]>
Commit | Line | Data |
---|---|---|
0b86a832 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8a4b83cc | 21 | #include <linux/buffer_head.h> |
f2d8d74d | 22 | #include <linux/blkdev.h> |
b765ead5 | 23 | #include <linux/iocontext.h> |
6f88a440 | 24 | #include <linux/capability.h> |
442a4f63 | 25 | #include <linux/ratelimit.h> |
59641015 | 26 | #include <linux/kthread.h> |
53b381b3 | 27 | #include <linux/raid/pq.h> |
803b2f54 | 28 | #include <linux/semaphore.h> |
8da4b8c4 | 29 | #include <linux/uuid.h> |
53b381b3 | 30 | #include <asm/div64.h> |
0b86a832 CM |
31 | #include "ctree.h" |
32 | #include "extent_map.h" | |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "print-tree.h" | |
36 | #include "volumes.h" | |
53b381b3 | 37 | #include "raid56.h" |
8b712842 | 38 | #include "async-thread.h" |
21adbd5c | 39 | #include "check-integrity.h" |
606686ee | 40 | #include "rcu-string.h" |
3fed40cc | 41 | #include "math.h" |
8dabb742 | 42 | #include "dev-replace.h" |
99994cde | 43 | #include "sysfs.h" |
0b86a832 | 44 | |
af902047 ZL |
45 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
46 | [BTRFS_RAID_RAID10] = { | |
47 | .sub_stripes = 2, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 0, /* 0 == as many as possible */ | |
50 | .devs_min = 4, | |
8789f4fe | 51 | .tolerated_failures = 1, |
af902047 ZL |
52 | .devs_increment = 2, |
53 | .ncopies = 2, | |
54 | }, | |
55 | [BTRFS_RAID_RAID1] = { | |
56 | .sub_stripes = 1, | |
57 | .dev_stripes = 1, | |
58 | .devs_max = 2, | |
59 | .devs_min = 2, | |
8789f4fe | 60 | .tolerated_failures = 1, |
af902047 ZL |
61 | .devs_increment = 2, |
62 | .ncopies = 2, | |
63 | }, | |
64 | [BTRFS_RAID_DUP] = { | |
65 | .sub_stripes = 1, | |
66 | .dev_stripes = 2, | |
67 | .devs_max = 1, | |
68 | .devs_min = 1, | |
8789f4fe | 69 | .tolerated_failures = 0, |
af902047 ZL |
70 | .devs_increment = 1, |
71 | .ncopies = 2, | |
72 | }, | |
73 | [BTRFS_RAID_RAID0] = { | |
74 | .sub_stripes = 1, | |
75 | .dev_stripes = 1, | |
76 | .devs_max = 0, | |
77 | .devs_min = 2, | |
8789f4fe | 78 | .tolerated_failures = 0, |
af902047 ZL |
79 | .devs_increment = 1, |
80 | .ncopies = 1, | |
81 | }, | |
82 | [BTRFS_RAID_SINGLE] = { | |
83 | .sub_stripes = 1, | |
84 | .dev_stripes = 1, | |
85 | .devs_max = 1, | |
86 | .devs_min = 1, | |
8789f4fe | 87 | .tolerated_failures = 0, |
af902047 ZL |
88 | .devs_increment = 1, |
89 | .ncopies = 1, | |
90 | }, | |
91 | [BTRFS_RAID_RAID5] = { | |
92 | .sub_stripes = 1, | |
93 | .dev_stripes = 1, | |
94 | .devs_max = 0, | |
95 | .devs_min = 2, | |
8789f4fe | 96 | .tolerated_failures = 1, |
af902047 ZL |
97 | .devs_increment = 1, |
98 | .ncopies = 2, | |
99 | }, | |
100 | [BTRFS_RAID_RAID6] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 3, | |
8789f4fe | 105 | .tolerated_failures = 2, |
af902047 ZL |
106 | .devs_increment = 1, |
107 | .ncopies = 3, | |
108 | }, | |
109 | }; | |
110 | ||
fb75d857 | 111 | const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { |
af902047 ZL |
112 | [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10, |
113 | [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1, | |
114 | [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP, | |
115 | [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0, | |
116 | [BTRFS_RAID_SINGLE] = 0, | |
117 | [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5, | |
118 | [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6, | |
119 | }; | |
120 | ||
621292ba DS |
121 | /* |
122 | * Table to convert BTRFS_RAID_* to the error code if minimum number of devices | |
123 | * condition is not met. Zero means there's no corresponding | |
124 | * BTRFS_ERROR_DEV_*_NOT_MET value. | |
125 | */ | |
126 | const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = { | |
127 | [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, | |
128 | [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, | |
129 | [BTRFS_RAID_DUP] = 0, | |
130 | [BTRFS_RAID_RAID0] = 0, | |
131 | [BTRFS_RAID_SINGLE] = 0, | |
132 | [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, | |
133 | [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, | |
134 | }; | |
135 | ||
2b82032c | 136 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 137 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 138 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 139 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 140 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 141 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
142 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
143 | enum btrfs_map_op op, | |
144 | u64 logical, u64 *length, | |
145 | struct btrfs_bio **bbio_ret, | |
146 | int mirror_num, int need_raid_map); | |
2b82032c | 147 | |
67a2c45e | 148 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 149 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
150 | struct list_head *btrfs_get_fs_uuids(void) |
151 | { | |
152 | return &fs_uuids; | |
153 | } | |
8a4b83cc | 154 | |
2208a378 ID |
155 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
156 | { | |
157 | struct btrfs_fs_devices *fs_devs; | |
158 | ||
78f2c9e6 | 159 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
160 | if (!fs_devs) |
161 | return ERR_PTR(-ENOMEM); | |
162 | ||
163 | mutex_init(&fs_devs->device_list_mutex); | |
164 | ||
165 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 166 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
167 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
168 | INIT_LIST_HEAD(&fs_devs->list); | |
169 | ||
170 | return fs_devs; | |
171 | } | |
172 | ||
173 | /** | |
174 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
175 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
176 | * generated. | |
177 | * | |
178 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
179 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
180 | * can be destroyed with kfree() right away. | |
181 | */ | |
182 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
183 | { | |
184 | struct btrfs_fs_devices *fs_devs; | |
185 | ||
186 | fs_devs = __alloc_fs_devices(); | |
187 | if (IS_ERR(fs_devs)) | |
188 | return fs_devs; | |
189 | ||
190 | if (fsid) | |
191 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
192 | else | |
193 | generate_random_uuid(fs_devs->fsid); | |
194 | ||
195 | return fs_devs; | |
196 | } | |
197 | ||
e4404d6e YZ |
198 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
199 | { | |
200 | struct btrfs_device *device; | |
201 | WARN_ON(fs_devices->opened); | |
202 | while (!list_empty(&fs_devices->devices)) { | |
203 | device = list_entry(fs_devices->devices.next, | |
204 | struct btrfs_device, dev_list); | |
205 | list_del(&device->dev_list); | |
606686ee | 206 | rcu_string_free(device->name); |
e4404d6e YZ |
207 | kfree(device); |
208 | } | |
209 | kfree(fs_devices); | |
210 | } | |
211 | ||
b8b8ff59 LC |
212 | static void btrfs_kobject_uevent(struct block_device *bdev, |
213 | enum kobject_action action) | |
214 | { | |
215 | int ret; | |
216 | ||
217 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
218 | if (ret) | |
efe120a0 | 219 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
220 | action, |
221 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
222 | &disk_to_dev(bdev->bd_disk)->kobj); | |
223 | } | |
224 | ||
143bede5 | 225 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
226 | { |
227 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 228 | |
2b82032c YZ |
229 | while (!list_empty(&fs_uuids)) { |
230 | fs_devices = list_entry(fs_uuids.next, | |
231 | struct btrfs_fs_devices, list); | |
232 | list_del(&fs_devices->list); | |
e4404d6e | 233 | free_fs_devices(fs_devices); |
8a4b83cc | 234 | } |
8a4b83cc CM |
235 | } |
236 | ||
12bd2fc0 ID |
237 | static struct btrfs_device *__alloc_device(void) |
238 | { | |
239 | struct btrfs_device *dev; | |
240 | ||
78f2c9e6 | 241 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
242 | if (!dev) |
243 | return ERR_PTR(-ENOMEM); | |
244 | ||
245 | INIT_LIST_HEAD(&dev->dev_list); | |
246 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 247 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
248 | |
249 | spin_lock_init(&dev->io_lock); | |
250 | ||
251 | spin_lock_init(&dev->reada_lock); | |
252 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 253 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 254 | btrfs_device_data_ordered_init(dev); |
cc8385b5 | 255 | INIT_RADIX_TREE(&dev->reada_zones, GFP_KERNEL); |
d0164adc | 256 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
257 | |
258 | return dev; | |
259 | } | |
260 | ||
a1b32a59 CM |
261 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
262 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
263 | { |
264 | struct btrfs_device *dev; | |
8a4b83cc | 265 | |
c6e30871 | 266 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 267 | if (dev->devid == devid && |
8f18cf13 | 268 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 269 | return dev; |
a443755f | 270 | } |
8a4b83cc CM |
271 | } |
272 | return NULL; | |
273 | } | |
274 | ||
a1b32a59 | 275 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 276 | { |
8a4b83cc CM |
277 | struct btrfs_fs_devices *fs_devices; |
278 | ||
c6e30871 | 279 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
280 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
281 | return fs_devices; | |
282 | } | |
283 | return NULL; | |
284 | } | |
285 | ||
beaf8ab3 SB |
286 | static int |
287 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
288 | int flush, struct block_device **bdev, | |
289 | struct buffer_head **bh) | |
290 | { | |
291 | int ret; | |
292 | ||
293 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
294 | ||
295 | if (IS_ERR(*bdev)) { | |
296 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
297 | goto error; |
298 | } | |
299 | ||
300 | if (flush) | |
301 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
302 | ret = set_blocksize(*bdev, 4096); | |
303 | if (ret) { | |
304 | blkdev_put(*bdev, flags); | |
305 | goto error; | |
306 | } | |
307 | invalidate_bdev(*bdev); | |
308 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
309 | if (IS_ERR(*bh)) { |
310 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
311 | blkdev_put(*bdev, flags); |
312 | goto error; | |
313 | } | |
314 | ||
315 | return 0; | |
316 | ||
317 | error: | |
318 | *bdev = NULL; | |
319 | *bh = NULL; | |
320 | return ret; | |
321 | } | |
322 | ||
ffbd517d CM |
323 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
324 | struct bio *head, struct bio *tail) | |
325 | { | |
326 | ||
327 | struct bio *old_head; | |
328 | ||
329 | old_head = pending_bios->head; | |
330 | pending_bios->head = head; | |
331 | if (pending_bios->tail) | |
332 | tail->bi_next = old_head; | |
333 | else | |
334 | pending_bios->tail = tail; | |
335 | } | |
336 | ||
8b712842 CM |
337 | /* |
338 | * we try to collect pending bios for a device so we don't get a large | |
339 | * number of procs sending bios down to the same device. This greatly | |
340 | * improves the schedulers ability to collect and merge the bios. | |
341 | * | |
342 | * But, it also turns into a long list of bios to process and that is sure | |
343 | * to eventually make the worker thread block. The solution here is to | |
344 | * make some progress and then put this work struct back at the end of | |
345 | * the list if the block device is congested. This way, multiple devices | |
346 | * can make progress from a single worker thread. | |
347 | */ | |
143bede5 | 348 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 349 | { |
0b246afa | 350 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
351 | struct bio *pending; |
352 | struct backing_dev_info *bdi; | |
ffbd517d | 353 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
354 | struct bio *tail; |
355 | struct bio *cur; | |
356 | int again = 0; | |
ffbd517d | 357 | unsigned long num_run; |
d644d8a1 | 358 | unsigned long batch_run = 0; |
b64a2851 | 359 | unsigned long limit; |
b765ead5 | 360 | unsigned long last_waited = 0; |
d84275c9 | 361 | int force_reg = 0; |
0e588859 | 362 | int sync_pending = 0; |
211588ad CM |
363 | struct blk_plug plug; |
364 | ||
365 | /* | |
366 | * this function runs all the bios we've collected for | |
367 | * a particular device. We don't want to wander off to | |
368 | * another device without first sending all of these down. | |
369 | * So, setup a plug here and finish it off before we return | |
370 | */ | |
371 | blk_start_plug(&plug); | |
8b712842 | 372 | |
efa7c9f9 | 373 | bdi = device->bdev->bd_bdi; |
b64a2851 CM |
374 | limit = btrfs_async_submit_limit(fs_info); |
375 | limit = limit * 2 / 3; | |
376 | ||
8b712842 CM |
377 | loop: |
378 | spin_lock(&device->io_lock); | |
379 | ||
a6837051 | 380 | loop_lock: |
d84275c9 | 381 | num_run = 0; |
ffbd517d | 382 | |
8b712842 CM |
383 | /* take all the bios off the list at once and process them |
384 | * later on (without the lock held). But, remember the | |
385 | * tail and other pointers so the bios can be properly reinserted | |
386 | * into the list if we hit congestion | |
387 | */ | |
d84275c9 | 388 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 389 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
390 | force_reg = 1; |
391 | } else { | |
ffbd517d | 392 | pending_bios = &device->pending_bios; |
d84275c9 CM |
393 | force_reg = 0; |
394 | } | |
ffbd517d CM |
395 | |
396 | pending = pending_bios->head; | |
397 | tail = pending_bios->tail; | |
8b712842 | 398 | WARN_ON(pending && !tail); |
8b712842 CM |
399 | |
400 | /* | |
401 | * if pending was null this time around, no bios need processing | |
402 | * at all and we can stop. Otherwise it'll loop back up again | |
403 | * and do an additional check so no bios are missed. | |
404 | * | |
405 | * device->running_pending is used to synchronize with the | |
406 | * schedule_bio code. | |
407 | */ | |
ffbd517d CM |
408 | if (device->pending_sync_bios.head == NULL && |
409 | device->pending_bios.head == NULL) { | |
8b712842 CM |
410 | again = 0; |
411 | device->running_pending = 0; | |
ffbd517d CM |
412 | } else { |
413 | again = 1; | |
414 | device->running_pending = 1; | |
8b712842 | 415 | } |
ffbd517d CM |
416 | |
417 | pending_bios->head = NULL; | |
418 | pending_bios->tail = NULL; | |
419 | ||
8b712842 CM |
420 | spin_unlock(&device->io_lock); |
421 | ||
d397712b | 422 | while (pending) { |
ffbd517d CM |
423 | |
424 | rmb(); | |
d84275c9 CM |
425 | /* we want to work on both lists, but do more bios on the |
426 | * sync list than the regular list | |
427 | */ | |
428 | if ((num_run > 32 && | |
429 | pending_bios != &device->pending_sync_bios && | |
430 | device->pending_sync_bios.head) || | |
431 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
432 | device->pending_bios.head)) { | |
ffbd517d CM |
433 | spin_lock(&device->io_lock); |
434 | requeue_list(pending_bios, pending, tail); | |
435 | goto loop_lock; | |
436 | } | |
437 | ||
8b712842 CM |
438 | cur = pending; |
439 | pending = pending->bi_next; | |
440 | cur->bi_next = NULL; | |
b64a2851 | 441 | |
ee863954 DS |
442 | /* |
443 | * atomic_dec_return implies a barrier for waitqueue_active | |
444 | */ | |
66657b31 | 445 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
446 | waitqueue_active(&fs_info->async_submit_wait)) |
447 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 448 | |
dac56212 | 449 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 450 | |
2ab1ba68 CM |
451 | /* |
452 | * if we're doing the sync list, record that our | |
453 | * plug has some sync requests on it | |
454 | * | |
455 | * If we're doing the regular list and there are | |
456 | * sync requests sitting around, unplug before | |
457 | * we add more | |
458 | */ | |
459 | if (pending_bios == &device->pending_sync_bios) { | |
460 | sync_pending = 1; | |
461 | } else if (sync_pending) { | |
462 | blk_finish_plug(&plug); | |
463 | blk_start_plug(&plug); | |
464 | sync_pending = 0; | |
465 | } | |
466 | ||
4e49ea4a | 467 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
468 | num_run++; |
469 | batch_run++; | |
853d8ec4 DS |
470 | |
471 | cond_resched(); | |
8b712842 CM |
472 | |
473 | /* | |
474 | * we made progress, there is more work to do and the bdi | |
475 | * is now congested. Back off and let other work structs | |
476 | * run instead | |
477 | */ | |
57fd5a5f | 478 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 479 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 480 | struct io_context *ioc; |
8b712842 | 481 | |
b765ead5 CM |
482 | ioc = current->io_context; |
483 | ||
484 | /* | |
485 | * the main goal here is that we don't want to | |
486 | * block if we're going to be able to submit | |
487 | * more requests without blocking. | |
488 | * | |
489 | * This code does two great things, it pokes into | |
490 | * the elevator code from a filesystem _and_ | |
491 | * it makes assumptions about how batching works. | |
492 | */ | |
493 | if (ioc && ioc->nr_batch_requests > 0 && | |
494 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
495 | (last_waited == 0 || | |
496 | ioc->last_waited == last_waited)) { | |
497 | /* | |
498 | * we want to go through our batch of | |
499 | * requests and stop. So, we copy out | |
500 | * the ioc->last_waited time and test | |
501 | * against it before looping | |
502 | */ | |
503 | last_waited = ioc->last_waited; | |
853d8ec4 | 504 | cond_resched(); |
b765ead5 CM |
505 | continue; |
506 | } | |
8b712842 | 507 | spin_lock(&device->io_lock); |
ffbd517d | 508 | requeue_list(pending_bios, pending, tail); |
a6837051 | 509 | device->running_pending = 1; |
8b712842 CM |
510 | |
511 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
512 | btrfs_queue_work(fs_info->submit_workers, |
513 | &device->work); | |
8b712842 CM |
514 | goto done; |
515 | } | |
d85c8a6f CM |
516 | /* unplug every 64 requests just for good measure */ |
517 | if (batch_run % 64 == 0) { | |
518 | blk_finish_plug(&plug); | |
519 | blk_start_plug(&plug); | |
520 | sync_pending = 0; | |
521 | } | |
8b712842 | 522 | } |
ffbd517d | 523 | |
51684082 CM |
524 | cond_resched(); |
525 | if (again) | |
526 | goto loop; | |
527 | ||
528 | spin_lock(&device->io_lock); | |
529 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
530 | goto loop_lock; | |
531 | spin_unlock(&device->io_lock); | |
532 | ||
8b712842 | 533 | done: |
211588ad | 534 | blk_finish_plug(&plug); |
8b712842 CM |
535 | } |
536 | ||
b2950863 | 537 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
538 | { |
539 | struct btrfs_device *device; | |
540 | ||
541 | device = container_of(work, struct btrfs_device, work); | |
542 | run_scheduled_bios(device); | |
543 | } | |
544 | ||
4fde46f0 AJ |
545 | |
546 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
547 | { | |
548 | struct btrfs_fs_devices *fs_devs; | |
549 | struct btrfs_device *dev; | |
550 | ||
551 | if (!cur_dev->name) | |
552 | return; | |
553 | ||
554 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
555 | int del = 1; | |
556 | ||
557 | if (fs_devs->opened) | |
558 | continue; | |
559 | if (fs_devs->seeding) | |
560 | continue; | |
561 | ||
562 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
563 | ||
564 | if (dev == cur_dev) | |
565 | continue; | |
566 | if (!dev->name) | |
567 | continue; | |
568 | ||
569 | /* | |
570 | * Todo: This won't be enough. What if the same device | |
571 | * comes back (with new uuid and) with its mapper path? | |
572 | * But for now, this does help as mostly an admin will | |
573 | * either use mapper or non mapper path throughout. | |
574 | */ | |
575 | rcu_read_lock(); | |
576 | del = strcmp(rcu_str_deref(dev->name), | |
577 | rcu_str_deref(cur_dev->name)); | |
578 | rcu_read_unlock(); | |
579 | if (!del) | |
580 | break; | |
581 | } | |
582 | ||
583 | if (!del) { | |
584 | /* delete the stale device */ | |
585 | if (fs_devs->num_devices == 1) { | |
586 | btrfs_sysfs_remove_fsid(fs_devs); | |
587 | list_del(&fs_devs->list); | |
588 | free_fs_devices(fs_devs); | |
589 | } else { | |
590 | fs_devs->num_devices--; | |
591 | list_del(&dev->dev_list); | |
592 | rcu_string_free(dev->name); | |
593 | kfree(dev); | |
594 | } | |
595 | break; | |
596 | } | |
597 | } | |
598 | } | |
599 | ||
60999ca4 DS |
600 | /* |
601 | * Add new device to list of registered devices | |
602 | * | |
603 | * Returns: | |
604 | * 1 - first time device is seen | |
605 | * 0 - device already known | |
606 | * < 0 - error | |
607 | */ | |
a1b32a59 | 608 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
609 | struct btrfs_super_block *disk_super, |
610 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
611 | { | |
612 | struct btrfs_device *device; | |
613 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 614 | struct rcu_string *name; |
60999ca4 | 615 | int ret = 0; |
8a4b83cc CM |
616 | u64 found_transid = btrfs_super_generation(disk_super); |
617 | ||
618 | fs_devices = find_fsid(disk_super->fsid); | |
619 | if (!fs_devices) { | |
2208a378 ID |
620 | fs_devices = alloc_fs_devices(disk_super->fsid); |
621 | if (IS_ERR(fs_devices)) | |
622 | return PTR_ERR(fs_devices); | |
623 | ||
8a4b83cc | 624 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 625 | |
8a4b83cc CM |
626 | device = NULL; |
627 | } else { | |
a443755f CM |
628 | device = __find_device(&fs_devices->devices, devid, |
629 | disk_super->dev_item.uuid); | |
8a4b83cc | 630 | } |
443f24fe | 631 | |
8a4b83cc | 632 | if (!device) { |
2b82032c YZ |
633 | if (fs_devices->opened) |
634 | return -EBUSY; | |
635 | ||
12bd2fc0 ID |
636 | device = btrfs_alloc_device(NULL, &devid, |
637 | disk_super->dev_item.uuid); | |
638 | if (IS_ERR(device)) { | |
8a4b83cc | 639 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 640 | return PTR_ERR(device); |
8a4b83cc | 641 | } |
606686ee JB |
642 | |
643 | name = rcu_string_strdup(path, GFP_NOFS); | |
644 | if (!name) { | |
8a4b83cc CM |
645 | kfree(device); |
646 | return -ENOMEM; | |
647 | } | |
606686ee | 648 | rcu_assign_pointer(device->name, name); |
90519d66 | 649 | |
e5e9a520 | 650 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 651 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 652 | fs_devices->num_devices++; |
e5e9a520 CM |
653 | mutex_unlock(&fs_devices->device_list_mutex); |
654 | ||
60999ca4 | 655 | ret = 1; |
2b82032c | 656 | device->fs_devices = fs_devices; |
606686ee | 657 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
658 | /* |
659 | * When FS is already mounted. | |
660 | * 1. If you are here and if the device->name is NULL that | |
661 | * means this device was missing at time of FS mount. | |
662 | * 2. If you are here and if the device->name is different | |
663 | * from 'path' that means either | |
664 | * a. The same device disappeared and reappeared with | |
665 | * different name. or | |
666 | * b. The missing-disk-which-was-replaced, has | |
667 | * reappeared now. | |
668 | * | |
669 | * We must allow 1 and 2a above. But 2b would be a spurious | |
670 | * and unintentional. | |
671 | * | |
672 | * Further in case of 1 and 2a above, the disk at 'path' | |
673 | * would have missed some transaction when it was away and | |
674 | * in case of 2a the stale bdev has to be updated as well. | |
675 | * 2b must not be allowed at all time. | |
676 | */ | |
677 | ||
678 | /* | |
0f23ae74 CM |
679 | * For now, we do allow update to btrfs_fs_device through the |
680 | * btrfs dev scan cli after FS has been mounted. We're still | |
681 | * tracking a problem where systems fail mount by subvolume id | |
682 | * when we reject replacement on a mounted FS. | |
b96de000 | 683 | */ |
0f23ae74 | 684 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
685 | /* |
686 | * That is if the FS is _not_ mounted and if you | |
687 | * are here, that means there is more than one | |
688 | * disk with same uuid and devid.We keep the one | |
689 | * with larger generation number or the last-in if | |
690 | * generation are equal. | |
691 | */ | |
0f23ae74 | 692 | return -EEXIST; |
77bdae4d | 693 | } |
b96de000 | 694 | |
606686ee | 695 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
696 | if (!name) |
697 | return -ENOMEM; | |
606686ee JB |
698 | rcu_string_free(device->name); |
699 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
700 | if (device->missing) { |
701 | fs_devices->missing_devices--; | |
702 | device->missing = 0; | |
703 | } | |
8a4b83cc CM |
704 | } |
705 | ||
77bdae4d AJ |
706 | /* |
707 | * Unmount does not free the btrfs_device struct but would zero | |
708 | * generation along with most of the other members. So just update | |
709 | * it back. We need it to pick the disk with largest generation | |
710 | * (as above). | |
711 | */ | |
712 | if (!fs_devices->opened) | |
713 | device->generation = found_transid; | |
714 | ||
4fde46f0 AJ |
715 | /* |
716 | * if there is new btrfs on an already registered device, | |
717 | * then remove the stale device entry. | |
718 | */ | |
02feae3c AJ |
719 | if (ret > 0) |
720 | btrfs_free_stale_device(device); | |
4fde46f0 | 721 | |
8a4b83cc | 722 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
723 | |
724 | return ret; | |
8a4b83cc CM |
725 | } |
726 | ||
e4404d6e YZ |
727 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
728 | { | |
729 | struct btrfs_fs_devices *fs_devices; | |
730 | struct btrfs_device *device; | |
731 | struct btrfs_device *orig_dev; | |
732 | ||
2208a378 ID |
733 | fs_devices = alloc_fs_devices(orig->fsid); |
734 | if (IS_ERR(fs_devices)) | |
735 | return fs_devices; | |
e4404d6e | 736 | |
adbbb863 | 737 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 738 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 739 | |
46224705 | 740 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 741 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
742 | struct rcu_string *name; |
743 | ||
12bd2fc0 ID |
744 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
745 | orig_dev->uuid); | |
746 | if (IS_ERR(device)) | |
e4404d6e YZ |
747 | goto error; |
748 | ||
606686ee JB |
749 | /* |
750 | * This is ok to do without rcu read locked because we hold the | |
751 | * uuid mutex so nothing we touch in here is going to disappear. | |
752 | */ | |
e755f780 | 753 | if (orig_dev->name) { |
78f2c9e6 DS |
754 | name = rcu_string_strdup(orig_dev->name->str, |
755 | GFP_KERNEL); | |
e755f780 AJ |
756 | if (!name) { |
757 | kfree(device); | |
758 | goto error; | |
759 | } | |
760 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 761 | } |
e4404d6e | 762 | |
e4404d6e YZ |
763 | list_add(&device->dev_list, &fs_devices->devices); |
764 | device->fs_devices = fs_devices; | |
765 | fs_devices->num_devices++; | |
766 | } | |
adbbb863 | 767 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
768 | return fs_devices; |
769 | error: | |
adbbb863 | 770 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
771 | free_fs_devices(fs_devices); |
772 | return ERR_PTR(-ENOMEM); | |
773 | } | |
774 | ||
9eaed21e | 775 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 776 | { |
c6e30871 | 777 | struct btrfs_device *device, *next; |
443f24fe | 778 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 779 | |
dfe25020 CM |
780 | mutex_lock(&uuid_mutex); |
781 | again: | |
46224705 | 782 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 783 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 784 | if (device->in_fs_metadata) { |
63a212ab | 785 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
786 | (!latest_dev || |
787 | device->generation > latest_dev->generation)) { | |
788 | latest_dev = device; | |
a6b0d5c8 | 789 | } |
2b82032c | 790 | continue; |
a6b0d5c8 | 791 | } |
2b82032c | 792 | |
8dabb742 SB |
793 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
794 | /* | |
795 | * In the first step, keep the device which has | |
796 | * the correct fsid and the devid that is used | |
797 | * for the dev_replace procedure. | |
798 | * In the second step, the dev_replace state is | |
799 | * read from the device tree and it is known | |
800 | * whether the procedure is really active or | |
801 | * not, which means whether this device is | |
802 | * used or whether it should be removed. | |
803 | */ | |
804 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
805 | continue; | |
806 | } | |
807 | } | |
2b82032c | 808 | if (device->bdev) { |
d4d77629 | 809 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
810 | device->bdev = NULL; |
811 | fs_devices->open_devices--; | |
812 | } | |
813 | if (device->writeable) { | |
814 | list_del_init(&device->dev_alloc_list); | |
815 | device->writeable = 0; | |
8dabb742 SB |
816 | if (!device->is_tgtdev_for_dev_replace) |
817 | fs_devices->rw_devices--; | |
2b82032c | 818 | } |
e4404d6e YZ |
819 | list_del_init(&device->dev_list); |
820 | fs_devices->num_devices--; | |
606686ee | 821 | rcu_string_free(device->name); |
e4404d6e | 822 | kfree(device); |
dfe25020 | 823 | } |
2b82032c YZ |
824 | |
825 | if (fs_devices->seed) { | |
826 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
827 | goto again; |
828 | } | |
829 | ||
443f24fe | 830 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 831 | |
dfe25020 | 832 | mutex_unlock(&uuid_mutex); |
dfe25020 | 833 | } |
a0af469b | 834 | |
1f78160c XG |
835 | static void __free_device(struct work_struct *work) |
836 | { | |
837 | struct btrfs_device *device; | |
838 | ||
839 | device = container_of(work, struct btrfs_device, rcu_work); | |
606686ee | 840 | rcu_string_free(device->name); |
1f78160c XG |
841 | kfree(device); |
842 | } | |
843 | ||
844 | static void free_device(struct rcu_head *head) | |
845 | { | |
846 | struct btrfs_device *device; | |
847 | ||
848 | device = container_of(head, struct btrfs_device, rcu); | |
849 | ||
850 | INIT_WORK(&device->rcu_work, __free_device); | |
851 | schedule_work(&device->rcu_work); | |
852 | } | |
853 | ||
14238819 AJ |
854 | static void btrfs_close_bdev(struct btrfs_device *device) |
855 | { | |
856 | if (device->bdev && device->writeable) { | |
857 | sync_blockdev(device->bdev); | |
858 | invalidate_bdev(device->bdev); | |
859 | } | |
860 | ||
861 | if (device->bdev) | |
862 | blkdev_put(device->bdev, device->mode); | |
863 | } | |
864 | ||
0ccd0528 | 865 | static void btrfs_prepare_close_one_device(struct btrfs_device *device) |
f448341a AJ |
866 | { |
867 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
868 | struct btrfs_device *new_device; | |
869 | struct rcu_string *name; | |
870 | ||
871 | if (device->bdev) | |
872 | fs_devices->open_devices--; | |
873 | ||
874 | if (device->writeable && | |
875 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
876 | list_del_init(&device->dev_alloc_list); | |
877 | fs_devices->rw_devices--; | |
878 | } | |
879 | ||
880 | if (device->missing) | |
881 | fs_devices->missing_devices--; | |
882 | ||
883 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
884 | device->uuid); | |
885 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
886 | ||
887 | /* Safe because we are under uuid_mutex */ | |
888 | if (device->name) { | |
889 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
890 | BUG_ON(!name); /* -ENOMEM */ | |
891 | rcu_assign_pointer(new_device->name, name); | |
892 | } | |
893 | ||
894 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
895 | new_device->fs_devices = device->fs_devices; | |
f448341a AJ |
896 | } |
897 | ||
2b82032c | 898 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 899 | { |
2037a093 | 900 | struct btrfs_device *device, *tmp; |
0ccd0528 AJ |
901 | struct list_head pending_put; |
902 | ||
903 | INIT_LIST_HEAD(&pending_put); | |
e4404d6e | 904 | |
2b82032c YZ |
905 | if (--fs_devices->opened > 0) |
906 | return 0; | |
8a4b83cc | 907 | |
c9513edb | 908 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 909 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
0ccd0528 AJ |
910 | btrfs_prepare_close_one_device(device); |
911 | list_add(&device->dev_list, &pending_put); | |
8a4b83cc | 912 | } |
c9513edb XG |
913 | mutex_unlock(&fs_devices->device_list_mutex); |
914 | ||
0ccd0528 AJ |
915 | /* |
916 | * btrfs_show_devname() is using the device_list_mutex, | |
917 | * sometimes call to blkdev_put() leads vfs calling | |
918 | * into this func. So do put outside of device_list_mutex, | |
919 | * as of now. | |
920 | */ | |
921 | while (!list_empty(&pending_put)) { | |
922 | device = list_first_entry(&pending_put, | |
923 | struct btrfs_device, dev_list); | |
924 | list_del(&device->dev_list); | |
925 | btrfs_close_bdev(device); | |
926 | call_rcu(&device->rcu, free_device); | |
927 | } | |
928 | ||
e4404d6e YZ |
929 | WARN_ON(fs_devices->open_devices); |
930 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
931 | fs_devices->opened = 0; |
932 | fs_devices->seeding = 0; | |
2b82032c | 933 | |
8a4b83cc CM |
934 | return 0; |
935 | } | |
936 | ||
2b82032c YZ |
937 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
938 | { | |
e4404d6e | 939 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
940 | int ret; |
941 | ||
942 | mutex_lock(&uuid_mutex); | |
943 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
944 | if (!fs_devices->opened) { |
945 | seed_devices = fs_devices->seed; | |
946 | fs_devices->seed = NULL; | |
947 | } | |
2b82032c | 948 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
949 | |
950 | while (seed_devices) { | |
951 | fs_devices = seed_devices; | |
952 | seed_devices = fs_devices->seed; | |
953 | __btrfs_close_devices(fs_devices); | |
954 | free_fs_devices(fs_devices); | |
955 | } | |
bc178622 ES |
956 | /* |
957 | * Wait for rcu kworkers under __btrfs_close_devices | |
958 | * to finish all blkdev_puts so device is really | |
959 | * free when umount is done. | |
960 | */ | |
961 | rcu_barrier(); | |
2b82032c YZ |
962 | return ret; |
963 | } | |
964 | ||
e4404d6e YZ |
965 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
966 | fmode_t flags, void *holder) | |
