]>
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 | |
2dfeca9b DS |
155 | /* |
156 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
157 | * @fsid: if not NULL, copy the uuid to fs_devices::fsid | |
158 | * | |
159 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
160 | * The returned struct is not linked onto any lists and can be destroyed with | |
161 | * kfree() right away. | |
162 | */ | |
163 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
2208a378 ID |
164 | { |
165 | struct btrfs_fs_devices *fs_devs; | |
166 | ||
78f2c9e6 | 167 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
168 | if (!fs_devs) |
169 | return ERR_PTR(-ENOMEM); | |
170 | ||
171 | mutex_init(&fs_devs->device_list_mutex); | |
172 | ||
173 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 174 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
175 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
176 | INIT_LIST_HEAD(&fs_devs->list); | |
2208a378 ID |
177 | if (fsid) |
178 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 ID |
179 | |
180 | return fs_devs; | |
181 | } | |
182 | ||
e4404d6e YZ |
183 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
184 | { | |
185 | struct btrfs_device *device; | |
186 | WARN_ON(fs_devices->opened); | |
187 | while (!list_empty(&fs_devices->devices)) { | |
188 | device = list_entry(fs_devices->devices.next, | |
189 | struct btrfs_device, dev_list); | |
190 | list_del(&device->dev_list); | |
606686ee | 191 | rcu_string_free(device->name); |
e4404d6e YZ |
192 | kfree(device); |
193 | } | |
194 | kfree(fs_devices); | |
195 | } | |
196 | ||
b8b8ff59 LC |
197 | static void btrfs_kobject_uevent(struct block_device *bdev, |
198 | enum kobject_action action) | |
199 | { | |
200 | int ret; | |
201 | ||
202 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
203 | if (ret) | |
efe120a0 | 204 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
205 | action, |
206 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
207 | &disk_to_dev(bdev->bd_disk)->kobj); | |
208 | } | |
209 | ||
143bede5 | 210 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
211 | { |
212 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 213 | |
2b82032c YZ |
214 | while (!list_empty(&fs_uuids)) { |
215 | fs_devices = list_entry(fs_uuids.next, | |
216 | struct btrfs_fs_devices, list); | |
217 | list_del(&fs_devices->list); | |
e4404d6e | 218 | free_fs_devices(fs_devices); |
8a4b83cc | 219 | } |
8a4b83cc CM |
220 | } |
221 | ||
12bd2fc0 ID |
222 | static struct btrfs_device *__alloc_device(void) |
223 | { | |
224 | struct btrfs_device *dev; | |
225 | ||
78f2c9e6 | 226 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
227 | if (!dev) |
228 | return ERR_PTR(-ENOMEM); | |
229 | ||
e0ae9994 DS |
230 | /* |
231 | * Preallocate a bio that's always going to be used for flushing device | |
232 | * barriers and matches the device lifespan | |
233 | */ | |
234 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
235 | if (!dev->flush_bio) { | |
236 | kfree(dev); | |
237 | return ERR_PTR(-ENOMEM); | |
238 | } | |
239 | bio_get(dev->flush_bio); | |
240 | ||
12bd2fc0 ID |
241 | INIT_LIST_HEAD(&dev->dev_list); |
242 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 243 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
244 | |
245 | spin_lock_init(&dev->io_lock); | |
246 | ||
247 | spin_lock_init(&dev->reada_lock); | |
248 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 249 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 250 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 251 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 252 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
253 | |
254 | return dev; | |
255 | } | |
256 | ||
35c70103 DS |
257 | /* |
258 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
259 | * return NULL. | |
260 | * | |
261 | * If devid and uuid are both specified, the match must be exact, otherwise | |
262 | * only devid is used. | |
263 | */ | |
264 | static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices, | |
265 | u64 devid, const u8 *uuid) | |
8a4b83cc | 266 | { |
35c70103 | 267 | struct list_head *head = &fs_devices->devices; |
8a4b83cc | 268 | struct btrfs_device *dev; |
8a4b83cc | 269 | |
c6e30871 | 270 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 271 | if (dev->devid == devid && |
8f18cf13 | 272 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 273 | return dev; |
a443755f | 274 | } |
8a4b83cc CM |
275 | } |
276 | return NULL; | |
277 | } | |
278 | ||
a1b32a59 | 279 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 280 | { |
8a4b83cc CM |
281 | struct btrfs_fs_devices *fs_devices; |
282 | ||
c6e30871 | 283 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
284 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
285 | return fs_devices; | |
286 | } | |
287 | return NULL; | |
288 | } | |
289 | ||
beaf8ab3 SB |
290 | static int |
291 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
292 | int flush, struct block_device **bdev, | |
293 | struct buffer_head **bh) | |
294 | { | |
295 | int ret; | |
296 | ||
297 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
298 | ||
299 | if (IS_ERR(*bdev)) { | |
300 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
301 | goto error; |
302 | } | |
303 | ||
304 | if (flush) | |
305 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
306 | ret = set_blocksize(*bdev, 4096); | |
307 | if (ret) { | |
308 | blkdev_put(*bdev, flags); | |
309 | goto error; | |
310 | } | |
311 | invalidate_bdev(*bdev); | |
312 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
313 | if (IS_ERR(*bh)) { |
314 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
315 | blkdev_put(*bdev, flags); |
316 | goto error; | |
317 | } | |
318 | ||
319 | return 0; | |
320 | ||
321 | error: | |
322 | *bdev = NULL; | |
323 | *bh = NULL; | |
324 | return ret; | |
325 | } | |
326 | ||
ffbd517d CM |
327 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
328 | struct bio *head, struct bio *tail) | |
329 | { | |
330 | ||
331 | struct bio *old_head; | |
332 | ||
333 | old_head = pending_bios->head; | |
334 | pending_bios->head = head; | |
335 | if (pending_bios->tail) | |
336 | tail->bi_next = old_head; | |
337 | else | |
338 | pending_bios->tail = tail; | |
339 | } | |
340 | ||
8b712842 CM |
341 | /* |
342 | * we try to collect pending bios for a device so we don't get a large | |
343 | * number of procs sending bios down to the same device. This greatly | |
344 | * improves the schedulers ability to collect and merge the bios. | |
345 | * | |
346 | * But, it also turns into a long list of bios to process and that is sure | |
347 | * to eventually make the worker thread block. The solution here is to | |
348 | * make some progress and then put this work struct back at the end of | |
349 | * the list if the block device is congested. This way, multiple devices | |
350 | * can make progress from a single worker thread. | |
351 | */ | |
143bede5 | 352 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 353 | { |
0b246afa | 354 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
355 | struct bio *pending; |
356 | struct backing_dev_info *bdi; | |
ffbd517d | 357 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
358 | struct bio *tail; |
359 | struct bio *cur; | |
360 | int again = 0; | |
ffbd517d | 361 | unsigned long num_run; |
d644d8a1 | 362 | unsigned long batch_run = 0; |
b64a2851 | 363 | unsigned long limit; |
b765ead5 | 364 | unsigned long last_waited = 0; |
d84275c9 | 365 | int force_reg = 0; |
0e588859 | 366 | int sync_pending = 0; |
211588ad CM |
367 | struct blk_plug plug; |
368 | ||
369 | /* | |
370 | * this function runs all the bios we've collected for | |
371 | * a particular device. We don't want to wander off to | |
372 | * another device without first sending all of these down. | |
373 | * So, setup a plug here and finish it off before we return | |
374 | */ | |
375 | blk_start_plug(&plug); | |
8b712842 | 376 | |
efa7c9f9 | 377 | bdi = device->bdev->bd_bdi; |
b64a2851 CM |
378 | limit = btrfs_async_submit_limit(fs_info); |
379 | limit = limit * 2 / 3; | |
380 | ||
8b712842 CM |
381 | loop: |
382 | spin_lock(&device->io_lock); | |
383 | ||
a6837051 | 384 | loop_lock: |
d84275c9 | 385 | num_run = 0; |
ffbd517d | 386 | |
8b712842 CM |
387 | /* take all the bios off the list at once and process them |
388 | * later on (without the lock held). But, remember the | |
389 | * tail and other pointers so the bios can be properly reinserted | |
390 | * into the list if we hit congestion | |
391 | */ | |
d84275c9 | 392 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 393 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
394 | force_reg = 1; |
395 | } else { | |
ffbd517d | 396 | pending_bios = &device->pending_bios; |
d84275c9 CM |
397 | force_reg = 0; |
398 | } | |
ffbd517d CM |
399 | |
400 | pending = pending_bios->head; | |
401 | tail = pending_bios->tail; | |
8b712842 | 402 | WARN_ON(pending && !tail); |
8b712842 CM |
403 | |
404 | /* | |
405 | * if pending was null this time around, no bios need processing | |
406 | * at all and we can stop. Otherwise it'll loop back up again | |
407 | * and do an additional check so no bios are missed. | |
408 | * | |
409 | * device->running_pending is used to synchronize with the | |
410 | * schedule_bio code. | |
411 | */ | |
ffbd517d CM |
412 | if (device->pending_sync_bios.head == NULL && |
413 | device->pending_bios.head == NULL) { | |
8b712842 CM |
414 | again = 0; |
415 | device->running_pending = 0; | |
ffbd517d CM |
416 | } else { |
417 | again = 1; | |
418 | device->running_pending = 1; | |
8b712842 | 419 | } |
ffbd517d CM |
420 | |
421 | pending_bios->head = NULL; | |
422 | pending_bios->tail = NULL; | |
423 | ||
8b712842 CM |
424 | spin_unlock(&device->io_lock); |
425 | ||
d397712b | 426 | while (pending) { |
ffbd517d CM |
427 | |
428 | rmb(); | |
d84275c9 CM |
429 | /* we want to work on both lists, but do more bios on the |
430 | * sync list than the regular list | |
431 | */ | |
432 | if ((num_run > 32 && | |
433 | pending_bios != &device->pending_sync_bios && | |
434 | device->pending_sync_bios.head) || | |
435 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
436 | device->pending_bios.head)) { | |
ffbd517d CM |
437 | spin_lock(&device->io_lock); |
438 | requeue_list(pending_bios, pending, tail); | |
439 | goto loop_lock; | |
440 | } | |
441 | ||
8b712842 CM |
442 | cur = pending; |
443 | pending = pending->bi_next; | |
444 | cur->bi_next = NULL; | |
b64a2851 | 445 | |
ee863954 DS |
446 | /* |
447 | * atomic_dec_return implies a barrier for waitqueue_active | |
448 | */ | |
66657b31 | 449 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
450 | waitqueue_active(&fs_info->async_submit_wait)) |
451 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 452 | |
dac56212 | 453 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 454 | |
2ab1ba68 CM |
455 | /* |
456 | * if we're doing the sync list, record that our | |
457 | * plug has some sync requests on it | |
458 | * | |
459 | * If we're doing the regular list and there are | |
460 | * sync requests sitting around, unplug before | |
461 | * we add more | |
462 | */ | |
463 | if (pending_bios == &device->pending_sync_bios) { | |
464 | sync_pending = 1; | |
465 | } else if (sync_pending) { | |
466 | blk_finish_plug(&plug); | |
467 | blk_start_plug(&plug); | |
468 | sync_pending = 0; | |
469 | } | |
470 | ||
4e49ea4a | 471 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
472 | num_run++; |
473 | batch_run++; | |
853d8ec4 DS |
474 | |
475 | cond_resched(); | |
8b712842 CM |
476 | |
477 | /* | |
478 | * we made progress, there is more work to do and the bdi | |
479 | * is now congested. Back off and let other work structs | |
480 | * run instead | |
481 | */ | |
57fd5a5f | 482 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 483 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 484 | struct io_context *ioc; |
8b712842 | 485 | |
b765ead5 CM |
486 | ioc = current->io_context; |
487 | ||
488 | /* | |
489 | * the main goal here is that we don't want to | |
490 | * block if we're going to be able to submit | |
491 | * more requests without blocking. | |
492 | * | |
493 | * This code does two great things, it pokes into | |
494 | * the elevator code from a filesystem _and_ | |
495 | * it makes assumptions about how batching works. | |
496 | */ | |
497 | if (ioc && ioc->nr_batch_requests > 0 && | |
498 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
499 | (last_waited == 0 || | |
500 | ioc->last_waited == last_waited)) { | |
501 | /* | |
502 | * we want to go through our batch of | |
503 | * requests and stop. So, we copy out | |
504 | * the ioc->last_waited time and test | |
505 | * against it before looping | |
506 | */ | |
507 | last_waited = ioc->last_waited; | |
853d8ec4 | 508 | cond_resched(); |
b765ead5 CM |
509 | continue; |
510 | } | |
8b712842 | 511 | spin_lock(&device->io_lock); |
ffbd517d | 512 | requeue_list(pending_bios, pending, tail); |
a6837051 | 513 | device->running_pending = 1; |
8b712842 CM |
