 | brelse(bh); | |
6927 | } | |
6928 | ||
6929 | /* Notify udev that device has changed */ | |
6930 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
6931 | ||
6932 | /* Update ctime/mtime for device path for libblkid */ | |
6933 | update_dev_time(device_path); | |
a8a6dab7 | 6934 | } |
935e5cc9 MX |
6935 | |
6936 | /* | |
6937 | * Update the size of all devices, which is used for writing out the | |
6938 | * super blocks. | |
6939 | */ | |
6940 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
6941 | { | |
6942 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6943 | struct btrfs_device *curr, *next; | |
6944 | ||
6945 | if (list_empty(&fs_devices->resized_devices)) | |
6946 | return; | |
6947 | ||
6948 | mutex_lock(&fs_devices->device_list_mutex); | |
6949 | lock_chunks(fs_info->dev_root); | |
6950 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
6951 | resized_list) { | |
6952 | list_del_init(&curr->resized_list); | |
6953 | curr->commit_total_bytes = curr->disk_total_bytes; | |
6954 | } | |
6955 | unlock_chunks(fs_info->dev_root); | |
6956 | mutex_unlock(&fs_devices->device_list_mutex); | |
6957 | } | |
ce7213c7 MX |
6958 | |
6959 | /* Must be invoked during the transaction commit */ | |
6960 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
6961 | struct btrfs_transaction *transaction) | |
6962 | { | |
6963 | struct extent_map *em; | |
6964 | struct map_lookup *map; | |
6965 | struct btrfs_device *dev; | |
6966 | int i; | |
6967 | ||
6968 | if (list_empty(&transaction->pending_chunks)) | |
6969 | return; | |
6970 | ||
6971 | /* In order to kick the device replace finish process */ | |
6972 | lock_chunks(root); | |
6973 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
95617d69 | 6974 | map = em->map_lookup; |
ce7213c7 MX |
6975 | |
6976 | for (i = 0; i < map->num_stripes; i++) { | |
6977 | dev = map->stripes[i].dev; | |
6978 | dev->commit_bytes_used = dev->bytes_used; | |
6979 | } | |
6980 | } | |
6981 | unlock_chunks(root); | |
6982 | } | |
5a13f430 AJ |
6983 | |
6984 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6985 | { | |
6986 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6987 | while (fs_devices) { | |
6988 | fs_devices->fs_info = fs_info; | |
6989 | fs_devices = fs_devices->seed; | |
6990 | } | |
6991 | } | |
6992 | ||
6993 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6994 | { | |
6995 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6996 | while (fs_devices) { | |
6997 | fs_devices->fs_info = NULL; | |
6998 | fs_devices = fs_devices->seed; | |
6999 | } | |
7000 | } | |
f190aa47 | 7001 | |
87ad58c5 | 7002 | static void btrfs_close_one_device(struct btrfs_device *device) |
f190aa47 AJ |
7003 | { |
7004 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
7005 | struct btrfs_device *new_device; | |
7006 | struct rcu_string *name; | |
7007 | ||
7008 | if (device->bdev) | |
7009 | fs_devices->open_devices--; | |
7010 | ||
7011 | if (device->writeable && | |
7012 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
7013 | list_del_init(&device->dev_alloc_list); | |
7014 | fs_devices->rw_devices--; | |
7015 | } | |
7016 | ||
7017 | if (device->missing) | |
7018 | fs_devices->missing_devices--; | |
7019 | ||
7020 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
7021 | device->uuid); | |
7022 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
7023 | ||
7024 | /* Safe because we are under uuid_mutex */ | |
7025 | if (device->name) { | |
7026 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
7027 | BUG_ON(!name); /* -ENOMEM */ | |
7028 | rcu_assign_pointer(new_device->name, name); | |
7029 | } | |
7030 | ||
7031 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
7032 | new_device->fs_devices = device->fs_devices; | |
7033 | ||
7034 | call_rcu(&device->rcu, free_device); | |
7035 | } |