8a4b83cc | 967 | { |
d5e2003c | 968 | struct request_queue *q; |
8a4b83cc CM |
969 | struct block_device *bdev; |
970 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 971 | struct btrfs_device *device; |
443f24fe | 972 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
973 | struct buffer_head *bh; |
974 | struct btrfs_super_block *disk_super; | |
a0af469b | 975 | u64 devid; |
2b82032c | 976 | int seeding = 1; |
a0af469b | 977 | int ret = 0; |
8a4b83cc | 978 | |
d4d77629 TH |
979 | flags |= FMODE_EXCL; |
980 | ||
c6e30871 | 981 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
982 | if (device->bdev) |
983 | continue; | |
dfe25020 CM |
984 | if (!device->name) |
985 | continue; | |
986 | ||
f63e0cca ES |
987 | /* Just open everything we can; ignore failures here */ |
988 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
989 | &bdev, &bh)) | |
beaf8ab3 | 990 | continue; |
a0af469b CM |
991 | |
992 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 993 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
994 | if (devid != device->devid) |
995 | goto error_brelse; | |
996 | ||
2b82032c YZ |
997 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
998 | BTRFS_UUID_SIZE)) | |
999 | goto error_brelse; | |
1000 | ||
1001 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
1002 | if (!latest_dev || |
1003 | device->generation > latest_dev->generation) | |
1004 | latest_dev = device; | |
a0af469b | 1005 | |
2b82032c YZ |
1006 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
1007 | device->writeable = 0; | |
1008 | } else { | |
1009 | device->writeable = !bdev_read_only(bdev); | |
1010 | seeding = 0; | |
1011 | } | |
1012 | ||
d5e2003c | 1013 | q = bdev_get_queue(bdev); |
90180da4 | 1014 | if (blk_queue_discard(q)) |
d5e2003c | 1015 | device->can_discard = 1; |
d5e2003c | 1016 | |
8a4b83cc | 1017 | device->bdev = bdev; |
dfe25020 | 1018 | device->in_fs_metadata = 0; |
15916de8 CM |
1019 | device->mode = flags; |
1020 | ||
c289811c CM |
1021 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1022 | fs_devices->rotating = 1; | |
1023 | ||
a0af469b | 1024 | fs_devices->open_devices++; |
55e50e45 ID |
1025 | if (device->writeable && |
1026 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
1027 | fs_devices->rw_devices++; |
1028 | list_add(&device->dev_alloc_list, | |
1029 | &fs_devices->alloc_list); | |
1030 | } | |
4f6c9328 | 1031 | brelse(bh); |
a0af469b | 1032 | continue; |
a061fc8d | 1033 | |
a0af469b CM |
1034 | error_brelse: |
1035 | brelse(bh); | |
d4d77629 | 1036 | blkdev_put(bdev, flags); |
a0af469b | 1037 | continue; |
8a4b83cc | 1038 | } |
a0af469b | 1039 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1040 | ret = -EINVAL; |
a0af469b CM |
1041 | goto out; |
1042 | } | |
2b82032c YZ |
1043 | fs_devices->seeding = seeding; |
1044 | fs_devices->opened = 1; | |
443f24fe | 1045 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1046 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1047 | out: |
2b82032c YZ |
1048 | return ret; |
1049 | } | |
1050 | ||
1051 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 1052 | fmode_t flags, void *holder) |
2b82032c YZ |
1053 | { |
1054 | int ret; | |
1055 | ||
1056 | mutex_lock(&uuid_mutex); | |
1057 | if (fs_devices->opened) { | |
e4404d6e YZ |
1058 | fs_devices->opened++; |
1059 | ret = 0; | |
2b82032c | 1060 | } else { |
15916de8 | 1061 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1062 | } |
8a4b83cc | 1063 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1064 | return ret; |
1065 | } | |
1066 | ||
6cf86a00 AJ |
1067 | void btrfs_release_disk_super(struct page *page) |
1068 | { | |
1069 | kunmap(page); | |
1070 | put_page(page); | |
1071 | } | |
1072 | ||
1073 | int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, | |
1074 | struct page **page, struct btrfs_super_block **disk_super) | |
1075 | { | |
1076 | void *p; | |
1077 | pgoff_t index; | |
1078 | ||
1079 | /* make sure our super fits in the device */ | |
1080 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1081 | return 1; | |
1082 | ||
1083 | /* make sure our super fits in the page */ | |
1084 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1085 | return 1; | |
1086 | ||
1087 | /* make sure our super doesn't straddle pages on disk */ | |
1088 | index = bytenr >> PAGE_SHIFT; | |
1089 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1090 | return 1; | |
1091 | ||
1092 | /* pull in the page with our super */ | |
1093 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1094 | index, GFP_KERNEL); | |
1095 | ||
1096 | if (IS_ERR_OR_NULL(*page)) | |
1097 | return 1; | |
1098 | ||
1099 | p = kmap(*page); | |
1100 | ||
1101 | /* align our pointer to the offset of the super block */ | |
1102 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1103 | ||
1104 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1105 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1106 | btrfs_release_disk_super(*page); | |
1107 | return 1; | |
1108 | } | |
1109 | ||
1110 | if ((*disk_super)->label[0] && | |
1111 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1112 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1113 | ||
1114 | return 0; | |
1115 | } | |
1116 | ||
6f60cbd3 DS |
1117 | /* |
1118 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1119 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1120 | * is read via pagecache | |
1121 | */ | |
97288f2c | 1122 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1123 | struct btrfs_fs_devices **fs_devices_ret) |
1124 | { | |
1125 | struct btrfs_super_block *disk_super; | |
1126 | struct block_device *bdev; | |
6f60cbd3 | 1127 | struct page *page; |
6f60cbd3 | 1128 | int ret = -EINVAL; |
8a4b83cc | 1129 | u64 devid; |
f2984462 | 1130 | u64 transid; |
02db0844 | 1131 | u64 total_devices; |
6f60cbd3 | 1132 | u64 bytenr; |
8a4b83cc | 1133 | |
6f60cbd3 DS |
1134 | /* |
1135 | * we would like to check all the supers, but that would make | |
1136 | * a btrfs mount succeed after a mkfs from a different FS. | |
1137 | * So, we need to add a special mount option to scan for | |
1138 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1139 | */ | |
1140 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1141 | flags |= FMODE_EXCL; |
10f6327b | 1142 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1143 | |
1144 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1145 | if (IS_ERR(bdev)) { |
1146 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1147 | goto error; |
6f60cbd3 DS |
1148 | } |
1149 | ||
6cf86a00 | 1150 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) |
6f60cbd3 DS |
1151 | goto error_bdev_put; |
1152 | ||
a343832f | 1153 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1154 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1155 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1156 | |
8a4b83cc | 1157 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1158 | if (ret > 0) { |
1159 | if (disk_super->label[0]) { | |
62e85577 | 1160 | pr_info("BTRFS: device label %s ", disk_super->label); |
60999ca4 | 1161 | } else { |
62e85577 | 1162 | pr_info("BTRFS: device fsid %pU ", disk_super->fsid); |
60999ca4 DS |
1163 | } |
1164 | ||
62e85577 | 1165 | pr_cont("devid %llu transid %llu %s\n", devid, transid, path); |
60999ca4 DS |
1166 | ret = 0; |
1167 | } | |
02db0844 JB |
1168 | if (!ret && fs_devices_ret) |
1169 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 | 1170 | |
6cf86a00 | 1171 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1172 | |
1173 | error_bdev_put: | |
d4d77629 | 1174 | blkdev_put(bdev, flags); |
8a4b83cc | 1175 | error: |
beaf8ab3 | 1176 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1177 | return ret; |
1178 | } | |
0b86a832 | 1179 | |
6d07bcec MX |
1180 | /* helper to account the used device space in the range */ |
1181 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1182 | u64 end, u64 *length) | |
1183 | { | |
1184 | struct btrfs_key key; | |
fb456252 | 1185 | struct btrfs_root *root = device->fs_info->dev_root; |
6d07bcec MX |
1186 | struct btrfs_dev_extent *dev_extent; |
1187 | struct btrfs_path *path; | |
1188 | u64 extent_end; | |
1189 | int ret; | |
1190 | int slot; | |
1191 | struct extent_buffer *l; | |
1192 | ||
1193 | *length = 0; | |
1194 | ||
63a212ab | 1195 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1196 | return 0; |
1197 | ||
1198 | path = btrfs_alloc_path(); | |
1199 | if (!path) | |
1200 | return -ENOMEM; | |
e4058b54 | 1201 | path->reada = READA_FORWARD; |
6d07bcec MX |
1202 | |
1203 | key.objectid = device->devid; | |
1204 | key.offset = start; | |
1205 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1206 | ||
1207 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1208 | if (ret < 0) | |
1209 | goto out; | |
1210 | if (ret > 0) { | |
1211 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1212 | if (ret < 0) | |
1213 | goto out; | |
1214 | } | |
1215 | ||
1216 | while (1) { | |
1217 | l = path->nodes[0]; | |
1218 | slot = path->slots[0]; | |
1219 | if (slot >= btrfs_header_nritems(l)) { | |
1220 | ret = btrfs_next_leaf(root, path); | |
1221 | if (ret == 0) | |
1222 | continue; | |
1223 | if (ret < 0) | |
1224 | goto out; | |
1225 | ||
1226 | break; | |
1227 | } | |
1228 | btrfs_item_key_to_cpu(l, &key, slot); | |
1229 | ||
1230 | if (key.objectid < device->devid) | |
1231 | goto next; | |
1232 | ||
1233 | if (key.objectid > device->devid) | |
1234 | break; | |
1235 | ||
962a298f | 1236 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1237 | goto next; |
1238 | ||
1239 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1240 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1241 | dev_extent); | |
1242 | if (key.offset <= start && extent_end > end) { | |
1243 | *length = end - start + 1; | |
1244 | break; | |
1245 | } else if (key.offset <= start && extent_end > start) | |
1246 | *length += extent_end - start; | |
1247 | else if (key.offset > start && extent_end <= end) | |
1248 | *length += extent_end - key.offset; | |
1249 | else if (key.offset > start && key.offset <= end) { | |
1250 | *length += end - key.offset + 1; | |
1251 | break; | |
1252 | } else if (key.offset > end) | |
1253 | break; | |
1254 | ||
1255 | next: | |
1256 | path->slots[0]++; | |
1257 | } | |
1258 | ret = 0; | |
1259 | out: | |
1260 | btrfs_free_path(path); | |
1261 | return ret; | |
1262 | } | |
1263 | ||
499f377f | 1264 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1265 | struct btrfs_device *device, |
1266 | u64 *start, u64 len) | |
1267 | { | |
fb456252 | 1268 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1269 | struct extent_map *em; |
499f377f | 1270 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1271 | int ret = 0; |
1b984508 | 1272 | u64 physical_start = *start; |
6df9a95e | 1273 | |
499f377f JM |
1274 | if (transaction) |
1275 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1276 | again: |
1277 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1278 | struct map_lookup *map; |
1279 | int i; | |
1280 | ||
95617d69 | 1281 | map = em->map_lookup; |
6df9a95e | 1282 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1283 | u64 end; |
1284 | ||
6df9a95e JB |
1285 | if (map->stripes[i].dev != device) |
1286 | continue; | |
1b984508 | 1287 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1288 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1289 | physical_start) |
6df9a95e | 1290 | continue; |
c152b63e FM |
1291 | /* |
1292 | * Make sure that while processing the pinned list we do | |
1293 | * not override our *start with a lower value, because | |
1294 | * we can have pinned chunks that fall within this | |
1295 | * device hole and that have lower physical addresses | |
1296 | * than the pending chunks we processed before. If we | |
1297 | * do not take this special care we can end up getting | |
1298 | * 2 pending chunks that start at the same physical | |
1299 | * device offsets because the end offset of a pinned | |
1300 | * chunk can be equal to the start offset of some | |
1301 | * pending chunk. | |
1302 | */ | |
1303 | end = map->stripes[i].physical + em->orig_block_len; | |
1304 | if (end > *start) { | |
1305 | *start = end; | |
1306 | ret = 1; | |
1307 | } | |
6df9a95e JB |
1308 | } |
1309 | } | |
499f377f JM |
1310 | if (search_list != &fs_info->pinned_chunks) { |
1311 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1312 | goto again; |
1313 | } | |
6df9a95e JB |
1314 | |
1315 | return ret; | |
1316 | } | |
1317 | ||
1318 | ||
0b86a832 | 1319 | /* |
499f377f JM |
1320 | * find_free_dev_extent_start - find free space in the specified device |
1321 | * @device: the device which we search the free space in | |
1322 | * @num_bytes: the size of the free space that we need | |
1323 | * @search_start: the position from which to begin the search | |
1324 | * @start: store the start of the free space. | |
1325 | * @len: the size of the free space. that we find, or the size | |
1326 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1327 | * |
0b86a832 CM |
1328 | * this uses a pretty simple search, the expectation is that it is |
1329 | * called very infrequently and that a given device has a small number | |
1330 | * of extents | |
7bfc837d MX |
1331 | * |
1332 | * @start is used to store the start of the free space if we find. But if we | |
1333 | * don't find suitable free space, it will be used to store the start position | |
1334 | * of the max free space. | |
1335 | * | |
1336 | * @len is used to store the size of the free space that we find. | |
1337 | * But if we don't find suitable free space, it is used to store the size of | |
1338 | * the max free space. | |
0b86a832 | 1339 | */ |
499f377f JM |
1340 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1341 | struct btrfs_device *device, u64 num_bytes, | |
1342 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1343 | { |
0b246afa JM |
1344 | struct btrfs_fs_info *fs_info = device->fs_info; |
1345 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1346 | struct btrfs_key key; |
7bfc837d | 1347 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1348 | struct btrfs_path *path; |
7bfc837d MX |
1349 | u64 hole_size; |
1350 | u64 max_hole_start; | |
1351 | u64 max_hole_size; | |
1352 | u64 extent_end; | |
0b86a832 CM |
1353 | u64 search_end = device->total_bytes; |
1354 | int ret; | |
7bfc837d | 1355 | int slot; |
0b86a832 | 1356 | struct extent_buffer *l; |
8cdc7c5b FM |
1357 | u64 min_search_start; |
1358 | ||
1359 | /* | |
1360 | * We don't want to overwrite the superblock on the drive nor any area | |
1361 | * used by the boot loader (grub for example), so we make sure to start | |
1362 | * at an offset of at least 1MB. | |
1363 | */ | |
0b246afa | 1364 | min_search_start = max(fs_info->alloc_start, 1024ull * 1024); |
8cdc7c5b | 1365 | search_start = max(search_start, min_search_start); |
0b86a832 | 1366 | |
6df9a95e JB |
1367 | path = btrfs_alloc_path(); |
1368 | if (!path) | |
1369 | return -ENOMEM; | |
f2ab7618 | 1370 | |
7bfc837d MX |
1371 | max_hole_start = search_start; |
1372 | max_hole_size = 0; | |
1373 | ||
f2ab7618 | 1374 | again: |
63a212ab | 1375 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1376 | ret = -ENOSPC; |
6df9a95e | 1377 | goto out; |
7bfc837d MX |
1378 | } |
1379 | ||
e4058b54 | 1380 | path->reada = READA_FORWARD; |
6df9a95e JB |
1381 | path->search_commit_root = 1; |
1382 | path->skip_locking = 1; | |
7bfc837d | 1383 | |
0b86a832 CM |
1384 | key.objectid = device->devid; |
1385 | key.offset = search_start; | |
1386 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1387 | |
125ccb0a | 1388 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1389 | if (ret < 0) |
7bfc837d | 1390 | goto out; |
1fcbac58 YZ |
1391 | if (ret > 0) { |
1392 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1393 | if (ret < 0) | |
7bfc837d | 1394 | goto out; |
1fcbac58 | 1395 | } |
7bfc837d | 1396 | |
0b86a832 CM |
1397 | while (1) { |
1398 | l = path->nodes[0]; | |
1399 | slot = path->slots[0]; | |
1400 | if (slot >= btrfs_header_nritems(l)) { | |
1401 | ret = btrfs_next_leaf(root, path); | |
1402 | if (ret == 0) | |
1403 | continue; | |
1404 | if (ret < 0) | |
7bfc837d MX |
1405 | goto out; |
1406 | ||
1407 | break; | |
0b86a832 CM |
1408 | } |
1409 | btrfs_item_key_to_cpu(l, &key, slot); | |
1410 | ||
1411 | if (key.objectid < device->devid) | |
1412 | goto next; | |
1413 | ||
1414 | if (key.objectid > device->devid) | |
7bfc837d | 1415 | break; |
0b86a832 | 1416 | |
962a298f | 1417 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1418 | goto next; |
9779b72f | 1419 | |
7bfc837d MX |
1420 | if (key.offset > search_start) { |
1421 | hole_size = key.offset - search_start; | |
9779b72f | 1422 | |
6df9a95e JB |
1423 | /* |
1424 | * Have to check before we set max_hole_start, otherwise | |
1425 | * we could end up sending back this offset anyway. | |
1426 | */ | |
499f377f | 1427 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1428 | &search_start, |
1b984508 FL |
1429 | hole_size)) { |
1430 | if (key.offset >= search_start) { | |
1431 | hole_size = key.offset - search_start; | |
1432 | } else { | |
1433 | WARN_ON_ONCE(1); | |
1434 | hole_size = 0; | |
1435 | } | |
1436 | } | |
6df9a95e | 1437 | |
7bfc837d MX |
1438 | if (hole_size > max_hole_size) { |
1439 | max_hole_start = search_start; | |
1440 | max_hole_size = hole_size; | |
1441 | } | |
9779b72f | 1442 | |
7bfc837d MX |
1443 | /* |
1444 | * If this free space is greater than which we need, | |
1445 | * it must be the max free space that we have found | |
1446 | * until now, so max_hole_start must point to the start | |
1447 | * of this free space and the length of this free space | |
1448 | * is stored in max_hole_size. Thus, we return | |
1449 | * max_hole_start and max_hole_size and go back to the | |
1450 | * caller. | |
1451 | */ | |
1452 | if (hole_size >= num_bytes) { | |
1453 | ret = 0; | |
1454 | goto out; | |
0b86a832 CM |
1455 | } |
1456 | } | |
0b86a832 | 1457 | |
0b86a832 | 1458 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1459 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1460 | dev_extent); | |
1461 | if (extent_end > search_start) | |
1462 | search_start = extent_end; | |
0b86a832 CM |
1463 | next: |
1464 | path->slots[0]++; | |
1465 | cond_resched(); | |
1466 | } | |
0b86a832 | 1467 | |
38c01b96 | 1468 | /* |
1469 | * At this point, search_start should be the end of | |
1470 | * allocated dev extents, and when shrinking the device, | |
1471 | * search_end may be smaller than search_start. | |
1472 | */ | |
f2ab7618 | 1473 | if (search_end > search_start) { |
38c01b96 | 1474 | hole_size = search_end - search_start; |
1475 | ||
499f377f | 1476 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1477 | hole_size)) { |
1478 | btrfs_release_path(path); | |
1479 | goto again; | |
1480 | } | |
0b86a832 | 1481 | |
f2ab7618 ZL |
1482 | if (hole_size > max_hole_size) { |
1483 | max_hole_start = search_start; | |
1484 | max_hole_size = hole_size; | |
1485 | } | |
6df9a95e JB |
1486 | } |
1487 | ||
7bfc837d | 1488 | /* See above. */ |
f2ab7618 | 1489 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1490 | ret = -ENOSPC; |
1491 | else | |
1492 | ret = 0; | |
1493 | ||
1494 | out: | |
2b82032c | 1495 | btrfs_free_path(path); |
7bfc837d | 1496 | *start = max_hole_start; |
b2117a39 | 1497 | if (len) |
7bfc837d | 1498 | *len = max_hole_size; |
0b86a832 CM |
1499 | return ret; |
1500 | } | |
1501 | ||
499f377f JM |
1502 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1503 | struct btrfs_device *device, u64 num_bytes, | |
1504 | u64 *start, u64 *len) | |
1505 | { | |
499f377f | 1506 | /* FIXME use last free of some kind */ |
499f377f | 1507 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1508 | num_bytes, 0, start, len); |
499f377f JM |
1509 | } |
1510 | ||
b2950863 | 1511 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1512 | struct btrfs_device *device, |
2196d6e8 | 1513 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1514 | { |
0b246afa JM |
1515 | struct btrfs_fs_info *fs_info = device->fs_info; |
1516 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1517 | int ret; |
1518 | struct btrfs_path *path; | |
8f18cf13 | 1519 | struct btrfs_key key; |
a061fc8d CM |
1520 | struct btrfs_key found_key; |
1521 | struct extent_buffer *leaf = NULL; | |
1522 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1523 | |
1524 | path = btrfs_alloc_path(); | |
1525 | if (!path) | |
1526 | return -ENOMEM; | |
1527 | ||
1528 | key.objectid = device->devid; | |
1529 | key.offset = start; | |
1530 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1531 | again: |
8f18cf13 | 1532 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1533 | if (ret > 0) { |
1534 | ret = btrfs_previous_item(root, path, key.objectid, | |
1535 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1536 | if (ret) |
1537 | goto out; | |
a061fc8d CM |
1538 | leaf = path->nodes[0]; |
1539 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1540 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1541 | struct btrfs_dev_extent); | |
1542 | BUG_ON(found_key.offset > start || found_key.offset + | |
1543 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1544 | key = found_key; |
1545 | btrfs_release_path(path); | |
1546 | goto again; | |
a061fc8d CM |
1547 | } else if (ret == 0) { |
1548 | leaf = path->nodes[0]; | |
1549 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1550 | struct btrfs_dev_extent); | |
79787eaa | 1551 | } else { |
0b246afa | 1552 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1553 | goto out; |
a061fc8d | 1554 | } |
8f18cf13 | 1555 | |
2196d6e8 MX |
1556 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1557 | ||
8f18cf13 | 1558 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1559 | if (ret) { |
0b246afa JM |
1560 | btrfs_handle_fs_error(fs_info, ret, |
1561 | "Failed to remove dev extent item"); | |
13212b54 | 1562 | } else { |
3204d33c | 1563 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1564 | } |
b0b802d7 | 1565 | out: |
8f18cf13 CM |
1566 | btrfs_free_path(path); |
1567 | return ret; | |
1568 | } | |
1569 | ||
48a3b636 ES |
1570 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1571 | struct btrfs_device *device, | |
1572 | u64 chunk_tree, u64 chunk_objectid, | |
1573 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1574 | { |
1575 | int ret; | |
1576 | struct btrfs_path *path; | |
0b246afa JM |
1577 | struct btrfs_fs_info *fs_info = device->fs_info; |
1578 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1579 | struct btrfs_dev_extent *extent; |
1580 | struct extent_buffer *leaf; | |
1581 | struct btrfs_key key; | |
1582 | ||
dfe25020 | 1583 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1584 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1585 | path = btrfs_alloc_path(); |
1586 | if (!path) | |
1587 | return -ENOMEM; | |
1588 | ||
0b86a832 | 1589 | key.objectid = device->devid; |
2b82032c | 1590 | key.offset = start; |
0b86a832 CM |
1591 | key.type = BTRFS_DEV_EXTENT_KEY; |
1592 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1593 | sizeof(*extent)); | |
2cdcecbc MF |
1594 | if (ret) |
1595 | goto out; | |
0b86a832 CM |
1596 | |
1597 | leaf = path->nodes[0]; | |
1598 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1599 | struct btrfs_dev_extent); | |
e17cade2 CM |
1600 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1601 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1602 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1603 | ||
0b246afa | 1604 | write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid); |
e17cade2 | 1605 | |
0b86a832 CM |
1606 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1607 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1608 | out: |
0b86a832 CM |
1609 | btrfs_free_path(path); |
1610 | return ret; | |
1611 | } | |
1612 | ||
6df9a95e | 1613 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1614 | { |
6df9a95e JB |
1615 | struct extent_map_tree *em_tree; |
1616 | struct extent_map *em; | |
1617 | struct rb_node *n; | |
1618 | u64 ret = 0; | |
0b86a832 | 1619 | |
6df9a95e JB |
1620 | em_tree = &fs_info->mapping_tree.map_tree; |
1621 | read_lock(&em_tree->lock); | |
1622 | n = rb_last(&em_tree->map); | |
1623 | if (n) { | |
1624 | em = rb_entry(n, struct extent_map, rb_node); | |
1625 | ret = em->start + em->len; | |
0b86a832 | 1626 | } |
6df9a95e JB |
1627 | read_unlock(&em_tree->lock); |
1628 | ||
0b86a832 CM |
1629 | return ret; |
1630 | } | |
1631 | ||
53f10659 ID |
1632 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1633 | u64 *devid_ret) | |
0b86a832 CM |
1634 | { |
1635 | int ret; | |
1636 | struct btrfs_key key; | |
1637 | struct btrfs_key found_key; | |
2b82032c YZ |
1638 | struct btrfs_path *path; |
1639 | ||
2b82032c YZ |
1640 | path = btrfs_alloc_path(); |
1641 | if (!path) | |
1642 | return -ENOMEM; | |
0b86a832 CM |
1643 | |
1644 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1645 | key.type = BTRFS_DEV_ITEM_KEY; | |
1646 | key.offset = (u64)-1; | |
1647 | ||
53f10659 | 1648 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1649 | if (ret < 0) |
1650 | goto error; | |
1651 | ||
79787eaa | 1652 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1653 | |
53f10659 ID |
1654 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1655 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1656 | BTRFS_DEV_ITEM_KEY); |
1657 | if (ret) { | |
53f10659 | 1658 | *devid_ret = 1; |
0b86a832 CM |
1659 | } else { |
1660 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1661 | path->slots[0]); | |
53f10659 | 1662 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1663 | } |
1664 | ret = 0; | |
1665 | error: | |
2b82032c | 1666 | btrfs_free_path(path); |
0b86a832 CM |
1667 | return ret; |
1668 | } | |
1669 | ||
1670 | /* | |
1671 | * the device information is stored in the chunk root | |
1672 | * the btrfs_device struct should be fully filled in | |
1673 | */ | |
48a3b636 | 1674 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
5b4aacef | 1675 | struct btrfs_fs_info *fs_info, |
48a3b636 | 1676 | struct btrfs_device *device) |
0b86a832 | 1677 | { |
5b4aacef | 1678 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
1679 | int ret; |
1680 | struct btrfs_path *path; | |
1681 | struct btrfs_dev_item *dev_item; | |
1682 | struct extent_buffer *leaf; | |
1683 | struct btrfs_key key; | |
1684 | unsigned long ptr; | |
0b86a832 | 1685 | |
0b86a832 CM |
1686 | path = btrfs_alloc_path(); |
1687 | if (!path) | |
1688 | return -ENOMEM; | |
1689 | ||
0b86a832 CM |
1690 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1691 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1692 | key.offset = device->devid; |
0b86a832 CM |
1693 | |
1694 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1695 | sizeof(*dev_item)); |
0b86a832 CM |
1696 | if (ret) |
1697 | goto out; | |
1698 | ||
1699 | leaf = path->nodes[0]; | |
1700 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1701 | ||
1702 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1703 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1704 | btrfs_set_device_type(leaf, dev_item, device->type); |
1705 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1706 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1707 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1708 | btrfs_set_device_total_bytes(leaf, dev_item, |
1709 | btrfs_device_get_disk_total_bytes(device)); | |
1710 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1711 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1712 | btrfs_set_device_group(leaf, dev_item, 0); |
1713 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1714 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1715 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1716 | |
410ba3a2 | 1717 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1718 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1719 | ptr = btrfs_device_fsid(dev_item); |
0b246afa | 1720 | write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1721 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1722 | |
2b82032c | 1723 | ret = 0; |
0b86a832 CM |
1724 | out: |
1725 | btrfs_free_path(path); | |
1726 | return ret; | |
1727 | } | |
8f18cf13 | 1728 | |
5a1972bd QW |
1729 | /* |
1730 | * Function to update ctime/mtime for a given device path. | |
1731 | * Mainly used for ctime/mtime based probe like libblkid. | |
1732 | */ | |
da353f6b | 1733 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1734 | { |
1735 | struct file *filp; | |
1736 | ||
1737 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1738 | if (IS_ERR(filp)) |
5a1972bd QW |
1739 | return; |
1740 | file_update_time(filp); | |
1741 | filp_close(filp, NULL); | |
5a1972bd QW |
1742 | } |
1743 | ||
5b4aacef | 1744 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1745 | struct btrfs_device *device) |
1746 | { | |
5b4aacef | 1747 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1748 | int ret; |
1749 | struct btrfs_path *path; | |
a061fc8d | 1750 | struct btrfs_key key; |
a061fc8d CM |
1751 | struct btrfs_trans_handle *trans; |
1752 | ||
a061fc8d CM |
1753 | path = btrfs_alloc_path(); |
1754 | if (!path) | |
1755 | return -ENOMEM; | |
1756 | ||
a22285a6 | 1757 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1758 | if (IS_ERR(trans)) { |
1759 | btrfs_free_path(path); | |
1760 | return PTR_ERR(trans); | |
1761 | } | |
a061fc8d CM |
1762 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1763 | key.type = BTRFS_DEV_ITEM_KEY; | |
1764 | key.offset = device->devid; | |
1765 | ||
1766 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1767 | if (ret < 0) | |
1768 | goto out; | |
1769 | ||
1770 | if (ret > 0) { | |
1771 | ret = -ENOENT; | |
1772 | goto out; | |
1773 | } | |
1774 | ||
1775 | ret = btrfs_del_item(trans, root, path); | |
1776 | if (ret) | |
1777 | goto out; | |
a061fc8d CM |
1778 | out: |
1779 | btrfs_free_path(path); | |
3a45bb20 | 1780 | btrfs_commit_transaction(trans); |
a061fc8d CM |
1781 | return ret; |
1782 | } | |
1783 | ||
3cc31a0d DS |
1784 | /* |
1785 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1786 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1787 | * replace. | |
1788 | */ | |
1789 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1790 | u64 num_devices) | |
a061fc8d | 1791 | { |
a061fc8d | 1792 | u64 all_avail; |
de98ced9 | 1793 | unsigned seq; |
418775a2 | 1794 | int i; |
a061fc8d | 1795 | |
de98ced9 | 1796 | do { |
bd45ffbc | 1797 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1798 | |
bd45ffbc AJ |
1799 | all_avail = fs_info->avail_data_alloc_bits | |
1800 | fs_info->avail_system_alloc_bits | | |
1801 | fs_info->avail_metadata_alloc_bits; | |
1802 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1803 | |
418775a2 DS |
1804 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1805 | if (!(all_avail & btrfs_raid_group[i])) | |
1806 | continue; | |
a061fc8d | 1807 | |
418775a2 DS |
1808 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1809 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1810 | |
418775a2 DS |
1811 | if (ret) |
1812 | return ret; | |
1813 | } | |
53b381b3 DW |
1814 | } |
1815 | ||
bd45ffbc | 1816 | return 0; |
f1fa7f26 AJ |
1817 | } |
1818 | ||
88acff64 AJ |
1819 | struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs, |
1820 | struct btrfs_device *device) | |
a061fc8d | 1821 | { |
2b82032c | 1822 | struct btrfs_device *next_device; |
88acff64 AJ |
1823 | |
1824 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1825 | if (next_device != device && | |
1826 | !next_device->missing && next_device->bdev) | |
1827 | return next_device; | |
1828 | } | |
1829 | ||
1830 | return NULL; | |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1835 | * and replace it with the provided or the next active device, in the context | |
1836 | * where this function called, there should be always be another device (or | |
1837 | * this_dev) which is active. | |
1838 | */ | |
1839 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1840 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1841 | { | |
1842 | struct btrfs_device *next_device; | |
1843 | ||
1844 | if (this_dev) | |
1845 | next_device = this_dev; | |
1846 | else | |
1847 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1848 | device); | |
1849 | ASSERT(next_device); | |
1850 | ||
1851 | if (fs_info->sb->s_bdev && | |
1852 | (fs_info->sb->s_bdev == device->bdev)) | |
1853 | fs_info->sb->s_bdev = next_device->bdev; | |
1854 | ||
1855 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1856 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1857 | } | |
1858 | ||
da353f6b DS |
1859 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1860 | u64 devid) | |
f1fa7f26 AJ |
1861 | { |
1862 | struct btrfs_device *device; | |
1f78160c | 1863 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1864 | u64 num_devices; |
a061fc8d | 1865 | int ret = 0; |
1f78160c | 1866 | bool clear_super = false; |
a061fc8d | 1867 | |
a061fc8d CM |
1868 | mutex_lock(&uuid_mutex); |
1869 | ||
0b246afa JM |
1870 | num_devices = fs_info->fs_devices->num_devices; |
1871 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
1872 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
8dabb742 SB |
1873 | WARN_ON(num_devices < 1); |
1874 | num_devices--; | |
1875 | } | |
0b246afa | 1876 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
8dabb742 | 1877 | |
0b246afa | 1878 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1879 | if (ret) |
a061fc8d | 1880 | goto out; |
a061fc8d | 1881 | |
2ff7e61e JM |
1882 | ret = btrfs_find_device_by_devspec(fs_info, devid, device_path, |
1883 | &device); | |
24fc572f | 1884 | if (ret) |
53b381b3 | 1885 | goto out; |
dfe25020 | 1886 | |
63a212ab | 1887 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1888 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1889 | goto out; |