514 | |
515 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
516 | btrfs_queue_work(fs_info->submit_workers, |
517 | &device->work); | |
8b712842 CM |
518 | goto done; |
519 | } | |
d85c8a6f CM |
520 | /* unplug every 64 requests just for good measure */ |
521 | if (batch_run % 64 == 0) { | |
522 | blk_finish_plug(&plug); | |
523 | blk_start_plug(&plug); | |
524 | sync_pending = 0; | |
525 | } | |
8b712842 | 526 | } |
ffbd517d | 527 | |
51684082 CM |
528 | cond_resched(); |
529 | if (again) | |
530 | goto loop; | |
531 | ||
532 | spin_lock(&device->io_lock); | |
533 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
534 | goto loop_lock; | |
535 | spin_unlock(&device->io_lock); | |
536 | ||
8b712842 | 537 | done: |
211588ad | 538 | blk_finish_plug(&plug); |
8b712842 CM |
539 | } |
540 | ||
b2950863 | 541 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
542 | { |
543 | struct btrfs_device *device; | |
544 | ||
545 | device = container_of(work, struct btrfs_device, work); | |
546 | run_scheduled_bios(device); | |
547 | } | |
548 | ||
4fde46f0 AJ |
549 | |
550 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
551 | { | |
552 | struct btrfs_fs_devices *fs_devs; | |
553 | struct btrfs_device *dev; | |
554 | ||
555 | if (!cur_dev->name) | |
556 | return; | |
557 | ||
558 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
559 | int del = 1; | |
560 | ||
561 | if (fs_devs->opened) | |
562 | continue; | |
563 | if (fs_devs->seeding) | |
564 | continue; | |
565 | ||
566 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
567 | ||
568 | if (dev == cur_dev) | |
569 | continue; | |
570 | if (!dev->name) | |
571 | continue; | |
572 | ||
573 | /* | |
574 | * Todo: This won't be enough. What if the same device | |
575 | * comes back (with new uuid and) with its mapper path? | |
576 | * But for now, this does help as mostly an admin will | |
577 | * either use mapper or non mapper path throughout. | |
578 | */ | |
579 | rcu_read_lock(); | |
580 | del = strcmp(rcu_str_deref(dev->name), | |
581 | rcu_str_deref(cur_dev->name)); | |
582 | rcu_read_unlock(); | |
583 | if (!del) | |
584 | break; | |
585 | } | |
586 | ||
587 | if (!del) { | |
588 | /* delete the stale device */ | |
589 | if (fs_devs->num_devices == 1) { | |
590 | btrfs_sysfs_remove_fsid(fs_devs); | |
591 | list_del(&fs_devs->list); | |
592 | free_fs_devices(fs_devs); | |
593 | } else { | |
594 | fs_devs->num_devices--; | |
595 | list_del(&dev->dev_list); | |
596 | rcu_string_free(dev->name); | |
597 | kfree(dev); | |
598 | } | |
599 | break; | |
600 | } | |
601 | } | |
602 | } | |
603 | ||
60999ca4 DS |
604 | /* |
605 | * Add new device to list of registered devices | |
606 | * | |
607 | * Returns: | |
608 | * 1 - first time device is seen | |
609 | * 0 - device already known | |
610 | * < 0 - error | |
611 | */ | |
a1b32a59 | 612 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
613 | struct btrfs_super_block *disk_super, |
614 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
615 | { | |
616 | struct btrfs_device *device; | |
617 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 618 | struct rcu_string *name; |
60999ca4 | 619 | int ret = 0; |
8a4b83cc CM |
620 | u64 found_transid = btrfs_super_generation(disk_super); |
621 | ||
622 | fs_devices = find_fsid(disk_super->fsid); | |
623 | if (!fs_devices) { | |
2208a378 ID |
624 | fs_devices = alloc_fs_devices(disk_super->fsid); |
625 | if (IS_ERR(fs_devices)) | |
626 | return PTR_ERR(fs_devices); | |
627 | ||
8a4b83cc | 628 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 629 | |
8a4b83cc CM |
630 | device = NULL; |
631 | } else { | |
35c70103 DS |
632 | device = find_device(fs_devices, devid, |
633 | disk_super->dev_item.uuid); | |
8a4b83cc | 634 | } |
443f24fe | 635 | |
8a4b83cc | 636 | if (!device) { |
2b82032c YZ |
637 | if (fs_devices->opened) |
638 | return -EBUSY; | |
639 | ||
12bd2fc0 ID |
640 | device = btrfs_alloc_device(NULL, &devid, |
641 | disk_super->dev_item.uuid); | |
642 | if (IS_ERR(device)) { | |
8a4b83cc | 643 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 644 | return PTR_ERR(device); |
8a4b83cc | 645 | } |
606686ee JB |
646 | |
647 | name = rcu_string_strdup(path, GFP_NOFS); | |
648 | if (!name) { | |
8a4b83cc CM |
649 | kfree(device); |
650 | return -ENOMEM; | |
651 | } | |
606686ee | 652 | rcu_assign_pointer(device->name, name); |
90519d66 | 653 | |
e5e9a520 | 654 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 655 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 656 | fs_devices->num_devices++; |
e5e9a520 CM |
657 | mutex_unlock(&fs_devices->device_list_mutex); |
658 | ||
60999ca4 | 659 | ret = 1; |
2b82032c | 660 | device->fs_devices = fs_devices; |
606686ee | 661 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
662 | /* |
663 | * When FS is already mounted. | |
664 | * 1. If you are here and if the device->name is NULL that | |
665 | * means this device was missing at time of FS mount. | |
666 | * 2. If you are here and if the device->name is different | |
667 | * from 'path' that means either | |
668 | * a. The same device disappeared and reappeared with | |
669 | * different name. or | |
670 | * b. The missing-disk-which-was-replaced, has | |
671 | * reappeared now. | |
672 | * | |
673 | * We must allow 1 and 2a above. But 2b would be a spurious | |
674 | * and unintentional. | |
675 | * | |
676 | * Further in case of 1 and 2a above, the disk at 'path' | |
677 | * would have missed some transaction when it was away and | |
678 | * in case of 2a the stale bdev has to be updated as well. | |
679 | * 2b must not be allowed at all time. | |
680 | */ | |
681 | ||
682 | /* | |
0f23ae74 CM |
683 | * For now, we do allow update to btrfs_fs_device through the |
684 | * btrfs dev scan cli after FS has been mounted. We're still | |
685 | * tracking a problem where systems fail mount by subvolume id | |
686 | * when we reject replacement on a mounted FS. | |
b96de000 | 687 | */ |
0f23ae74 | 688 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
689 | /* |
690 | * That is if the FS is _not_ mounted and if you | |
691 | * are here, that means there is more than one | |
692 | * disk with same uuid and devid.We keep the one | |
693 | * with larger generation number or the last-in if | |
694 | * generation are equal. | |
695 | */ | |
0f23ae74 | 696 | return -EEXIST; |
77bdae4d | 697 | } |
b96de000 | 698 | |
606686ee | 699 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
700 | if (!name) |
701 | return -ENOMEM; | |
606686ee JB |
702 | rcu_string_free(device->name); |
703 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
704 | if (device->missing) { |
705 | fs_devices->missing_devices--; | |
706 | device->missing = 0; | |
707 | } | |
8a4b83cc CM |
708 | } |
709 | ||
77bdae4d AJ |
710 | /* |
711 | * Unmount does not free the btrfs_device struct but would zero | |
712 | * generation along with most of the other members. So just update | |
713 | * it back. We need it to pick the disk with largest generation | |
714 | * (as above). | |
715 | */ | |
716 | if (!fs_devices->opened) | |
717 | device->generation = found_transid; | |
718 | ||
4fde46f0 AJ |
719 | /* |
720 | * if there is new btrfs on an already registered device, | |
721 | * then remove the stale device entry. | |
722 | */ | |
02feae3c AJ |
723 | if (ret > 0) |
724 | btrfs_free_stale_device(device); | |
4fde46f0 | 725 | |
8a4b83cc | 726 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
727 | |
728 | return ret; | |
8a4b83cc CM |
729 | } |
730 | ||
e4404d6e YZ |
731 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
732 | { | |
733 | struct btrfs_fs_devices *fs_devices; | |
734 | struct btrfs_device *device; | |
735 | struct btrfs_device *orig_dev; | |
736 | ||
2208a378 ID |
737 | fs_devices = alloc_fs_devices(orig->fsid); |
738 | if (IS_ERR(fs_devices)) | |
739 | return fs_devices; | |
e4404d6e | 740 | |
adbbb863 | 741 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 742 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 743 | |
46224705 | 744 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 745 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
746 | struct rcu_string *name; |
747 | ||
12bd2fc0 ID |
748 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
749 | orig_dev->uuid); | |
750 | if (IS_ERR(device)) | |
e4404d6e YZ |
751 | goto error; |
752 | ||
606686ee JB |
753 | /* |
754 | * This is ok to do without rcu read locked because we hold the | |
755 | * uuid mutex so nothing we touch in here is going to disappear. | |
756 | */ | |
e755f780 | 757 | if (orig_dev->name) { |
78f2c9e6 DS |
758 | name = rcu_string_strdup(orig_dev->name->str, |
759 | GFP_KERNEL); | |
e755f780 AJ |
760 | if (!name) { |
761 | kfree(device); | |
762 | goto error; | |
763 | } | |
764 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 765 | } |
e4404d6e | 766 | |
e4404d6e YZ |
767 | list_add(&device->dev_list, &fs_devices->devices); |
768 | device->fs_devices = fs_devices; | |
769 | fs_devices->num_devices++; | |
770 | } | |
adbbb863 | 771 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
772 | return fs_devices; |
773 | error: | |
adbbb863 | 774 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
775 | free_fs_devices(fs_devices); |
776 | return ERR_PTR(-ENOMEM); | |
777 | } | |
778 | ||
9eaed21e | 779 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 780 | { |
c6e30871 | 781 | struct btrfs_device *device, *next; |
443f24fe | 782 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 783 | |
dfe25020 CM |
784 | mutex_lock(&uuid_mutex); |
785 | again: | |
46224705 | 786 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 787 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 788 | if (device->in_fs_metadata) { |
63a212ab | 789 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
790 | (!latest_dev || |
791 | device->generation > latest_dev->generation)) { | |
792 | latest_dev = device; | |
a6b0d5c8 | 793 | } |
2b82032c | 794 | continue; |
a6b0d5c8 | 795 | } |
2b82032c | 796 | |
8dabb742 SB |
797 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
798 | /* | |
799 | * In the first step, keep the device which has | |
800 | * the correct fsid and the devid that is used | |
801 | * for the dev_replace procedure. | |
802 | * In the second step, the dev_replace state is | |
803 | * read from the device tree and it is known | |
804 | * whether the procedure is really active or | |
805 | * not, which means whether this device is | |
806 | * used or whether it should be removed. | |
807 | */ | |
808 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
809 | continue; | |
810 | } | |
811 | } | |
2b82032c | 812 | if (device->bdev) { |
d4d77629 | 813 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
814 | device->bdev = NULL; |
815 | fs_devices->open_devices--; | |
816 | } | |
817 | if (device->writeable) { | |
818 | list_del_init(&device->dev_alloc_list); | |
819 | device->writeable = 0; | |
8dabb742 SB |
820 | if (!device->is_tgtdev_for_dev_replace) |
821 | fs_devices->rw_devices--; | |
2b82032c | 822 | } |
e4404d6e YZ |
823 | list_del_init(&device->dev_list); |
824 | fs_devices->num_devices--; | |
606686ee | 825 | rcu_string_free(device->name); |
e4404d6e | 826 | kfree(device); |
dfe25020 | 827 | } |
2b82032c YZ |
828 | |
829 | if (fs_devices->seed) { | |
830 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
831 | goto again; |
832 | } | |
833 | ||
443f24fe | 834 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 835 | |
dfe25020 | 836 | mutex_unlock(&uuid_mutex); |
dfe25020 | 837 | } |
a0af469b | 838 | |
1f78160c XG |
839 | static void __free_device(struct work_struct *work) |
840 | { | |
841 | struct btrfs_device *device; | |
842 | ||
843 | device = container_of(work, struct btrfs_device, rcu_work); | |
606686ee | 844 | rcu_string_free(device->name); |
e0ae9994 | 845 | bio_put(device->flush_bio); |
1f78160c XG |
846 | kfree(device); |
847 | } | |
848 | ||
849 | static void free_device(struct rcu_head *head) | |
850 | { | |
851 | struct btrfs_device *device; | |
852 | ||
853 | device = container_of(head, struct btrfs_device, rcu); | |
854 | ||
855 | INIT_WORK(&device->rcu_work, __free_device); | |
856 | schedule_work(&device->rcu_work); | |
857 | } | |
858 | ||
14238819 AJ |
859 | static void btrfs_close_bdev(struct btrfs_device *device) |
860 | { | |
861 | if (device->bdev && device->writeable) { | |
862 | sync_blockdev(device->bdev); | |
863 | invalidate_bdev(device->bdev); | |
864 | } | |
865 | ||
866 | if (device->bdev) | |
867 | blkdev_put(device->bdev, device->mode); | |
868 | } | |
869 | ||
0ccd0528 | 870 | static void btrfs_prepare_close_one_device(struct btrfs_device *device) |
f448341a AJ |
871 | { |
872 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
873 | struct btrfs_device *new_device; | |
874 | struct rcu_string *name; | |
875 | ||
876 | if (device->bdev) | |
877 | fs_devices->open_devices--; | |
878 | ||
879 | if (device->writeable && | |
880 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
881 | list_del_init(&device->dev_alloc_list); | |
882 | fs_devices->rw_devices--; | |
883 | } | |