63a212ab SB |
1890 | } |
1891 | ||
0b246afa | 1892 | if (device->writeable && fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1893 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1894 | goto out; |
2b82032c YZ |
1895 | } |
1896 | ||
1897 | if (device->writeable) { | |
34441361 | 1898 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1899 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1900 | device->fs_devices->rw_devices--; |
34441361 | 1901 | mutex_unlock(&fs_info->chunk_mutex); |
1f78160c | 1902 | clear_super = true; |
dfe25020 | 1903 | } |
a061fc8d | 1904 | |
d7901554 | 1905 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1906 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1907 | mutex_lock(&uuid_mutex); |
a061fc8d | 1908 | if (ret) |
9b3517e9 | 1909 | goto error_undo; |
a061fc8d | 1910 | |
63a212ab SB |
1911 | /* |
1912 | * TODO: the superblock still includes this device in its num_devices | |
1913 | * counter although write_all_supers() is not locked out. This | |
1914 | * could give a filesystem state which requires a degraded mount. | |
1915 | */ | |
0b246afa | 1916 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1917 | if (ret) |
9b3517e9 | 1918 | goto error_undo; |
a061fc8d | 1919 | |
2b82032c | 1920 | device->in_fs_metadata = 0; |
0b246afa | 1921 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1922 | |
1923 | /* | |
1924 | * the device list mutex makes sure that we don't change | |
1925 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1926 | * the device supers. Whoever is writing all supers, should |
1927 | * lock the device list mutex before getting the number of | |
1928 | * devices in the super block (super_copy). Conversely, | |
1929 | * whoever updates the number of devices in the super block | |
1930 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1931 | */ |
1f78160c XG |
1932 | |
1933 | cur_devices = device->fs_devices; | |
0b246afa | 1934 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c | 1935 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1936 | |
e4404d6e | 1937 | device->fs_devices->num_devices--; |
02db0844 | 1938 | device->fs_devices->total_devices--; |
2b82032c | 1939 | |
cd02dca5 | 1940 | if (device->missing) |
3a7d55c8 | 1941 | device->fs_devices->missing_devices--; |
cd02dca5 | 1942 | |
0b246afa | 1943 | btrfs_assign_next_active_device(fs_info, device, NULL); |
2b82032c | 1944 | |
0bfaa9c5 | 1945 | if (device->bdev) { |
e4404d6e | 1946 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1947 | /* remove sysfs entry */ |
0b246afa | 1948 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
0bfaa9c5 | 1949 | } |
99994cde | 1950 | |
0b246afa JM |
1951 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
1952 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
1953 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2b82032c | 1954 | |
cea67ab9 JM |
1955 | /* |
1956 | * at this point, the device is zero sized and detached from | |
1957 | * the devices list. All that's left is to zero out the old | |
1958 | * supers and free the device. | |
1959 | */ | |
1960 | if (device->writeable) | |
1961 | btrfs_scratch_superblocks(device->bdev, device->name->str); | |
1962 | ||
1963 | btrfs_close_bdev(device); | |
1964 | call_rcu(&device->rcu, free_device); | |
1965 | ||
1f78160c | 1966 | if (cur_devices->open_devices == 0) { |
e4404d6e | 1967 | struct btrfs_fs_devices *fs_devices; |
0b246afa | 1968 | fs_devices = fs_info->fs_devices; |
e4404d6e | 1969 | while (fs_devices) { |
8321cf25 RS |
1970 | if (fs_devices->seed == cur_devices) { |
1971 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1972 | break; |
8321cf25 | 1973 | } |
e4404d6e | 1974 | fs_devices = fs_devices->seed; |
2b82032c | 1975 | } |
1f78160c | 1976 | cur_devices->seed = NULL; |
1f78160c | 1977 | __btrfs_close_devices(cur_devices); |
1f78160c | 1978 | free_fs_devices(cur_devices); |
2b82032c YZ |
1979 | } |
1980 | ||
0b246afa JM |
1981 | fs_info->num_tolerated_disk_barrier_failures = |
1982 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
5af3e8cc | 1983 | |
a061fc8d CM |
1984 | out: |
1985 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1986 | return ret; |
24fc572f | 1987 | |
9b3517e9 ID |
1988 | error_undo: |
1989 | if (device->writeable) { | |
34441361 | 1990 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 1991 | list_add(&device->dev_alloc_list, |
0b246afa | 1992 | &fs_info->fs_devices->alloc_list); |
c3929c36 | 1993 | device->fs_devices->rw_devices++; |
34441361 | 1994 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 1995 | } |
24fc572f | 1996 | goto out; |
a061fc8d CM |
1997 | } |
1998 | ||
084b6e7c QW |
1999 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
2000 | struct btrfs_device *srcdev) | |
e93c89c1 | 2001 | { |
d51908ce AJ |
2002 | struct btrfs_fs_devices *fs_devices; |
2003 | ||
e93c89c1 | 2004 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 2005 | |
25e8e911 AJ |
2006 | /* |
2007 | * in case of fs with no seed, srcdev->fs_devices will point | |
2008 | * to fs_devices of fs_info. However when the dev being replaced is | |
2009 | * a seed dev it will point to the seed's local fs_devices. In short | |
2010 | * srcdev will have its correct fs_devices in both the cases. | |
2011 | */ | |
2012 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2013 | |
e93c89c1 SB |
2014 | list_del_rcu(&srcdev->dev_list); |
2015 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 2016 | fs_devices->num_devices--; |
82372bc8 | 2017 | if (srcdev->missing) |
d51908ce | 2018 | fs_devices->missing_devices--; |
e93c89c1 | 2019 | |
48b3b9d4 | 2020 | if (srcdev->writeable) |
82372bc8 | 2021 | fs_devices->rw_devices--; |
1357272f | 2022 | |
82372bc8 | 2023 | if (srcdev->bdev) |
d51908ce | 2024 | fs_devices->open_devices--; |
084b6e7c QW |
2025 | } |
2026 | ||
2027 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2028 | struct btrfs_device *srcdev) | |
2029 | { | |
2030 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2031 | |
48b3b9d4 AJ |
2032 | if (srcdev->writeable) { |
2033 | /* zero out the old super if it is writable */ | |
2034 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2035 | } | |
14238819 AJ |
2036 | |
2037 | btrfs_close_bdev(srcdev); | |
2038 | ||
e93c89c1 | 2039 | call_rcu(&srcdev->rcu, free_device); |
94d5f0c2 AJ |
2040 | |
2041 | /* | |
2042 | * unless fs_devices is seed fs, num_devices shouldn't go | |
2043 | * zero | |
2044 | */ | |
2045 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
2046 | ||
2047 | /* if this is no devs we rather delete the fs_devices */ | |
2048 | if (!fs_devices->num_devices) { | |
2049 | struct btrfs_fs_devices *tmp_fs_devices; | |
2050 | ||
2051 | tmp_fs_devices = fs_info->fs_devices; | |
2052 | while (tmp_fs_devices) { | |
2053 | if (tmp_fs_devices->seed == fs_devices) { | |
2054 | tmp_fs_devices->seed = fs_devices->seed; | |
2055 | break; | |
2056 | } | |
2057 | tmp_fs_devices = tmp_fs_devices->seed; | |
2058 | } | |
2059 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2060 | __btrfs_close_devices(fs_devices); |
2061 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2062 | } |
e93c89c1 SB |
2063 | } |
2064 | ||
2065 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2066 | struct btrfs_device *tgtdev) | |
2067 | { | |
67a2c45e | 2068 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2069 | WARN_ON(!tgtdev); |
2070 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2071 | |
32576040 | 2072 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2073 | |
779bf3fe | 2074 | if (tgtdev->bdev) |
e93c89c1 | 2075 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2076 | |
e93c89c1 | 2077 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2078 | |
88acff64 | 2079 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2080 | |
e93c89c1 | 2081 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2082 | |
2083 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2084 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2085 | |
2086 | /* | |
2087 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2088 | * may lead to a call to btrfs_show_devname() which will try | |
2089 | * to hold device_list_mutex. And here this device | |
2090 | * is already out of device list, so we don't have to hold | |
2091 | * the device_list_mutex lock. | |
2092 | */ | |
2093 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2094 | |
2095 | btrfs_close_bdev(tgtdev); | |
779bf3fe | 2096 | call_rcu(&tgtdev->rcu, free_device); |
e93c89c1 SB |
2097 | } |
2098 | ||
2ff7e61e | 2099 | static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2100 | const char *device_path, |
48a3b636 | 2101 | struct btrfs_device **device) |
7ba15b7d SB |
2102 | { |
2103 | int ret = 0; | |
2104 | struct btrfs_super_block *disk_super; | |
2105 | u64 devid; | |
2106 | u8 *dev_uuid; | |
2107 | struct block_device *bdev; | |
2108 | struct buffer_head *bh; | |
2109 | ||
2110 | *device = NULL; | |
2111 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2112 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2113 | if (ret) |
2114 | return ret; | |
2115 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2116 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2117 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2118 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2119 | brelse(bh); |
2120 | if (!*device) | |
2121 | ret = -ENOENT; | |
2122 | blkdev_put(bdev, FMODE_READ); | |
2123 | return ret; | |
2124 | } | |
2125 | ||
2ff7e61e | 2126 | int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2127 | const char *device_path, |
7ba15b7d SB |
2128 | struct btrfs_device **device) |
2129 | { | |
2130 | *device = NULL; | |
2131 | if (strcmp(device_path, "missing") == 0) { | |
2132 | struct list_head *devices; | |
2133 | struct btrfs_device *tmp; | |
2134 | ||
0b246afa | 2135 | devices = &fs_info->fs_devices->devices; |
7ba15b7d SB |
2136 | /* |
2137 | * It is safe to read the devices since the volume_mutex | |
2138 | * is held by the caller. | |
2139 | */ | |
2140 | list_for_each_entry(tmp, devices, dev_list) { | |
2141 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2142 | *device = tmp; | |
2143 | break; | |
2144 | } | |
2145 | } | |
2146 | ||
d74a6259 AJ |
2147 | if (!*device) |
2148 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2149 | |
2150 | return 0; | |
2151 | } else { | |
2ff7e61e | 2152 | return btrfs_find_device_by_path(fs_info, device_path, device); |
7ba15b7d SB |
2153 | } |
2154 | } | |
2155 | ||
5c5c0df0 DS |
2156 | /* |
2157 | * Lookup a device given by device id, or the path if the id is 0. | |
2158 | */ | |
2ff7e61e | 2159 | int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid, |
da353f6b DS |
2160 | const char *devpath, |
2161 | struct btrfs_device **device) | |
24e0474b AJ |
2162 | { |
2163 | int ret; | |
2164 | ||
5c5c0df0 | 2165 | if (devid) { |
24e0474b | 2166 | ret = 0; |
0b246afa | 2167 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2168 | if (!*device) |
2169 | ret = -ENOENT; | |
2170 | } else { | |
5c5c0df0 | 2171 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2172 | return -EINVAL; |
2173 | ||
2ff7e61e | 2174 | ret = btrfs_find_device_missing_or_by_path(fs_info, devpath, |
24e0474b AJ |
2175 | device); |
2176 | } | |
2177 | return ret; | |
2178 | } | |
2179 | ||
2b82032c YZ |
2180 | /* |
2181 | * does all the dirty work required for changing file system's UUID. | |
2182 | */ | |
2ff7e61e | 2183 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2184 | { |
0b246afa | 2185 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2186 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2187 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2188 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2189 | struct btrfs_device *device; |
2190 | u64 super_flags; | |
2191 | ||
2192 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2193 | if (!fs_devices->seeding) |
2b82032c YZ |
2194 | return -EINVAL; |
2195 | ||
2208a378 ID |
2196 | seed_devices = __alloc_fs_devices(); |
2197 | if (IS_ERR(seed_devices)) | |
2198 | return PTR_ERR(seed_devices); | |
2b82032c | 2199 | |
e4404d6e YZ |
2200 | old_devices = clone_fs_devices(fs_devices); |
2201 | if (IS_ERR(old_devices)) { | |
2202 | kfree(seed_devices); | |
2203 | return PTR_ERR(old_devices); | |
2b82032c | 2204 | } |
e4404d6e | 2205 | |
2b82032c YZ |
2206 | list_add(&old_devices->list, &fs_uuids); |
2207 | ||
e4404d6e YZ |
2208 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2209 | seed_devices->opened = 1; | |
2210 | INIT_LIST_HEAD(&seed_devices->devices); | |
2211 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2212 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2213 | |
0b246afa | 2214 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c XG |
2215 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2216 | synchronize_rcu); | |
2196d6e8 MX |
2217 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2218 | device->fs_devices = seed_devices; | |
c9513edb | 2219 | |
34441361 | 2220 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2221 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2222 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2223 | |
2b82032c YZ |
2224 | fs_devices->seeding = 0; |
2225 | fs_devices->num_devices = 0; | |
2226 | fs_devices->open_devices = 0; | |
69611ac8 | 2227 | fs_devices->missing_devices = 0; |
69611ac8 | 2228 | fs_devices->rotating = 0; |
e4404d6e | 2229 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2230 | |
2231 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2232 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2233 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
0b246afa | 2234 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
f7171750 | 2235 | |
2b82032c YZ |
2236 | super_flags = btrfs_super_flags(disk_super) & |
2237 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2238 | btrfs_set_super_flags(disk_super, super_flags); | |
2239 | ||
2240 | return 0; | |
2241 | } | |
2242 | ||
2243 | /* | |
01327610 | 2244 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2245 | */ |
2246 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2247 | struct btrfs_fs_info *fs_info) |
2b82032c | 2248 | { |
5b4aacef | 2249 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2250 | struct btrfs_path *path; |
2251 | struct extent_buffer *leaf; | |
2252 | struct btrfs_dev_item *dev_item; | |
2253 | struct btrfs_device *device; | |
2254 | struct btrfs_key key; | |
2255 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2256 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2257 | u64 devid; | |
2258 | int ret; | |
2259 | ||
2260 | path = btrfs_alloc_path(); | |
2261 | if (!path) | |
2262 | return -ENOMEM; | |
2263 | ||
2b82032c YZ |
2264 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2265 | key.offset = 0; | |
2266 | key.type = BTRFS_DEV_ITEM_KEY; | |
2267 | ||
2268 | while (1) { | |
2269 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2270 | if (ret < 0) | |
2271 | goto error; | |
2272 | ||
2273 | leaf = path->nodes[0]; | |
2274 | next_slot: | |
2275 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2276 | ret = btrfs_next_leaf(root, path); | |
2277 | if (ret > 0) | |
2278 | break; | |
2279 | if (ret < 0) | |
2280 | goto error; | |
2281 | leaf = path->nodes[0]; | |
2282 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2283 | btrfs_release_path(path); |
2b82032c YZ |
2284 | continue; |
2285 | } | |
2286 | ||
2287 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2288 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2289 | key.type != BTRFS_DEV_ITEM_KEY) | |
2290 | break; | |
2291 | ||
2292 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2293 | struct btrfs_dev_item); | |
2294 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2295 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2296 | BTRFS_UUID_SIZE); |
1473b24e | 2297 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2298 | BTRFS_UUID_SIZE); |
0b246afa | 2299 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2300 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2301 | |
2302 | if (device->fs_devices->seeding) { | |
2303 | btrfs_set_device_generation(leaf, dev_item, | |
2304 | device->generation); | |
2305 | btrfs_mark_buffer_dirty(leaf); | |
2306 | } | |
2307 | ||
2308 | path->slots[0]++; | |
2309 | goto next_slot; | |
2310 | } | |
2311 | ret = 0; | |
2312 | error: | |
2313 | btrfs_free_path(path); | |
2314 | return ret; | |
2315 | } | |
2316 | ||
da353f6b | 2317 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2318 | { |
5112febb | 2319 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2320 | struct request_queue *q; |
788f20eb CM |
2321 | struct btrfs_trans_handle *trans; |
2322 | struct btrfs_device *device; | |
2323 | struct block_device *bdev; | |
788f20eb | 2324 | struct list_head *devices; |
0b246afa | 2325 | struct super_block *sb = fs_info->sb; |
606686ee | 2326 | struct rcu_string *name; |
3c1dbdf5 | 2327 | u64 tmp; |
2b82032c | 2328 | int seeding_dev = 0; |
788f20eb CM |
2329 | int ret = 0; |
2330 | ||
0b246afa | 2331 | if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding) |
f8c5d0b4 | 2332 | return -EROFS; |
788f20eb | 2333 | |
a5d16333 | 2334 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2335 | fs_info->bdev_holder); |
7f59203a JB |
2336 | if (IS_ERR(bdev)) |
2337 | return PTR_ERR(bdev); | |
a2135011 | 2338 | |
0b246afa | 2339 | if (fs_info->fs_devices->seeding) { |
2b82032c YZ |
2340 | seeding_dev = 1; |
2341 | down_write(&sb->s_umount); | |
2342 | mutex_lock(&uuid_mutex); | |
2343 | } | |
2344 | ||
8c8bee1d | 2345 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2346 | |
0b246afa | 2347 | devices = &fs_info->fs_devices->devices; |
d25628bd | 2348 | |
0b246afa | 2349 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
c6e30871 | 2350 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2351 | if (device->bdev == bdev) { |
2352 | ret = -EEXIST; | |
d25628bd | 2353 | mutex_unlock( |
0b246afa | 2354 | &fs_info->fs_devices->device_list_mutex); |
2b82032c | 2355 | goto error; |
788f20eb CM |
2356 | } |
2357 | } | |
0b246afa | 2358 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2359 | |
0b246afa | 2360 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2361 | if (IS_ERR(device)) { |
788f20eb | 2362 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2363 | ret = PTR_ERR(device); |
2b82032c | 2364 | goto error; |
788f20eb CM |
2365 | } |
2366 | ||
78f2c9e6 | 2367 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2368 | if (!name) { |
788f20eb | 2369 | kfree(device); |
2b82032c YZ |
2370 | ret = -ENOMEM; |
2371 | goto error; | |
788f20eb | 2372 | } |
606686ee | 2373 | rcu_assign_pointer(device->name, name); |
2b82032c | 2374 | |
a22285a6 | 2375 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2376 | if (IS_ERR(trans)) { |
606686ee | 2377 | rcu_string_free(device->name); |
98d5dc13 TI |
2378 | kfree(device); |
2379 | ret = PTR_ERR(trans); | |
2380 | goto error; | |
2381 | } | |
2382 | ||
d5e2003c JB |
2383 | q = bdev_get_queue(bdev); |
2384 | if (blk_queue_discard(q)) | |
2385 | device->can_discard = 1; | |
2b82032c | 2386 | device->writeable = 1; |
2b82032c | 2387 | device->generation = trans->transid; |
0b246afa JM |
2388 | device->io_width = fs_info->sectorsize; |
2389 | device->io_align = fs_info->sectorsize; | |
2390 | device->sector_size = fs_info->sectorsize; | |
788f20eb | 2391 | device->total_bytes = i_size_read(bdev->bd_inode); |
2cc3c559 | 2392 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2393 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2394 | device->fs_info = fs_info; |
788f20eb | 2395 | device->bdev = bdev; |
dfe25020 | 2396 | device->in_fs_metadata = 1; |
63a212ab | 2397 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2398 | device->mode = FMODE_EXCL; |
27087f37 | 2399 | device->dev_stats_valid = 1; |
2b82032c | 2400 | set_blocksize(device->bdev, 4096); |
788f20eb | 2401 | |
2b82032c YZ |
2402 | if (seeding_dev) { |
2403 | sb->s_flags &= ~MS_RDONLY; | |
2ff7e61e | 2404 | ret = btrfs_prepare_sprout(fs_info); |
79787eaa | 2405 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2406 | } |
788f20eb | 2407 | |
0b246afa | 2408 | device->fs_devices = fs_info->fs_devices; |
e5e9a520 | 2409 | |
0b246afa | 2410 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
34441361 | 2411 | mutex_lock(&fs_info->chunk_mutex); |
0b246afa | 2412 | list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices); |
2b82032c | 2413 | list_add(&device->dev_alloc_list, |
0b246afa JM |
2414 | &fs_info->fs_devices->alloc_list); |
2415 | fs_info->fs_devices->num_devices++; | |
2416 | fs_info->fs_devices->open_devices++; | |
2417 | fs_info->fs_devices->rw_devices++; | |
2418 | fs_info->fs_devices->total_devices++; | |
2419 | fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2420 | |
0b246afa JM |
2421 | spin_lock(&fs_info->free_chunk_lock); |
2422 | fs_info->free_chunk_space += device->total_bytes; | |
2423 | spin_unlock(&fs_info->free_chunk_lock); | |
2bf64758 | 2424 | |
c289811c | 2425 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
0b246afa | 2426 | fs_info->fs_devices->rotating = 1; |
c289811c | 2427 | |
0b246afa JM |
2428 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2429 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
3c1dbdf5 | 2430 | tmp + device->total_bytes); |
788f20eb | 2431 | |
0b246afa JM |
2432 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2433 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2434 | |
2435 | /* add sysfs device entry */ | |
0b246afa | 2436 | btrfs_sysfs_add_device_link(fs_info->fs_devices, device); |
0d39376a | 2437 | |
2196d6e8 MX |
2438 | /* |
2439 | * we've got more storage, clear any full flags on the space | |
2440 | * infos | |
2441 | */ | |
0b246afa | 2442 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2443 | |
34441361 | 2444 | mutex_unlock(&fs_info->chunk_mutex); |
0b246afa | 2445 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2446 | |
2b82032c | 2447 | if (seeding_dev) { |
34441361 | 2448 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2449 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2450 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2451 | if (ret) { |
66642832 | 2452 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2453 | goto error_trans; |
005d6427 | 2454 | } |
2196d6e8 MX |
2455 | } |
2456 | ||
0b246afa | 2457 | ret = btrfs_add_device(trans, fs_info, device); |
2196d6e8 | 2458 | if (ret) { |
66642832 | 2459 | btrfs_abort_transaction(trans, ret); |
2196d6e8 MX |
2460 | goto error_trans; |
2461 | } | |
2462 | ||
2463 | if (seeding_dev) { | |
2464 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2465 | ||
0b246afa | 2466 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2467 | if (ret) { |
66642832 | 2468 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2469 | goto error_trans; |
005d6427 | 2470 | } |
b2373f25 AJ |
2471 | |
2472 | /* Sprouting would change fsid of the mounted root, | |
2473 | * so rename the fsid on the sysfs | |
2474 | */ | |
2475 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa JM |
2476 | fs_info->fsid); |
2477 | if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf)) | |
2478 | btrfs_warn(fs_info, | |
2479 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2480 | } |
2481 | ||
0b246afa JM |
2482 | fs_info->num_tolerated_disk_barrier_failures = |
2483 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3a45bb20 | 2484 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2485 | |
2b82032c YZ |
2486 | if (seeding_dev) { |
2487 | mutex_unlock(&uuid_mutex); | |
2488 | up_write(&sb->s_umount); | |
788f20eb | 2489 | |
79787eaa JM |
2490 | if (ret) /* transaction commit */ |
2491 | return ret; | |
2492 | ||
2ff7e61e | 2493 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2494 | if (ret < 0) |
0b246afa | 2495 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2496 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2497 | trans = btrfs_attach_transaction(root); |
2498 | if (IS_ERR(trans)) { | |
2499 | if (PTR_ERR(trans) == -ENOENT) | |
2500 | return 0; | |
2501 | return PTR_ERR(trans); | |
2502 | } | |
3a45bb20 | 2503 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2504 | } |
c9e9f97b | 2505 | |
5a1972bd QW |
2506 | /* Update ctime/mtime for libblkid */ |
2507 | update_dev_time(device_path); | |
2b82032c | 2508 | return ret; |
79787eaa JM |
2509 | |
2510 | error_trans: | |
3a45bb20 | 2511 | btrfs_end_transaction(trans); |
606686ee | 2512 | rcu_string_free(device->name); |
0b246afa | 2513 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
79787eaa | 2514 | kfree(device); |
2b82032c | 2515 | error: |
e525fd89 | 2516 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2517 | if (seeding_dev) { |
2518 | mutex_unlock(&uuid_mutex); | |
2519 | up_write(&sb->s_umount); | |
2520 | } | |
c9e9f97b | 2521 | return ret; |
788f20eb CM |
2522 | } |
2523 | ||
2ff7e61e | 2524 | int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, |
da353f6b | 2525 | const char *device_path, |
1c43366d | 2526 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2527 | struct btrfs_device **device_out) |
2528 | { | |
2529 | struct request_queue *q; | |
2530 | struct btrfs_device *device; | |
2531 | struct block_device *bdev; | |
e93c89c1 SB |
2532 | struct list_head *devices; |
2533 | struct rcu_string *name; | |
12bd2fc0 | 2534 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2535 | int ret = 0; |
2536 | ||
2537 | *device_out = NULL; | |
1c43366d MX |
2538 | if (fs_info->fs_devices->seeding) { |
2539 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2540 | return -EINVAL; |
1c43366d | 2541 | } |
e93c89c1 SB |
2542 | |
2543 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2544 | fs_info->bdev_holder); | |
1c43366d MX |
2545 | if (IS_ERR(bdev)) { |
2546 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2547 | return PTR_ERR(bdev); |
1c43366d | 2548 | } |
e93c89c1 SB |
2549 | |
2550 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2551 | ||
2552 | devices = &fs_info->fs_devices->devices; | |
2553 | list_for_each_entry(device, devices, dev_list) { | |
2554 | if (device->bdev == bdev) { | |
5d163e0e JM |
2555 | btrfs_err(fs_info, |
2556 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2557 | ret = -EEXIST; |
2558 | goto error; | |
2559 | } | |
2560 | } | |
2561 | ||
1c43366d | 2562 | |
7cc8e58d MX |
2563 | if (i_size_read(bdev->bd_inode) < |
2564 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2565 | btrfs_err(fs_info, |
2566 | "target device is smaller than source device!"); | |
1c43366d MX |
2567 | ret = -EINVAL; |
2568 | goto error; | |
2569 | } | |
2570 | ||
2571 | ||
12bd2fc0 ID |
2572 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2573 | if (IS_ERR(device)) { | |
2574 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2575 | goto error; |
2576 | } | |
2577 | ||
2578 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2579 | if (!name) { | |
2580 | kfree(device); | |
2581 | ret = -ENOMEM; | |
2582 | goto error; | |
2583 | } | |
2584 | rcu_assign_pointer(device->name, name); | |
2585 | ||
2586 | q = bdev_get_queue(bdev); | |
2587 | if (blk_queue_discard(q)) | |
2588 | device->can_discard = 1; | |
0b246afa | 2589 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 | 2590 | device->writeable = 1; |
e93c89c1 | 2591 | device->generation = 0; |
0b246afa JM |
2592 | device->io_width = fs_info->sectorsize; |
2593 | device->io_align = fs_info->sectorsize; | |
2594 | device->sector_size = fs_info->sectorsize; | |
7cc8e58d MX |
2595 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2596 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2597 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2598 | ASSERT(list_empty(&srcdev->resized_list)); |
2599 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2600 | device->commit_bytes_used = device->bytes_used; |
fb456252 | 2601 | device->fs_info = fs_info; |
e93c89c1 SB |
2602 | device->bdev = bdev; |
2603 | device->in_fs_metadata = 1; | |
2604 | device->is_tgtdev_for_dev_replace = 1; | |
2605 | device->mode = FMODE_EXCL; | |
27087f37 | 2606 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2607 | set_blocksize(device->bdev, 4096); |
2608 | device->fs_devices = fs_info->fs_devices; | |
2609 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2610 | fs_info->fs_devices->num_devices++; | |
2611 | fs_info->fs_devices->open_devices++; | |
0b246afa | 2612 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 SB |
2613 | |
2614 | *device_out = device; | |
2615 | return ret; | |
2616 | ||
2617 | error: | |
2618 | blkdev_put(bdev, FMODE_EXCL); | |
2619 | return ret; | |
2620 | } | |
2621 | ||
2622 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2623 | struct btrfs_device *tgtdev) | |
2624 | { | |
da17066c JM |
2625 | u32 sectorsize = fs_info->sectorsize; |
2626 | ||
e93c89c1 | 2627 | WARN_ON(fs_info->fs_devices->rw_devices == 0); |
da17066c JM |
2628 | tgtdev->io_width = sectorsize; |
2629 | tgtdev->io_align = sectorsize; | |
2630 | tgtdev->sector_size = sectorsize; | |
fb456252 | 2631 | tgtdev->fs_info = fs_info; |
e93c89c1 SB |
2632 | tgtdev->in_fs_metadata = 1; |
2633 | } | |
2634 | ||
d397712b CM |
2635 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2636 | struct btrfs_device *device) | |
0b86a832 CM |
2637 | { |
2638 | int ret; | |
2639 | struct btrfs_path *path; | |
0b246afa | 2640 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2641 | struct btrfs_dev_item *dev_item; |
2642 | struct extent_buffer *leaf; | |
2643 | struct btrfs_key key; | |
2644 | ||
0b86a832 CM |
2645 | path = btrfs_alloc_path(); |
2646 | if (!path) | |
2647 | return -ENOMEM; | |
2648 | ||
2649 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2650 | key.type = BTRFS_DEV_ITEM_KEY; | |
2651 | key.offset = device->devid; | |
2652 | ||
2653 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2654 | if (ret < 0) | |
2655 | goto out; | |
2656 | ||
2657 | if (ret > 0) { | |
2658 | ret = -ENOENT; | |
2659 | goto out; | |
2660 | } | |
2661 | ||
2662 | leaf = path->nodes[0]; | |
2663 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2664 | ||
2665 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2666 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2667 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2668 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2669 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2670 | btrfs_set_device_total_bytes(leaf, dev_item, |
2671 | btrfs_device_get_disk_total_bytes(device)); | |
2672 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2673 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2674 | btrfs_mark_buffer_dirty(leaf); |
2675 | ||
2676 | out: | |
2677 | btrfs_free_path(path); | |
2678 | return ret; | |
2679 | } | |
2680 | ||
2196d6e8 | 2681 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2682 | struct btrfs_device *device, u64 new_size) |
2683 | { | |
0b246afa JM |
2684 | struct btrfs_fs_info *fs_info = device->fs_info; |
2685 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2686 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2687 | u64 old_total; |
2688 | u64 diff; | |
8f18cf13 | 2689 | |
2b82032c YZ |
2690 | if (!device->writeable) |
2691 | return -EACCES; | |
2196d6e8 | 2692 | |
34441361 | 2693 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2694 | old_total = btrfs_super_total_bytes(super_copy); |
2695 | diff = new_size - device->total_bytes; | |
2696 | ||
63a212ab | 2697 | if (new_size <= device->total_bytes || |
2196d6e8 | 2698 | device->is_tgtdev_for_dev_replace) { |
34441361 | 2699 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2700 | return -EINVAL; |
2196d6e8 | 2701 | } |
2b82032c | 2702 | |
0b246afa | 2703 | fs_devices = fs_info->fs_devices; |
2b82032c | 2704 | |
8f18cf13 | 2705 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2706 | device->fs_devices->total_rw_bytes += diff; |
2707 | ||
7cc8e58d MX |
2708 | btrfs_device_set_total_bytes(device, new_size); |
2709 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2710 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2711 | if (list_empty(&device->resized_list)) |
2712 | list_add_tail(&device->resized_list, | |
2713 | &fs_devices->resized_devices); | |
34441361 | 2714 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2715 | |
8f18cf13 CM |
2716 | return btrfs_update_device(trans, device); |
2717 | } | |
2718 | ||
2719 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
5b4aacef | 2720 | struct btrfs_fs_info *fs_info, u64 chunk_objectid, |
8f18cf13 CM |
2721 | u64 chunk_offset) |
2722 | { | |
5b4aacef | 2723 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2724 | int ret; |
2725 | struct btrfs_path *path; | |
2726 | struct btrfs_key key; | |
2727 | ||
8f18cf13 CM |
2728 | path = btrfs_alloc_path(); |
2729 | if (!path) | |
2730 | return -ENOMEM; | |
2731 | ||
2732 | key.objectid = chunk_objectid; | |
2733 | key.offset = chunk_offset; | |
2734 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2735 | ||
2736 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2737 | if (ret < 0) |
2738 | goto out; | |
2739 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2740 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2741 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2742 | ret = -ENOENT; |
2743 | goto out; | |
2744 | } | |
8f18cf13 CM |
2745 | |
2746 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2747 | if (ret < 0) |
0b246afa JM |
2748 | btrfs_handle_fs_error(fs_info, ret, |
2749 | "Failed to delete chunk item."); | |
79787eaa | 2750 | out: |
8f18cf13 | 2751 | btrfs_free_path(path); |
65a246c5 | 2752 | return ret; |
8f18cf13 CM |
2753 | } |
2754 | ||
6bccf3ab JM |
2755 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, |
2756 | u64 chunk_objectid, u64 chunk_offset) | |
8f18cf13 | 2757 | { |
0b246afa | 2758 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2759 | struct btrfs_disk_key *disk_key; |
2760 | struct btrfs_chunk *chunk; | |
2761 | u8 *ptr; | |
2762 | int ret = 0; | |
2763 | u32 num_stripes; | |
2764 | u32 array_size; | |
2765 | u32 len = 0; | |
2766 | u32 cur; | |
2767 | struct btrfs_key key; | |
2768 | ||
34441361 | 2769 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2770 | array_size = btrfs_super_sys_array_size(super_copy); |
2771 | ||
2772 | ptr = super_copy->sys_chunk_array; | |
2773 | cur = 0; | |
2774 | ||
2775 | while (cur < array_size) { | |
2776 | disk_key = (struct btrfs_disk_key *)ptr; | |
2777 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2778 | ||
2779 | len = sizeof(*disk_key); | |
2780 | ||
2781 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2782 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2783 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2784 | len += btrfs_chunk_item_size(num_stripes); | |
2785 | } else { | |
2786 | ret = -EIO; | |
2787 | break; | |
2788 | } | |
2789 | if (key.objectid == chunk_objectid && | |
2790 | key.offset == chunk_offset) { | |
2791 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2792 | array_size -= len; | |
2793 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2794 | } else { | |
2795 | ptr += len; | |
2796 | cur += len; | |
2797 | } | |
2798 | } | |
34441361 | 2799 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2800 | return ret; |
2801 | } | |
2802 | ||
592d92ee LB |
2803 | static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info, |
2804 | u64 logical, u64 length) | |
2805 | { | |
2806 | struct extent_map_tree *em_tree; | |
2807 | struct extent_map *em; | |
2808 | ||
2809 | em_tree = &fs_info->mapping_tree.map_tree; | |
2810 | read_lock(&em_tree->lock); | |
2811 | em = lookup_extent_mapping(em_tree, logical, length); | |
2812 | read_unlock(&em_tree->lock); | |
2813 | ||
2814 | if (!em) { | |
2815 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2816 | logical, length); | |
2817 | return ERR_PTR(-EINVAL); | |
2818 | } | |
2819 | ||
2820 | if (em->start > logical || em->start + em->len < logical) { | |
2821 | btrfs_crit(fs_info, | |
2822 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2823 | logical, length, em->start, em->start + em->len); | |
2824 | free_extent_map(em); | |
2825 | return ERR_PTR(-EINVAL); | |
2826 | } | |
2827 | ||
2828 | /* callers are responsible for dropping em's ref. */ | |
2829 | return em; | |
2830 | } | |
2831 | ||
47ab2a6c | 2832 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
5b4aacef | 2833 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2834 | { |
8f18cf13 CM |
2835 | struct extent_map *em; |
2836 | struct map_lookup *map; | |
2196d6e8 | 2837 | u64 dev_extent_len = 0; |
47ab2a6c | 2838 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2839 | int i, ret = 0; |
0b246afa | 2840 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2841 | |
592d92ee LB |
2842 | em = get_chunk_map(fs_info, chunk_offset, 1); |
2843 | if (IS_ERR(em)) { | |
47ab2a6c JB |
2844 | /* |
2845 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2846 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2847 | * do anything we still error out. |
2848 | */ | |
2849 | ASSERT(0); | |
592d92ee | 2850 | return PTR_ERR(em); |
47ab2a6c | 2851 | } |
95617d69 | 2852 | map = em->map_lookup; |
34441361 | 2853 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 2854 | check_system_chunk(trans, fs_info, map->type); |
34441361 | 2855 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2856 | |
57ba4cb8 FM |
2857 | /* |
2858 | * Take the device list mutex to prevent races with the final phase of | |
2859 | * a device replace operation that replaces the device object associated | |
2860 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2861 | */ | |
2862 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2863 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2864 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2865 | ret = btrfs_free_dev_extent(trans, device, |
2866 | map->stripes[i].physical, | |
2867 | &dev_extent_len); | |
47ab2a6c | 2868 | if (ret) { |
57ba4cb8 | 2869 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2870 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2871 | goto out; |
2872 | } | |
a061fc8d | 2873 | |
2196d6e8 | 2874 | if (device->bytes_used > 0) { |
34441361 | 2875 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2876 | btrfs_device_set_bytes_used(device, |
2877 | device->bytes_used - dev_extent_len); | |
0b246afa JM |
2878 | spin_lock(&fs_info->free_chunk_lock); |
2879 | fs_info->free_chunk_space += dev_extent_len; | |
2880 | spin_unlock(&fs_info->free_chunk_lock); | |
2881 | btrfs_clear_space_info_full(fs_info); | |
34441361 | 2882 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2883 | } |
a061fc8d | 2884 | |
dfe25020 CM |
2885 | if (map->stripes[i].dev) { |
2886 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2887 | if (ret) { |
57ba4cb8 | 2888 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2889 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2890 | goto out; |
2891 | } | |
dfe25020 | 2892 | } |
8f18cf13 | 2893 | } |
57ba4cb8 FM |
2894 | mutex_unlock(&fs_devices->device_list_mutex); |
2895 | ||
0b246afa | 2896 | ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset); |
47ab2a6c | 2897 | if (ret) { |
66642832 | 2898 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2899 | goto out; |
2900 | } | |
8f18cf13 | 2901 | |
6bccf3ab | 2902 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2903 | |
8f18cf13 | 2904 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
6bccf3ab JM |
2905 | ret = btrfs_del_sys_chunk(fs_info, chunk_objectid, |
2906 | chunk_offset); | |
47ab2a6c | 2907 | if (ret) { |
66642832 | 2908 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2909 | goto out; |
2910 | } | |
8f18cf13 CM |
2911 | } |
2912 | ||
6bccf3ab | 2913 | ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em); |
47ab2a6c | 2914 | if (ret) { |
66642832 | 2915 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2916 | goto out; |
2917 | } | |
2b82032c | 2918 | |
47ab2a6c | 2919 | out: |
2b82032c YZ |
2920 | /* once for us */ |
2921 | free_extent_map(em); | |
47ab2a6c JB |
2922 | return ret; |
2923 | } | |
2b82032c | 2924 | |
5b4aacef | 2925 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2926 | { |
5b4aacef | 2927 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2928 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2929 | int ret; |
2b82032c | 2930 | |
67c5e7d4 FM |
2931 | /* |
2932 | * Prevent races with automatic removal of unused block groups. | |
2933 | * After we relocate and before we remove the chunk with offset | |
2934 | * chunk_offset, automatic removal of the block group can kick in, | |
2935 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2936 | * | |
2937 | * Make sure to acquire this mutex before doing a tree search (dev | |
2938 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2939 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2940 | * we release the path used to search the chunk/dev tree and before | |
2941 | * the current task acquires this mutex and calls us. | |
2942 | */ | |
0b246afa | 2943 | ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex)); |
67c5e7d4 | 2944 | |
0b246afa | 2945 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2946 | if (ret) |
2947 | return -ENOSPC; | |
2948 | ||
2949 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 2950 | btrfs_scrub_pause(fs_info); |
0b246afa | 2951 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2952 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2953 | if (ret) |
2954 | return ret; | |
2955 | ||
19c4d2f9 CM |
2956 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2957 | chunk_offset); | |
2958 | if (IS_ERR(trans)) { | |
2959 | ret = PTR_ERR(trans); | |
2960 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2961 | return ret; | |
2962 | } | |
2963 | ||
47ab2a6c | 2964 | /* |
19c4d2f9 CM |
2965 | * step two, delete the device extents and the |
2966 | * chunk tree entries | |
47ab2a6c | 2967 | */ |
5b4aacef | 2968 | ret = btrfs_remove_chunk(trans, fs_info, chunk_offset); |
3a45bb20 | 2969 | btrfs_end_transaction(trans); |
19c4d2f9 | 2970 | return ret; |
2b82032c YZ |
2971 | } |
2972 | ||
2ff7e61e | 2973 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2974 | { |
0b246afa | 2975 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2976 | struct btrfs_path *path; |
2977 | struct extent_buffer *leaf; | |
2978 | struct btrfs_chunk *chunk; | |
2979 | struct btrfs_key key; | |
2980 | struct btrfs_key found_key; | |
2b82032c | 2981 | u64 chunk_type; |
ba1bf481 JB |
2982 | bool retried = false; |
2983 | int failed = 0; | |
2b82032c YZ |
2984 | int ret; |
2985 | ||
2986 | path = btrfs_alloc_path(); | |
2987 | if (!path) | |
2988 | return -ENOMEM; | |
2989 | ||
ba1bf481 | 2990 | again: |
2b82032c YZ |
2991 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2992 | key.offset = (u64)-1; | |
2993 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2994 | ||
2995 | while (1) { | |
0b246afa | 2996 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2997 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2998 | if (ret < 0) { |
0b246afa | 2999 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3000 | goto error; |
67c5e7d4 | 3001 | } |
79787eaa | 3002 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3003 | |
3004 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3005 | key.type); | |
67c5e7d4 | 3006 | if (ret) |
0b246afa | 3007 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3008 | if (ret < 0) |
3009 | goto error; | |
3010 | if (ret > 0) | |
3011 | break; | |
1a40e23b | 3012 | |
2b82032c YZ |
3013 | leaf = path->nodes[0]; |
3014 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3015 | |
2b82032c YZ |
3016 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3017 | struct btrfs_chunk); | |
3018 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3019 | btrfs_release_path(path); |
8f18cf13 | 3020 | |
2b82032c | 3021 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3022 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3023 | if (ret == -ENOSPC) |
3024 | failed++; | |
14586651 HS |
3025 | else |
3026 | BUG_ON(ret); | |
2b82032c | 3027 | } |
0b246afa | 3028 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3029 | |
2b82032c YZ |
3030 | if (found_key.offset == 0) |
3031 | break; | |
3032 | key.offset = found_key.offset - 1; | |
3033 | } | |
3034 | ret = 0; | |
ba1bf481 JB |
3035 | if (failed && !retried) { |
3036 | failed = 0; | |
3037 | retried = true; | |
3038 | goto again; | |
fae7f21c | 3039 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3040 | ret = -ENOSPC; |
3041 | } | |
2b82032c YZ |
3042 | error: |
3043 | btrfs_free_path(path); | |
3044 | return ret; | |
8f18cf13 CM |
3045 | } |
3046 | ||
6bccf3ab | 3047 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3048 | struct btrfs_balance_control *bctl) |
3049 | { | |
6bccf3ab | 3050 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3051 | struct btrfs_trans_handle *trans; |
3052 | struct btrfs_balance_item *item; | |
3053 | struct btrfs_disk_balance_args disk_bargs; | |
3054 | struct btrfs_path *path; | |
3055 | struct extent_buffer *leaf; | |
3056 | struct btrfs_key key; | |
3057 | int ret, err; | |
3058 | ||
3059 | path = btrfs_alloc_path(); | |
3060 | if (!path) | |
3061 | return -ENOMEM; | |
3062 | ||
3063 | trans = btrfs_start_transaction(root, 0); | |
3064 | if (IS_ERR(trans)) { | |
3065 | btrfs_free_path(path); | |
3066 | return PTR_ERR(trans); | |
3067 | } | |
3068 | ||
3069 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3070 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3071 | key.offset = 0; |
3072 | ||
3073 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3074 | sizeof(*item)); | |
3075 | if (ret) | |
3076 | goto out; | |
3077 | ||
3078 | leaf = path->nodes[0]; | |
3079 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3080 | ||
b159fa28 | 3081 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3082 | |
3083 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3084 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3085 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3086 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3087 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3088 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3089 | ||
3090 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3091 | ||
3092 | btrfs_mark_buffer_dirty(leaf); | |
3093 | out: | |
3094 | btrfs_free_path(path); | |
3a45bb20 | 3095 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3096 | if (err && !ret) |
3097 | ret = err; | |
3098 | return ret; | |
3099 | } | |
3100 | ||
6bccf3ab | 3101 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3102 | { |
6bccf3ab | 3103 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3104 | struct btrfs_trans_handle *trans; |
3105 | struct btrfs_path *path; | |
3106 | struct btrfs_key key; | |
3107 | int ret, err; | |
3108 | ||
3109 | path = btrfs_alloc_path(); | |
3110 | if (!path) | |
3111 | return -ENOMEM; | |
3112 | ||
3113 | trans = btrfs_start_transaction(root, 0); | |
3114 | if (IS_ERR(trans)) { | |
3115 | btrfs_free_path(path); | |
3116 | return PTR_ERR(trans); | |
3117 | } | |
3118 | ||
3119 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3120 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3121 | key.offset = 0; |
3122 | ||
3123 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3124 | if (ret < 0) | |
3125 | goto out; | |
3126 | if (ret > 0) { | |
3127 | ret = -ENOENT; | |
3128 | goto out; | |
3129 | } | |
3130 | ||
3131 | ret = btrfs_del_item(trans, root, path); | |
3132 | out: | |
3133 | btrfs_free_path(path); | |
3a45bb20 | 3134 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3135 | if (err && !ret) |
3136 | ret = err; | |
3137 | return ret; | |
3138 | } | |
3139 | ||
59641015 ID |
3140 | /* |
3141 | * This is a heuristic used to reduce the number of chunks balanced on | |
3142 | * resume after balance was interrupted. | |
3143 | */ | |
3144 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3145 | { | |
3146 | /* | |
3147 | * Turn on soft mode for chunk types that were being converted. | |
3148 | */ | |
3149 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3150 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3151 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3152 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3153 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3154 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3155 | ||
3156 | /* | |
3157 | * Turn on usage filter if is not already used. The idea is | |
3158 | * that chunks that we have already balanced should be | |
3159 | * reasonably full. Don't do it for chunks that are being | |
3160 | * converted - that will keep us from relocating unconverted | |
3161 | * (albeit full) chunks. | |
3162 | */ | |
3163 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3164 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3165 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3166 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3167 | bctl->data.usage = 90; | |
3168 | } | |
3169 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3170 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3171 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3172 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3173 | bctl->sys.usage = 90; | |
3174 | } | |
3175 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3176 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3177 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3178 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3179 | bctl->meta.usage = 90; | |
3180 | } | |
3181 | } | |
3182 | ||
c9e9f97b ID |
3183 | /* |
3184 | * Should be called with both balance and volume mutexes held to | |
3185 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3186 | * restriper. Same goes for unset_balance_control. | |
3187 | */ | |
3188 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3189 | { | |
3190 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3191 | ||
3192 | BUG_ON(fs_info->balance_ctl); | |
3193 | ||
3194 | spin_lock(&fs_info->balance_lock); | |
3195 | fs_info->balance_ctl = bctl; | |
3196 | spin_unlock(&fs_info->balance_lock); | |
3197 | } | |
3198 | ||
3199 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3200 | { | |
3201 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3202 | ||
3203 | BUG_ON(!fs_info->balance_ctl); | |
3204 | ||
3205 | spin_lock(&fs_info->balance_lock); | |
3206 | fs_info->balance_ctl = NULL; | |
3207 | spin_unlock(&fs_info->balance_lock); | |
3208 | ||
3209 | kfree(bctl); | |
3210 | } | |
3211 | ||
ed25e9b2 ID |
3212 | /* |
3213 | * Balance filters. Return 1 if chunk should be filtered out | |
3214 | * (should not be balanced). | |
3215 | */ | |
899c81ea | 3216 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3217 | struct btrfs_balance_args *bargs) |
3218 | { | |
899c81ea ID |
3219 | chunk_type = chunk_to_extended(chunk_type) & |
3220 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3221 | |
899c81ea | 3222 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3223 | return 0; |
3224 | ||
3225 | return 1; | |
3226 | } | |
3227 | ||
dba72cb3 | 3228 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3229 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3230 | { |
3231 | struct btrfs_block_group_cache *cache; | |
3232 | u64 chunk_used; | |
3233 | u64 user_thresh_min; | |
3234 | u64 user_thresh_max; | |
3235 | int ret = 1; | |
3236 | ||
3237 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3238 | chunk_used = btrfs_block_group_used(&cache->item); | |
3239 | ||
3240 | if (bargs->usage_min == 0) | |
3241 | user_thresh_min = 0; | |
3242 | else | |
3243 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3244 | bargs->usage_min); | |
3245 | ||
3246 | if (bargs->usage_max == 0) | |
3247 | user_thresh_max = 1; | |
3248 | else if (bargs->usage_max > 100) | |
3249 | user_thresh_max = cache->key.offset; | |
3250 | else | |
3251 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3252 | bargs->usage_max); | |
3253 | ||
3254 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3255 | ret = 0; | |
3256 | ||
3257 | btrfs_put_block_group(cache); | |
3258 | return ret; | |
3259 | } | |
3260 | ||
dba72cb3 | 3261 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3262 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3263 | { |
3264 | struct btrfs_block_group_cache *cache; | |
3265 | u64 chunk_used, user_thresh; | |
3266 | int ret = 1; | |
3267 | ||
3268 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3269 | chunk_used = btrfs_block_group_used(&cache->item); | |
3270 | ||
bc309467 | 3271 | if (bargs->usage_min == 0) |
3e39cea6 | 3272 | user_thresh = 1; |
a105bb88 ID |
3273 | else if (bargs->usage > 100) |
3274 | user_thresh = cache->key.offset; | |
3275 | else | |
3276 | user_thresh = div_factor_fine(cache->key.offset, | |
3277 | bargs->usage); | |
3278 | ||
5ce5b3c0 ID |
3279 | if (chunk_used < user_thresh) |
3280 | ret = 0; | |
3281 | ||
3282 | btrfs_put_block_group(cache); | |
3283 | return ret; | |
3284 | } | |
3285 | ||
409d404b ID |
3286 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3287 | struct btrfs_chunk *chunk, | |
3288 | struct btrfs_balance_args *bargs) | |
3289 | { | |
3290 | struct btrfs_stripe *stripe; | |
3291 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3292 | int i; | |
3293 | ||
3294 | for (i = 0; i < num_stripes; i++) { | |
3295 | stripe = btrfs_stripe_nr(chunk, i); | |
3296 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3297 | return 0; | |
3298 | } | |
3299 | ||
3300 | return 1; | |
3301 | } | |
3302 | ||
94e60d5a ID |
3303 | /* [pstart, pend) */ |
3304 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3305 | struct btrfs_chunk *chunk, | |
3306 | u64 chunk_offset, | |
3307 | struct btrfs_balance_args *bargs) | |
3308 | { | |
3309 | struct btrfs_stripe *stripe; | |
3310 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3311 | u64 stripe_offset; | |
3312 | u64 stripe_length; | |
3313 | int factor; | |
3314 | int i; | |
3315 | ||
3316 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3317 | return 0; | |
3318 | ||
3319 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3320 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3321 | factor = num_stripes / 2; | |
3322 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3323 | factor = num_stripes - 1; | |
3324 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3325 | factor = num_stripes - 2; | |
3326 | } else { | |
3327 | factor = num_stripes; | |
3328 | } | |
94e60d5a ID |
3329 | |
3330 | for (i = 0; i < num_stripes; i++) { | |
3331 | stripe = btrfs_stripe_nr(chunk, i); | |
3332 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3333 | continue; | |
3334 | ||
3335 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3336 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3337 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3338 | |
3339 | if (stripe_offset < bargs->pend && | |
3340 | stripe_offset + stripe_length > bargs->pstart) | |
3341 | return 0; | |
3342 | } | |
3343 | ||
3344 | return 1; | |
3345 | } | |
3346 | ||
ea67176a ID |
3347 | /* [vstart, vend) */ |
3348 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3349 | struct btrfs_chunk *chunk, | |
3350 | u64 chunk_offset, | |
3351 | struct btrfs_balance_args *bargs) | |
3352 | { | |
3353 | if (chunk_offset < bargs->vend && | |
3354 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3355 | /* at least part of the chunk is inside this vrange */ | |
3356 | return 0; | |
3357 | ||
3358 | return 1; | |
3359 | } | |
3360 | ||
dee32d0a GAP |
3361 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3362 | struct btrfs_chunk *chunk, | |
3363 | struct btrfs_balance_args *bargs) | |
3364 | { | |
3365 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3366 | ||
3367 | if (bargs->stripes_min <= num_stripes | |
3368 | && num_stripes <= bargs->stripes_max) | |
3369 | return 0; | |
3370 | ||
3371 | return 1; | |
3372 | } | |
3373 | ||
899c81ea | 3374 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3375 | struct btrfs_balance_args *bargs) |
3376 | { | |
3377 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3378 | return 0; | |
3379 | ||
899c81ea ID |
3380 | chunk_type = chunk_to_extended(chunk_type) & |
3381 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3382 | |
899c81ea | 3383 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3384 | return 1; |
3385 | ||
3386 | return 0; | |
3387 | } | |
3388 | ||
2ff7e61e | 3389 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3390 | struct extent_buffer *leaf, |
3391 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3392 | { | |
0b246afa | 3393 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3394 | struct btrfs_balance_args *bargs = NULL; |
3395 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3396 | ||
3397 | /* type filter */ | |
3398 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3399 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3400 | return 0; | |
3401 | } | |
3402 | ||
3403 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3404 | bargs = &bctl->data; | |
3405 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3406 | bargs = &bctl->sys; | |
3407 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3408 | bargs = &bctl->meta; | |
3409 | ||
ed25e9b2 ID |
3410 | /* profiles filter */ |
3411 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3412 | chunk_profiles_filter(chunk_type, bargs)) { | |
3413 | return 0; | |
5ce5b3c0 ID |
3414 | } |
3415 | ||
3416 | /* usage filter */ | |
3417 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3418 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3419 | return 0; |
bc309467 | 3420 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3421 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3422 | return 0; |
409d404b ID |
3423 | } |
3424 | ||
3425 | /* devid filter */ | |
3426 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3427 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3428 | return 0; | |
94e60d5a ID |
3429 | } |
3430 | ||
3431 | /* drange filter, makes sense only with devid filter */ | |
3432 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3433 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3434 | return 0; | |
ea67176a ID |
3435 | } |
3436 | ||
3437 | /* vrange filter */ | |
3438 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3439 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3440 | return 0; | |
ed25e9b2 ID |
3441 | } |
3442 | ||
dee32d0a GAP |
3443 | /* stripes filter */ |
3444 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3445 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3446 | return 0; | |
3447 | } | |
3448 | ||
cfa4c961 ID |
3449 | /* soft profile changing mode */ |
3450 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3451 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3452 | return 0; | |
3453 | } | |
3454 | ||
7d824b6f DS |
3455 | /* |
3456 | * limited by count, must be the last filter | |
3457 | */ | |
3458 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3459 | if (bargs->limit == 0) | |
3460 | return 0; | |
3461 | else | |
3462 | bargs->limit--; | |
12907fc7 DS |
3463 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3464 | /* | |
3465 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3466 | * determined here because we do not have the global information |
12907fc7 DS |
3467 | * about the count of all chunks that satisfy the filters. |
3468 | */ | |
3469 | if (bargs->limit_max == 0) | |
3470 | return 0; | |
3471 | else | |
3472 | bargs->limit_max--; | |
7d824b6f DS |
3473 | } |
3474 | ||
f43ffb60 ID |
3475 | return 1; |
3476 | } | |
3477 | ||
c9e9f97b | 3478 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3479 | { |
19a39dce | 3480 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3481 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3482 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3483 | struct list_head *devices; | |
ec44a35c CM |
3484 | struct btrfs_device *device; |
3485 | u64 old_size; | |
3486 | u64 size_to_free; | |
12907fc7 | 3487 | u64 chunk_type; |
f43ffb60 | 3488 | struct btrfs_chunk *chunk; |
5a488b9d | 3489 | struct btrfs_path *path = NULL; |
ec44a35c | 3490 | struct btrfs_key key; |
ec44a35c | 3491 | struct btrfs_key found_key; |
c9e9f97b | 3492 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3493 | struct extent_buffer *leaf; |
3494 | int slot; | |
c9e9f97b ID |
3495 | int ret; |
3496 | int enospc_errors = 0; | |
19a39dce | 3497 | bool counting = true; |
12907fc7 | 3498 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3499 | u64 limit_data = bctl->data.limit; |
3500 | u64 limit_meta = bctl->meta.limit; | |
3501 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3502 | u32 count_data = 0; |
3503 | u32 count_meta = 0; | |
3504 | u32 count_sys = 0; | |
2c9fe835 | 3505 | int chunk_reserved = 0; |
cf25ce51 | 3506 | u64 bytes_used = 0; |
ec44a35c | 3507 | |
ec44a35c | 3508 | /* step one make some room on all the devices */ |
c9e9f97b | 3509 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3510 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3511 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3512 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3513 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
2b82032c | 3514 | if (!device->writeable || |
7cc8e58d MX |
3515 | btrfs_device_get_total_bytes(device) - |
3516 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3517 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3518 | continue; |
3519 | ||
3520 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3521 | if (ret == -ENOSPC) |
3522 | break; | |
5a488b9d LB |
3523 | if (ret) { |
3524 | /* btrfs_shrink_device never returns ret > 0 */ | |
3525 | WARN_ON(ret > 0); | |
3526 | goto error; | |
3527 | } | |
ec44a35c | 3528 | |
a22285a6 | 3529 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3530 | if (IS_ERR(trans)) { |
3531 | ret = PTR_ERR(trans); | |
3532 | btrfs_info_in_rcu(fs_info, | |
3533 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3534 | rcu_str_deref(device->name), ret, | |
3535 | old_size, old_size - size_to_free); | |
3536 | goto error; | |
3537 | } | |
ec44a35c CM |
3538 | |
3539 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3540 | if (ret) { |
3a45bb20 | 3541 | btrfs_end_transaction(trans); |
5a488b9d LB |
3542 | /* btrfs_grow_device never returns ret > 0 */ |
3543 | WARN_ON(ret > 0); | |
3544 | btrfs_info_in_rcu(fs_info, | |
3545 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3546 | rcu_str_deref(device->name), ret, | |
3547 | old_size, old_size - size_to_free); | |
3548 | goto error; | |
3549 | } | |
ec44a35c | 3550 | |
3a45bb20 | 3551 | btrfs_end_transaction(trans); |
ec44a35c CM |
3552 | } |
3553 | ||
3554 | /* step two, relocate all the chunks */ | |
3555 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3556 | if (!path) { |
3557 | ret = -ENOMEM; | |
3558 | goto error; | |
3559 | } | |
19a39dce ID |
3560 | |
3561 | /* zero out stat counters */ | |
3562 | spin_lock(&fs_info->balance_lock); | |
3563 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3564 | spin_unlock(&fs_info->balance_lock); | |
3565 | again: | |
7d824b6f | 3566 | if (!counting) { |
12907fc7 DS |
3567 | /* |
3568 | * The single value limit and min/max limits use the same bytes | |
3569 | * in the | |
3570 | */ | |
7d824b6f DS |
3571 | bctl->data.limit = limit_data; |
3572 | bctl->meta.limit = limit_meta; | |
3573 | bctl->sys.limit = limit_sys; | |
3574 | } | |
ec44a35c CM |
3575 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3576 | key.offset = (u64)-1; | |
3577 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3578 | ||
d397712b | 3579 | while (1) { |
19a39dce | 3580 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3581 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3582 | ret = -ECANCELED; |
3583 | goto error; | |
3584 | } | |
3585 | ||
67c5e7d4 | 3586 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3587 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3588 | if (ret < 0) { |
3589 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3590 | goto error; |
67c5e7d4 | 3591 | } |
ec44a35c CM |
3592 | |
3593 | /* | |
3594 | * this shouldn't happen, it means the last relocate | |
3595 | * failed | |
3596 | */ | |
3597 | if (ret == 0) | |
c9e9f97b | 3598 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3599 | |
3600 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3601 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3602 | if (ret) { |
67c5e7d4 | 3603 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3604 | ret = 0; |
ec44a35c | 3605 | break; |
c9e9f97b | 3606 | } |
7d9eb12c | 3607 | |
f43ffb60 ID |
3608 | leaf = path->nodes[0]; |
3609 | slot = path->slots[0]; | |
3610 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3611 | |
67c5e7d4 FM |
3612 | if (found_key.objectid != key.objectid) { |
3613 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3614 | break; |
67c5e7d4 | 3615 | } |
7d9eb12c | 3616 | |
f43ffb60 | 3617 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3618 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3619 | |
19a39dce ID |
3620 | if (!counting) { |
3621 | spin_lock(&fs_info->balance_lock); | |
3622 | bctl->stat.considered++; | |
3623 | spin_unlock(&fs_info->balance_lock); | |
3624 | } | |
3625 | ||
2ff7e61e | 3626 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3627 | found_key.offset); |
2c9fe835 | 3628 | |
b3b4aa74 | 3629 | btrfs_release_path(path); |
67c5e7d4 FM |
3630 | if (!ret) { |
3631 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3632 | goto loop; |
67c5e7d4 | 3633 | } |
f43ffb60 | 3634 | |
19a39dce | 3635 | if (counting) { |
67c5e7d4 | 3636 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3637 | spin_lock(&fs_info->balance_lock); |
3638 | bctl->stat.expected++; | |
3639 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3640 | |
3641 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3642 | count_data++; | |
3643 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3644 | count_sys++; | |
3645 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3646 | count_meta++; | |
3647 | ||
3648 | goto loop; | |
3649 | } | |
3650 | ||
3651 | /* | |
3652 | * Apply limit_min filter, no need to check if the LIMITS | |
3653 | * filter is used, limit_min is 0 by default | |
3654 | */ | |
3655 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3656 | count_data < bctl->data.limit_min) | |
3657 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3658 | count_meta < bctl->meta.limit_min) | |
3659 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3660 | count_sys < bctl->sys.limit_min)) { | |
3661 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3662 | goto loop; |
3663 | } | |
3664 | ||
cf25ce51 LB |
3665 | ASSERT(fs_info->data_sinfo); |
3666 | spin_lock(&fs_info->data_sinfo->lock); | |
3667 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3668 | spin_unlock(&fs_info->data_sinfo->lock); | |
3669 | ||
3670 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3671 | !chunk_reserved && !bytes_used) { | |
2c9fe835 ZL |
3672 | trans = btrfs_start_transaction(chunk_root, 0); |
3673 | if (IS_ERR(trans)) { | |
3674 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3675 | ret = PTR_ERR(trans); | |
3676 | goto error; | |
3677 | } | |
3678 | ||
2ff7e61e | 3679 | ret = btrfs_force_chunk_alloc(trans, fs_info, |
2c9fe835 | 3680 | BTRFS_BLOCK_GROUP_DATA); |
3a45bb20 | 3681 | btrfs_end_transaction(trans); |
2c9fe835 ZL |