884 | ||
885 | if (device->missing) | |
886 | fs_devices->missing_devices--; | |
887 | ||
888 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
889 | device->uuid); | |
890 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
891 | ||
892 | /* Safe because we are under uuid_mutex */ | |
893 | if (device->name) { | |
894 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
895 | BUG_ON(!name); /* -ENOMEM */ | |
896 | rcu_assign_pointer(new_device->name, name); | |
897 | } | |
898 | ||
899 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
900 | new_device->fs_devices = device->fs_devices; | |
f448341a AJ |
901 | } |
902 | ||
2b82032c | 903 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 904 | { |
2037a093 | 905 | struct btrfs_device *device, *tmp; |
0ccd0528 AJ |
906 | struct list_head pending_put; |
907 | ||
908 | INIT_LIST_HEAD(&pending_put); | |
e4404d6e | 909 | |
2b82032c YZ |
910 | if (--fs_devices->opened > 0) |
911 | return 0; | |
8a4b83cc | 912 | |
c9513edb | 913 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 914 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
0ccd0528 AJ |
915 | btrfs_prepare_close_one_device(device); |
916 | list_add(&device->dev_list, &pending_put); | |
8a4b83cc | 917 | } |
c9513edb XG |
918 | mutex_unlock(&fs_devices->device_list_mutex); |
919 | ||
0ccd0528 AJ |
920 | /* |
921 | * btrfs_show_devname() is using the device_list_mutex, | |
922 | * sometimes call to blkdev_put() leads vfs calling | |
923 | * into this func. So do put outside of device_list_mutex, | |
924 | * as of now. | |
925 | */ | |
926 | while (!list_empty(&pending_put)) { | |
927 | device = list_first_entry(&pending_put, | |
928 | struct btrfs_device, dev_list); | |
929 | list_del(&device->dev_list); | |
930 | btrfs_close_bdev(device); | |
931 | call_rcu(&device->rcu, free_device); | |
932 | } | |
933 | ||
e4404d6e YZ |
934 | WARN_ON(fs_devices->open_devices); |
935 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
936 | fs_devices->opened = 0; |
937 | fs_devices->seeding = 0; | |
2b82032c | 938 | |
8a4b83cc CM |
939 | return 0; |
940 | } | |
941 | ||
2b82032c YZ |
942 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
943 | { | |
e4404d6e | 944 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
945 | int ret; |
946 | ||
947 | mutex_lock(&uuid_mutex); | |
948 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
949 | if (!fs_devices->opened) { |
950 | seed_devices = fs_devices->seed; | |
951 | fs_devices->seed = NULL; | |
952 | } | |
2b82032c | 953 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
954 | |
955 | while (seed_devices) { | |
956 | fs_devices = seed_devices; | |
957 | seed_devices = fs_devices->seed; | |
958 | __btrfs_close_devices(fs_devices); | |
959 | free_fs_devices(fs_devices); | |
960 | } | |
bc178622 ES |
961 | /* |
962 | * Wait for rcu kworkers under __btrfs_close_devices | |
963 | * to finish all blkdev_puts so device is really | |
964 | * free when umount is done. | |
965 | */ | |
966 | rcu_barrier(); | |
2b82032c YZ |
967 | return ret; |
968 | } | |
969 | ||
e4404d6e YZ |
970 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
971 | fmode_t flags, void *holder) | |
8a4b83cc | 972 | { |
d5e2003c | 973 | struct request_queue *q; |
8a4b83cc CM |
974 | struct block_device *bdev; |
975 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 976 | struct btrfs_device *device; |
443f24fe | 977 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
978 | struct buffer_head *bh; |
979 | struct btrfs_super_block *disk_super; | |
a0af469b | 980 | u64 devid; |
2b82032c | 981 | int seeding = 1; |
a0af469b | 982 | int ret = 0; |
8a4b83cc | 983 | |
d4d77629 TH |
984 | flags |= FMODE_EXCL; |
985 | ||
c6e30871 | 986 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
987 | if (device->bdev) |
988 | continue; | |
dfe25020 CM |
989 | if (!device->name) |
990 | continue; | |
991 | ||
f63e0cca ES |
992 | /* Just open everything we can; ignore failures here */ |
993 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
994 | &bdev, &bh)) | |
beaf8ab3 | 995 | continue; |
a0af469b CM |
996 | |
997 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 998 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
999 | if (devid != device->devid) |
1000 | goto error_brelse; | |
1001 | ||
2b82032c YZ |
1002 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
1003 | BTRFS_UUID_SIZE)) | |
1004 | goto error_brelse; | |
1005 | ||
1006 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
1007 | if (!latest_dev || |
1008 | device->generation > latest_dev->generation) | |
1009 | latest_dev = device; | |
a0af469b | 1010 | |
2b82032c YZ |
1011 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
1012 | device->writeable = 0; | |
1013 | } else { | |
1014 | device->writeable = !bdev_read_only(bdev); | |
1015 | seeding = 0; | |
1016 | } | |
1017 | ||
d5e2003c | 1018 | q = bdev_get_queue(bdev); |
90180da4 | 1019 | if (blk_queue_discard(q)) |
d5e2003c | 1020 | device->can_discard = 1; |
e884f4f0 AJ |
1021 | if (!blk_queue_nonrot(q)) |
1022 | fs_devices->rotating = 1; | |
d5e2003c | 1023 | |
8a4b83cc | 1024 | device->bdev = bdev; |
dfe25020 | 1025 | device->in_fs_metadata = 0; |
15916de8 CM |
1026 | device->mode = flags; |
1027 | ||
a0af469b | 1028 | fs_devices->open_devices++; |
55e50e45 ID |
1029 | if (device->writeable && |
1030 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
1031 | fs_devices->rw_devices++; |
1032 | list_add(&device->dev_alloc_list, | |
1033 | &fs_devices->alloc_list); | |
1034 | } | |
4f6c9328 | 1035 | brelse(bh); |
a0af469b | 1036 | continue; |
a061fc8d | 1037 | |
a0af469b CM |
1038 | error_brelse: |
1039 | brelse(bh); | |
d4d77629 | 1040 | blkdev_put(bdev, flags); |
a0af469b | 1041 | continue; |
8a4b83cc | 1042 | } |
a0af469b | 1043 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1044 | ret = -EINVAL; |
a0af469b CM |
1045 | goto out; |
1046 | } | |
2b82032c YZ |
1047 | fs_devices->seeding = seeding; |
1048 | fs_devices->opened = 1; | |
443f24fe | 1049 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1050 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1051 | out: |
2b82032c YZ |
1052 | return ret; |
1053 | } | |
1054 | ||
1055 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 1056 | fmode_t flags, void *holder) |
2b82032c YZ |
1057 | { |
1058 | int ret; | |
1059 | ||
1060 | mutex_lock(&uuid_mutex); | |
1061 | if (fs_devices->opened) { | |
e4404d6e YZ |
1062 | fs_devices->opened++; |
1063 | ret = 0; | |
2b82032c | 1064 | } else { |
15916de8 | 1065 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1066 | } |
8a4b83cc | 1067 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1068 | return ret; |
1069 | } | |
1070 | ||
6cf86a00 AJ |
1071 | void btrfs_release_disk_super(struct page *page) |
1072 | { | |
1073 | kunmap(page); | |
1074 | put_page(page); | |
1075 | } | |
1076 | ||
1077 | int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, | |
1078 | struct page **page, struct btrfs_super_block **disk_super) | |
1079 | { | |
1080 | void *p; | |
1081 | pgoff_t index; | |
1082 | ||
1083 | /* make sure our super fits in the device */ | |
1084 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1085 | return 1; | |
1086 | ||
1087 | /* make sure our super fits in the page */ | |
1088 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1089 | return 1; | |
1090 | ||
1091 | /* make sure our super doesn't straddle pages on disk */ | |
1092 | index = bytenr >> PAGE_SHIFT; | |
1093 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1094 | return 1; | |
1095 | ||
1096 | /* pull in the page with our super */ | |
1097 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1098 | index, GFP_KERNEL); | |
1099 | ||
1100 | if (IS_ERR_OR_NULL(*page)) | |
1101 | return 1; | |
1102 | ||
1103 | p = kmap(*page); | |
1104 | ||
1105 | /* align our pointer to the offset of the super block */ | |
1106 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1107 | ||
1108 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1109 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1110 | btrfs_release_disk_super(*page); | |
1111 | return 1; | |
1112 | } | |
1113 | ||
1114 | if ((*disk_super)->label[0] && | |
1115 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1116 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1117 | ||
1118 | return 0; | |
1119 | } | |
1120 | ||
6f60cbd3 DS |
1121 | /* |
1122 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1123 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1124 | * is read via pagecache | |
1125 | */ | |
97288f2c | 1126 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1127 | struct btrfs_fs_devices **fs_devices_ret) |
1128 | { | |
1129 | struct btrfs_super_block *disk_super; | |
1130 | struct block_device *bdev; | |
6f60cbd3 | 1131 | struct page *page; |
6f60cbd3 | 1132 | int ret = -EINVAL; |
8a4b83cc | 1133 | u64 devid; |
f2984462 | 1134 | u64 transid; |
02db0844 | 1135 | u64 total_devices; |
6f60cbd3 | 1136 | u64 bytenr; |
8a4b83cc | 1137 | |
6f60cbd3 DS |
1138 | /* |
1139 | * we would like to check all the supers, but that would make | |
1140 | * a btrfs mount succeed after a mkfs from a different FS. | |
1141 | * So, we need to add a special mount option to scan for | |
1142 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1143 | */ | |
1144 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1145 | flags |= FMODE_EXCL; |
10f6327b | 1146 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1147 | |
1148 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1149 | if (IS_ERR(bdev)) { |
1150 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1151 | goto error; |
6f60cbd3 DS |
1152 | } |
1153 | ||
6cf86a00 | 1154 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) |
6f60cbd3 DS |
1155 | goto error_bdev_put; |
1156 | ||
a343832f | 1157 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1158 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1159 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1160 | |
8a4b83cc | 1161 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1162 | if (ret > 0) { |
1163 | if (disk_super->label[0]) { | |
62e85577 | 1164 | pr_info("BTRFS: device label %s ", disk_super->label); |
60999ca4 | 1165 | } else { |
62e85577 | 1166 | pr_info("BTRFS: device fsid %pU ", disk_super->fsid); |
60999ca4 DS |
1167 | } |
1168 | ||
62e85577 | 1169 | pr_cont("devid %llu transid %llu %s\n", devid, transid, path); |
60999ca4 DS |
1170 | ret = 0; |
1171 | } | |
02db0844 JB |
1172 | if (!ret && fs_devices_ret) |
1173 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 | 1174 | |
6cf86a00 | 1175 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1176 | |
1177 | error_bdev_put: | |
d4d77629 | 1178 | blkdev_put(bdev, flags); |
8a4b83cc | 1179 | error: |
beaf8ab3 | 1180 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1181 | return ret; |
1182 | } | |
0b86a832 | 1183 | |
6d07bcec MX |
1184 | /* helper to account the used device space in the range */ |
1185 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1186 | u64 end, u64 *length) | |
1187 | { | |
1188 | struct btrfs_key key; | |
fb456252 | 1189 | struct btrfs_root *root = device->fs_info->dev_root; |
6d07bcec MX |
1190 | struct btrfs_dev_extent *dev_extent; |
1191 | struct btrfs_path *path; | |
1192 | u64 extent_end; | |
1193 | int ret; | |
1194 | int slot; | |
1195 | struct extent_buffer *l; | |
1196 | ||
1197 | *length = 0; | |
1198 | ||
63a212ab | 1199 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1200 | return 0; |
1201 | ||
1202 | path = btrfs_alloc_path(); | |
1203 | if (!path) | |
1204 | return -ENOMEM; | |
e4058b54 | 1205 | path->reada = READA_FORWARD; |
6d07bcec MX |
1206 | |
1207 | key.objectid = device->devid; | |
1208 | key.offset = start; | |
1209 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1210 | ||
1211 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1212 | if (ret < 0) | |
1213 | goto out; | |
1214 | if (ret > 0) { | |
1215 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1216 | if (ret < 0) | |
1217 | goto out; | |
1218 | } | |
1219 | ||
1220 | while (1) { | |
1221 | l = path->nodes[0]; | |
1222 | slot = path->slots[0]; | |
1223 | if (slot >= btrfs_header_nritems(l)) { | |
1224 | ret = btrfs_next_leaf(root, path); | |
1225 | if (ret == 0) | |
1226 | continue; | |
1227 | if (ret < 0) | |
1228 | goto out; | |
1229 | ||
1230 | break; | |
1231 | } | |
1232 | btrfs_item_key_to_cpu(l, &key, slot); | |
1233 | ||
1234 | if (key.objectid < device->devid) | |
1235 | goto next; | |
1236 | ||
1237 | if (key.objectid > device->devid) | |
1238 | break; | |
1239 | ||
962a298f | 1240 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1241 | goto next; |
1242 | ||
1243 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1244 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1245 | dev_extent); | |
1246 | if (key.offset <= start && extent_end > end) { | |
1247 | *length = end - start + 1; | |
1248 | break; | |
1249 | } else if (key.offset <= start && extent_end > start) | |
1250 | *length += extent_end - start; | |
1251 | else if (key.offset > start && extent_end <= end) | |