3682 | if (ret < 0) { |
3683 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3684 | goto error; | |
3685 | } | |
2c9fe835 ZL |
3686 | chunk_reserved = 1; |
3687 | } | |
3688 | ||
5b4aacef | 3689 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3690 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3691 | if (ret && ret != -ENOSPC) |
3692 | goto error; | |
19a39dce | 3693 | if (ret == -ENOSPC) { |
c9e9f97b | 3694 | enospc_errors++; |
19a39dce ID |
3695 | } else { |
3696 | spin_lock(&fs_info->balance_lock); | |
3697 | bctl->stat.completed++; | |
3698 | spin_unlock(&fs_info->balance_lock); | |
3699 | } | |
f43ffb60 | 3700 | loop: |
795a3321 ID |
3701 | if (found_key.offset == 0) |
3702 | break; | |
ba1bf481 | 3703 | key.offset = found_key.offset - 1; |
ec44a35c | 3704 | } |
c9e9f97b | 3705 | |
19a39dce ID |
3706 | if (counting) { |
3707 | btrfs_release_path(path); | |
3708 | counting = false; | |
3709 | goto again; | |
3710 | } | |
ec44a35c CM |
3711 | error: |
3712 | btrfs_free_path(path); | |
c9e9f97b | 3713 | if (enospc_errors) { |
efe120a0 | 3714 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3715 | enospc_errors); |
c9e9f97b ID |
3716 | if (!ret) |
3717 | ret = -ENOSPC; | |
3718 | } | |
3719 | ||
ec44a35c CM |
3720 | return ret; |
3721 | } | |
3722 | ||
0c460c0d ID |
3723 | /** |
3724 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3725 | * @flags: profile to validate | |
3726 | * @extended: if true @flags is treated as an extended profile | |
3727 | */ | |
3728 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3729 | { | |
3730 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3731 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3732 | ||
3733 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3734 | ||
3735 | /* 1) check that all other bits are zeroed */ | |
3736 | if (flags & ~mask) | |
3737 | return 0; | |
3738 | ||
3739 | /* 2) see if profile is reduced */ | |
3740 | if (flags == 0) | |
3741 | return !extended; /* "0" is valid for usual profiles */ | |
3742 | ||
3743 | /* true if exactly one bit set */ | |
3744 | return (flags & (flags - 1)) == 0; | |
3745 | } | |
3746 | ||
837d5b6e ID |
3747 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3748 | { | |
a7e99c69 ID |
3749 | /* cancel requested || normal exit path */ |
3750 | return atomic_read(&fs_info->balance_cancel_req) || | |
3751 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3752 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3753 | } |
3754 | ||
c9e9f97b ID |
3755 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3756 | { | |
0940ebf6 ID |
3757 | int ret; |
3758 | ||
c9e9f97b | 3759 | unset_balance_control(fs_info); |
6bccf3ab | 3760 | ret = del_balance_item(fs_info); |
0f788c58 | 3761 | if (ret) |
34d97007 | 3762 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f ID |
3763 | |
3764 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3765 | } |
3766 | ||
bdcd3c97 AM |
3767 | /* Non-zero return value signifies invalidity */ |
3768 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3769 | u64 allowed) | |
3770 | { | |
3771 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3772 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3773 | (bctl_arg->target & ~allowed))); | |
3774 | } | |
3775 | ||
c9e9f97b ID |
3776 | /* |
3777 | * Should be called with both balance and volume mutexes held | |
3778 | */ | |
3779 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3780 | struct btrfs_ioctl_balance_args *bargs) | |
3781 | { | |
3782 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
14506127 | 3783 | u64 meta_target, data_target; |
f43ffb60 | 3784 | u64 allowed; |
e4837f8f | 3785 | int mixed = 0; |
c9e9f97b | 3786 | int ret; |
8dabb742 | 3787 | u64 num_devices; |
de98ced9 | 3788 | unsigned seq; |
c9e9f97b | 3789 | |
837d5b6e | 3790 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3791 | atomic_read(&fs_info->balance_pause_req) || |
3792 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3793 | ret = -EINVAL; |
3794 | goto out; | |
3795 | } | |
3796 | ||
e4837f8f ID |
3797 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3798 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3799 | mixed = 1; | |
3800 | ||
f43ffb60 ID |
3801 | /* |
3802 | * In case of mixed groups both data and meta should be picked, | |
3803 | * and identical options should be given for both of them. | |
3804 | */ | |
e4837f8f ID |
3805 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3806 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3807 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3808 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3809 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3810 | btrfs_err(fs_info, |
3811 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3812 | ret = -EINVAL; |
3813 | goto out; | |
3814 | } | |
3815 | } | |
3816 | ||
8dabb742 | 3817 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3818 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3819 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3820 | BUG_ON(num_devices < 1); | |
3821 | num_devices--; | |
3822 | } | |
73beece9 | 3823 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
88be159c AH |
3824 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3825 | if (num_devices > 1) | |
e4d8ec0f | 3826 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3827 | if (num_devices > 2) |
3828 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3829 | if (num_devices > 3) | |
3830 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3831 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3832 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3833 | btrfs_err(fs_info, |
3834 | "unable to start balance with target data profile %llu", | |
3835 | bctl->data.target); | |
e4d8ec0f ID |
3836 | ret = -EINVAL; |
3837 | goto out; | |
3838 | } | |
bdcd3c97 | 3839 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3840 | btrfs_err(fs_info, |
5d163e0e JM |
3841 | "unable to start balance with target metadata profile %llu", |
3842 | bctl->meta.target); | |
e4d8ec0f ID |
3843 | ret = -EINVAL; |
3844 | goto out; | |
3845 | } | |
bdcd3c97 | 3846 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3847 | btrfs_err(fs_info, |
5d163e0e JM |
3848 | "unable to start balance with target system profile %llu", |
3849 | bctl->sys.target); | |
e4d8ec0f ID |
3850 | ret = -EINVAL; |
3851 | goto out; | |
3852 | } | |
3853 | ||
e4d8ec0f ID |
3854 | /* allow to reduce meta or sys integrity only if force set */ |
3855 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3856 | BTRFS_BLOCK_GROUP_RAID10 | |
3857 | BTRFS_BLOCK_GROUP_RAID5 | | |
3858 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3859 | do { |
3860 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3861 | ||
3862 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3863 | (fs_info->avail_system_alloc_bits & allowed) && | |
3864 | !(bctl->sys.target & allowed)) || | |
3865 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3866 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3867 | !(bctl->meta.target & allowed))) { | |
3868 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3869 | btrfs_info(fs_info, |
3870 | "force reducing metadata integrity"); | |
de98ced9 | 3871 | } else { |
5d163e0e JM |
3872 | btrfs_err(fs_info, |
3873 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3874 | ret = -EINVAL; |
3875 | goto out; | |
3876 | } | |
e4d8ec0f | 3877 | } |
de98ced9 | 3878 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3879 | |
14506127 AB |
3880 | /* if we're not converting, the target field is uninitialized */ |
3881 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3882 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3883 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3884 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3885 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3886 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 3887 | btrfs_warn(fs_info, |
5d163e0e | 3888 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
14506127 | 3889 | meta_target, data_target); |
ee592d07 ST |
3890 | } |
3891 | ||
5af3e8cc | 3892 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
943c6e99 ZL |
3893 | fs_info->num_tolerated_disk_barrier_failures = min( |
3894 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3895 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3896 | bctl->sys.target)); | |
5af3e8cc SB |
3897 | } |
3898 | ||
6bccf3ab | 3899 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3900 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3901 | goto out; |
3902 | ||
59641015 ID |
3903 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3904 | BUG_ON(ret == -EEXIST); | |
3905 | set_balance_control(bctl); | |
3906 | } else { | |
3907 | BUG_ON(ret != -EEXIST); | |
3908 | spin_lock(&fs_info->balance_lock); | |
3909 | update_balance_args(bctl); | |
3910 | spin_unlock(&fs_info->balance_lock); | |
3911 | } | |
c9e9f97b | 3912 | |
837d5b6e | 3913 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3914 | mutex_unlock(&fs_info->balance_mutex); |
3915 | ||
3916 | ret = __btrfs_balance(fs_info); | |
3917 | ||
3918 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3919 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3920 | |
bf023ecf ID |
3921 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3922 | fs_info->num_tolerated_disk_barrier_failures = | |
3923 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3924 | } | |
3925 | ||
c9e9f97b ID |
3926 | if (bargs) { |
3927 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3928 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3929 | } |
3930 | ||
3a01aa7a ID |
3931 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3932 | balance_need_close(fs_info)) { | |
3933 | __cancel_balance(fs_info); | |
3934 | } | |
3935 | ||
837d5b6e | 3936 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3937 | |
3938 | return ret; | |
3939 | out: | |
59641015 ID |
3940 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3941 | __cancel_balance(fs_info); | |
ed0fb78f | 3942 | else { |
59641015 | 3943 | kfree(bctl); |
ed0fb78f ID |
3944 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3945 | } | |
59641015 ID |
3946 | return ret; |
3947 | } | |
3948 | ||
3949 | static int balance_kthread(void *data) | |
3950 | { | |
2b6ba629 | 3951 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3952 | int ret = 0; |
59641015 ID |
3953 | |
3954 | mutex_lock(&fs_info->volume_mutex); | |
3955 | mutex_lock(&fs_info->balance_mutex); | |
3956 | ||
2b6ba629 | 3957 | if (fs_info->balance_ctl) { |
efe120a0 | 3958 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3959 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3960 | } |
59641015 ID |
3961 | |
3962 | mutex_unlock(&fs_info->balance_mutex); | |
3963 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3964 | |
59641015 ID |
3965 | return ret; |
3966 | } | |
3967 | ||
2b6ba629 ID |
3968 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3969 | { | |
3970 | struct task_struct *tsk; | |
3971 | ||
3972 | spin_lock(&fs_info->balance_lock); | |
3973 | if (!fs_info->balance_ctl) { | |
3974 | spin_unlock(&fs_info->balance_lock); | |
3975 | return 0; | |
3976 | } | |
3977 | spin_unlock(&fs_info->balance_lock); | |
3978 | ||
3cdde224 | 3979 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 3980 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3981 | return 0; |
3982 | } | |
3983 | ||
3984 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3985 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3986 | } |
3987 | ||
68310a5e | 3988 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3989 | { |
59641015 ID |
3990 | struct btrfs_balance_control *bctl; |
3991 | struct btrfs_balance_item *item; | |
3992 | struct btrfs_disk_balance_args disk_bargs; | |
3993 | struct btrfs_path *path; | |
3994 | struct extent_buffer *leaf; | |
3995 | struct btrfs_key key; | |
3996 | int ret; | |
3997 | ||
3998 | path = btrfs_alloc_path(); | |
3999 | if (!path) | |
4000 | return -ENOMEM; | |
4001 | ||
59641015 | 4002 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4003 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4004 | key.offset = 0; |
4005 | ||
68310a5e | 4006 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4007 | if (ret < 0) |
68310a5e | 4008 | goto out; |
59641015 ID |
4009 | if (ret > 0) { /* ret = -ENOENT; */ |
4010 | ret = 0; | |
68310a5e ID |
4011 | goto out; |
4012 | } | |
4013 | ||
4014 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4015 | if (!bctl) { | |
4016 | ret = -ENOMEM; | |
4017 | goto out; | |
59641015 ID |
4018 | } |
4019 | ||
4020 | leaf = path->nodes[0]; | |
4021 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4022 | ||
68310a5e ID |
4023 | bctl->fs_info = fs_info; |
4024 | bctl->flags = btrfs_balance_flags(leaf, item); | |
4025 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4026 | |
4027 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4028 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4029 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4030 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4031 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4032 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4033 | ||
ed0fb78f ID |
4034 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
4035 | ||
68310a5e ID |
4036 | mutex_lock(&fs_info->volume_mutex); |
4037 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 4038 | |
68310a5e ID |
4039 | set_balance_control(bctl); |
4040 | ||
4041 | mutex_unlock(&fs_info->balance_mutex); | |
4042 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4043 | out: |
4044 | btrfs_free_path(path); | |
ec44a35c CM |
4045 | return ret; |
4046 | } | |
4047 | ||
837d5b6e ID |
4048 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4049 | { | |
4050 | int ret = 0; | |
4051 | ||
4052 | mutex_lock(&fs_info->balance_mutex); | |
4053 | if (!fs_info->balance_ctl) { | |
4054 | mutex_unlock(&fs_info->balance_mutex); | |
4055 | return -ENOTCONN; | |
4056 | } | |
4057 | ||
4058 | if (atomic_read(&fs_info->balance_running)) { | |
4059 | atomic_inc(&fs_info->balance_pause_req); | |
4060 | mutex_unlock(&fs_info->balance_mutex); | |
4061 | ||
4062 | wait_event(fs_info->balance_wait_q, | |
4063 | atomic_read(&fs_info->balance_running) == 0); | |
4064 | ||
4065 | mutex_lock(&fs_info->balance_mutex); | |
4066 | /* we are good with balance_ctl ripped off from under us */ | |
4067 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4068 | atomic_dec(&fs_info->balance_pause_req); | |
4069 | } else { | |
4070 | ret = -ENOTCONN; | |
4071 | } | |
4072 | ||
4073 | mutex_unlock(&fs_info->balance_mutex); | |
4074 | return ret; | |
4075 | } | |
4076 | ||
a7e99c69 ID |
4077 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4078 | { | |
e649e587 ID |
4079 | if (fs_info->sb->s_flags & MS_RDONLY) |
4080 | return -EROFS; | |
4081 | ||
a7e99c69 ID |
4082 | mutex_lock(&fs_info->balance_mutex); |
4083 | if (!fs_info->balance_ctl) { | |
4084 | mutex_unlock(&fs_info->balance_mutex); | |
4085 | return -ENOTCONN; | |
4086 | } | |
4087 | ||
4088 | atomic_inc(&fs_info->balance_cancel_req); | |
4089 | /* | |
4090 | * if we are running just wait and return, balance item is | |
4091 | * deleted in btrfs_balance in this case | |
4092 | */ | |
4093 | if (atomic_read(&fs_info->balance_running)) { | |
4094 | mutex_unlock(&fs_info->balance_mutex); | |
4095 | wait_event(fs_info->balance_wait_q, | |
4096 | atomic_read(&fs_info->balance_running) == 0); | |
4097 | mutex_lock(&fs_info->balance_mutex); | |
4098 | } else { | |
4099 | /* __cancel_balance needs volume_mutex */ | |
4100 | mutex_unlock(&fs_info->balance_mutex); | |
4101 | mutex_lock(&fs_info->volume_mutex); | |
4102 | mutex_lock(&fs_info->balance_mutex); | |
4103 | ||
4104 | if (fs_info->balance_ctl) | |
4105 | __cancel_balance(fs_info); | |
4106 | ||
4107 | mutex_unlock(&fs_info->volume_mutex); | |
4108 | } | |
4109 | ||
4110 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4111 | atomic_dec(&fs_info->balance_cancel_req); | |
4112 | mutex_unlock(&fs_info->balance_mutex); | |
4113 | return 0; | |
4114 | } | |
4115 | ||
803b2f54 SB |
4116 | static int btrfs_uuid_scan_kthread(void *data) |
4117 | { | |
4118 | struct btrfs_fs_info *fs_info = data; | |
4119 | struct btrfs_root *root = fs_info->tree_root; | |
4120 | struct btrfs_key key; | |
4121 | struct btrfs_key max_key; | |
4122 | struct btrfs_path *path = NULL; | |
4123 | int ret = 0; | |
4124 | struct extent_buffer *eb; | |
4125 | int slot; | |
4126 | struct btrfs_root_item root_item; | |
4127 | u32 item_size; | |
f45388f3 | 4128 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4129 | |
4130 | path = btrfs_alloc_path(); | |
4131 | if (!path) { | |
4132 | ret = -ENOMEM; | |
4133 | goto out; | |
4134 | } | |
4135 | ||
4136 | key.objectid = 0; | |
4137 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4138 | key.offset = 0; | |
4139 | ||
4140 | max_key.objectid = (u64)-1; | |
4141 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
4142 | max_key.offset = (u64)-1; | |
4143 | ||
803b2f54 | 4144 | while (1) { |
6174d3cb | 4145 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4146 | if (ret) { |
4147 | if (ret > 0) | |
4148 | ret = 0; | |
4149 | break; | |
4150 | } | |
4151 | ||
4152 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4153 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4154 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4155 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4156 | goto skip; | |
4157 | ||
4158 | eb = path->nodes[0]; | |
4159 | slot = path->slots[0]; | |
4160 | item_size = btrfs_item_size_nr(eb, slot); | |
4161 | if (item_size < sizeof(root_item)) | |
4162 | goto skip; | |
4163 | ||
803b2f54 SB |
4164 | read_extent_buffer(eb, &root_item, |
4165 | btrfs_item_ptr_offset(eb, slot), | |
4166 | (int)sizeof(root_item)); | |
4167 | if (btrfs_root_refs(&root_item) == 0) | |
4168 | goto skip; | |
f45388f3 FDBM |
4169 | |
4170 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4171 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4172 | if (trans) | |
4173 | goto update_tree; | |
4174 | ||
4175 | btrfs_release_path(path); | |
803b2f54 SB |
4176 | /* |
4177 | * 1 - subvol uuid item | |
4178 | * 1 - received_subvol uuid item | |
4179 | */ | |
4180 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4181 | if (IS_ERR(trans)) { | |
4182 | ret = PTR_ERR(trans); | |
4183 | break; | |
4184 | } | |
f45388f3 FDBM |
4185 | continue; |
4186 | } else { | |
4187 | goto skip; | |
4188 | } | |
4189 | update_tree: | |
4190 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
6bccf3ab | 4191 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4192 | root_item.uuid, |
4193 | BTRFS_UUID_KEY_SUBVOL, | |
4194 | key.objectid); | |
4195 | if (ret < 0) { | |
efe120a0 | 4196 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4197 | ret); |
803b2f54 SB |
4198 | break; |
4199 | } | |
4200 | } | |
4201 | ||
4202 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
6bccf3ab | 4203 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4204 | root_item.received_uuid, |
4205 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4206 | key.objectid); | |
4207 | if (ret < 0) { | |
efe120a0 | 4208 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4209 | ret); |
803b2f54 SB |
4210 | break; |
4211 | } | |
4212 | } | |
4213 | ||
f45388f3 | 4214 | skip: |
803b2f54 | 4215 | if (trans) { |
3a45bb20 | 4216 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4217 | trans = NULL; |
803b2f54 SB |
4218 | if (ret) |
4219 | break; | |
4220 | } | |
4221 | ||
803b2f54 SB |
4222 | btrfs_release_path(path); |
4223 | if (key.offset < (u64)-1) { | |
4224 | key.offset++; | |
4225 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4226 | key.offset = 0; | |
4227 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4228 | } else if (key.objectid < (u64)-1) { | |
4229 | key.offset = 0; | |
4230 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4231 | key.objectid++; | |
4232 | } else { | |
4233 | break; | |
4234 | } | |
4235 | cond_resched(); | |
4236 | } | |
4237 | ||
4238 | out: | |
4239 | btrfs_free_path(path); | |
f45388f3 | 4240 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4241 | btrfs_end_transaction(trans); |
803b2f54 | 4242 | if (ret) |
efe120a0 | 4243 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4244 | else |
afcdd129 | 4245 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4246 | up(&fs_info->uuid_tree_rescan_sem); |
4247 | return 0; | |
4248 | } | |
4249 | ||
70f80175 SB |
4250 | /* |
4251 | * Callback for btrfs_uuid_tree_iterate(). | |
4252 | * returns: | |
4253 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4254 | * < 0 if an error occurred. |
70f80175 SB |
4255 | * > 0 if the check failed, which means the caller shall remove the entry. |
4256 | */ | |
4257 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4258 | u8 *uuid, u8 type, u64 subid) | |
4259 | { | |
4260 | struct btrfs_key key; | |
4261 | int ret = 0; | |
4262 | struct btrfs_root *subvol_root; | |
4263 | ||
4264 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4265 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4266 | goto out; | |
4267 | ||
4268 | key.objectid = subid; | |
4269 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4270 | key.offset = (u64)-1; | |
4271 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4272 | if (IS_ERR(subvol_root)) { | |
4273 | ret = PTR_ERR(subvol_root); | |
4274 | if (ret == -ENOENT) | |
4275 | ret = 1; | |
4276 | goto out; | |
4277 | } | |
4278 | ||
4279 | switch (type) { | |
4280 | case BTRFS_UUID_KEY_SUBVOL: | |
4281 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4282 | ret = 1; | |
4283 | break; | |
4284 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4285 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4286 | BTRFS_UUID_SIZE)) | |
4287 | ret = 1; | |
4288 | break; | |
4289 | } | |
4290 | ||
4291 | out: | |
4292 | return ret; | |
4293 | } | |
4294 | ||
4295 | static int btrfs_uuid_rescan_kthread(void *data) | |
4296 | { | |
4297 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4298 | int ret; | |
4299 | ||
4300 | /* | |
4301 | * 1st step is to iterate through the existing UUID tree and | |
4302 | * to delete all entries that contain outdated data. | |
4303 | * 2nd step is to add all missing entries to the UUID tree. | |
4304 | */ | |
4305 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4306 | if (ret < 0) { | |
efe120a0 | 4307 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4308 | up(&fs_info->uuid_tree_rescan_sem); |
4309 | return ret; | |
4310 | } | |
4311 | return btrfs_uuid_scan_kthread(data); | |
4312 | } | |
4313 | ||
f7a81ea4 SB |
4314 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4315 | { | |
4316 | struct btrfs_trans_handle *trans; | |
4317 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4318 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4319 | struct task_struct *task; |
4320 | int ret; | |
f7a81ea4 SB |
4321 | |
4322 | /* | |
4323 | * 1 - root node | |
4324 | * 1 - root item | |
4325 | */ | |
4326 | trans = btrfs_start_transaction(tree_root, 2); | |
4327 | if (IS_ERR(trans)) | |
4328 | return PTR_ERR(trans); | |
4329 | ||
4330 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4331 | BTRFS_UUID_TREE_OBJECTID); | |
4332 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4333 | ret = PTR_ERR(uuid_root); |
66642832 | 4334 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4335 | btrfs_end_transaction(trans); |
6d13f549 | 4336 | return ret; |
f7a81ea4 SB |
4337 | } |
4338 | ||
4339 | fs_info->uuid_root = uuid_root; | |
4340 | ||
3a45bb20 | 4341 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4342 | if (ret) |
4343 | return ret; | |
4344 | ||
4345 | down(&fs_info->uuid_tree_rescan_sem); | |
4346 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4347 | if (IS_ERR(task)) { | |
70f80175 | 4348 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4349 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4350 | up(&fs_info->uuid_tree_rescan_sem); |
4351 | return PTR_ERR(task); | |
4352 | } | |
4353 | ||
4354 | return 0; | |
f7a81ea4 | 4355 | } |
803b2f54 | 4356 | |
70f80175 SB |
4357 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4358 | { | |
4359 | struct task_struct *task; | |
4360 | ||
4361 | down(&fs_info->uuid_tree_rescan_sem); | |
4362 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4363 | if (IS_ERR(task)) { | |
4364 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4365 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4366 | up(&fs_info->uuid_tree_rescan_sem); |
4367 | return PTR_ERR(task); | |
4368 | } | |
4369 | ||
4370 | return 0; | |
4371 | } | |
4372 | ||
8f18cf13 CM |
4373 | /* |
4374 | * shrinking a device means finding all of the device extents past | |
4375 | * the new size, and then following the back refs to the chunks. | |
4376 | * The chunk relocation code actually frees the device extent | |
4377 | */ | |
4378 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4379 | { | |
0b246afa JM |
4380 | struct btrfs_fs_info *fs_info = device->fs_info; |
4381 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4382 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4383 | struct btrfs_dev_extent *dev_extent = NULL; |
4384 | struct btrfs_path *path; | |
4385 | u64 length; | |
8f18cf13 CM |
4386 | u64 chunk_offset; |
4387 | int ret; | |
4388 | int slot; | |
ba1bf481 JB |
4389 | int failed = 0; |
4390 | bool retried = false; | |
53e489bc | 4391 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4392 | struct extent_buffer *l; |
4393 | struct btrfs_key key; | |
0b246afa | 4394 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4395 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4396 | u64 old_size = btrfs_device_get_total_bytes(device); |
4397 | u64 diff = old_size - new_size; | |
8f18cf13 | 4398 | |
63a212ab SB |
4399 | if (device->is_tgtdev_for_dev_replace) |
4400 | return -EINVAL; | |
4401 | ||
8f18cf13 CM |
4402 | path = btrfs_alloc_path(); |
4403 | if (!path) | |
4404 | return -ENOMEM; | |
4405 | ||
e4058b54 | 4406 | path->reada = READA_FORWARD; |
8f18cf13 | 4407 | |
34441361 | 4408 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4409 | |
7cc8e58d | 4410 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4411 | if (device->writeable) { |
2b82032c | 4412 | device->fs_devices->total_rw_bytes -= diff; |
0b246afa JM |
4413 | spin_lock(&fs_info->free_chunk_lock); |
4414 | fs_info->free_chunk_space -= diff; | |
4415 | spin_unlock(&fs_info->free_chunk_lock); | |
2bf64758 | 4416 | } |
34441361 | 4417 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4418 | |
ba1bf481 | 4419 | again: |
8f18cf13 CM |
4420 | key.objectid = device->devid; |
4421 | key.offset = (u64)-1; | |
4422 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4423 | ||
213e64da | 4424 | do { |
0b246afa | 4425 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4426 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4427 | if (ret < 0) { |
0b246afa | 4428 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4429 | goto done; |
67c5e7d4 | 4430 | } |
8f18cf13 CM |
4431 | |
4432 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4433 | if (ret) |
0b246afa | 4434 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4435 | if (ret < 0) |
4436 | goto done; | |
4437 | if (ret) { | |
4438 | ret = 0; | |
b3b4aa74 | 4439 | btrfs_release_path(path); |
bf1fb512 | 4440 | break; |
8f18cf13 CM |
4441 | } |
4442 | ||
4443 | l = path->nodes[0]; | |
4444 | slot = path->slots[0]; | |
4445 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4446 | ||
ba1bf481 | 4447 | if (key.objectid != device->devid) { |
0b246afa | 4448 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4449 | btrfs_release_path(path); |
bf1fb512 | 4450 | break; |
ba1bf481 | 4451 | } |
8f18cf13 CM |
4452 | |
4453 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4454 | length = btrfs_dev_extent_length(l, dev_extent); | |
4455 | ||
ba1bf481 | 4456 | if (key.offset + length <= new_size) { |
0b246afa | 4457 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4458 | btrfs_release_path(path); |
d6397bae | 4459 | break; |
ba1bf481 | 4460 | } |
8f18cf13 | 4461 | |
8f18cf13 | 4462 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4463 | btrfs_release_path(path); |
8f18cf13 | 4464 | |
0b246afa JM |
4465 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4466 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4467 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4468 | goto done; |
ba1bf481 JB |
4469 | if (ret == -ENOSPC) |
4470 | failed++; | |
213e64da | 4471 | } while (key.offset-- > 0); |
ba1bf481 JB |
4472 | |
4473 | if (failed && !retried) { | |
4474 | failed = 0; | |
4475 | retried = true; | |
4476 | goto again; | |
4477 | } else if (failed && retried) { | |
4478 | ret = -ENOSPC; | |
ba1bf481 | 4479 | goto done; |
8f18cf13 CM |
4480 | } |
4481 | ||
d6397bae | 4482 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4483 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4484 | if (IS_ERR(trans)) { |
4485 | ret = PTR_ERR(trans); | |
4486 | goto done; | |
4487 | } | |
4488 | ||
34441361 | 4489 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4490 | |
4491 | /* | |
4492 | * We checked in the above loop all device extents that were already in | |
4493 | * the device tree. However before we have updated the device's | |
4494 | * total_bytes to the new size, we might have had chunk allocations that | |
4495 | * have not complete yet (new block groups attached to transaction | |
4496 | * handles), and therefore their device extents were not yet in the | |
4497 | * device tree and we missed them in the loop above. So if we have any | |
4498 | * pending chunk using a device extent that overlaps the device range | |
4499 | * that we can not use anymore, commit the current transaction and | |
4500 | * repeat the search on the device tree - this way we guarantee we will | |
4501 | * not have chunks using device extents that end beyond 'new_size'. | |
4502 | */ | |
4503 | if (!checked_pending_chunks) { | |
4504 | u64 start = new_size; | |
4505 | u64 len = old_size - new_size; | |
4506 | ||
499f377f JM |
4507 | if (contains_pending_extent(trans->transaction, device, |
4508 | &start, len)) { | |
34441361 | 4509 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4510 | checked_pending_chunks = true; |
4511 | failed = 0; | |
4512 | retried = false; | |
3a45bb20 | 4513 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4514 | if (ret) |
4515 | goto done; | |
4516 | goto again; | |
4517 | } | |
4518 | } | |
4519 | ||
7cc8e58d | 4520 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4521 | if (list_empty(&device->resized_list)) |
4522 | list_add_tail(&device->resized_list, | |
0b246afa | 4523 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4524 | |
d6397bae CB |
4525 | WARN_ON(diff > old_total); |
4526 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
34441361 | 4527 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4528 | |
4529 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4530 | ret = btrfs_update_device(trans, device); | |
3a45bb20 | 4531 | btrfs_end_transaction(trans); |
8f18cf13 CM |
4532 | done: |
4533 | btrfs_free_path(path); | |
53e489bc | 4534 | if (ret) { |
34441361 | 4535 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4536 | btrfs_device_set_total_bytes(device, old_size); |
4537 | if (device->writeable) | |
4538 | device->fs_devices->total_rw_bytes += diff; | |
0b246afa JM |
4539 | spin_lock(&fs_info->free_chunk_lock); |
4540 | fs_info->free_chunk_space += diff; | |
4541 | spin_unlock(&fs_info->free_chunk_lock); | |
34441361 | 4542 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4543 | } |
8f18cf13 CM |
4544 | return ret; |
4545 | } | |
4546 | ||
2ff7e61e | 4547 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4548 | struct btrfs_key *key, |
4549 | struct btrfs_chunk *chunk, int item_size) | |
4550 | { | |
0b246afa | 4551 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4552 | struct btrfs_disk_key disk_key; |
4553 | u32 array_size; | |
4554 | u8 *ptr; | |
4555 | ||
34441361 | 4556 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4557 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4558 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4559 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4560 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4561 | return -EFBIG; |
fe48a5c0 | 4562 | } |
0b86a832 CM |
4563 | |
4564 | ptr = super_copy->sys_chunk_array + array_size; | |
4565 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4566 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4567 | ptr += sizeof(disk_key); | |
4568 | memcpy(ptr, chunk, item_size); | |
4569 | item_size += sizeof(disk_key); | |
4570 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4571 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4572 | |
0b86a832 CM |
4573 | return 0; |
4574 | } | |
4575 | ||
73c5de00 AJ |
4576 | /* |
4577 | * sort the devices in descending order by max_avail, total_avail | |
4578 | */ | |
4579 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4580 | { |
73c5de00 AJ |
4581 | const struct btrfs_device_info *di_a = a; |
4582 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4583 | |
73c5de00 | 4584 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4585 | return -1; |
73c5de00 | 4586 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4587 | return 1; |
73c5de00 AJ |
4588 | if (di_a->total_avail > di_b->total_avail) |
4589 | return -1; | |
4590 | if (di_a->total_avail < di_b->total_avail) | |
4591 | return 1; | |
4592 | return 0; | |
b2117a39 | 4593 | } |
0b86a832 | 4594 | |
53b381b3 DW |
4595 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4596 | { | |