1252 | *length += extent_end - key.offset; | |
1253 | else if (key.offset > start && key.offset <= end) { | |
1254 | *length += end - key.offset + 1; | |
1255 | break; | |
1256 | } else if (key.offset > end) | |
1257 | break; | |
1258 | ||
1259 | next: | |
1260 | path->slots[0]++; | |
1261 | } | |
1262 | ret = 0; | |
1263 | out: | |
1264 | btrfs_free_path(path); | |
1265 | return ret; | |
1266 | } | |
1267 | ||
499f377f | 1268 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1269 | struct btrfs_device *device, |
1270 | u64 *start, u64 len) | |
1271 | { | |
fb456252 | 1272 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1273 | struct extent_map *em; |
499f377f | 1274 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1275 | int ret = 0; |
1b984508 | 1276 | u64 physical_start = *start; |
6df9a95e | 1277 | |
499f377f JM |
1278 | if (transaction) |
1279 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1280 | again: |
1281 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1282 | struct map_lookup *map; |
1283 | int i; | |
1284 | ||
95617d69 | 1285 | map = em->map_lookup; |
6df9a95e | 1286 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1287 | u64 end; |
1288 | ||
6df9a95e JB |
1289 | if (map->stripes[i].dev != device) |
1290 | continue; | |
1b984508 | 1291 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1292 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1293 | physical_start) |
6df9a95e | 1294 | continue; |
c152b63e FM |
1295 | /* |
1296 | * Make sure that while processing the pinned list we do | |
1297 | * not override our *start with a lower value, because | |
1298 | * we can have pinned chunks that fall within this | |
1299 | * device hole and that have lower physical addresses | |
1300 | * than the pending chunks we processed before. If we | |
1301 | * do not take this special care we can end up getting | |
1302 | * 2 pending chunks that start at the same physical | |
1303 | * device offsets because the end offset of a pinned | |
1304 | * chunk can be equal to the start offset of some | |
1305 | * pending chunk. | |
1306 | */ | |
1307 | end = map->stripes[i].physical + em->orig_block_len; | |
1308 | if (end > *start) { | |
1309 | *start = end; | |
1310 | ret = 1; | |
1311 | } | |
6df9a95e JB |
1312 | } |
1313 | } | |
499f377f JM |
1314 | if (search_list != &fs_info->pinned_chunks) { |
1315 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1316 | goto again; |
1317 | } | |
6df9a95e JB |
1318 | |
1319 | return ret; | |
1320 | } | |
1321 | ||
1322 | ||
0b86a832 | 1323 | /* |
499f377f JM |
1324 | * find_free_dev_extent_start - find free space in the specified device |
1325 | * @device: the device which we search the free space in | |
1326 | * @num_bytes: the size of the free space that we need | |
1327 | * @search_start: the position from which to begin the search | |
1328 | * @start: store the start of the free space. | |
1329 | * @len: the size of the free space. that we find, or the size | |
1330 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1331 | * |
0b86a832 CM |
1332 | * this uses a pretty simple search, the expectation is that it is |
1333 | * called very infrequently and that a given device has a small number | |
1334 | * of extents | |
7bfc837d MX |
1335 | * |
1336 | * @start is used to store the start of the free space if we find. But if we | |
1337 | * don't find suitable free space, it will be used to store the start position | |
1338 | * of the max free space. | |
1339 | * | |
1340 | * @len is used to store the size of the free space that we find. | |
1341 | * But if we don't find suitable free space, it is used to store the size of | |
1342 | * the max free space. | |
0b86a832 | 1343 | */ |
499f377f JM |
1344 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1345 | struct btrfs_device *device, u64 num_bytes, | |
1346 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1347 | { |
0b246afa JM |
1348 | struct btrfs_fs_info *fs_info = device->fs_info; |
1349 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1350 | struct btrfs_key key; |
7bfc837d | 1351 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1352 | struct btrfs_path *path; |
7bfc837d MX |
1353 | u64 hole_size; |
1354 | u64 max_hole_start; | |
1355 | u64 max_hole_size; | |
1356 | u64 extent_end; | |
0b86a832 CM |
1357 | u64 search_end = device->total_bytes; |
1358 | int ret; | |
7bfc837d | 1359 | int slot; |
0b86a832 | 1360 | struct extent_buffer *l; |
8cdc7c5b FM |
1361 | |
1362 | /* | |
1363 | * We don't want to overwrite the superblock on the drive nor any area | |
1364 | * used by the boot loader (grub for example), so we make sure to start | |
1365 | * at an offset of at least 1MB. | |
1366 | */ | |
0d0c71b3 | 1367 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1368 | |
6df9a95e JB |
1369 | path = btrfs_alloc_path(); |
1370 | if (!path) | |
1371 | return -ENOMEM; | |
f2ab7618 | 1372 | |
7bfc837d MX |
1373 | max_hole_start = search_start; |
1374 | max_hole_size = 0; | |
1375 | ||
f2ab7618 | 1376 | again: |
63a212ab | 1377 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1378 | ret = -ENOSPC; |
6df9a95e | 1379 | goto out; |
7bfc837d MX |
1380 | } |
1381 | ||
e4058b54 | 1382 | path->reada = READA_FORWARD; |
6df9a95e JB |
1383 | path->search_commit_root = 1; |
1384 | path->skip_locking = 1; | |
7bfc837d | 1385 | |
0b86a832 CM |
1386 | key.objectid = device->devid; |
1387 | key.offset = search_start; | |
1388 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1389 | |
125ccb0a | 1390 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1391 | if (ret < 0) |
7bfc837d | 1392 | goto out; |
1fcbac58 YZ |
1393 | if (ret > 0) { |
1394 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1395 | if (ret < 0) | |
7bfc837d | 1396 | goto out; |
1fcbac58 | 1397 | } |
7bfc837d | 1398 | |
0b86a832 CM |
1399 | while (1) { |
1400 | l = path->nodes[0]; | |
1401 | slot = path->slots[0]; | |
1402 | if (slot >= btrfs_header_nritems(l)) { | |
1403 | ret = btrfs_next_leaf(root, path); | |
1404 | if (ret == 0) | |
1405 | continue; | |
1406 | if (ret < 0) | |
7bfc837d MX |
1407 | goto out; |
1408 | ||
1409 | break; | |
0b86a832 CM |
1410 | } |
1411 | btrfs_item_key_to_cpu(l, &key, slot); | |
1412 | ||
1413 | if (key.objectid < device->devid) | |
1414 | goto next; | |
1415 | ||
1416 | if (key.objectid > device->devid) | |
7bfc837d | 1417 | break; |
0b86a832 | 1418 | |
962a298f | 1419 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1420 | goto next; |
9779b72f | 1421 | |
7bfc837d MX |
1422 | if (key.offset > search_start) { |
1423 | hole_size = key.offset - search_start; | |
9779b72f | 1424 | |
6df9a95e JB |
1425 | /* |
1426 | * Have to check before we set max_hole_start, otherwise | |
1427 | * we could end up sending back this offset anyway. | |
1428 | */ | |
499f377f | 1429 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1430 | &search_start, |
1b984508 FL |
1431 | hole_size)) { |
1432 | if (key.offset >= search_start) { | |
1433 | hole_size = key.offset - search_start; | |
1434 | } else { | |
1435 | WARN_ON_ONCE(1); | |
1436 | hole_size = 0; | |
1437 | } | |
1438 | } | |
6df9a95e | 1439 | |
7bfc837d MX |
1440 | if (hole_size > max_hole_size) { |
1441 | max_hole_start = search_start; | |
1442 | max_hole_size = hole_size; | |
1443 | } | |
9779b72f | 1444 | |
7bfc837d MX |
1445 | /* |
1446 | * If this free space is greater than which we need, | |
1447 | * it must be the max free space that we have found | |
1448 | * until now, so max_hole_start must point to the start | |
1449 | * of this free space and the length of this free space | |
1450 | * is stored in max_hole_size. Thus, we return | |
1451 | * max_hole_start and max_hole_size and go back to the | |
1452 | * caller. | |
1453 | */ | |
1454 | if (hole_size >= num_bytes) { | |
1455 | ret = 0; | |
1456 | goto out; | |
0b86a832 CM |
1457 | } |
1458 | } | |
0b86a832 | 1459 | |
0b86a832 | 1460 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1461 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1462 | dev_extent); | |
1463 | if (extent_end > search_start) | |
1464 | search_start = extent_end; | |
0b86a832 CM |
1465 | next: |
1466 | path->slots[0]++; | |
1467 | cond_resched(); | |
1468 | } | |
0b86a832 | 1469 | |
38c01b96 | 1470 | /* |
1471 | * At this point, search_start should be the end of | |
1472 | * allocated dev extents, and when shrinking the device, | |
1473 | * search_end may be smaller than search_start. | |
1474 | */ | |
f2ab7618 | 1475 | if (search_end > search_start) { |
38c01b96 | 1476 | hole_size = search_end - search_start; |
1477 | ||
499f377f | 1478 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1479 | hole_size)) { |
1480 | btrfs_release_path(path); | |
1481 | goto again; | |
1482 | } | |
0b86a832 | 1483 | |
f2ab7618 ZL |
1484 | if (hole_size > max_hole_size) { |
1485 | max_hole_start = search_start; | |
1486 | max_hole_size = hole_size; | |
1487 | } | |
6df9a95e JB |
1488 | } |
1489 | ||
7bfc837d | 1490 | /* See above. */ |
f2ab7618 | 1491 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1492 | ret = -ENOSPC; |
1493 | else | |
1494 | ret = 0; | |
1495 | ||
1496 | out: | |
2b82032c | 1497 | btrfs_free_path(path); |
7bfc837d | 1498 | *start = max_hole_start; |
b2117a39 | 1499 | if (len) |
7bfc837d | 1500 | *len = max_hole_size; |
0b86a832 CM |
1501 | return ret; |
1502 | } | |
1503 | ||
499f377f JM |
1504 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1505 | struct btrfs_device *device, u64 num_bytes, | |
1506 | u64 *start, u64 *len) | |
1507 | { | |
499f377f | 1508 | /* FIXME use last free of some kind */ |
499f377f | 1509 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1510 | num_bytes, 0, start, len); |
499f377f JM |
1511 | } |
1512 | ||
b2950863 | 1513 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1514 | struct btrfs_device *device, |
2196d6e8 | 1515 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1516 | { |
0b246afa JM |
1517 | struct btrfs_fs_info *fs_info = device->fs_info; |
1518 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1519 | int ret; |
1520 | struct btrfs_path *path; | |
8f18cf13 | 1521 | struct btrfs_key key; |
a061fc8d CM |
1522 | struct btrfs_key found_key; |
1523 | struct extent_buffer *leaf = NULL; | |
1524 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1525 | |
1526 | path = btrfs_alloc_path(); | |
1527 | if (!path) | |
1528 | return -ENOMEM; | |
1529 | ||
1530 | key.objectid = device->devid; | |
1531 | key.offset = start; | |
1532 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1533 | again: |
8f18cf13 | 1534 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1535 | if (ret > 0) { |
1536 | ret = btrfs_previous_item(root, path, key.objectid, | |
1537 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1538 | if (ret) |
1539 | goto out; | |
a061fc8d CM |
1540 | leaf = path->nodes[0]; |
1541 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1542 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1543 | struct btrfs_dev_extent); | |
1544 | BUG_ON(found_key.offset > start || found_key.offset + | |
1545 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1546 | key = found_key; |
1547 | btrfs_release_path(path); | |
1548 | goto again; | |
a061fc8d CM |
1549 | } else if (ret == 0) { |
1550 | leaf = path->nodes[0]; | |
1551 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1552 | struct btrfs_dev_extent); | |
79787eaa | 1553 | } else { |
0b246afa | 1554 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1555 | goto out; |
a061fc8d | 1556 | } |
8f18cf13 | 1557 | |
2196d6e8 MX |
1558 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1559 | ||
8f18cf13 | 1560 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1561 | if (ret) { |
0b246afa JM |
1562 | btrfs_handle_fs_error(fs_info, ret, |
1563 | "Failed to remove dev extent item"); | |
13212b54 | 1564 | } else { |
3204d33c | 1565 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1566 | } |
b0b802d7 | 1567 | out: |
8f18cf13 CM |
1568 | btrfs_free_path(path); |
1569 | return ret; | |
1570 | } | |
1571 | ||
48a3b636 ES |
1572 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1573 | struct btrfs_device *device, | |
1574 | u64 chunk_tree, u64 chunk_objectid, | |
1575 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1576 | { |
1577 | int ret; | |
1578 | struct btrfs_path *path; | |
0b246afa JM |
1579 | struct btrfs_fs_info *fs_info = device->fs_info; |
1580 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1581 | struct btrfs_dev_extent *extent; |
1582 | struct extent_buffer *leaf; | |
1583 | struct btrfs_key key; | |
1584 | ||
dfe25020 | 1585 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1586 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1587 | path = btrfs_alloc_path(); |
1588 | if (!path) | |
1589 | return -ENOMEM; | |
1590 | ||
0b86a832 | 1591 | key.objectid = device->devid; |
2b82032c | 1592 | key.offset = start; |
0b86a832 CM |
1593 | key.type = BTRFS_DEV_EXTENT_KEY; |
1594 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1595 | sizeof(*extent)); | |
2cdcecbc MF |
1596 | if (ret) |
1597 | goto out; | |
0b86a832 CM |
1598 | |
1599 | leaf = path->nodes[0]; | |
1600 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1601 | struct btrfs_dev_extent); | |
e17cade2 CM |
1602 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1603 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1604 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1605 | ||
0b246afa | 1606 | write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid); |
e17cade2 | 1607 | |
0b86a832 CM |
1608 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1609 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1610 | out: |
0b86a832 CM |
1611 | btrfs_free_path(path); |
1612 | return ret; | |
1613 | } | |
1614 | ||
6df9a95e | 1615 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1616 | { |
6df9a95e JB |
1617 | struct extent_map_tree *em_tree; |
1618 | struct extent_map *em; | |
1619 | struct rb_node *n; | |
1620 | u64 ret = 0; | |
0b86a832 | 1621 | |
6df9a95e JB |
1622 | em_tree = &fs_info->mapping_tree.map_tree; |
1623 | read_lock(&em_tree->lock); | |
1624 | n = rb_last(&em_tree->map); | |
1625 | if (n) { | |
1626 | em = rb_entry(n, struct extent_map, rb_node); | |
1627 | ret = em->start + em->len; | |
0b86a832 | 1628 | } |
6df9a95e JB |
1629 | read_unlock(&em_tree->lock); |
1630 | ||
0b86a832 CM |
1631 | return ret; |
1632 | } | |
1633 | ||
53f10659 ID |
1634 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1635 | u64 *devid_ret) | |
0b86a832 CM |
1636 | { |
1637 | int ret; | |
1638 | struct btrfs_key key; | |
1639 | struct btrfs_key found_key; | |
2b82032c YZ |
1640 | struct btrfs_path *path; |
1641 | ||
2b82032c YZ |
1642 | path = btrfs_alloc_path(); |
1643 | if (!path) | |
1644 | return -ENOMEM; | |
0b86a832 CM |
1645 | |
1646 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1647 | key.type = BTRFS_DEV_ITEM_KEY; | |
1648 | key.offset = (u64)-1; | |
1649 | ||
53f10659 | 1650 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1651 | if (ret < 0) |
1652 | goto error; | |
1653 | ||
79787eaa | 1654 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1655 | |
53f10659 ID |
1656 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1657 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1658 | BTRFS_DEV_ITEM_KEY); |
1659 | if (ret) { | |
53f10659 | 1660 | *devid_ret = 1; |
0b86a832 CM |
1661 | } else { |
1662 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1663 | path->slots[0]); | |
53f10659 | 1664 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1665 | } |
1666 | ret = 0; | |
1667 | error: | |
2b82032c | 1668 | btrfs_free_path(path); |
0b86a832 CM |
1669 | return ret; |
1670 | } | |
1671 | ||
1672 | /* | |
1673 | * the device information is stored in the chunk root | |
1674 | * the btrfs_device struct should be fully filled in | |
1675 | */ | |
48a3b636 | 1676 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
5b4aacef | 1677 | struct btrfs_fs_info *fs_info, |
48a3b636 | 1678 | struct btrfs_device *device) |
0b86a832 | 1679 | { |
5b4aacef | 1680 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
1681 | int ret; |
1682 | struct btrfs_path *path; | |
1683 | struct btrfs_dev_item *dev_item; | |
1684 | struct extent_buffer *leaf; | |
1685 | struct btrfs_key key; | |
1686 | unsigned long ptr; | |
0b86a832 | 1687 | |
0b86a832 CM |
1688 | path = btrfs_alloc_path(); |
1689 | if (!path) | |
1690 | return -ENOMEM; | |
1691 | ||
0b86a832 CM |
1692 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1693 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1694 | key.offset = device->devid; |
0b86a832 CM |
1695 | |
1696 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1697 | sizeof(*dev_item)); |
0b86a832 CM |
1698 | if (ret) |
1699 | goto out; | |
1700 | ||
1701 | leaf = path->nodes[0]; | |
1702 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1703 | ||
1704 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1705 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1706 | btrfs_set_device_type(leaf, dev_item, device->type); |
1707 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1708 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1709 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1710 | btrfs_set_device_total_bytes(leaf, dev_item, |
1711 | btrfs_device_get_disk_total_bytes(device)); | |
1712 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1713 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1714 | btrfs_set_device_group(leaf, dev_item, 0); |
1715 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1716 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1717 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1718 | |
410ba3a2 | 1719 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1720 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1721 | ptr = btrfs_device_fsid(dev_item); |
0b246afa | 1722 | write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1723 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1724 | |
2b82032c | 1725 | ret = 0; |
0b86a832 CM |
1726 | out: |
1727 | btrfs_free_path(path); | |
1728 | return ret; | |
1729 | } | |
8f18cf13 | 1730 | |
5a1972bd QW |
1731 | /* |
1732 | * Function to update ctime/mtime for a given device path. | |
1733 | * Mainly used for ctime/mtime based probe like libblkid. | |
1734 | */ | |
da353f6b | 1735 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1736 | { |
1737 | struct file *filp; | |
1738 | ||
1739 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1740 | if (IS_ERR(filp)) |
5a1972bd QW |
1741 | return; |
1742 | file_update_time(filp); | |
1743 | filp_close(filp, NULL); | |
5a1972bd QW |
1744 | } |
1745 | ||
5b4aacef | 1746 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1747 | struct btrfs_device *device) |
1748 | { | |
5b4aacef | 1749 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1750 | int ret; |
1751 | struct btrfs_path *path; | |
a061fc8d | 1752 | struct btrfs_key key; |
a061fc8d CM |
1753 | struct btrfs_trans_handle *trans; |
1754 | ||
a061fc8d CM |
1755 | path = btrfs_alloc_path(); |
1756 | if (!path) | |
1757 | return -ENOMEM; | |
1758 | ||
a22285a6 | 1759 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1760 | if (IS_ERR(trans)) { |
1761 | btrfs_free_path(path); | |
1762 | return PTR_ERR(trans); | |
1763 | } | |
a061fc8d CM |
1764 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1765 | key.type = BTRFS_DEV_ITEM_KEY; | |
1766 | key.offset = device->devid; | |
1767 | ||
1768 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1769 | if (ret < 0) | |
1770 | goto out; | |
1771 | ||
1772 | if (ret > 0) { | |
1773 | ret = -ENOENT; | |
1774 | goto out; | |
1775 | } | |
1776 | ||
1777 | ret = btrfs_del_item(trans, root, path); | |
1778 | if (ret) | |
1779 | goto out; | |
a061fc8d CM |
1780 | out: |
1781 | btrfs_free_path(path); | |
3a45bb20 | 1782 | btrfs_commit_transaction(trans); |
a061fc8d CM |
1783 | return ret; |
1784 | } | |
1785 | ||
3cc31a0d DS |
1786 | /* |
1787 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1788 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1789 | * replace. | |
1790 | */ | |
1791 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1792 | u64 num_devices) | |
a061fc8d | 1793 | { |
a061fc8d | 1794 | u64 all_avail; |
de98ced9 | 1795 | unsigned seq; |
418775a2 | 1796 | int i; |
a061fc8d | 1797 | |
de98ced9 | 1798 | do { |
bd45ffbc | 1799 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1800 | |
bd45ffbc AJ |
1801 | all_avail = fs_info->avail_data_alloc_bits | |
1802 | fs_info->avail_system_alloc_bits | | |
1803 | fs_info->avail_metadata_alloc_bits; | |
1804 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1805 | |
418775a2 DS |
1806 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1807 | if (!(all_avail & btrfs_raid_group[i])) | |
1808 | continue; | |
a061fc8d | 1809 | |
418775a2 DS |
1810 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1811 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1812 | |
418775a2 DS |
1813 | if (ret) |
1814 | return ret; | |
1815 | } | |
53b381b3 DW |
1816 | } |
1817 | ||
bd45ffbc | 1818 | return 0; |
f1fa7f26 AJ |
1819 | } |
1820 | ||
88acff64 AJ |
1821 | struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs, |
1822 | struct btrfs_device *device) | |
a061fc8d | 1823 | { |
2b82032c | 1824 | struct btrfs_device *next_device; |
88acff64 AJ |
1825 | |
1826 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1827 | if (next_device != device && | |
1828 | !next_device->missing && next_device->bdev) | |
1829 | return next_device; | |
1830 | } | |
1831 | ||
1832 | return NULL; | |
1833 | } | |
1834 | ||
1835 | /* | |
1836 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1837 | * and replace it with the provided or the next active device, in the context | |
1838 | * where this function called, there should be always be another device (or | |
1839 | * this_dev) which is active. | |
1840 | */ | |
1841 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1842 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1843 | { | |
1844 | struct btrfs_device *next_device; | |
1845 | ||
1846 | if (this_dev) | |
1847 | next_device = this_dev; | |
1848 | else | |
1849 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1850 | device); | |
1851 | ASSERT(next_device); | |
1852 | ||
1853 | if (fs_info->sb->s_bdev && | |
1854 | (fs_info->sb->s_bdev == device->bdev)) | |
1855 | fs_info->sb->s_bdev = next_device->bdev; | |
1856 | ||
1857 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1858 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1859 | } | |
1860 | ||
da353f6b DS |
1861 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1862 | u64 devid) | |
f1fa7f26 AJ |
1863 | { |
1864 | struct btrfs_device *device; | |
1f78160c | 1865 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1866 | u64 num_devices; |
a061fc8d CM |
1867 | int ret = 0; |
1868 | ||
a061fc8d CM |
1869 | mutex_lock(&uuid_mutex); |
1870 | ||
0b246afa JM |
1871 | num_devices = fs_info->fs_devices->num_devices; |
1872 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
1873 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
8dabb742 SB |
1874 | WARN_ON(num_devices < 1); |
1875 | num_devices--; | |
1876 | } | |
0b246afa | 1877 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
8dabb742 | 1878 | |
0b246afa | 1879 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1880 | if (ret) |
a061fc8d | 1881 | goto out; |
a061fc8d | 1882 | |
2ff7e61e JM |
1883 | ret = btrfs_find_device_by_devspec(fs_info, devid, device_path, |
1884 | &device); | |
24fc572f | 1885 | if (ret) |
53b381b3 | 1886 | goto out; |
dfe25020 | 1887 | |
63a212ab | 1888 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1889 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1890 | goto out; |
63a212ab SB |
1891 | } |
1892 | ||
0b246afa | 1893 | if (device->writeable && fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1894 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1895 | goto out; |
2b82032c YZ |
1896 | } |
1897 | ||
1898 | if (device->writeable) { | |
34441361 | 1899 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1900 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1901 | device->fs_devices->rw_devices--; |
34441361 | 1902 | mutex_unlock(&fs_info->chunk_mutex); |
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 | ||
a061fc8d CM |
1981 | out: |
1982 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1983 | return ret; |
24fc572f | 1984 | |
9b3517e9 ID |
1985 | error_undo: |
1986 | if (device->writeable) { | |
34441361 | 1987 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 1988 | list_add(&device->dev_alloc_list, |
0b246afa | 1989 | &fs_info->fs_devices->alloc_list); |
c3929c36 | 1990 | device->fs_devices->rw_devices++; |
34441361 | 1991 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 1992 | } |
24fc572f | 1993 | goto out; |
a061fc8d CM |
1994 | } |
1995 | ||
084b6e7c QW |
1996 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1997 | struct btrfs_device *srcdev) | |
e93c89c1 | 1998 | { |
d51908ce AJ |
1999 | struct btrfs_fs_devices *fs_devices; |
2000 | ||
e93c89c1 | 2001 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 2002 | |
25e8e911 AJ |
2003 | /* |
2004 | * in case of fs with no seed, srcdev->fs_devices will point | |
2005 | * to fs_devices of fs_info. However when the dev being replaced is | |
2006 | * a seed dev it will point to the seed's local fs_devices. In short | |
2007 | * srcdev will have its correct fs_devices in both the cases. | |
2008 | */ | |
2009 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2010 | |
e93c89c1 SB |
2011 | list_del_rcu(&srcdev->dev_list); |
2012 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 2013 | fs_devices->num_devices--; |
82372bc8 | 2014 | if (srcdev->missing) |
d51908ce | 2015 | fs_devices->missing_devices--; |
e93c89c1 | 2016 | |
48b3b9d4 | 2017 | if (srcdev->writeable) |
82372bc8 | 2018 | fs_devices->rw_devices--; |
1357272f | 2019 | |
82372bc8 | 2020 | if (srcdev->bdev) |
d51908ce | 2021 | fs_devices->open_devices--; |
084b6e7c QW |
2022 | } |
2023 | ||