4597 | /* TODO allow them to set a preferred stripe size */ | |
ee22184b | 4598 | return SZ_64K; |
53b381b3 DW |
4599 | } |
4600 | ||
4601 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4602 | { | |
ffe2d203 | 4603 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4604 | return; |
4605 | ||
ceda0864 | 4606 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4607 | } |
4608 | ||
da17066c | 4609 | #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \ |
23f8f9b7 GH |
4610 | - sizeof(struct btrfs_chunk)) \ |
4611 | / sizeof(struct btrfs_stripe) + 1) | |
4612 | ||
4613 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4614 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4615 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4616 | / sizeof(struct btrfs_stripe) + 1) | |
4617 | ||
73c5de00 | 4618 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4619 | u64 start, u64 type) |
b2117a39 | 4620 | { |
2ff7e61e | 4621 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 AJ |
4622 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
4623 | struct list_head *cur; | |
4624 | struct map_lookup *map = NULL; | |
4625 | struct extent_map_tree *em_tree; | |
4626 | struct extent_map *em; | |
4627 | struct btrfs_device_info *devices_info = NULL; | |
4628 | u64 total_avail; | |
4629 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4630 | int data_stripes; /* number of stripes that count for |
4631 | block group size */ | |
73c5de00 AJ |
4632 | int sub_stripes; /* sub_stripes info for map */ |
4633 | int dev_stripes; /* stripes per dev */ | |
4634 | int devs_max; /* max devs to use */ | |
4635 | int devs_min; /* min devs needed */ | |
4636 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4637 | int ncopies; /* how many copies to data has */ | |
4638 | int ret; | |
4639 | u64 max_stripe_size; | |
4640 | u64 max_chunk_size; | |
4641 | u64 stripe_size; | |
4642 | u64 num_bytes; | |
53b381b3 | 4643 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4644 | int ndevs; |
4645 | int i; | |
4646 | int j; | |
31e50229 | 4647 | int index; |
593060d7 | 4648 | |
0c460c0d | 4649 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4650 | |
73c5de00 AJ |
4651 | if (list_empty(&fs_devices->alloc_list)) |
4652 | return -ENOSPC; | |
b2117a39 | 4653 | |
31e50229 | 4654 | index = __get_raid_index(type); |
73c5de00 | 4655 | |
31e50229 LB |
4656 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4657 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4658 | devs_max = btrfs_raid_array[index].devs_max; | |
4659 | devs_min = btrfs_raid_array[index].devs_min; | |
4660 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4661 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4662 | |
9b3f68b9 | 4663 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4664 | max_stripe_size = SZ_1G; |
73c5de00 | 4665 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 GH |
4666 | if (!devs_max) |
4667 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4668 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4669 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4670 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4671 | max_stripe_size = SZ_1G; | |
1100373f | 4672 | else |
ee22184b | 4673 | max_stripe_size = SZ_256M; |
73c5de00 | 4674 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4675 | if (!devs_max) |
4676 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4677 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4678 | max_stripe_size = SZ_32M; |
73c5de00 | 4679 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4680 | if (!devs_max) |
4681 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4682 | } else { |
351fd353 | 4683 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4684 | type); |
4685 | BUG_ON(1); | |
9b3f68b9 CM |
4686 | } |
4687 | ||
2b82032c YZ |
4688 | /* we don't want a chunk larger than 10% of writeable space */ |
4689 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4690 | max_chunk_size); | |
9b3f68b9 | 4691 | |
31e818fe | 4692 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4693 | GFP_NOFS); |
4694 | if (!devices_info) | |
4695 | return -ENOMEM; | |
0cad8a11 | 4696 | |
73c5de00 | 4697 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4698 | |
9f680ce0 | 4699 | /* |
73c5de00 AJ |
4700 | * in the first pass through the devices list, we gather information |
4701 | * about the available holes on each device. | |
9f680ce0 | 4702 | */ |
73c5de00 AJ |
4703 | ndevs = 0; |
4704 | while (cur != &fs_devices->alloc_list) { | |
4705 | struct btrfs_device *device; | |
4706 | u64 max_avail; | |
4707 | u64 dev_offset; | |
b2117a39 | 4708 | |
73c5de00 | 4709 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4710 | |
73c5de00 | 4711 | cur = cur->next; |
b2117a39 | 4712 | |
73c5de00 | 4713 | if (!device->writeable) { |
31b1a2bd | 4714 | WARN(1, KERN_ERR |
efe120a0 | 4715 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4716 | continue; |
4717 | } | |
b2117a39 | 4718 | |
63a212ab SB |
4719 | if (!device->in_fs_metadata || |
4720 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4721 | continue; |
b2117a39 | 4722 | |
73c5de00 AJ |
4723 | if (device->total_bytes > device->bytes_used) |
4724 | total_avail = device->total_bytes - device->bytes_used; | |
4725 | else | |
4726 | total_avail = 0; | |
38c01b96 | 4727 | |
4728 | /* If there is no space on this device, skip it. */ | |
4729 | if (total_avail == 0) | |
4730 | continue; | |
b2117a39 | 4731 | |
6df9a95e | 4732 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4733 | max_stripe_size * dev_stripes, |
4734 | &dev_offset, &max_avail); | |
4735 | if (ret && ret != -ENOSPC) | |
4736 | goto error; | |
b2117a39 | 4737 | |
73c5de00 AJ |
4738 | if (ret == 0) |
4739 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4740 | |
73c5de00 AJ |
4741 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4742 | continue; | |
b2117a39 | 4743 | |
063d006f ES |
4744 | if (ndevs == fs_devices->rw_devices) { |
4745 | WARN(1, "%s: found more than %llu devices\n", | |
4746 | __func__, fs_devices->rw_devices); | |
4747 | break; | |
4748 | } | |
73c5de00 AJ |
4749 | devices_info[ndevs].dev_offset = dev_offset; |
4750 | devices_info[ndevs].max_avail = max_avail; | |
4751 | devices_info[ndevs].total_avail = total_avail; | |
4752 | devices_info[ndevs].dev = device; | |
4753 | ++ndevs; | |
4754 | } | |
b2117a39 | 4755 | |
73c5de00 AJ |
4756 | /* |
4757 | * now sort the devices by hole size / available space | |
4758 | */ | |
4759 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4760 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4761 | |
73c5de00 AJ |
4762 | /* round down to number of usable stripes */ |
4763 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4764 | |
73c5de00 AJ |
4765 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4766 | ret = -ENOSPC; | |
4767 | goto error; | |
b2117a39 | 4768 | } |
9f680ce0 | 4769 | |
73c5de00 AJ |
4770 | if (devs_max && ndevs > devs_max) |
4771 | ndevs = devs_max; | |
4772 | /* | |
4773 | * the primary goal is to maximize the number of stripes, so use as many | |
4774 | * devices as possible, even if the stripes are not maximum sized. | |
4775 | */ | |
4776 | stripe_size = devices_info[ndevs-1].max_avail; | |
4777 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4778 | |
53b381b3 DW |
4779 | /* |
4780 | * this will have to be fixed for RAID1 and RAID10 over | |
4781 | * more drives | |
4782 | */ | |
4783 | data_stripes = num_stripes / ncopies; | |
4784 | ||
53b381b3 DW |
4785 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4786 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
da17066c | 4787 | info->stripesize); |
53b381b3 DW |
4788 | data_stripes = num_stripes - 1; |
4789 | } | |
4790 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4791 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
da17066c | 4792 | info->stripesize); |
53b381b3 DW |
4793 | data_stripes = num_stripes - 2; |
4794 | } | |
86db2578 CM |
4795 | |
4796 | /* | |
4797 | * Use the number of data stripes to figure out how big this chunk | |
4798 | * is really going to be in terms of logical address space, | |
4799 | * and compare that answer with the max chunk size | |
4800 | */ | |
4801 | if (stripe_size * data_stripes > max_chunk_size) { | |
4802 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4803 | |
4804 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4805 | |
4806 | /* bump the answer up to a 16MB boundary */ | |
4807 | stripe_size = (stripe_size + mask) & ~mask; | |
4808 | ||
4809 | /* but don't go higher than the limits we found | |
4810 | * while searching for free extents | |
4811 | */ | |
4812 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4813 | stripe_size = devices_info[ndevs-1].max_avail; | |
4814 | } | |
4815 | ||
b8b93add | 4816 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4817 | |
4818 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4819 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4820 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4821 | |
4822 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4823 | if (!map) { | |
4824 | ret = -ENOMEM; | |
4825 | goto error; | |
4826 | } | |
4827 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4828 | |
73c5de00 AJ |
4829 | for (i = 0; i < ndevs; ++i) { |
4830 | for (j = 0; j < dev_stripes; ++j) { | |
4831 | int s = i * dev_stripes + j; | |
4832 | map->stripes[s].dev = devices_info[i].dev; | |
4833 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4834 | j * stripe_size; | |
6324fbf3 | 4835 | } |
6324fbf3 | 4836 | } |
da17066c | 4837 | map->sector_size = info->sectorsize; |
53b381b3 DW |
4838 | map->stripe_len = raid_stripe_len; |
4839 | map->io_align = raid_stripe_len; | |
4840 | map->io_width = raid_stripe_len; | |
2b82032c | 4841 | map->type = type; |
2b82032c | 4842 | map->sub_stripes = sub_stripes; |
0b86a832 | 4843 | |
53b381b3 | 4844 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4845 | |
6bccf3ab | 4846 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4847 | |
172ddd60 | 4848 | em = alloc_extent_map(); |
2b82032c | 4849 | if (!em) { |
298a8f9c | 4850 | kfree(map); |
b2117a39 MX |
4851 | ret = -ENOMEM; |
4852 | goto error; | |
593060d7 | 4853 | } |
298a8f9c | 4854 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4855 | em->map_lookup = map; |
2b82032c | 4856 | em->start = start; |
73c5de00 | 4857 | em->len = num_bytes; |
2b82032c YZ |
4858 | em->block_start = 0; |
4859 | em->block_len = em->len; | |
6df9a95e | 4860 | em->orig_block_len = stripe_size; |
593060d7 | 4861 | |
0b246afa | 4862 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4863 | write_lock(&em_tree->lock); |
09a2a8f9 | 4864 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4865 | if (!ret) { |
4866 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
490b54d6 | 4867 | refcount_inc(&em->refs); |
6df9a95e | 4868 | } |
890871be | 4869 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4870 | if (ret) { |
4871 | free_extent_map(em); | |
1dd4602f | 4872 | goto error; |
0f5d42b2 | 4873 | } |
0b86a832 | 4874 | |
2ff7e61e | 4875 | ret = btrfs_make_block_group(trans, info, 0, type, |
04487488 JB |
4876 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, |
4877 | start, num_bytes); | |
6df9a95e JB |
4878 | if (ret) |
4879 | goto error_del_extent; | |
2b82032c | 4880 | |
7cc8e58d MX |
4881 | for (i = 0; i < map->num_stripes; i++) { |
4882 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4883 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4884 | } | |
43530c46 | 4885 | |
0b246afa JM |
4886 | spin_lock(&info->free_chunk_lock); |
4887 | info->free_chunk_space -= (stripe_size * map->num_stripes); | |
4888 | spin_unlock(&info->free_chunk_lock); | |
1c116187 | 4889 | |
0f5d42b2 | 4890 | free_extent_map(em); |
0b246afa | 4891 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4892 | |
b2117a39 | 4893 | kfree(devices_info); |
2b82032c | 4894 | return 0; |
b2117a39 | 4895 | |
6df9a95e | 4896 | error_del_extent: |
0f5d42b2 JB |
4897 | write_lock(&em_tree->lock); |
4898 | remove_extent_mapping(em_tree, em); | |
4899 | write_unlock(&em_tree->lock); | |
4900 | ||
4901 | /* One for our allocation */ | |
4902 | free_extent_map(em); | |
4903 | /* One for the tree reference */ | |
4904 | free_extent_map(em); | |
495e64f4 FM |
4905 | /* One for the pending_chunks list reference */ |
4906 | free_extent_map(em); | |
b2117a39 | 4907 | error: |
b2117a39 MX |
4908 | kfree(devices_info); |
4909 | return ret; | |
2b82032c YZ |
4910 | } |
4911 | ||
6df9a95e | 4912 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
6bccf3ab | 4913 | struct btrfs_fs_info *fs_info, |
6df9a95e | 4914 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4915 | { |
6bccf3ab JM |
4916 | struct btrfs_root *extent_root = fs_info->extent_root; |
4917 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4918 | struct btrfs_key key; |
2b82032c YZ |
4919 | struct btrfs_device *device; |
4920 | struct btrfs_chunk *chunk; | |
4921 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4922 | struct extent_map *em; |
4923 | struct map_lookup *map; | |
4924 | size_t item_size; | |
4925 | u64 dev_offset; | |
4926 | u64 stripe_size; | |
4927 | int i = 0; | |
140e639f | 4928 | int ret = 0; |
2b82032c | 4929 | |
592d92ee LB |
4930 | em = get_chunk_map(fs_info, chunk_offset, chunk_size); |
4931 | if (IS_ERR(em)) | |
4932 | return PTR_ERR(em); | |
6df9a95e | 4933 | |
95617d69 | 4934 | map = em->map_lookup; |
6df9a95e JB |
4935 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4936 | stripe_size = em->orig_block_len; | |
4937 | ||
2b82032c | 4938 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4939 | if (!chunk) { |
4940 | ret = -ENOMEM; | |
4941 | goto out; | |
4942 | } | |
4943 | ||
50460e37 FM |
4944 | /* |
4945 | * Take the device list mutex to prevent races with the final phase of | |
4946 | * a device replace operation that replaces the device object associated | |
4947 | * with the map's stripes, because the device object's id can change | |
4948 | * at any time during that final phase of the device replace operation | |
4949 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4950 | */ | |
0b246afa | 4951 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
4952 | for (i = 0; i < map->num_stripes; i++) { |
4953 | device = map->stripes[i].dev; | |
4954 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4955 | |
0b86a832 | 4956 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4957 | if (ret) |
50460e37 | 4958 | break; |
6df9a95e JB |
4959 | ret = btrfs_alloc_dev_extent(trans, device, |
4960 | chunk_root->root_key.objectid, | |
4961 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4962 | chunk_offset, dev_offset, | |
4963 | stripe_size); | |
4964 | if (ret) | |
50460e37 FM |
4965 | break; |
4966 | } | |
4967 | if (ret) { | |
0b246afa | 4968 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 4969 | goto out; |
2b82032c YZ |
4970 | } |
4971 | ||
2b82032c | 4972 | stripe = &chunk->stripe; |
6df9a95e JB |
4973 | for (i = 0; i < map->num_stripes; i++) { |
4974 | device = map->stripes[i].dev; | |
4975 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4976 | |
e17cade2 CM |
4977 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4978 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4979 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4980 | stripe++; |
0b86a832 | 4981 | } |
0b246afa | 4982 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4983 | |
2b82032c | 4984 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4985 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4986 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4987 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4988 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4989 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4990 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 4991 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 4992 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4993 | |
2b82032c YZ |
4994 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4995 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4996 | key.offset = chunk_offset; | |
0b86a832 | 4997 | |
2b82032c | 4998 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4999 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5000 | /* | |
5001 | * TODO: Cleanup of inserted chunk root in case of | |
5002 | * failure. | |
5003 | */ | |
2ff7e61e | 5004 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5005 | } |
1abe9b8a | 5006 | |
6df9a95e | 5007 | out: |
0b86a832 | 5008 | kfree(chunk); |
6df9a95e | 5009 | free_extent_map(em); |
4ed1d16e | 5010 | return ret; |
2b82032c | 5011 | } |
0b86a832 | 5012 | |
2b82032c YZ |
5013 | /* |
5014 | * Chunk allocation falls into two parts. The first part does works | |
5015 | * that make the new allocated chunk useable, but not do any operation | |
5016 | * that modifies the chunk tree. The second part does the works that | |
5017 | * require modifying the chunk tree. This division is important for the | |
5018 | * bootstrap process of adding storage to a seed btrfs. | |
5019 | */ | |
5020 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 5021 | struct btrfs_fs_info *fs_info, u64 type) |
2b82032c YZ |
5022 | { |
5023 | u64 chunk_offset; | |
2b82032c | 5024 | |
0b246afa JM |
5025 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
5026 | chunk_offset = find_next_chunk(fs_info); | |
72b468c8 | 5027 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5028 | } |
5029 | ||
d397712b | 5030 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5031 | struct btrfs_fs_info *fs_info) |
2b82032c | 5032 | { |
2ff7e61e | 5033 | struct btrfs_root *extent_root = fs_info->extent_root; |
2b82032c YZ |
5034 | u64 chunk_offset; |
5035 | u64 sys_chunk_offset; | |
2b82032c | 5036 | u64 alloc_profile; |
2b82032c YZ |
5037 | int ret; |
5038 | ||
6df9a95e | 5039 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 5040 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
72b468c8 | 5041 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5042 | if (ret) |
5043 | return ret; | |
2b82032c | 5044 | |
0b246afa | 5045 | sys_chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 5046 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
72b468c8 | 5047 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5048 | return ret; |
2b82032c YZ |
5049 | } |
5050 | ||
d20983b4 MX |
5051 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5052 | { | |
5053 | int max_errors; | |
5054 | ||
5055 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5056 | BTRFS_BLOCK_GROUP_RAID10 | | |
5057 | BTRFS_BLOCK_GROUP_RAID5 | | |
5058 | BTRFS_BLOCK_GROUP_DUP)) { | |
5059 | max_errors = 1; | |
5060 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5061 | max_errors = 2; | |
5062 | } else { | |
5063 | max_errors = 0; | |
005d6427 | 5064 | } |
2b82032c | 5065 | |
d20983b4 | 5066 | return max_errors; |
2b82032c YZ |
5067 | } |
5068 | ||
2ff7e61e | 5069 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5070 | { |
5071 | struct extent_map *em; | |
5072 | struct map_lookup *map; | |
2b82032c | 5073 | int readonly = 0; |
d20983b4 | 5074 | int miss_ndevs = 0; |
2b82032c YZ |
5075 | int i; |
5076 | ||
592d92ee LB |
5077 | em = get_chunk_map(fs_info, chunk_offset, 1); |
5078 | if (IS_ERR(em)) | |
2b82032c YZ |
5079 | return 1; |
5080 | ||
95617d69 | 5081 | map = em->map_lookup; |
2b82032c | 5082 | for (i = 0; i < map->num_stripes; i++) { |
d20983b4 MX |
5083 | if (map->stripes[i].dev->missing) { |
5084 | miss_ndevs++; | |
5085 | continue; | |
5086 | } | |
5087 | ||
2b82032c YZ |
5088 | if (!map->stripes[i].dev->writeable) { |
5089 | readonly = 1; | |
d20983b4 | 5090 | goto end; |
2b82032c YZ |
5091 | } |
5092 | } | |
d20983b4 MX |
5093 | |
5094 | /* | |
5095 | * If the number of missing devices is larger than max errors, | |
5096 | * we can not write the data into that chunk successfully, so | |
5097 | * set it readonly. | |
5098 | */ | |
5099 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5100 | readonly = 1; | |
5101 | end: | |
0b86a832 | 5102 | free_extent_map(em); |
2b82032c | 5103 | return readonly; |
0b86a832 CM |
5104 | } |
5105 | ||
5106 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5107 | { | |
a8067e02 | 5108 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5109 | } |
5110 | ||
5111 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5112 | { | |
5113 | struct extent_map *em; | |
5114 | ||
d397712b | 5115 | while (1) { |
890871be | 5116 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5117 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5118 | if (em) | |
5119 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5120 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5121 | if (!em) |
5122 | break; | |
0b86a832 CM |
5123 | /* once for us */ |
5124 | free_extent_map(em); | |
5125 | /* once for the tree */ | |
5126 | free_extent_map(em); | |
5127 | } | |
5128 | } | |
5129 | ||
5d964051 | 5130 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5131 | { |
5132 | struct extent_map *em; | |
5133 | struct map_lookup *map; | |
f188591e CM |
5134 | int ret; |
5135 | ||
592d92ee LB |
5136 | em = get_chunk_map(fs_info, logical, len); |
5137 | if (IS_ERR(em)) | |
5138 | /* | |
5139 | * We could return errors for these cases, but that could get | |
5140 | * ugly and we'd probably do the same thing which is just not do | |
5141 | * anything else and exit, so return 1 so the callers don't try | |
5142 | * to use other copies. | |
5143 | */ | |
fb7669b5 | 5144 | return 1; |
fb7669b5 | 5145 | |
95617d69 | 5146 | map = em->map_lookup; |
f188591e CM |
5147 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5148 | ret = map->num_stripes; | |
321aecc6 CM |
5149 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5150 | ret = map->sub_stripes; | |
53b381b3 DW |
5151 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5152 | ret = 2; | |
5153 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5154 | ret = 3; | |
f188591e CM |
5155 | else |
5156 | ret = 1; | |
5157 | free_extent_map(em); | |
ad6d620e | 5158 | |
73beece9 | 5159 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
6fad823f LB |
5160 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5161 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5162 | ret++; |
73beece9 | 5163 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5164 | |
f188591e CM |
5165 | return ret; |
5166 | } | |
5167 | ||
2ff7e61e | 5168 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5169 | struct btrfs_mapping_tree *map_tree, |
5170 | u64 logical) | |
5171 | { | |
5172 | struct extent_map *em; | |
5173 | struct map_lookup *map; | |
0b246afa | 5174 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5175 | |
592d92ee | 5176 | em = get_chunk_map(fs_info, logical, len); |
539b50d2 | 5177 | WARN_ON(IS_ERR(em)); |
53b381b3 | 5178 | |
95617d69 | 5179 | map = em->map_lookup; |
ffe2d203 | 5180 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 5181 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
5182 | free_extent_map(em); |
5183 | return len; | |
5184 | } | |
5185 | ||
592d92ee | 5186 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5187 | u64 logical, u64 len, int mirror_num) |
5188 | { | |
5189 | struct extent_map *em; | |
5190 | struct map_lookup *map; | |
53b381b3 DW |
5191 | int ret = 0; |
5192 | ||
592d92ee | 5193 | em = get_chunk_map(fs_info, logical, len); |
539b50d2 | 5194 | WARN_ON(IS_ERR(em)); |
53b381b3 | 5195 | |
95617d69 | 5196 | map = em->map_lookup; |
ffe2d203 | 5197 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
5198 | ret = 1; |
5199 | free_extent_map(em); | |
5200 | return ret; | |
5201 | } | |
5202 | ||
30d9861f SB |
5203 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5204 | struct map_lookup *map, int first, int num, | |
5205 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5206 | { |
5207 | int i; | |
30d9861f SB |
5208 | int tolerance; |
5209 | struct btrfs_device *srcdev; | |
5210 | ||
5211 | if (dev_replace_is_ongoing && | |
5212 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5213 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5214 | srcdev = fs_info->dev_replace.srcdev; | |
5215 | else | |
5216 | srcdev = NULL; | |
5217 | ||
5218 | /* | |
5219 | * try to avoid the drive that is the source drive for a | |
5220 | * dev-replace procedure, only choose it if no other non-missing | |
5221 | * mirror is available | |
5222 | */ | |
5223 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5224 | if (map->stripes[optimal].dev->bdev && | |
5225 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5226 | return optimal; | |
5227 | for (i = first; i < first + num; i++) { | |
5228 | if (map->stripes[i].dev->bdev && | |
5229 | (tolerance || map->stripes[i].dev != srcdev)) | |
5230 | return i; | |
5231 | } | |
dfe25020 | 5232 | } |
30d9861f | 5233 | |
dfe25020 CM |
5234 | /* we couldn't find one that doesn't fail. Just return something |
5235 | * and the io error handling code will clean up eventually | |
5236 | */ | |
5237 | return optimal; | |
5238 | } | |
5239 | ||
53b381b3 DW |
5240 | static inline int parity_smaller(u64 a, u64 b) |
5241 | { | |
5242 | return a > b; | |
5243 | } | |
5244 | ||
5245 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5246 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5247 | { |
5248 | struct btrfs_bio_stripe s; | |
5249 | int i; | |
5250 | u64 l; | |
5251 | int again = 1; | |
5252 | ||
5253 | while (again) { | |
5254 | again = 0; | |
cc7539ed | 5255 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5256 | if (parity_smaller(bbio->raid_map[i], |
5257 | bbio->raid_map[i+1])) { | |
53b381b3 | 5258 | s = bbio->stripes[i]; |
8e5cfb55 | 5259 | l = bbio->raid_map[i]; |
53b381b3 | 5260 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5261 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5262 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5263 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5264 | |
53b381b3 DW |
5265 | again = 1; |
5266 | } | |
5267 | } | |
5268 | } | |
5269 | } | |
5270 | ||
6e9606d2 ZL |
5271 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5272 | { | |
5273 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5274 | /* the size of the btrfs_bio */ |
6e9606d2 | 5275 | sizeof(struct btrfs_bio) + |
e57cf21e | 5276 | /* plus the variable array for the stripes */ |
6e9606d2 | 5277 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5278 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5279 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5280 | /* |
5281 | * plus the raid_map, which includes both the tgt dev | |
5282 | * and the stripes | |
5283 | */ | |
5284 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5285 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5286 | |
5287 | atomic_set(&bbio->error, 0); | |
140475ae | 5288 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5289 | |
5290 | return bbio; | |
5291 | } | |
5292 | ||
5293 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5294 | { | |
140475ae ER |
5295 | WARN_ON(!refcount_read(&bbio->refs)); |
5296 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5297 | } |
5298 | ||
5299 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5300 | { | |
5301 | if (!bbio) | |
5302 | return; | |
140475ae | 5303 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5304 | kfree(bbio); |
5305 | } | |
5306 | ||
0b3d4cd3 LB |
5307 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5308 | /* | |
5309 | * Please note that, discard won't be sent to target device of device | |
5310 | * replace. | |
5311 | */ | |
5312 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5313 | u64 logical, u64 length, | |
5314 | struct btrfs_bio **bbio_ret) | |
5315 | { | |
5316 | struct extent_map *em; | |
5317 | struct map_lookup *map; | |
5318 | struct btrfs_bio *bbio; | |
5319 | u64 offset; | |
5320 | u64 stripe_nr; | |
5321 | u64 stripe_nr_end; | |
5322 | u64 stripe_end_offset; | |
5323 | u64 stripe_cnt; | |
5324 | u64 stripe_len; | |
5325 | u64 stripe_offset; | |
5326 | u64 num_stripes; | |
5327 | u32 stripe_index; | |
5328 | u32 factor = 0; | |
5329 | u32 sub_stripes = 0; | |
5330 | u64 stripes_per_dev = 0; | |
5331 | u32 remaining_stripes = 0; | |
5332 | u32 last_stripe = 0; | |
5333 | int ret = 0; | |
5334 | int i; | |
5335 | ||
5336 | /* discard always return a bbio */ | |
5337 | ASSERT(bbio_ret); | |
5338 | ||
5339 | em = get_chunk_map(fs_info, logical, length); | |
5340 | if (IS_ERR(em)) | |
5341 | return PTR_ERR(em); | |
5342 | ||
5343 | map = em->map_lookup; | |
5344 | /* we don't discard raid56 yet */ | |
5345 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5346 | ret = -EOPNOTSUPP; | |
5347 | goto out; | |
5348 | } | |
5349 | ||
5350 | offset = logical - em->start; | |
5351 | length = min_t(u64, em->len - offset, length); | |
5352 | ||
5353 | stripe_len = map->stripe_len; | |
5354 | /* | |
5355 | * stripe_nr counts the total number of stripes we have to stride | |
5356 | * to get to this block | |
5357 | */ | |
5358 | stripe_nr = div64_u64(offset, stripe_len); | |
5359 | ||
5360 | /* stripe_offset is the offset of this block in its stripe */ | |
5361 | stripe_offset = offset - stripe_nr * stripe_len; | |
5362 | ||
5363 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
5364 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); | |
5365 | stripe_cnt = stripe_nr_end - stripe_nr; | |
5366 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5367 | (offset + length); | |
5368 | /* | |
5369 | * after this, stripe_nr is the number of stripes on this | |
5370 | * device we have to walk to find the data, and stripe_index is | |
5371 | * the number of our device in the stripe array | |
5372 | */ | |
5373 | num_stripes = 1; | |
5374 | stripe_index = 0; | |
5375 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5376 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5377 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5378 | sub_stripes = 1; | |
5379 | else | |
5380 | sub_stripes = map->sub_stripes; | |
5381 | ||
5382 | factor = map->num_stripes / sub_stripes; | |
5383 | num_stripes = min_t(u64, map->num_stripes, | |
5384 | sub_stripes * stripe_cnt); | |
5385 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5386 | stripe_index *= sub_stripes; | |
5387 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5388 | &remaining_stripes); | |
5389 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5390 | last_stripe *= sub_stripes; | |
5391 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5392 | BTRFS_BLOCK_GROUP_DUP)) { | |
5393 | num_stripes = map->num_stripes; | |
5394 | } else { | |
5395 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5396 | &stripe_index); | |
5397 | } | |
5398 | ||
5399 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5400 | if (!bbio) { | |
5401 | ret = -ENOMEM; | |
5402 | goto out; | |
5403 | } | |
5404 | ||
5405 | for (i = 0; i < num_stripes; i++) { | |
5406 | bbio->stripes[i].physical = | |
5407 | map->stripes[stripe_index].physical + | |
5408 | stripe_offset + stripe_nr * map->stripe_len; | |
5409 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5410 | ||
5411 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5412 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5413 | bbio->stripes[i].length = stripes_per_dev * | |
5414 | map->stripe_len; | |
5415 | ||
5416 | if (i / sub_stripes < remaining_stripes) | |
5417 | bbio->stripes[i].length += | |
5418 | map->stripe_len; | |
5419 | ||
5420 | /* | |
5421 | * Special for the first stripe and | |
5422 | * the last stripe: | |
5423 | * | |
5424 | * |-------|...|-------| | |
5425 | * |----------| | |
5426 | * off end_off | |
5427 | */ | |
5428 | if (i < sub_stripes) | |
5429 | bbio->stripes[i].length -= | |
5430 | stripe_offset; | |
5431 | ||
5432 | if (stripe_index >= last_stripe && | |
5433 | stripe_index <= (last_stripe + | |
5434 | sub_stripes - 1)) | |
5435 | bbio->stripes[i].length -= | |
5436 | stripe_end_offset; | |
5437 | ||
5438 | if (i == sub_stripes - 1) | |
5439 | stripe_offset = 0; | |
5440 | } else { | |
5441 | bbio->stripes[i].length = length; | |
5442 | } | |
5443 | ||
5444 | stripe_index++; | |
5445 | if (stripe_index == map->num_stripes) { | |
5446 | stripe_index = 0; | |
5447 | stripe_nr++; | |
5448 | } | |
5449 | } | |
5450 | ||
5451 | *bbio_ret = bbio; | |
5452 | bbio->map_type = map->type; | |
5453 | bbio->num_stripes = num_stripes; | |
5454 | out: | |
5455 | free_extent_map(em); | |
5456 | return ret; | |
5457 | } | |
5458 | ||
5ab56090 LB |
5459 | /* |
5460 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5461 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5462 | * left cursor can also be read from the target drive. | |
5463 | * | |
5464 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5465 | * array of stripes. | |
5466 | * For READ, it also needs to be supported using the same mirror number. | |
5467 | * | |
5468 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5469 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5470 | * case. | |
5471 | */ | |
5472 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5473 | u64 logical, u64 length, | |
5474 | u64 srcdev_devid, int *mirror_num, | |
5475 | u64 *physical) | |
5476 | { | |
5477 | struct btrfs_bio *bbio = NULL; | |
5478 | int num_stripes; | |
5479 | int index_srcdev = 0; | |
5480 | int found = 0; | |
5481 | u64 physical_of_found = 0; | |
5482 | int i; | |
5483 | int ret = 0; | |
5484 | ||
5485 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5486 | logical, &length, &bbio, 0, 0); | |
5487 | if (ret) { | |
5488 | ASSERT(bbio == NULL); | |
5489 | return ret; | |
5490 | } | |
5491 | ||
5492 | num_stripes = bbio->num_stripes; | |
5493 | if (*mirror_num > num_stripes) { | |
5494 | /* | |