2024 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2025 | struct btrfs_device *srcdev) | |
2026 | { | |
2027 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2028 | |
48b3b9d4 AJ |
2029 | if (srcdev->writeable) { |
2030 | /* zero out the old super if it is writable */ | |
2031 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2032 | } | |
14238819 AJ |
2033 | |
2034 | btrfs_close_bdev(srcdev); | |
2035 | ||
e93c89c1 | 2036 | call_rcu(&srcdev->rcu, free_device); |
94d5f0c2 AJ |
2037 | |
2038 | /* | |
2039 | * unless fs_devices is seed fs, num_devices shouldn't go | |
2040 | * zero | |
2041 | */ | |
2042 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
2043 | ||
2044 | /* if this is no devs we rather delete the fs_devices */ | |
2045 | if (!fs_devices->num_devices) { | |
2046 | struct btrfs_fs_devices *tmp_fs_devices; | |
2047 | ||
2048 | tmp_fs_devices = fs_info->fs_devices; | |
2049 | while (tmp_fs_devices) { | |
2050 | if (tmp_fs_devices->seed == fs_devices) { | |
2051 | tmp_fs_devices->seed = fs_devices->seed; | |
2052 | break; | |
2053 | } | |
2054 | tmp_fs_devices = tmp_fs_devices->seed; | |
2055 | } | |
2056 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2057 | __btrfs_close_devices(fs_devices); |
2058 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2059 | } |
e93c89c1 SB |
2060 | } |
2061 | ||
2062 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2063 | struct btrfs_device *tgtdev) | |
2064 | { | |
67a2c45e | 2065 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2066 | WARN_ON(!tgtdev); |
2067 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2068 | |
32576040 | 2069 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2070 | |
779bf3fe | 2071 | if (tgtdev->bdev) |
e93c89c1 | 2072 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2073 | |
e93c89c1 | 2074 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2075 | |
88acff64 | 2076 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2077 | |
e93c89c1 | 2078 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2079 | |
2080 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2081 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2082 | |
2083 | /* | |
2084 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2085 | * may lead to a call to btrfs_show_devname() which will try | |
2086 | * to hold device_list_mutex. And here this device | |
2087 | * is already out of device list, so we don't have to hold | |
2088 | * the device_list_mutex lock. | |
2089 | */ | |
2090 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2091 | |
2092 | btrfs_close_bdev(tgtdev); | |
779bf3fe | 2093 | call_rcu(&tgtdev->rcu, free_device); |
e93c89c1 SB |
2094 | } |
2095 | ||
2ff7e61e | 2096 | static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2097 | const char *device_path, |
48a3b636 | 2098 | struct btrfs_device **device) |
7ba15b7d SB |
2099 | { |
2100 | int ret = 0; | |
2101 | struct btrfs_super_block *disk_super; | |
2102 | u64 devid; | |
2103 | u8 *dev_uuid; | |
2104 | struct block_device *bdev; | |
2105 | struct buffer_head *bh; | |
2106 | ||
2107 | *device = NULL; | |
2108 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2109 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2110 | if (ret) |
2111 | return ret; | |
2112 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2113 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2114 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2115 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2116 | brelse(bh); |
2117 | if (!*device) | |
2118 | ret = -ENOENT; | |
2119 | blkdev_put(bdev, FMODE_READ); | |
2120 | return ret; | |
2121 | } | |
2122 | ||
2ff7e61e | 2123 | int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2124 | const char *device_path, |
7ba15b7d SB |
2125 | struct btrfs_device **device) |
2126 | { | |
2127 | *device = NULL; | |
2128 | if (strcmp(device_path, "missing") == 0) { | |
2129 | struct list_head *devices; | |
2130 | struct btrfs_device *tmp; | |
2131 | ||
0b246afa | 2132 | devices = &fs_info->fs_devices->devices; |
7ba15b7d SB |
2133 | /* |
2134 | * It is safe to read the devices since the volume_mutex | |
2135 | * is held by the caller. | |
2136 | */ | |
2137 | list_for_each_entry(tmp, devices, dev_list) { | |
2138 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2139 | *device = tmp; | |
2140 | break; | |
2141 | } | |
2142 | } | |
2143 | ||
d74a6259 AJ |
2144 | if (!*device) |
2145 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2146 | |
2147 | return 0; | |
2148 | } else { | |
2ff7e61e | 2149 | return btrfs_find_device_by_path(fs_info, device_path, device); |
7ba15b7d SB |
2150 | } |
2151 | } | |
2152 | ||
5c5c0df0 DS |
2153 | /* |
2154 | * Lookup a device given by device id, or the path if the id is 0. | |
2155 | */ | |
2ff7e61e | 2156 | int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid, |
da353f6b DS |
2157 | const char *devpath, |
2158 | struct btrfs_device **device) | |
24e0474b AJ |
2159 | { |
2160 | int ret; | |
2161 | ||
5c5c0df0 | 2162 | if (devid) { |
24e0474b | 2163 | ret = 0; |
0b246afa | 2164 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2165 | if (!*device) |
2166 | ret = -ENOENT; | |
2167 | } else { | |
5c5c0df0 | 2168 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2169 | return -EINVAL; |
2170 | ||
2ff7e61e | 2171 | ret = btrfs_find_device_missing_or_by_path(fs_info, devpath, |
24e0474b AJ |
2172 | device); |
2173 | } | |
2174 | return ret; | |
2175 | } | |
2176 | ||
2b82032c YZ |
2177 | /* |
2178 | * does all the dirty work required for changing file system's UUID. | |
2179 | */ | |
2ff7e61e | 2180 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2181 | { |
0b246afa | 2182 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2183 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2184 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2185 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2186 | struct btrfs_device *device; |
2187 | u64 super_flags; | |
2188 | ||
2189 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2190 | if (!fs_devices->seeding) |
2b82032c YZ |
2191 | return -EINVAL; |
2192 | ||
2dfeca9b | 2193 | seed_devices = alloc_fs_devices(NULL); |
2208a378 ID |
2194 | if (IS_ERR(seed_devices)) |
2195 | return PTR_ERR(seed_devices); | |
2b82032c | 2196 | |
e4404d6e YZ |
2197 | old_devices = clone_fs_devices(fs_devices); |
2198 | if (IS_ERR(old_devices)) { | |
2199 | kfree(seed_devices); | |
2200 | return PTR_ERR(old_devices); | |
2b82032c | 2201 | } |
e4404d6e | 2202 | |
2b82032c YZ |
2203 | list_add(&old_devices->list, &fs_uuids); |
2204 | ||
e4404d6e YZ |
2205 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2206 | seed_devices->opened = 1; | |
2207 | INIT_LIST_HEAD(&seed_devices->devices); | |
2208 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2209 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2210 | |
0b246afa | 2211 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c XG |
2212 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2213 | synchronize_rcu); | |
2196d6e8 MX |
2214 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2215 | device->fs_devices = seed_devices; | |
c9513edb | 2216 | |
34441361 | 2217 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2218 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2219 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2220 | |
2b82032c YZ |
2221 | fs_devices->seeding = 0; |
2222 | fs_devices->num_devices = 0; | |
2223 | fs_devices->open_devices = 0; | |
69611ac8 | 2224 | fs_devices->missing_devices = 0; |
69611ac8 | 2225 | fs_devices->rotating = 0; |
e4404d6e | 2226 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2227 | |
2228 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2229 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2230 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
0b246afa | 2231 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
f7171750 | 2232 | |
2b82032c YZ |
2233 | super_flags = btrfs_super_flags(disk_super) & |
2234 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2235 | btrfs_set_super_flags(disk_super, super_flags); | |
2236 | ||
2237 | return 0; | |
2238 | } | |
2239 | ||
2240 | /* | |
01327610 | 2241 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2242 | */ |
2243 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2244 | struct btrfs_fs_info *fs_info) |
2b82032c | 2245 | { |
5b4aacef | 2246 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2247 | struct btrfs_path *path; |
2248 | struct extent_buffer *leaf; | |
2249 | struct btrfs_dev_item *dev_item; | |
2250 | struct btrfs_device *device; | |
2251 | struct btrfs_key key; | |
2252 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2253 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2254 | u64 devid; | |
2255 | int ret; | |
2256 | ||
2257 | path = btrfs_alloc_path(); | |
2258 | if (!path) | |
2259 | return -ENOMEM; | |
2260 | ||
2b82032c YZ |
2261 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2262 | key.offset = 0; | |
2263 | key.type = BTRFS_DEV_ITEM_KEY; | |
2264 | ||
2265 | while (1) { | |
2266 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2267 | if (ret < 0) | |
2268 | goto error; | |
2269 | ||
2270 | leaf = path->nodes[0]; | |
2271 | next_slot: | |
2272 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2273 | ret = btrfs_next_leaf(root, path); | |
2274 | if (ret > 0) | |
2275 | break; | |
2276 | if (ret < 0) | |
2277 | goto error; | |
2278 | leaf = path->nodes[0]; | |
2279 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2280 | btrfs_release_path(path); |
2b82032c YZ |
2281 | continue; |
2282 | } | |
2283 | ||
2284 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2285 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2286 | key.type != BTRFS_DEV_ITEM_KEY) | |
2287 | break; | |
2288 | ||
2289 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2290 | struct btrfs_dev_item); | |
2291 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2292 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2293 | BTRFS_UUID_SIZE); |
1473b24e | 2294 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2295 | BTRFS_UUID_SIZE); |
0b246afa | 2296 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2297 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2298 | |
2299 | if (device->fs_devices->seeding) { | |
2300 | btrfs_set_device_generation(leaf, dev_item, | |
2301 | device->generation); | |
2302 | btrfs_mark_buffer_dirty(leaf); | |
2303 | } | |
2304 | ||
2305 | path->slots[0]++; | |
2306 | goto next_slot; | |
2307 | } | |
2308 | ret = 0; | |
2309 | error: | |
2310 | btrfs_free_path(path); | |
2311 | return ret; | |
2312 | } | |
2313 | ||
da353f6b | 2314 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2315 | { |
5112febb | 2316 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2317 | struct request_queue *q; |
788f20eb CM |
2318 | struct btrfs_trans_handle *trans; |
2319 | struct btrfs_device *device; | |
2320 | struct block_device *bdev; | |
788f20eb | 2321 | struct list_head *devices; |
0b246afa | 2322 | struct super_block *sb = fs_info->sb; |
606686ee | 2323 | struct rcu_string *name; |
3c1dbdf5 | 2324 | u64 tmp; |
2b82032c | 2325 | int seeding_dev = 0; |
788f20eb CM |
2326 | int ret = 0; |
2327 | ||
0b246afa | 2328 | if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding) |
f8c5d0b4 | 2329 | return -EROFS; |
788f20eb | 2330 | |
a5d16333 | 2331 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2332 | fs_info->bdev_holder); |
7f59203a JB |
2333 | if (IS_ERR(bdev)) |
2334 | return PTR_ERR(bdev); | |
a2135011 | 2335 | |
0b246afa | 2336 | if (fs_info->fs_devices->seeding) { |
2b82032c YZ |
2337 | seeding_dev = 1; |
2338 | down_write(&sb->s_umount); | |
2339 | mutex_lock(&uuid_mutex); | |
2340 | } | |
2341 | ||
8c8bee1d | 2342 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2343 | |
0b246afa | 2344 | devices = &fs_info->fs_devices->devices; |
d25628bd | 2345 | |
0b246afa | 2346 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
c6e30871 | 2347 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2348 | if (device->bdev == bdev) { |
2349 | ret = -EEXIST; | |
d25628bd | 2350 | mutex_unlock( |
0b246afa | 2351 | &fs_info->fs_devices->device_list_mutex); |
2b82032c | 2352 | goto error; |
788f20eb CM |
2353 | } |
2354 | } | |
0b246afa | 2355 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2356 | |
0b246afa | 2357 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2358 | if (IS_ERR(device)) { |
788f20eb | 2359 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2360 | ret = PTR_ERR(device); |
2b82032c | 2361 | goto error; |
788f20eb CM |
2362 | } |
2363 | ||
78f2c9e6 | 2364 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2365 | if (!name) { |
788f20eb | 2366 | kfree(device); |
2b82032c YZ |
2367 | ret = -ENOMEM; |
2368 | goto error; | |
788f20eb | 2369 | } |
606686ee | 2370 | rcu_assign_pointer(device->name, name); |
2b82032c | 2371 | |
a22285a6 | 2372 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2373 | if (IS_ERR(trans)) { |
606686ee | 2374 | rcu_string_free(device->name); |
98d5dc13 TI |
2375 | kfree(device); |
2376 | ret = PTR_ERR(trans); | |
2377 | goto error; | |
2378 | } | |
2379 | ||
d5e2003c JB |
2380 | q = bdev_get_queue(bdev); |
2381 | if (blk_queue_discard(q)) | |
2382 | device->can_discard = 1; | |
2b82032c | 2383 | device->writeable = 1; |
2b82032c | 2384 | device->generation = trans->transid; |
0b246afa JM |
2385 | device->io_width = fs_info->sectorsize; |
2386 | device->io_align = fs_info->sectorsize; | |
2387 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2388 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2389 | fs_info->sectorsize); | |
2cc3c559 | 2390 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2391 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2392 | device->fs_info = fs_info; |
788f20eb | 2393 | device->bdev = bdev; |
dfe25020 | 2394 | device->in_fs_metadata = 1; |
63a212ab | 2395 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2396 | device->mode = FMODE_EXCL; |
27087f37 | 2397 | device->dev_stats_valid = 1; |
2b82032c | 2398 | set_blocksize(device->bdev, 4096); |
788f20eb | 2399 | |
2b82032c YZ |
2400 | if (seeding_dev) { |
2401 | sb->s_flags &= ~MS_RDONLY; | |
2ff7e61e | 2402 | ret = btrfs_prepare_sprout(fs_info); |
79787eaa | 2403 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2404 | } |
788f20eb | 2405 | |
0b246afa | 2406 | device->fs_devices = fs_info->fs_devices; |
e5e9a520 | 2407 | |
0b246afa | 2408 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
34441361 | 2409 | mutex_lock(&fs_info->chunk_mutex); |
0b246afa | 2410 | list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices); |
2b82032c | 2411 | list_add(&device->dev_alloc_list, |
0b246afa JM |
2412 | &fs_info->fs_devices->alloc_list); |
2413 | fs_info->fs_devices->num_devices++; | |
2414 | fs_info->fs_devices->open_devices++; | |
2415 | fs_info->fs_devices->rw_devices++; | |
2416 | fs_info->fs_devices->total_devices++; | |
2417 | fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2418 | |
a5ed45f8 | 2419 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2420 | |
e884f4f0 | 2421 | if (!blk_queue_nonrot(q)) |
0b246afa | 2422 | fs_info->fs_devices->rotating = 1; |
c289811c | 2423 | |
0b246afa JM |
2424 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2425 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
7dfb8be1 | 2426 | round_down(tmp + device->total_bytes, fs_info->sectorsize)); |
788f20eb | 2427 | |
0b246afa JM |
2428 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2429 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2430 | |
2431 | /* add sysfs device entry */ | |
0b246afa | 2432 | btrfs_sysfs_add_device_link(fs_info->fs_devices, device); |
0d39376a | 2433 | |
2196d6e8 MX |
2434 | /* |
2435 | * we've got more storage, clear any full flags on the space | |
2436 | * infos | |
2437 | */ | |
0b246afa | 2438 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2439 | |
34441361 | 2440 | mutex_unlock(&fs_info->chunk_mutex); |
0b246afa | 2441 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2442 | |
2b82032c | 2443 | if (seeding_dev) { |
34441361 | 2444 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2445 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2446 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2447 | if (ret) { |
66642832 | 2448 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2449 | goto error_trans; |
005d6427 | 2450 | } |
2196d6e8 MX |
2451 | } |
2452 | ||
0b246afa | 2453 | ret = btrfs_add_device(trans, fs_info, device); |
2196d6e8 | 2454 | if (ret) { |
66642832 | 2455 | btrfs_abort_transaction(trans, ret); |
2196d6e8 MX |
2456 | goto error_trans; |
2457 | } | |
2458 | ||
2459 | if (seeding_dev) { | |
2460 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2461 | ||
0b246afa | 2462 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2463 | if (ret) { |
66642832 | 2464 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2465 | goto error_trans; |
005d6427 | 2466 | } |
b2373f25 AJ |
2467 | |
2468 | /* Sprouting would change fsid of the mounted root, | |
2469 | * so rename the fsid on the sysfs | |
2470 | */ | |
2471 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa JM |
2472 | fs_info->fsid); |
2473 | if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf)) | |
2474 | btrfs_warn(fs_info, | |
2475 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2476 | } |
2477 | ||
3a45bb20 | 2478 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2479 | |
2b82032c YZ |
2480 | if (seeding_dev) { |
2481 | mutex_unlock(&uuid_mutex); | |
2482 | up_write(&sb->s_umount); | |
788f20eb | 2483 | |
79787eaa JM |
2484 | if (ret) /* transaction commit */ |
2485 | return ret; | |
2486 | ||
2ff7e61e | 2487 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2488 | if (ret < 0) |
0b246afa | 2489 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2490 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2491 | trans = btrfs_attach_transaction(root); |
2492 | if (IS_ERR(trans)) { | |
2493 | if (PTR_ERR(trans) == -ENOENT) | |
2494 | return 0; | |
2495 | return PTR_ERR(trans); | |
2496 | } | |
3a45bb20 | 2497 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2498 | } |
c9e9f97b | 2499 | |
5a1972bd QW |
2500 | /* Update ctime/mtime for libblkid */ |
2501 | update_dev_time(device_path); | |
2b82032c | 2502 | return ret; |
79787eaa JM |
2503 | |
2504 | error_trans: | |
3a45bb20 | 2505 | btrfs_end_transaction(trans); |
606686ee | 2506 | rcu_string_free(device->name); |
0b246afa | 2507 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
79787eaa | 2508 | kfree(device); |
2b82032c | 2509 | error: |
e525fd89 | 2510 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2511 | if (seeding_dev) { |
2512 | mutex_unlock(&uuid_mutex); | |
2513 | up_write(&sb->s_umount); | |
2514 | } | |
c9e9f97b | 2515 | return ret; |
788f20eb CM |
2516 | } |
2517 | ||
2ff7e61e | 2518 | int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, |
da353f6b | 2519 | const char *device_path, |
1c43366d | 2520 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2521 | struct btrfs_device **device_out) |
2522 | { | |
2523 | struct request_queue *q; | |
2524 | struct btrfs_device *device; | |
2525 | struct block_device *bdev; | |
e93c89c1 SB |
2526 | struct list_head *devices; |
2527 | struct rcu_string *name; | |
12bd2fc0 | 2528 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2529 | int ret = 0; |
2530 | ||
2531 | *device_out = NULL; | |
1c43366d MX |
2532 | if (fs_info->fs_devices->seeding) { |
2533 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2534 | return -EINVAL; |
1c43366d | 2535 | } |
e93c89c1 SB |
2536 | |
2537 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2538 | fs_info->bdev_holder); | |
1c43366d MX |
2539 | if (IS_ERR(bdev)) { |
2540 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2541 | return PTR_ERR(bdev); |
1c43366d | 2542 | } |
e93c89c1 SB |
2543 | |
2544 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2545 | ||
2546 | devices = &fs_info->fs_devices->devices; | |
2547 | list_for_each_entry(device, devices, dev_list) { | |
2548 | if (device->bdev == bdev) { | |
5d163e0e JM |
2549 | btrfs_err(fs_info, |
2550 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2551 | ret = -EEXIST; |
2552 | goto error; | |
2553 | } | |
2554 | } | |
2555 | ||
1c43366d | 2556 | |
7cc8e58d MX |
2557 | if (i_size_read(bdev->bd_inode) < |
2558 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2559 | btrfs_err(fs_info, |
2560 | "target device is smaller than source device!"); | |
1c43366d MX |
2561 | ret = -EINVAL; |
2562 | goto error; | |
2563 | } | |
2564 | ||
2565 | ||
12bd2fc0 ID |
2566 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2567 | if (IS_ERR(device)) { | |
2568 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2569 | goto error; |
2570 | } | |
2571 | ||
6165572c | 2572 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
e93c89c1 SB |
2573 | if (!name) { |
2574 | kfree(device); | |
2575 | ret = -ENOMEM; | |
2576 | goto error; | |
2577 | } | |
2578 | rcu_assign_pointer(device->name, name); | |
2579 | ||
2580 | q = bdev_get_queue(bdev); | |
2581 | if (blk_queue_discard(q)) | |
2582 | device->can_discard = 1; | |
0b246afa | 2583 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 | 2584 | device->writeable = 1; |
e93c89c1 | 2585 | device->generation = 0; |
0b246afa JM |
2586 | device->io_width = fs_info->sectorsize; |
2587 | device->io_align = fs_info->sectorsize; | |
2588 | device->sector_size = fs_info->sectorsize; | |
7cc8e58d MX |
2589 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2590 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2591 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2592 | ASSERT(list_empty(&srcdev->resized_list)); |
2593 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2594 | device->commit_bytes_used = device->bytes_used; |
fb456252 | 2595 | device->fs_info = fs_info; |
e93c89c1 SB |
2596 | device->bdev = bdev; |
2597 | device->in_fs_metadata = 1; | |
2598 | device->is_tgtdev_for_dev_replace = 1; | |
2599 | device->mode = FMODE_EXCL; | |
27087f37 | 2600 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2601 | set_blocksize(device->bdev, 4096); |
2602 | device->fs_devices = fs_info->fs_devices; | |
2603 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2604 | fs_info->fs_devices->num_devices++; | |
2605 | fs_info->fs_devices->open_devices++; | |
0b246afa | 2606 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 SB |
2607 | |
2608 | *device_out = device; | |
2609 | return ret; | |
2610 | ||
2611 | error: | |
2612 | blkdev_put(bdev, FMODE_EXCL); | |
2613 | return ret; | |
2614 | } | |
2615 | ||
2616 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2617 | struct btrfs_device *tgtdev) | |
2618 | { | |
da17066c JM |
2619 | u32 sectorsize = fs_info->sectorsize; |
2620 | ||
e93c89c1 | 2621 | WARN_ON(fs_info->fs_devices->rw_devices == 0); |
da17066c JM |
2622 | tgtdev->io_width = sectorsize; |
2623 | tgtdev->io_align = sectorsize; | |
2624 | tgtdev->sector_size = sectorsize; | |
fb456252 | 2625 | tgtdev->fs_info = fs_info; |
e93c89c1 SB |
2626 | tgtdev->in_fs_metadata = 1; |
2627 | } | |
2628 | ||
d397712b CM |
2629 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2630 | struct btrfs_device *device) | |
0b86a832 CM |
2631 | { |
2632 | int ret; | |
2633 | struct btrfs_path *path; | |
0b246afa | 2634 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2635 | struct btrfs_dev_item *dev_item; |
2636 | struct extent_buffer *leaf; | |
2637 | struct btrfs_key key; | |
2638 | ||
0b86a832 CM |
2639 | path = btrfs_alloc_path(); |
2640 | if (!path) | |
2641 | return -ENOMEM; | |
2642 | ||
2643 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2644 | key.type = BTRFS_DEV_ITEM_KEY; | |
2645 | key.offset = device->devid; | |
2646 | ||
2647 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2648 | if (ret < 0) | |
2649 | goto out; | |
2650 | ||
2651 | if (ret > 0) { | |
2652 | ret = -ENOENT; | |
2653 | goto out; | |
2654 | } | |
2655 | ||
2656 | leaf = path->nodes[0]; | |
2657 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2658 | ||
2659 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2660 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2661 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2662 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2663 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2664 | btrfs_set_device_total_bytes(leaf, dev_item, |
2665 | btrfs_device_get_disk_total_bytes(device)); | |
2666 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2667 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2668 | btrfs_mark_buffer_dirty(leaf); |
2669 | ||
2670 | out: | |
2671 | btrfs_free_path(path); | |
2672 | return ret; | |
2673 | } | |
2674 | ||
2196d6e8 | 2675 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2676 | struct btrfs_device *device, u64 new_size) |
2677 | { | |
0b246afa JM |
2678 | struct btrfs_fs_info *fs_info = device->fs_info; |
2679 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2680 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2681 | u64 old_total; |
2682 | u64 diff; | |
8f18cf13 | 2683 | |
2b82032c YZ |
2684 | if (!device->writeable) |
2685 | return -EACCES; | |
2196d6e8 | 2686 | |
7dfb8be1 NB |
2687 | new_size = round_down(new_size, fs_info->sectorsize); |
2688 | ||
34441361 | 2689 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2690 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2691 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2692 | |
63a212ab | 2693 | if (new_size <= device->total_bytes || |
2196d6e8 | 2694 | device->is_tgtdev_for_dev_replace) { |
34441361 | 2695 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2696 | return -EINVAL; |
2196d6e8 | 2697 | } |
2b82032c | 2698 | |
0b246afa | 2699 | fs_devices = fs_info->fs_devices; |
2b82032c | 2700 | |
7dfb8be1 NB |
2701 | btrfs_set_super_total_bytes(super_copy, |
2702 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2703 | device->fs_devices->total_rw_bytes += diff; |
2704 | ||
7cc8e58d MX |
2705 | btrfs_device_set_total_bytes(device, new_size); |