5495 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5496 | * that means that the requested area is not left of the left | |
5497 | * cursor | |
5498 | */ | |
5499 | btrfs_put_bbio(bbio); | |
5500 | return -EIO; | |
5501 | } | |
5502 | ||
5503 | /* | |
5504 | * process the rest of the function using the mirror_num of the source | |
5505 | * drive. Therefore look it up first. At the end, patch the device | |
5506 | * pointer to the one of the target drive. | |
5507 | */ | |
5508 | for (i = 0; i < num_stripes; i++) { | |
5509 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5510 | continue; | |
5511 | ||
5512 | /* | |
5513 | * In case of DUP, in order to keep it simple, only add the | |
5514 | * mirror with the lowest physical address | |
5515 | */ | |
5516 | if (found && | |
5517 | physical_of_found <= bbio->stripes[i].physical) | |
5518 | continue; | |
5519 | ||
5520 | index_srcdev = i; | |
5521 | found = 1; | |
5522 | physical_of_found = bbio->stripes[i].physical; | |
5523 | } | |
5524 | ||
5525 | btrfs_put_bbio(bbio); | |
5526 | ||
5527 | ASSERT(found); | |
5528 | if (!found) | |
5529 | return -EIO; | |
5530 | ||
5531 | *mirror_num = index_srcdev + 1; | |
5532 | *physical = physical_of_found; | |
5533 | return ret; | |
5534 | } | |
5535 | ||
73c0f228 LB |
5536 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5537 | struct btrfs_bio **bbio_ret, | |
5538 | struct btrfs_dev_replace *dev_replace, | |
5539 | int *num_stripes_ret, int *max_errors_ret) | |
5540 | { | |
5541 | struct btrfs_bio *bbio = *bbio_ret; | |
5542 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5543 | int tgtdev_indexes = 0; | |
5544 | int num_stripes = *num_stripes_ret; | |
5545 | int max_errors = *max_errors_ret; | |
5546 | int i; | |
5547 | ||
5548 | if (op == BTRFS_MAP_WRITE) { | |
5549 | int index_where_to_add; | |
5550 | ||
5551 | /* | |
5552 | * duplicate the write operations while the dev replace | |
5553 | * procedure is running. Since the copying of the old disk to | |
5554 | * the new disk takes place at run time while the filesystem is | |
5555 | * mounted writable, the regular write operations to the old | |
5556 | * disk have to be duplicated to go to the new disk as well. | |
5557 | * | |
5558 | * Note that device->missing is handled by the caller, and that | |
5559 | * the write to the old disk is already set up in the stripes | |
5560 | * array. | |
5561 | */ | |
5562 | index_where_to_add = num_stripes; | |
5563 | for (i = 0; i < num_stripes; i++) { | |
5564 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5565 | /* write to new disk, too */ | |
5566 | struct btrfs_bio_stripe *new = | |
5567 | bbio->stripes + index_where_to_add; | |
5568 | struct btrfs_bio_stripe *old = | |
5569 | bbio->stripes + i; | |
5570 | ||
5571 | new->physical = old->physical; | |
5572 | new->length = old->length; | |
5573 | new->dev = dev_replace->tgtdev; | |
5574 | bbio->tgtdev_map[i] = index_where_to_add; | |
5575 | index_where_to_add++; | |
5576 | max_errors++; | |
5577 | tgtdev_indexes++; | |
5578 | } | |
5579 | } | |
5580 | num_stripes = index_where_to_add; | |
5581 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5582 | int index_srcdev = 0; | |
5583 | int found = 0; | |
5584 | u64 physical_of_found = 0; | |
5585 | ||
5586 | /* | |
5587 | * During the dev-replace procedure, the target drive can also | |
5588 | * be used to read data in case it is needed to repair a corrupt | |
5589 | * block elsewhere. This is possible if the requested area is | |
5590 | * left of the left cursor. In this area, the target drive is a | |
5591 | * full copy of the source drive. | |
5592 | */ | |
5593 | for (i = 0; i < num_stripes; i++) { | |
5594 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5595 | /* | |
5596 | * In case of DUP, in order to keep it simple, | |
5597 | * only add the mirror with the lowest physical | |
5598 | * address | |
5599 | */ | |
5600 | if (found && | |
5601 | physical_of_found <= | |
5602 | bbio->stripes[i].physical) | |
5603 | continue; | |
5604 | index_srcdev = i; | |
5605 | found = 1; | |
5606 | physical_of_found = bbio->stripes[i].physical; | |
5607 | } | |
5608 | } | |
5609 | if (found) { | |
5610 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5611 | bbio->stripes + num_stripes; | |
5612 | ||
5613 | tgtdev_stripe->physical = physical_of_found; | |
5614 | tgtdev_stripe->length = | |
5615 | bbio->stripes[index_srcdev].length; | |
5616 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5617 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5618 | ||
5619 | tgtdev_indexes++; | |
5620 | num_stripes++; | |
5621 | } | |
5622 | } | |
5623 | ||
5624 | *num_stripes_ret = num_stripes; | |
5625 | *max_errors_ret = max_errors; | |
5626 | bbio->num_tgtdevs = tgtdev_indexes; | |
5627 | *bbio_ret = bbio; | |
5628 | } | |
5629 | ||
2b19a1fe LB |
5630 | static bool need_full_stripe(enum btrfs_map_op op) |
5631 | { | |
5632 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5633 | } | |
5634 | ||
cf8cddd3 CH |
5635 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5636 | enum btrfs_map_op op, | |
f2d8d74d | 5637 | u64 logical, u64 *length, |
a1d3c478 | 5638 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5639 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5640 | { |
5641 | struct extent_map *em; | |
5642 | struct map_lookup *map; | |
0b86a832 | 5643 | u64 offset; |
593060d7 CM |
5644 | u64 stripe_offset; |
5645 | u64 stripe_nr; | |
53b381b3 | 5646 | u64 stripe_len; |
9d644a62 | 5647 | u32 stripe_index; |
cea9e445 | 5648 | int i; |
de11cc12 | 5649 | int ret = 0; |
f2d8d74d | 5650 | int num_stripes; |
a236aed1 | 5651 | int max_errors = 0; |
2c8cdd6e | 5652 | int tgtdev_indexes = 0; |
a1d3c478 | 5653 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5654 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5655 | int dev_replace_is_ongoing = 0; | |
5656 | int num_alloc_stripes; | |
ad6d620e SB |
5657 | int patch_the_first_stripe_for_dev_replace = 0; |
5658 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5659 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5660 | |
0b3d4cd3 LB |
5661 | if (op == BTRFS_MAP_DISCARD) |
5662 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5663 | *length, bbio_ret); | |
5664 | ||
592d92ee LB |
5665 | em = get_chunk_map(fs_info, logical, *length); |
5666 | if (IS_ERR(em)) | |
5667 | return PTR_ERR(em); | |
0b86a832 | 5668 | |
95617d69 | 5669 | map = em->map_lookup; |
0b86a832 | 5670 | offset = logical - em->start; |
593060d7 | 5671 | |
53b381b3 | 5672 | stripe_len = map->stripe_len; |
593060d7 CM |
5673 | stripe_nr = offset; |
5674 | /* | |
5675 | * stripe_nr counts the total number of stripes we have to stride | |
5676 | * to get to this block | |
5677 | */ | |
47c5713f | 5678 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5679 | |
53b381b3 | 5680 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5681 | if (offset < stripe_offset) { |
5d163e0e JM |
5682 | btrfs_crit(fs_info, |
5683 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5684 | stripe_offset, offset, em->start, logical, |
5685 | stripe_len); | |
5686 | free_extent_map(em); | |
5687 | return -EINVAL; | |
5688 | } | |
593060d7 CM |
5689 | |
5690 | /* stripe_offset is the offset of this block in its stripe*/ | |
5691 | stripe_offset = offset - stripe_offset; | |
5692 | ||
53b381b3 | 5693 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5694 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5695 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5696 | raid56_full_stripe_start = offset; | |
5697 | ||
5698 | /* allow a write of a full stripe, but make sure we don't | |
5699 | * allow straddling of stripes | |
5700 | */ | |
47c5713f DS |
5701 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5702 | full_stripe_len); | |
53b381b3 DW |
5703 | raid56_full_stripe_start *= full_stripe_len; |
5704 | } | |
5705 | ||
0b3d4cd3 | 5706 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5707 | u64 max_len; |
5708 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5709 | For other RAID types and for RAID[56] reads, just allow a single | |
5710 | stripe (on a single disk). */ | |
ffe2d203 | 5711 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5712 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5713 | max_len = stripe_len * nr_data_stripes(map) - |
5714 | (offset - raid56_full_stripe_start); | |
5715 | } else { | |
5716 | /* we limit the length of each bio to what fits in a stripe */ | |
5717 | max_len = stripe_len - stripe_offset; | |
5718 | } | |
5719 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5720 | } else { |
5721 | *length = em->len - offset; | |
5722 | } | |
f2d8d74d | 5723 | |
53b381b3 DW |
5724 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5725 | it cares about is the length */ | |
a1d3c478 | 5726 | if (!bbio_ret) |
cea9e445 CM |
5727 | goto out; |
5728 | ||
73beece9 | 5729 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5730 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5731 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5732 | btrfs_dev_replace_unlock(dev_replace, 0); |
5733 | else | |
5734 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5735 | |
ad6d620e | 5736 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5737 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5738 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5739 | dev_replace->srcdev->devid, | |
5740 | &mirror_num, | |
5741 | &physical_to_patch_in_first_stripe); | |
5742 | if (ret) | |
ad6d620e | 5743 | goto out; |
5ab56090 LB |
5744 | else |
5745 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5746 | } else if (mirror_num > map->num_stripes) { |
5747 | mirror_num = 0; | |
5748 | } | |
5749 | ||
f2d8d74d | 5750 | num_stripes = 1; |
cea9e445 | 5751 | stripe_index = 0; |
fce3bb9a | 5752 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5753 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5754 | &stripe_index); | |
0b3d4cd3 | 5755 | if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS) |
28e1cc7d | 5756 | mirror_num = 1; |
fce3bb9a | 5757 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
0b3d4cd3 | 5758 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
f2d8d74d | 5759 | num_stripes = map->num_stripes; |
2fff734f | 5760 | else if (mirror_num) |
f188591e | 5761 | stripe_index = mirror_num - 1; |
dfe25020 | 5762 | else { |
30d9861f | 5763 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5764 | map->num_stripes, |
30d9861f SB |
5765 | current->pid % map->num_stripes, |
5766 | dev_replace_is_ongoing); | |
a1d3c478 | 5767 | mirror_num = stripe_index + 1; |
dfe25020 | 5768 | } |
2fff734f | 5769 | |
611f0e00 | 5770 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
0b3d4cd3 | 5771 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) { |
f2d8d74d | 5772 | num_stripes = map->num_stripes; |
a1d3c478 | 5773 | } else if (mirror_num) { |
f188591e | 5774 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5775 | } else { |
5776 | mirror_num = 1; | |
5777 | } | |
2fff734f | 5778 | |
321aecc6 | 5779 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5780 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5781 | |
47c5713f | 5782 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5783 | stripe_index *= map->sub_stripes; |
5784 | ||
cf8cddd3 | 5785 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
f2d8d74d | 5786 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5787 | else if (mirror_num) |
5788 | stripe_index += mirror_num - 1; | |
dfe25020 | 5789 | else { |
3e74317a | 5790 | int old_stripe_index = stripe_index; |
30d9861f SB |
5791 | stripe_index = find_live_mirror(fs_info, map, |
5792 | stripe_index, | |
dfe25020 | 5793 | map->sub_stripes, stripe_index + |
30d9861f SB |
5794 | current->pid % map->sub_stripes, |
5795 | dev_replace_is_ongoing); | |
3e74317a | 5796 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5797 | } |
53b381b3 | 5798 | |
ffe2d203 | 5799 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5800 | if (need_raid_map && |
cf8cddd3 | 5801 | (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS || |
af8e2d1d | 5802 | mirror_num > 1)) { |
53b381b3 | 5803 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5804 | stripe_nr = div_u64(raid56_full_stripe_start, |
5805 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5806 | |
5807 | /* RAID[56] write or recovery. Return all stripes */ | |
5808 | num_stripes = map->num_stripes; | |
5809 | max_errors = nr_parity_stripes(map); | |
5810 | ||
53b381b3 DW |
5811 | *length = map->stripe_len; |
5812 | stripe_index = 0; | |
5813 | stripe_offset = 0; | |
5814 | } else { | |
5815 | /* | |
5816 | * Mirror #0 or #1 means the original data block. | |
5817 | * Mirror #2 is RAID5 parity block. | |
5818 | * Mirror #3 is RAID6 Q block. | |
5819 | */ | |
47c5713f DS |
5820 | stripe_nr = div_u64_rem(stripe_nr, |
5821 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5822 | if (mirror_num > 1) |
5823 | stripe_index = nr_data_stripes(map) + | |
5824 | mirror_num - 2; | |
5825 | ||
5826 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5827 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5828 | &stripe_index); | |
0b3d4cd3 LB |
5829 | if ((op != BTRFS_MAP_WRITE && |
5830 | op != BTRFS_MAP_GET_READ_MIRRORS) && | |
5831 | mirror_num <= 1) | |
28e1cc7d | 5832 | mirror_num = 1; |
53b381b3 | 5833 | } |
8790d502 CM |
5834 | } else { |
5835 | /* | |
47c5713f DS |
5836 | * after this, stripe_nr is the number of stripes on this |
5837 | * device we have to walk to find the data, and stripe_index is | |
5838 | * the number of our device in the stripe array | |
8790d502 | 5839 | */ |
47c5713f DS |
5840 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5841 | &stripe_index); | |
a1d3c478 | 5842 | mirror_num = stripe_index + 1; |
8790d502 | 5843 | } |
e042d1ec | 5844 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5845 | btrfs_crit(fs_info, |
5846 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5847 | stripe_index, map->num_stripes); |
5848 | ret = -EINVAL; | |
5849 | goto out; | |
5850 | } | |
cea9e445 | 5851 | |
472262f3 | 5852 | num_alloc_stripes = num_stripes; |
6fad823f | 5853 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5854 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5855 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5856 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5857 | num_alloc_stripes++; |
2c8cdd6e | 5858 | tgtdev_indexes = num_stripes; |
ad6d620e | 5859 | } |
2c8cdd6e | 5860 | |
6e9606d2 | 5861 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5862 | if (!bbio) { |
5863 | ret = -ENOMEM; | |
5864 | goto out; | |
5865 | } | |
6fad823f | 5866 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5867 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5868 | |
8e5cfb55 | 5869 | /* build raid_map */ |
2b19a1fe LB |
5870 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5871 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5872 | u64 tmp; |
9d644a62 | 5873 | unsigned rot; |
8e5cfb55 ZL |
5874 | |
5875 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5876 | sizeof(struct btrfs_bio_stripe) * | |
5877 | num_alloc_stripes + | |
5878 | sizeof(int) * tgtdev_indexes); | |
5879 | ||
5880 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5881 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5882 | |
5883 | /* Fill in the logical address of each stripe */ | |
5884 | tmp = stripe_nr * nr_data_stripes(map); | |
5885 | for (i = 0; i < nr_data_stripes(map); i++) | |
5886 | bbio->raid_map[(i+rot) % num_stripes] = | |
5887 | em->start + (tmp + i) * map->stripe_len; | |
5888 | ||
5889 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5890 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5891 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5892 | RAID6_Q_STRIPE; | |
5893 | } | |
5894 | ||
b89203f7 | 5895 | |
0b3d4cd3 LB |
5896 | for (i = 0; i < num_stripes; i++) { |
5897 | bbio->stripes[i].physical = | |
5898 | map->stripes[stripe_index].physical + | |
5899 | stripe_offset + | |
5900 | stripe_nr * map->stripe_len; | |
5901 | bbio->stripes[i].dev = | |
5902 | map->stripes[stripe_index].dev; | |
5903 | stripe_index++; | |
593060d7 | 5904 | } |
de11cc12 | 5905 | |
2b19a1fe | 5906 | if (need_full_stripe(op)) |
d20983b4 | 5907 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5908 | |
8e5cfb55 ZL |
5909 | if (bbio->raid_map) |
5910 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5911 | |
73c0f228 | 5912 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5913 | need_full_stripe(op)) { |
73c0f228 LB |
5914 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5915 | &max_errors); | |
472262f3 SB |
5916 | } |
5917 | ||
de11cc12 | 5918 | *bbio_ret = bbio; |
10f11900 | 5919 | bbio->map_type = map->type; |
de11cc12 LZ |
5920 | bbio->num_stripes = num_stripes; |
5921 | bbio->max_errors = max_errors; | |
5922 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5923 | |
5924 | /* | |
5925 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5926 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5927 | * available as a mirror | |
5928 | */ | |
5929 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5930 | WARN_ON(num_stripes > 1); | |
5931 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5932 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5933 | bbio->mirror_num = map->num_stripes + 1; | |
5934 | } | |
cea9e445 | 5935 | out: |
73beece9 LB |
5936 | if (dev_replace_is_ongoing) { |
5937 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5938 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5939 | } | |
0b86a832 | 5940 | free_extent_map(em); |
de11cc12 | 5941 | return ret; |
0b86a832 CM |
5942 | } |
5943 | ||
cf8cddd3 | 5944 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 5945 | u64 logical, u64 *length, |
a1d3c478 | 5946 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5947 | { |
b3d3fa51 | 5948 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5949 | mirror_num, 0); |
f2d8d74d CM |
5950 | } |
5951 | ||
af8e2d1d | 5952 | /* For Scrub/replace */ |
cf8cddd3 | 5953 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d MX |
5954 | u64 logical, u64 *length, |
5955 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5956 | int need_raid_map) |
af8e2d1d | 5957 | { |
b3d3fa51 | 5958 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5959 | mirror_num, need_raid_map); |
af8e2d1d MX |
5960 | } |
5961 | ||
ab8d0fc4 | 5962 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
5963 | u64 chunk_start, u64 physical, u64 devid, |
5964 | u64 **logical, int *naddrs, int *stripe_len) | |
5965 | { | |
a512bbf8 YZ |
5966 | struct extent_map *em; |
5967 | struct map_lookup *map; | |
5968 | u64 *buf; | |
5969 | u64 bytenr; | |
5970 | u64 length; | |
5971 | u64 stripe_nr; | |
53b381b3 | 5972 | u64 rmap_len; |
a512bbf8 YZ |
5973 | int i, j, nr = 0; |
5974 | ||
592d92ee LB |
5975 | em = get_chunk_map(fs_info, chunk_start, 1); |
5976 | if (IS_ERR(em)) | |
835d974f | 5977 | return -EIO; |
835d974f | 5978 | |
95617d69 | 5979 | map = em->map_lookup; |
a512bbf8 | 5980 | length = em->len; |
53b381b3 DW |
5981 | rmap_len = map->stripe_len; |
5982 | ||
a512bbf8 | 5983 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5984 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5985 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5986 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5987 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5988 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5989 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5990 | } | |
a512bbf8 | 5991 | |
31e818fe | 5992 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5993 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5994 | |
5995 | for (i = 0; i < map->num_stripes; i++) { | |
5996 | if (devid && map->stripes[i].dev->devid != devid) | |
5997 | continue; | |
5998 | if (map->stripes[i].physical > physical || | |
5999 | map->stripes[i].physical + length <= physical) | |
6000 | continue; | |
6001 | ||
6002 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 6003 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6004 | |
6005 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6006 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6007 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6008 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6009 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6010 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6011 | * Alternatively, just use rmap_len below instead of | |
6012 | * map->stripe_len */ | |
6013 | ||
6014 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6015 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6016 | for (j = 0; j < nr; j++) { |
6017 | if (buf[j] == bytenr) | |
6018 | break; | |
6019 | } | |
934d375b CM |
6020 | if (j == nr) { |
6021 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6022 | buf[nr++] = bytenr; |
934d375b | 6023 | } |
a512bbf8 YZ |
6024 | } |
6025 | ||
a512bbf8 YZ |
6026 | *logical = buf; |
6027 | *naddrs = nr; | |
53b381b3 | 6028 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6029 | |
6030 | free_extent_map(em); | |
6031 | return 0; | |
f2d8d74d CM |
6032 | } |
6033 | ||
4246a0b6 | 6034 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6035 | { |
326e1dbb MS |
6036 | bio->bi_private = bbio->private; |
6037 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6038 | bio_endio(bio); |
326e1dbb | 6039 | |
6e9606d2 | 6040 | btrfs_put_bbio(bbio); |
8408c716 MX |
6041 | } |
6042 | ||
4246a0b6 | 6043 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6044 | { |
9be3395b | 6045 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6046 | int is_orig_bio = 0; |
8790d502 | 6047 | |
4246a0b6 | 6048 | if (bio->bi_error) { |
a1d3c478 | 6049 | atomic_inc(&bbio->error); |
4246a0b6 | 6050 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { |
442a4f63 | 6051 | unsigned int stripe_index = |
9be3395b | 6052 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6053 | struct btrfs_device *dev; |
442a4f63 SB |
6054 | |
6055 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6056 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6057 | if (dev->bdev) { |
37226b21 | 6058 | if (bio_op(bio) == REQ_OP_WRITE) |
597a60fa SB |
6059 | btrfs_dev_stat_inc(dev, |
6060 | BTRFS_DEV_STAT_WRITE_ERRS); | |
6061 | else | |
6062 | btrfs_dev_stat_inc(dev, | |
6063 | BTRFS_DEV_STAT_READ_ERRS); | |
70fd7614 | 6064 | if (bio->bi_opf & REQ_PREFLUSH) |
597a60fa SB |
6065 | btrfs_dev_stat_inc(dev, |
6066 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
6067 | btrfs_dev_stat_print_on_error(dev); | |
6068 | } | |
442a4f63 SB |
6069 | } |
6070 | } | |
8790d502 | 6071 | |
a1d3c478 | 6072 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6073 | is_orig_bio = 1; |
6074 | ||
c404e0dc MX |
6075 | btrfs_bio_counter_dec(bbio->fs_info); |
6076 | ||
a1d3c478 | 6077 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6078 | if (!is_orig_bio) { |
6079 | bio_put(bio); | |
a1d3c478 | 6080 | bio = bbio->orig_bio; |
7d2b4daa | 6081 | } |
c7b22bb1 | 6082 | |
9be3395b | 6083 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6084 | /* only send an error to the higher layers if it is |
53b381b3 | 6085 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6086 | */ |
a1d3c478 | 6087 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4246a0b6 | 6088 | bio->bi_error = -EIO; |
5dbc8fca | 6089 | } else { |
1259ab75 CM |
6090 | /* |
6091 | * this bio is actually up to date, we didn't | |
6092 | * go over the max number of errors | |
6093 | */ | |
4246a0b6 | 6094 | bio->bi_error = 0; |
1259ab75 | 6095 | } |
c55f1396 | 6096 | |
4246a0b6 | 6097 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6098 | } else if (!is_orig_bio) { |
8790d502 CM |
6099 | bio_put(bio); |
6100 | } | |
8790d502 CM |
6101 | } |
6102 | ||
8b712842 CM |
6103 | /* |
6104 | * see run_scheduled_bios for a description of why bios are collected for | |
6105 | * async submit. | |
6106 | * | |
6107 | * This will add one bio to the pending list for a device and make sure | |
6108 | * the work struct is scheduled. | |
6109 | */ | |
2ff7e61e | 6110 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6111 | struct bio *bio) |
8b712842 | 6112 | { |
0b246afa | 6113 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6114 | int should_queue = 1; |
ffbd517d | 6115 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6116 | |
53b381b3 | 6117 | if (device->missing || !device->bdev) { |
4246a0b6 | 6118 | bio_io_error(bio); |
53b381b3 DW |
6119 | return; |
6120 | } | |
6121 | ||
8b712842 | 6122 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6123 | if (bio_op(bio) == REQ_OP_READ) { |
492bb6de | 6124 | bio_get(bio); |
4e49ea4a | 6125 | btrfsic_submit_bio(bio); |
492bb6de | 6126 | bio_put(bio); |
143bede5 | 6127 | return; |
8b712842 CM |
6128 | } |
6129 | ||
6130 | /* | |
0986fe9e | 6131 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
6132 | * higher layers. Otherwise, the async bio makes it appear we have |
6133 | * made progress against dirty pages when we've really just put it | |
6134 | * on a queue for later | |
6135 | */ | |
0b246afa | 6136 | atomic_inc(&fs_info->nr_async_bios); |
492bb6de | 6137 | WARN_ON(bio->bi_next); |
8b712842 | 6138 | bio->bi_next = NULL; |
8b712842 CM |
6139 | |
6140 | spin_lock(&device->io_lock); | |
67f055c7 | 6141 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6142 | pending_bios = &device->pending_sync_bios; |
6143 | else | |
6144 | pending_bios = &device->pending_bios; | |
8b712842 | 6145 | |
ffbd517d CM |
6146 | if (pending_bios->tail) |
6147 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6148 | |
ffbd517d CM |
6149 | pending_bios->tail = bio; |
6150 | if (!pending_bios->head) | |
6151 | pending_bios->head = bio; | |
8b712842 CM |
6152 | if (device->running_pending) |
6153 | should_queue = 0; | |
6154 | ||
6155 | spin_unlock(&device->io_lock); | |
6156 | ||
6157 | if (should_queue) | |
0b246afa | 6158 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6159 | } |
6160 | ||
2ff7e61e JM |
6161 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6162 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6163 | { |
6164 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6165 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6166 | |
6167 | bio->bi_private = bbio; | |
9be3395b | 6168 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6169 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6170 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6171 | #ifdef DEBUG |
6172 | { | |
6173 | struct rcu_string *name; | |
6174 | ||
6175 | rcu_read_lock(); | |
6176 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6177 | btrfs_debug(fs_info, |
6178 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6179 | bio_op(bio), bio->bi_opf, | |
6180 | (u64)bio->bi_iter.bi_sector, | |
6181 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6182 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6183 | rcu_read_unlock(); |
6184 | } | |
6185 | #endif | |
6186 | bio->bi_bdev = dev->bdev; | |
c404e0dc | 6187 | |
2ff7e61e | 6188 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6189 | |
de1ee92a | 6190 | if (async) |
2ff7e61e | 6191 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6192 | else |
4e49ea4a | 6193 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6194 | } |
6195 | ||
de1ee92a JB |
6196 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6197 | { | |
6198 | atomic_inc(&bbio->error); | |
6199 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6200 | /* Should be the original bio. */ |
8408c716 MX |
6201 | WARN_ON(bio != bbio->orig_bio); |
6202 | ||
9be3395b | 6203 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6204 | bio->bi_iter.bi_sector = logical >> 9; |
4246a0b6 CH |
6205 | bio->bi_error = -EIO; |
6206 | btrfs_end_bbio(bbio, bio); | |
de1ee92a JB |
6207 | } |
6208 | } | |
6209 | ||
2ff7e61e | 6210 | int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
8b712842 | 6211 | int mirror_num, int async_submit) |
0b86a832 | 6212 | { |
0b86a832 | 6213 | struct btrfs_device *dev; |
8790d502 | 6214 | struct bio *first_bio = bio; |
4f024f37 | 6215 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6216 | u64 length = 0; |
6217 | u64 map_length; | |
0b86a832 | 6218 | int ret; |
08da757d ZL |
6219 | int dev_nr; |
6220 | int total_devs; | |
a1d3c478 | 6221 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6222 | |
4f024f37 | 6223 | length = bio->bi_iter.bi_size; |
0b86a832 | 6224 | map_length = length; |
cea9e445 | 6225 | |
0b246afa JM |
6226 | btrfs_bio_counter_inc_blocked(fs_info); |
6227 | ret = __btrfs_map_block(fs_info, bio_op(bio), logical, | |
37226b21 | 6228 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6229 | if (ret) { |
0b246afa | 6230 | btrfs_bio_counter_dec(fs_info); |
79787eaa | 6231 | return ret; |
c404e0dc | 6232 | } |
cea9e445 | 6233 | |
a1d3c478 | 6234 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6235 | bbio->orig_bio = first_bio; |
6236 | bbio->private = first_bio->bi_private; | |
6237 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6238 | bbio->fs_info = fs_info; |
53b381b3 DW |
6239 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6240 | ||
ad1ba2a0 | 6241 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6242 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6243 | /* In this case, map_length has been set to the length of |
6244 | a single stripe; not the whole write */ | |
37226b21 | 6245 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6246 | ret = raid56_parity_write(fs_info, bio, bbio, |
6247 | map_length); | |
53b381b3 | 6248 | } else { |
2ff7e61e JM |
6249 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6250 | map_length, mirror_num, 1); | |
53b381b3 | 6251 | } |
4245215d | 6252 | |
0b246afa | 6253 | btrfs_bio_counter_dec(fs_info); |
c404e0dc | 6254 | return ret; |
53b381b3 DW |
6255 | } |
6256 | ||
cea9e445 | 6257 | if (map_length < length) { |
0b246afa | 6258 | btrfs_crit(fs_info, |
5d163e0e JM |
6259 | "mapping failed logical %llu bio len %llu len %llu", |
6260 | logical, length, map_length); | |
cea9e445 CM |
6261 | BUG(); |
6262 | } | |
a1d3c478 | 6263 | |
08da757d | 6264 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6265 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6266 | if (!dev || !dev->bdev || |
a967efb3 | 6267 | (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) { |
de1ee92a | 6268 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6269 | continue; |
6270 | } | |
6271 | ||
a1d3c478 | 6272 | if (dev_nr < total_devs - 1) { |
9be3395b | 6273 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 6274 | BUG_ON(!bio); /* -ENOMEM */ |
326e1dbb | 6275 | } else |
a1d3c478 | 6276 | bio = first_bio; |
de1ee92a | 6277 | |
2ff7e61e JM |
6278 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6279 | dev_nr, async_submit); | |
8790d502 | 6280 | } |
0b246afa | 6281 | btrfs_bio_counter_dec(fs_info); |
0b86a832 CM |
6282 | return 0; |
6283 | } | |
6284 | ||
aa1b8cd4 | 6285 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6286 | u8 *uuid, u8 *fsid) |
0b86a832 | 6287 | { |
2b82032c YZ |
6288 | struct btrfs_device *device; |
6289 | struct btrfs_fs_devices *cur_devices; | |
6290 | ||
aa1b8cd4 | 6291 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6292 | while (cur_devices) { |
6293 | if (!fsid || | |
6294 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6295 | device = __find_device(&cur_devices->devices, | |
6296 | devid, uuid); | |
6297 | if (device) | |
6298 | return device; | |
6299 | } | |
6300 | cur_devices = cur_devices->seed; | |
6301 | } | |
6302 | return NULL; | |
0b86a832 CM |
6303 | } |
6304 | ||
2ff7e61e | 6305 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6306 | u64 devid, u8 *dev_uuid) |
6307 | { | |
6308 | struct btrfs_device *device; | |
dfe25020 | 6309 | |
12bd2fc0 ID |
6310 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6311 | if (IS_ERR(device)) | |
7cbd8a83 | 6312 | return NULL; |
12bd2fc0 ID |
6313 | |
6314 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6315 | device->fs_devices = fs_devices; |
dfe25020 | 6316 | fs_devices->num_devices++; |
12bd2fc0 ID |
6317 | |
6318 | device->missing = 1; | |
cd02dca5 | 6319 | fs_devices->missing_devices++; |
12bd2fc0 | 6320 | |
dfe25020 CM |
6321 | return device; |
6322 | } | |
6323 | ||
12bd2fc0 ID |
6324 | /** |
6325 | * btrfs_alloc_device - allocate struct btrfs_device | |
6326 | * @fs_info: used only for generating a new devid, can be NULL if | |
6327 | * devid is provided (i.e. @devid != NULL). | |
6328 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6329 | * is generated. | |
6330 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6331 | * is generated. | |
6332 | * | |
6333 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6334 | * on error. Returned struct is not linked onto any lists and can be | |
6335 | * destroyed with kfree() right away. | |
6336 | */ | |
6337 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6338 | const u64 *devid, | |
6339 | const u8 *uuid) | |
6340 | { | |
6341 | struct btrfs_device *dev; | |
6342 | u64 tmp; | |
6343 | ||
fae7f21c | 6344 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6345 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6346 | |
6347 | dev = __alloc_device(); | |
6348 | if (IS_ERR(dev)) | |
6349 | return dev; | |
6350 | ||
6351 | if (devid) | |
6352 | tmp = *devid; | |
6353 | else { | |
6354 | int ret; | |
6355 | ||
6356 | ret = find_next_devid(fs_info, &tmp); | |
6357 | if (ret) { | |
6358 | kfree(dev); | |
6359 | return ERR_PTR(ret); | |
6360 | } | |
6361 | } | |
6362 | dev->devid = tmp; | |
6363 | ||
6364 | if (uuid) | |
6365 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6366 | else | |
6367 | generate_random_uuid(dev->uuid); | |
6368 | ||
9e0af237 LB |
6369 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6370 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6371 | |
6372 | return dev; | |
6373 | } | |
6374 | ||
e06cd3dd | 6375 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6376 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6377 | struct extent_buffer *leaf, |
6378 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6379 | { |
0b86a832 | 6380 | u64 length; |
f04b772b | 6381 | u64 stripe_len; |
e06cd3dd LB |
6382 | u16 num_stripes; |
6383 | u16 sub_stripes; | |
6384 | u64 type; | |
0b86a832 | 6385 | |
e17cade2 | 6386 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6387 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6388 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6389 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6390 | type = btrfs_chunk_type(leaf, chunk); | |
6391 | ||
f04b772b | 6392 | if (!num_stripes) { |
0b246afa | 6393 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6394 | num_stripes); |
6395 | return -EIO; | |
6396 | } | |
0b246afa JM |
6397 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6398 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6399 | return -EIO; |
6400 | } | |
0b246afa JM |
6401 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6402 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6403 | btrfs_chunk_sector_size(leaf, chunk)); |
6404 | return -EIO; | |
6405 | } | |
0b246afa JM |
6406 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6407 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6408 | return -EIO; |
6409 | } | |
3d8da678 | 6410 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6411 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6412 | stripe_len); |
6413 | return -EIO; | |
6414 | } | |
6415 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6416 | type) { |
0b246afa | 6417 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6418 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6419 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6420 | btrfs_chunk_type(leaf, chunk)); | |
6421 | return -EIO; | |
6422 | } | |
e06cd3dd LB |
6423 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6424 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6425 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6426 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6427 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6428 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6429 | num_stripes != 1)) { | |
0b246afa | 6430 | btrfs_err(fs_info, |
e06cd3dd LB |
6431 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6432 | num_stripes, sub_stripes, | |
6433 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6434 | return -EIO; | |
6435 | } | |
6436 | ||
6437 | return 0; | |
6438 | } | |
6439 | ||
2ff7e61e | 6440 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6441 | struct extent_buffer *leaf, |
6442 | struct btrfs_chunk *chunk) | |
6443 | { | |
0b246afa | 6444 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6445 | struct map_lookup *map; |
6446 | struct extent_map *em; | |
6447 | u64 logical; | |
6448 | u64 length; | |
6449 | u64 stripe_len; | |
6450 | u64 devid; | |
6451 | u8 uuid[BTRFS_UUID_SIZE]; | |
6452 | int num_stripes; | |
6453 | int ret; | |
6454 | int i; | |
6455 | ||
6456 | logical = key->offset; | |
6457 | length = btrfs_chunk_length(leaf, chunk); | |
6458 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6459 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
6460 | ||
2ff7e61e | 6461 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6462 | if (ret) |
6463 | return ret; | |
a061fc8d | 6464 | |
890871be | 6465 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6466 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6467 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6468 | |
6469 | /* already mapped? */ | |
6470 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6471 | free_extent_map(em); | |
0b86a832 CM |
6472 | return 0; |
6473 | } else if (em) { | |
6474 | free_extent_map(em); | |
6475 | } | |
0b86a832 | 6476 | |
172ddd60 | 6477 | em = alloc_extent_map(); |
0b86a832 CM |
6478 | if (!em) |
6479 | return -ENOMEM; | |
593060d7 | 6480 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6481 | if (!map) { |
6482 | free_extent_map(em); | |
6483 | return -ENOMEM; | |
6484 | } | |
6485 | ||
298a8f9c | 6486 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6487 | em->map_lookup = map; |
0b86a832 CM |
6488 | em->start = logical; |
6489 | em->len = length; | |
70c8a91c | 6490 | em->orig_start = 0; |
0b86a832 | 6491 | em->block_start = 0; |
c8b97818 | 6492 | em->block_len = em->len; |
0b86a832 | 6493 | |
593060d7 CM |
6494 | map->num_stripes = num_stripes; |
6495 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6496 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6497 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6498 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6499 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6500 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6501 | for (i = 0; i < num_stripes; i++) { |
6502 | map->stripes[i].physical = | |
6503 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6504 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6505 | read_extent_buffer(leaf, uuid, (unsigned long) |
6506 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6507 | BTRFS_UUID_SIZE); | |
0b246afa | 6508 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6509 | uuid, NULL); |
3cdde224 | 6510 | if (!map->stripes[i].dev && |
0b246afa | 6511 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 CM |
6512 | free_extent_map(em); |
6513 | return -EIO; | |
6514 | } | |
dfe25020 CM |
6515 | if (!map->stripes[i].dev) { |
6516 | map->stripes[i].dev = | |
2ff7e61e JM |
6517 | add_missing_dev(fs_info->fs_devices, devid, |
6518 | uuid); | |
dfe25020 | 6519 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6520 | free_extent_map(em); |
6521 | return -EIO; | |
6522 | } | |
0b246afa | 6523 | btrfs_warn(fs_info, "devid %llu uuid %pU is missing", |
5d163e0e | 6524 | devid, uuid); |
dfe25020 CM |
6525 | } |
6526 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6527 | } |
6528 | ||
890871be | 6529 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6530 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6531 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6532 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6533 | free_extent_map(em); |
6534 | ||
6535 | return 0; | |
6536 | } | |
6537 | ||
143bede5 | 6538 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6539 | struct btrfs_dev_item *dev_item, |
6540 | struct btrfs_device *device) | |
6541 | { | |
6542 | unsigned long ptr; | |
0b86a832 CM |
6543 | |
6544 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6545 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6546 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6547 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6548 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6549 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6550 | device->type = btrfs_device_type(leaf, dev_item); |
6551 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6552 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6553 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6554 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6555 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6556 | |
410ba3a2 | 6557 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6558 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6559 | } |
6560 | ||
2ff7e61e | 6561 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6562 | u8 *fsid) |
2b82032c YZ |
6563 | { |
6564 | struct btrfs_fs_devices *fs_devices; | |
6565 | int ret; | |
6566 | ||
b367e47f | 6567 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c | 6568 | |
0b246afa | 6569 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6570 | while (fs_devices) { |
5f375835 MX |
6571 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6572 | return fs_devices; | |
6573 | ||
2b82032c YZ |
6574 | fs_devices = fs_devices->seed; |
6575 | } | |
6576 | ||
6577 | fs_devices = find_fsid(fsid); | |
6578 | if (!fs_devices) { | |
0b246afa | 6579 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6580 | return ERR_PTR(-ENOENT); |
6581 | ||
6582 | fs_devices = alloc_fs_devices(fsid); | |
6583 | if (IS_ERR(fs_devices)) | |
6584 | return fs_devices; | |
6585 | ||
6586 | fs_devices->seeding = 1; | |
6587 | fs_devices->opened = 1; | |
6588 | return fs_devices; | |
2b82032c | 6589 | } |
e4404d6e YZ |
6590 | |
6591 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6592 | if (IS_ERR(fs_devices)) |
6593 | return fs_devices; | |
2b82032c | 6594 | |
97288f2c | 6595 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
0b246afa | 6596 | fs_info->bdev_holder); |
48d28232 JL |
6597 | if (ret) { |
6598 | free_fs_devices(fs_devices); | |
5f375835 | 6599 | fs_devices = ERR_PTR(ret); |
2b82032c | 6600 | goto out; |
48d28232 | 6601 | } |
2b82032c YZ |
6602 | |
6603 | if (!fs_devices->seeding) { | |
6604 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6605 | free_fs_devices(fs_devices); |
5f375835 | 6606 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6607 | goto out; |
6608 | } | |
6609 | ||
0b246afa JM |
6610 | fs_devices->seed = fs_info->fs_devices->seed; |
6611 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6612 | out: |
5f375835 | 6613 | return fs_devices; |
2b82032c YZ |
6614 | } |
6615 | ||
2ff7e61e | 6616 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6617 | struct extent_buffer *leaf, |
6618 | struct btrfs_dev_item *dev_item) | |
6619 | { | |
0b246afa | 6620 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6621 | struct btrfs_device *device; |
6622 | u64 devid; | |
6623 | int ret; | |
2b82032c | 6624 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6625 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6626 | ||
0b86a832 | 6627 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6628 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6629 | BTRFS_UUID_SIZE); |
1473b24e | 6630 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6631 | BTRFS_UUID_SIZE); |
6632 | ||
0b246afa | 6633 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) { |
2ff7e61e | 6634 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6635 | if (IS_ERR(fs_devices)) |
6636 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6637 | } |
6638 | ||
0b246afa | 6639 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6640 | if (!device) { |
0b246afa | 6641 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
2b82032c YZ |
6642 | return -EIO; |
6643 | ||
2ff7e61e | 6644 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
5f375835 MX |
6645 | if (!device) |
6646 | return -ENOMEM; | |
0b246afa | 6647 | btrfs_warn(fs_info, "devid %llu uuid %pU missing", |
33b97e43 | 6648 | devid, dev_uuid); |
5f375835 | 6649 | } else { |
0b246afa | 6650 | if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6651 | return -EIO; |
6652 | ||
6653 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6654 | /* |
6655 | * this happens when a device that was properly setup | |
6656 | * in the device info lists suddenly goes bad. | |
6657 | * device->bdev is NULL, and so we have to set | |
6658 | * device->missing to one here | |
6659 | */ | |
5f375835 | 6660 | device->fs_devices->missing_devices++; |
cd02dca5 | 6661 | device->missing = 1; |
2b82032c | 6662 | } |
5f375835 MX |
6663 | |
6664 | /* Move the device to its own fs_devices */ | |
6665 | if (device->fs_devices != fs_devices) { | |
6666 | ASSERT(device->missing); | |
6667 | ||
6668 | list_move(&device->dev_list, &fs_devices->devices); | |
6669 | device->fs_devices->num_devices--; | |
6670 | fs_devices->num_devices++; | |
6671 | ||
6672 | device->fs_devices->missing_devices--; | |
6673 | fs_devices->missing_devices++; | |
6674 | ||
6675 | device->fs_devices = fs_devices; | |
6676 | } | |
2b82032c YZ |
6677 | } |
6678 | ||
0b246afa | 6679 | if (device->fs_devices != fs_info->fs_devices) { |
2b82032c YZ |
6680 | BUG_ON(device->writeable); |
6681 | if (device->generation != | |
6682 | btrfs_device_generation(leaf, dev_item)) | |
6683 | return -EINVAL; | |
6324fbf3 | 6684 | } |
0b86a832 CM |
6685 | |
6686 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6687 | device->in_fs_metadata = 1; |
63a212ab | 6688 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6689 | device->fs_devices->total_rw_bytes += device->total_bytes; |
0b246afa JM |
6690 | spin_lock(&fs_info->free_chunk_lock); |
6691 | fs_info->free_chunk_space += device->total_bytes - | |
2bf64758 | 6692 | device->bytes_used; |
0b246afa | 6693 | spin_unlock(&fs_info->free_chunk_lock); |
2bf64758 | 6694 | } |
0b86a832 | 6695 | ret = 0; |
0b86a832 CM |
6696 | return ret; |
6697 | } | |
6698 | ||
6bccf3ab | 6699 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6700 | { |
6bccf3ab | 6701 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6702 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6703 | struct extent_buffer *sb; |
0b86a832 | 6704 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6705 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6706 | u8 *array_ptr; |
6707 | unsigned long sb_array_offset; | |
84eed90f | 6708 | int ret = 0; |
0b86a832 CM |
6709 | u32 num_stripes; |
6710 | u32 array_size; | |
6711 | u32 len = 0; | |
1ffb22cf | 6712 | u32 cur_offset; |
e06cd3dd | 6713 | u64 type; |
84eed90f | 6714 | struct btrfs_key key; |
0b86a832 | 6715 | |
0b246afa | 6716 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6717 | /* |
6718 | * This will create extent buffer of nodesize, superblock size is | |
6719 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6720 | * overallocate but we can keep it as-is, only the first page is used. | |
6721 | */ | |
2ff7e61e | 6722 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6723 | if (IS_ERR(sb)) |
6724 | return PTR_ERR(sb); | |
4db8c528 | 6725 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6726 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6727 | /* |
01327610 | 6728 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6729 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6730 | * pages up-to-date when the page is larger: extent does not cover the |
6731 | * whole page and consequently check_page_uptodate does not find all | |
6732 | * the page's extents up-to-date (the hole beyond sb), | |
6733 | * write_extent_buffer then triggers a WARN_ON. | |
6734 | * | |
6735 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6736 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6737 | * to silence the warning eg. on PowerPC 64. | |
6738 | */ | |
09cbfeaf | 6739 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6740 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6741 | |
a061fc8d | 6742 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6743 | array_size = btrfs_super_sys_array_size(super_copy); |
6744 | ||
1ffb22cf DS |
6745 | array_ptr = super_copy->sys_chunk_array; |
6746 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6747 | cur_offset = 0; | |
0b86a832 | 6748 | |
1ffb22cf DS |
6749 | while (cur_offset < array_size) { |
6750 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6751 | len = sizeof(*disk_key); |
6752 | if (cur_offset + len > array_size) | |
6753 | goto out_short_read; | |
6754 | ||
0b86a832 CM |
6755 | btrfs_disk_key_to_cpu(&key, disk_key); |
6756 | ||
1ffb22cf DS |
6757 | array_ptr += len; |
6758 | sb_array_offset += len; | |
6759 | cur_offset += len; | |
0b86a832 | 6760 | |
0d81ba5d | 6761 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6762 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6763 | /* |
6764 | * At least one btrfs_chunk with one stripe must be | |
6765 | * present, exact stripe count check comes afterwards | |
6766 | */ | |
6767 | len = btrfs_chunk_item_size(1); | |
6768 | if (cur_offset + len > array_size) | |
6769 | goto out_short_read; | |
6770 | ||
6771 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6772 | if (!num_stripes) { |
ab8d0fc4 JM |
6773 | btrfs_err(fs_info, |
6774 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6775 | num_stripes, cur_offset); |
6776 | ret = -EIO; | |
6777 | break; | |
6778 | } | |
6779 | ||
e06cd3dd LB |
6780 | type = btrfs_chunk_type(sb, chunk); |
6781 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6782 | btrfs_err(fs_info, |
e06cd3dd LB |
6783 | "invalid chunk type %llu in sys_array at offset %u", |
6784 | type, cur_offset); | |
6785 | ret = -EIO; | |
6786 | break; | |
6787 | } | |
6788 | ||
e3540eab DS |
6789 | len = btrfs_chunk_item_size(num_stripes); |
6790 | if (cur_offset + len > array_size) | |
6791 | goto out_short_read; | |
6792 | ||
2ff7e61e | 6793 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6794 | if (ret) |
6795 | break; | |
0b86a832 | 6796 | } else { |
ab8d0fc4 JM |
6797 | btrfs_err(fs_info, |
6798 | "unexpected item type %u in sys_array at offset %u", | |
6799 | (u32)key.type, cur_offset); | |
84eed90f CM |
6800 | ret = -EIO; |
6801 | break; | |
0b86a832 | 6802 | } |
1ffb22cf DS |
6803 | array_ptr += len; |
6804 | sb_array_offset += len; | |
6805 | cur_offset += len; | |
0b86a832 | 6806 | } |
d865177a | 6807 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6808 | free_extent_buffer_stale(sb); |
84eed90f | 6809 | return ret; |
e3540eab DS |
6810 | |
6811 | out_short_read: | |
ab8d0fc4 | 6812 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6813 | len, cur_offset); |
d865177a | 6814 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6815 | free_extent_buffer_stale(sb); |
e3540eab | 6816 | return -EIO; |
0b86a832 CM |
6817 | } |
6818 | ||
5b4aacef | 6819 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6820 | { |
5b4aacef | 6821 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6822 | struct btrfs_path *path; |
6823 | struct extent_buffer *leaf; | |
6824 | struct btrfs_key key; | |
6825 | struct btrfs_key found_key; | |
6826 | int ret; | |
6827 | int slot; | |
99e3ecfc | 6828 | u64 total_dev = 0; |
0b86a832 | 6829 | |
0b86a832 CM |
6830 | path = btrfs_alloc_path(); |
6831 | if (!path) | |
6832 | return -ENOMEM; | |
6833 | ||
b367e47f | 6834 | mutex_lock(&uuid_mutex); |
34441361 | 6835 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6836 | |
395927a9 FDBM |
6837 | /* |
6838 | * Read all device items, and then all the chunk items. All | |
6839 | * device items are found before any chunk item (their object id | |
6840 | * is smaller than the lowest possible object id for a chunk | |
6841 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6842 | */ |
6843 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6844 | key.offset = 0; | |
6845 | key.type = 0; | |
0b86a832 | 6846 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6847 | if (ret < 0) |
6848 | goto error; | |
d397712b | 6849 | while (1) { |
0b86a832 CM |
6850 | leaf = path->nodes[0]; |
6851 | slot = path->slots[0]; | |
6852 | if (slot >= btrfs_header_nritems(leaf)) { | |
6853 | ret = btrfs_next_leaf(root, path); | |
6854 | if (ret == 0) | |
6855 | continue; | |
6856 | if (ret < 0) | |
6857 | goto error; | |
6858 | break; | |
6859 | } | |
6860 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6861 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6862 | struct btrfs_dev_item *dev_item; | |
6863 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6864 | struct btrfs_dev_item); |
2ff7e61e | 6865 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
6866 | if (ret) |
6867 | goto error; | |
99e3ecfc | 6868 | total_dev++; |
0b86a832 CM |
6869 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6870 | struct btrfs_chunk *chunk; | |
6871 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 6872 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
6873 | if (ret) |
6874 | goto error; | |
0b86a832 CM |
6875 | } |
6876 | path->slots[0]++; | |
6877 | } | |
99e3ecfc LB |
6878 | |
6879 | /* | |
6880 | * After loading chunk tree, we've got all device information, | |
6881 | * do another round of validation checks. | |
6882 | */ | |
0b246afa JM |
6883 | if (total_dev != fs_info->fs_devices->total_devices) { |
6884 | btrfs_err(fs_info, | |
99e3ecfc | 6885 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 6886 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
6887 | total_dev); |
6888 | ret = -EINVAL; | |
6889 | goto error; | |
6890 | } | |
0b246afa JM |
6891 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
6892 | fs_info->fs_devices->total_rw_bytes) { | |
6893 | btrfs_err(fs_info, | |
99e3ecfc | 6894 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
6895 | btrfs_super_total_bytes(fs_info->super_copy), |
6896 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
6897 | ret = -EINVAL; |
6898 | goto error; | |
6899 | } | |
0b86a832 CM |
6900 | ret = 0; |
6901 | error: | |
34441361 | 6902 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
6903 | mutex_unlock(&uuid_mutex); |
6904 | ||
2b82032c | 6905 | btrfs_free_path(path); |
0b86a832 CM |
6906 | return ret; |
6907 | } | |
442a4f63 | 6908 | |
cb517eab MX |
6909 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6910 | { | |
6911 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6912 | struct btrfs_device *device; | |
6913 | ||
29cc83f6 LB |
6914 | while (fs_devices) { |
6915 | mutex_lock(&fs_devices->device_list_mutex); | |
6916 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 6917 | device->fs_info = fs_info; |
29cc83f6 LB |
6918 | mutex_unlock(&fs_devices->device_list_mutex); |
6919 | ||
6920 | fs_devices = fs_devices->seed; | |
6921 | } | |
cb517eab MX |
6922 | } |
6923 | ||
733f4fbb SB |
6924 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6925 | { | |
6926 | int i; | |
6927 | ||
6928 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6929 | btrfs_dev_stat_reset(dev, i); | |
6930 | } | |
6931 | ||
6932 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6933 | { | |
6934 | struct btrfs_key key; | |
6935 | struct btrfs_key found_key; | |
6936 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6937 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6938 | struct extent_buffer *eb; | |
6939 | int slot; | |
6940 | int ret = 0; | |
6941 | struct btrfs_device *device; | |
6942 | struct btrfs_path *path = NULL; | |
6943 | int i; | |
6944 | ||
6945 | path = btrfs_alloc_path(); | |
6946 | if (!path) { | |
6947 | ret = -ENOMEM; | |
6948 | goto out; | |
6949 | } | |
6950 | ||
6951 | mutex_lock(&fs_devices->device_list_mutex); | |
6952 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6953 | int item_size; | |
6954 | struct btrfs_dev_stats_item *ptr; | |
6955 | ||
242e2956 DS |
6956 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6957 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6958 | key.offset = device->devid; |
6959 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6960 | if (ret) { | |
733f4fbb SB |
6961 | __btrfs_reset_dev_stats(device); |
6962 | device->dev_stats_valid = 1; | |
6963 | btrfs_release_path(path); | |
6964 | continue; | |
6965 | } | |
6966 | slot = path->slots[0]; | |
6967 | eb = path->nodes[0]; | |
6968 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6969 | item_size = btrfs_item_size_nr(eb, slot); | |
6970 | ||
6971 | ptr = btrfs_item_ptr(eb, slot, | |
6972 | struct btrfs_dev_stats_item); | |
6973 | ||
6974 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6975 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6976 | btrfs_dev_stat_set(device, i, | |
6977 | btrfs_dev_stats_value(eb, ptr, i)); | |
6978 | else | |
6979 | btrfs_dev_stat_reset(device, i); | |
6980 | } | |
6981 | ||
6982 | device->dev_stats_valid = 1; | |
6983 | btrfs_dev_stat_print_on_load(device); | |
6984 | btrfs_release_path(path); | |
6985 | } | |
6986 | mutex_unlock(&fs_devices->device_list_mutex); | |
6987 | ||
6988 | out: | |
6989 | btrfs_free_path(path); | |
6990 | return ret < 0 ? ret : 0; | |
6991 | } | |
6992 | ||
6993 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6bccf3ab | 6994 | struct btrfs_fs_info *fs_info, |
733f4fbb SB |
6995 | struct btrfs_device *device) |
6996 | { | |
6bccf3ab | 6997 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
6998 | struct btrfs_path *path; |
6999 | struct btrfs_key key; | |
7000 | struct extent_buffer *eb; | |
7001 | struct btrfs_dev_stats_item *ptr; | |
7002 | int ret; | |
7003 | int i; | |
7004 | ||
242e2956 DS |
7005 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7006 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7007 | key.offset = device->devid; |
7008 | ||
7009 | path = btrfs_alloc_path(); | |
fa252992 DS |
7010 | if (!path) |
7011 | return -ENOMEM; | |
733f4fbb SB |
7012 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7013 | if (ret < 0) { | |
0b246afa | 7014 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7015 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7016 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7017 | goto out; |
7018 | } | |
7019 | ||
7020 | if (ret == 0 && | |
7021 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7022 | /* need to delete old one and insert a new one */ | |
7023 | ret = btrfs_del_item(trans, dev_root, path); | |
7024 | if (ret != 0) { | |
0b246afa | 7025 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7026 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7027 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7028 | goto out; |
7029 | } | |
7030 | ret = 1; | |
7031 | } | |
7032 | ||
7033 | if (ret == 1) { | |
7034 | /* need to insert a new item */ | |
7035 | btrfs_release_path(path); | |
7036 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7037 | &key, sizeof(*ptr)); | |
7038 | if (ret < 0) { | |
0b246afa | 7039 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7040 | "insert dev_stats item for device %s failed %d", |
7041 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7042 | goto out; |
7043 | } | |
7044 | } | |
7045 | ||
7046 | eb = path->nodes[0]; | |
7047 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7048 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7049 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7050 | btrfs_dev_stat_read(device, i)); | |
7051 | btrfs_mark_buffer_dirty(eb); | |
7052 | ||
7053 | out: | |
7054 | btrfs_free_path(path); | |
7055 | return ret; | |
7056 | } | |
7057 | ||
7058 | /* | |
7059 | * called from commit_transaction. Writes all changed device stats to disk. | |
7060 | */ | |
7061 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7062 | struct btrfs_fs_info *fs_info) | |
7063 | { | |
733f4fbb SB |
7064 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7065 | struct btrfs_device *device; | |
addc3fa7 | 7066 | int stats_cnt; |
733f4fbb SB |
7067 | int ret = 0; |
7068 | ||
7069 | mutex_lock(&fs_devices->device_list_mutex); | |
7070 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 7071 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
7072 | continue; |
7073 | ||
addc3fa7 | 7074 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
6bccf3ab | 7075 | ret = update_dev_stat_item(trans, fs_info, device); |
733f4fbb | 7076 | if (!ret) |
addc3fa7 | 7077 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7078 | } |
7079 | mutex_unlock(&fs_devices->device_list_mutex); | |
7080 | ||
7081 | return ret; | |
7082 | } | |
7083 | ||
442a4f63 SB |
7084 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7085 | { | |
7086 | btrfs_dev_stat_inc(dev, index); | |
7087 | btrfs_dev_stat_print_on_error(dev); | |
7088 | } | |
7089 | ||
48a3b636 | 7090 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7091 | { |
733f4fbb SB |
7092 | if (!dev->dev_stats_valid) |
7093 | return; | |
fb456252 | 7094 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7095 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7096 | rcu_str_deref(dev->name), |
442a4f63 SB |
7097 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7098 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7099 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7100 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7101 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7102 | } |
c11d2c23 | 7103 | |
733f4fbb SB |
7104 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7105 | { | |
a98cdb85 SB |
7106 | int i; |
7107 | ||
7108 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7109 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7110 | break; | |
7111 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7112 | return; /* all values == 0, suppress message */ | |
7113 | ||
fb456252 | 7114 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7115 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7116 | rcu_str_deref(dev->name), |
733f4fbb SB |
7117 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7118 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7119 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7120 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7121 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7122 | } | |
7123 | ||
2ff7e61e | 7124 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7125 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7126 | { |
7127 | struct btrfs_device *dev; | |
0b246afa | 7128 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7129 | int i; |
7130 | ||
7131 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7132 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7133 | mutex_unlock(&fs_devices->device_list_mutex); |
7134 | ||
7135 | if (!dev) { | |
0b246afa | 7136 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7137 | return -ENODEV; |
733f4fbb | 7138 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7139 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7140 | return -ENODEV; |
b27f7c0c | 7141 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7142 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7143 | if (stats->nr_items > i) | |
7144 | stats->values[i] = | |
7145 | btrfs_dev_stat_read_and_reset(dev, i); | |
7146 | else | |
7147 | btrfs_dev_stat_reset(dev, i); | |
7148 | } | |
7149 | } else { | |
7150 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7151 | if (stats->nr_items > i) | |
7152 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7153 | } | |
7154 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7155 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7156 | return 0; | |
7157 | } | |
a8a6dab7 | 7158 | |
da353f6b | 7159 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7160 | { |
7161 | struct buffer_head *bh; | |
7162 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7163 | int copy_num; |
a8a6dab7 | 7164 | |
12b1c263 AJ |
7165 | if (!bdev) |
7166 | return; | |
a8a6dab7 | 7167 | |
12b1c263 AJ |
7168 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7169 | copy_num++) { | |
a8a6dab7 | 7170 | |
12b1c263 AJ |
7171 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7172 | continue; | |
7173 | ||
7174 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7175 | ||
7176 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7177 | set_buffer_dirty(bh); | |
7178 | sync_dirty_buffer(bh); | |
7179 | brelse(bh); | |
7180 | } | |
7181 | ||
7182 | /* Notify udev that device has changed */ | |
7183 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7184 | ||
7185 | /* Update ctime/mtime for device path for libblkid */ | |
7186 | update_dev_time(device_path); | |
a8a6dab7 | 7187 | } |
935e5cc9 MX |
7188 | |
7189 | /* | |
7190 | * Update the size of all devices, which is used for writing out the | |
7191 | * super blocks. | |
7192 | */ | |
7193 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7194 | { | |
7195 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7196 | struct btrfs_device *curr, *next; | |
7197 | ||
7198 | if (list_empty(&fs_devices->resized_devices)) | |
7199 | return; | |
7200 | ||
7201 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7202 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7203 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7204 | resized_list) { | |
7205 | list_del_init(&curr->resized_list); | |
7206 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7207 | } | |
34441361 | 7208 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7209 | mutex_unlock(&fs_devices->device_list_mutex); |
7210 | } | |
ce7213c7 MX |
7211 | |
7212 | /* Must be invoked during the transaction commit */ | |
2ff7e61e | 7213 | void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info, |
ce7213c7 MX |
7214 | struct btrfs_transaction *transaction) |
7215 | { | |
7216 | struct extent_map *em; | |
7217 | struct map_lookup *map; | |
7218 | struct btrfs_device *dev; | |
7219 | int i; | |
7220 | ||
7221 | if (list_empty(&transaction->pending_chunks)) | |
7222 | return; | |
7223 | ||
7224 | /* In order to kick the device replace finish process */ | |
34441361 | 7225 | mutex_lock(&fs_info->chunk_mutex); |
ce7213c7 | 7226 | list_for_each_entry(em, &transaction->pending_chunks, list) { |
95617d69 | 7227 | map = em->map_lookup; |
ce7213c7 MX |
7228 | |
7229 | for (i = 0; i < map->num_stripes; i++) { | |
7230 | dev = map->stripes[i].dev; | |
7231 | dev->commit_bytes_used = dev->bytes_used; | |
7232 | } | |
7233 | } | |
34441361 | 7234 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7235 | } |
5a13f430 AJ |
7236 | |
7237 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7238 | { | |
7239 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7240 | while (fs_devices) { | |
7241 | fs_devices->fs_info = fs_info; | |
7242 | fs_devices = fs_devices->seed; | |
7243 | } | |
7244 | } | |
7245 | ||
7246 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7247 | { | |
7248 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7249 | while (fs_devices) { | |
7250 | fs_devices->fs_info = NULL; | |
7251 | fs_devices = fs_devices->seed; | |
7252 | } | |
7253 | } |