2706 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2707 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2708 | if (list_empty(&device->resized_list)) |
2709 | list_add_tail(&device->resized_list, | |
2710 | &fs_devices->resized_devices); | |
34441361 | 2711 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2712 | |
8f18cf13 CM |
2713 | return btrfs_update_device(trans, device); |
2714 | } | |
2715 | ||
2716 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
5b4aacef | 2717 | struct btrfs_fs_info *fs_info, u64 chunk_objectid, |
8f18cf13 CM |
2718 | u64 chunk_offset) |
2719 | { | |
5b4aacef | 2720 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2721 | int ret; |
2722 | struct btrfs_path *path; | |
2723 | struct btrfs_key key; | |
2724 | ||
8f18cf13 CM |
2725 | path = btrfs_alloc_path(); |
2726 | if (!path) | |
2727 | return -ENOMEM; | |
2728 | ||
2729 | key.objectid = chunk_objectid; | |
2730 | key.offset = chunk_offset; | |
2731 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2732 | ||
2733 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2734 | if (ret < 0) |
2735 | goto out; | |
2736 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2737 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2738 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2739 | ret = -ENOENT; |
2740 | goto out; | |
2741 | } | |
8f18cf13 CM |
2742 | |
2743 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2744 | if (ret < 0) |
0b246afa JM |
2745 | btrfs_handle_fs_error(fs_info, ret, |
2746 | "Failed to delete chunk item."); | |
79787eaa | 2747 | out: |
8f18cf13 | 2748 | btrfs_free_path(path); |
65a246c5 | 2749 | return ret; |
8f18cf13 CM |
2750 | } |
2751 | ||
6bccf3ab JM |
2752 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, |
2753 | u64 chunk_objectid, u64 chunk_offset) | |
8f18cf13 | 2754 | { |
0b246afa | 2755 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2756 | struct btrfs_disk_key *disk_key; |
2757 | struct btrfs_chunk *chunk; | |
2758 | u8 *ptr; | |
2759 | int ret = 0; | |
2760 | u32 num_stripes; | |
2761 | u32 array_size; | |
2762 | u32 len = 0; | |
2763 | u32 cur; | |
2764 | struct btrfs_key key; | |
2765 | ||
34441361 | 2766 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2767 | array_size = btrfs_super_sys_array_size(super_copy); |
2768 | ||
2769 | ptr = super_copy->sys_chunk_array; | |
2770 | cur = 0; | |
2771 | ||
2772 | while (cur < array_size) { | |
2773 | disk_key = (struct btrfs_disk_key *)ptr; | |
2774 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2775 | ||
2776 | len = sizeof(*disk_key); | |
2777 | ||
2778 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2779 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2780 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2781 | len += btrfs_chunk_item_size(num_stripes); | |
2782 | } else { | |
2783 | ret = -EIO; | |
2784 | break; | |
2785 | } | |
2786 | if (key.objectid == chunk_objectid && | |
2787 | key.offset == chunk_offset) { | |
2788 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2789 | array_size -= len; | |
2790 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2791 | } else { | |
2792 | ptr += len; | |
2793 | cur += len; | |
2794 | } | |
2795 | } | |
34441361 | 2796 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2797 | return ret; |
2798 | } | |
2799 | ||
592d92ee LB |
2800 | static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info, |
2801 | u64 logical, u64 length) | |
2802 | { | |
2803 | struct extent_map_tree *em_tree; | |
2804 | struct extent_map *em; | |
2805 | ||
2806 | em_tree = &fs_info->mapping_tree.map_tree; | |
2807 | read_lock(&em_tree->lock); | |
2808 | em = lookup_extent_mapping(em_tree, logical, length); | |
2809 | read_unlock(&em_tree->lock); | |
2810 | ||
2811 | if (!em) { | |
2812 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2813 | logical, length); | |
2814 | return ERR_PTR(-EINVAL); | |
2815 | } | |
2816 | ||
2817 | if (em->start > logical || em->start + em->len < logical) { | |
2818 | btrfs_crit(fs_info, | |
2819 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2820 | logical, length, em->start, em->start + em->len); | |
2821 | free_extent_map(em); | |
2822 | return ERR_PTR(-EINVAL); | |
2823 | } | |
2824 | ||
2825 | /* callers are responsible for dropping em's ref. */ | |
2826 | return em; | |
2827 | } | |
2828 | ||
47ab2a6c | 2829 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
5b4aacef | 2830 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2831 | { |
8f18cf13 CM |
2832 | struct extent_map *em; |
2833 | struct map_lookup *map; | |
2196d6e8 | 2834 | u64 dev_extent_len = 0; |
47ab2a6c | 2835 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2836 | int i, ret = 0; |
0b246afa | 2837 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2838 | |
592d92ee LB |
2839 | em = get_chunk_map(fs_info, chunk_offset, 1); |
2840 | if (IS_ERR(em)) { | |
47ab2a6c JB |
2841 | /* |
2842 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2843 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2844 | * do anything we still error out. |
2845 | */ | |
2846 | ASSERT(0); | |
592d92ee | 2847 | return PTR_ERR(em); |
47ab2a6c | 2848 | } |
95617d69 | 2849 | map = em->map_lookup; |
34441361 | 2850 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 2851 | check_system_chunk(trans, fs_info, map->type); |
34441361 | 2852 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2853 | |
57ba4cb8 FM |
2854 | /* |
2855 | * Take the device list mutex to prevent races with the final phase of | |
2856 | * a device replace operation that replaces the device object associated | |
2857 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2858 | */ | |
2859 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2860 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2861 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2862 | ret = btrfs_free_dev_extent(trans, device, |
2863 | map->stripes[i].physical, | |
2864 | &dev_extent_len); | |
47ab2a6c | 2865 | if (ret) { |
57ba4cb8 | 2866 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2867 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2868 | goto out; |
2869 | } | |
a061fc8d | 2870 | |
2196d6e8 | 2871 | if (device->bytes_used > 0) { |
34441361 | 2872 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2873 | btrfs_device_set_bytes_used(device, |
2874 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2875 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2876 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2877 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2878 | } |
a061fc8d | 2879 | |
dfe25020 CM |
2880 | if (map->stripes[i].dev) { |
2881 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2882 | if (ret) { |
57ba4cb8 | 2883 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2884 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2885 | goto out; |
2886 | } | |
dfe25020 | 2887 | } |
8f18cf13 | 2888 | } |
57ba4cb8 FM |
2889 | mutex_unlock(&fs_devices->device_list_mutex); |
2890 | ||
0b246afa | 2891 | ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset); |
47ab2a6c | 2892 | if (ret) { |
66642832 | 2893 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2894 | goto out; |
2895 | } | |
8f18cf13 | 2896 | |
6bccf3ab | 2897 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2898 | |
8f18cf13 | 2899 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
6bccf3ab JM |
2900 | ret = btrfs_del_sys_chunk(fs_info, chunk_objectid, |
2901 | chunk_offset); | |
47ab2a6c | 2902 | if (ret) { |
66642832 | 2903 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2904 | goto out; |
2905 | } | |
8f18cf13 CM |
2906 | } |
2907 | ||
6bccf3ab | 2908 | ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em); |
47ab2a6c | 2909 | if (ret) { |
66642832 | 2910 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2911 | goto out; |
2912 | } | |
2b82032c | 2913 | |
47ab2a6c | 2914 | out: |
2b82032c YZ |
2915 | /* once for us */ |
2916 | free_extent_map(em); | |
47ab2a6c JB |
2917 | return ret; |
2918 | } | |
2b82032c | 2919 | |
5b4aacef | 2920 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2921 | { |
5b4aacef | 2922 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2923 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2924 | int ret; |
2b82032c | 2925 | |
67c5e7d4 FM |
2926 | /* |
2927 | * Prevent races with automatic removal of unused block groups. | |
2928 | * After we relocate and before we remove the chunk with offset | |
2929 | * chunk_offset, automatic removal of the block group can kick in, | |
2930 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2931 | * | |
2932 | * Make sure to acquire this mutex before doing a tree search (dev | |
2933 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2934 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2935 | * we release the path used to search the chunk/dev tree and before | |
2936 | * the current task acquires this mutex and calls us. | |
2937 | */ | |
0b246afa | 2938 | ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex)); |
67c5e7d4 | 2939 | |
0b246afa | 2940 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2941 | if (ret) |
2942 | return -ENOSPC; | |
2943 | ||
2944 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 2945 | btrfs_scrub_pause(fs_info); |
0b246afa | 2946 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2947 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2948 | if (ret) |
2949 | return ret; | |
2950 | ||
19c4d2f9 CM |
2951 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2952 | chunk_offset); | |
2953 | if (IS_ERR(trans)) { | |
2954 | ret = PTR_ERR(trans); | |
2955 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2956 | return ret; | |
2957 | } | |
2958 | ||
47ab2a6c | 2959 | /* |
19c4d2f9 CM |
2960 | * step two, delete the device extents and the |
2961 | * chunk tree entries | |
47ab2a6c | 2962 | */ |
5b4aacef | 2963 | ret = btrfs_remove_chunk(trans, fs_info, chunk_offset); |
3a45bb20 | 2964 | btrfs_end_transaction(trans); |
19c4d2f9 | 2965 | return ret; |
2b82032c YZ |
2966 | } |
2967 | ||
2ff7e61e | 2968 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2969 | { |
0b246afa | 2970 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2971 | struct btrfs_path *path; |
2972 | struct extent_buffer *leaf; | |
2973 | struct btrfs_chunk *chunk; | |
2974 | struct btrfs_key key; | |
2975 | struct btrfs_key found_key; | |
2b82032c | 2976 | u64 chunk_type; |
ba1bf481 JB |
2977 | bool retried = false; |
2978 | int failed = 0; | |
2b82032c YZ |
2979 | int ret; |
2980 | ||
2981 | path = btrfs_alloc_path(); | |
2982 | if (!path) | |
2983 | return -ENOMEM; | |
2984 | ||
ba1bf481 | 2985 | again: |
2b82032c YZ |
2986 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2987 | key.offset = (u64)-1; | |
2988 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2989 | ||
2990 | while (1) { | |
0b246afa | 2991 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2992 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2993 | if (ret < 0) { |
0b246afa | 2994 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2995 | goto error; |
67c5e7d4 | 2996 | } |
79787eaa | 2997 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2998 | |
2999 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3000 | key.type); | |
67c5e7d4 | 3001 | if (ret) |
0b246afa | 3002 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3003 | if (ret < 0) |
3004 | goto error; | |
3005 | if (ret > 0) | |
3006 | break; | |
1a40e23b | 3007 | |
2b82032c YZ |
3008 | leaf = path->nodes[0]; |
3009 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3010 | |
2b82032c YZ |
3011 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3012 | struct btrfs_chunk); | |
3013 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3014 | btrfs_release_path(path); |
8f18cf13 | 3015 | |
2b82032c | 3016 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3017 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3018 | if (ret == -ENOSPC) |
3019 | failed++; | |
14586651 HS |
3020 | else |
3021 | BUG_ON(ret); | |
2b82032c | 3022 | } |
0b246afa | 3023 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3024 | |
2b82032c YZ |
3025 | if (found_key.offset == 0) |
3026 | break; | |
3027 | key.offset = found_key.offset - 1; | |
3028 | } | |
3029 | ret = 0; | |
ba1bf481 JB |
3030 | if (failed && !retried) { |
3031 | failed = 0; | |
3032 | retried = true; | |
3033 | goto again; | |
fae7f21c | 3034 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3035 | ret = -ENOSPC; |
3036 | } | |
2b82032c YZ |
3037 | error: |
3038 | btrfs_free_path(path); | |
3039 | return ret; | |
8f18cf13 CM |
3040 | } |
3041 | ||
6bccf3ab | 3042 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3043 | struct btrfs_balance_control *bctl) |
3044 | { | |
6bccf3ab | 3045 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3046 | struct btrfs_trans_handle *trans; |
3047 | struct btrfs_balance_item *item; | |
3048 | struct btrfs_disk_balance_args disk_bargs; | |
3049 | struct btrfs_path *path; | |
3050 | struct extent_buffer *leaf; | |
3051 | struct btrfs_key key; | |
3052 | int ret, err; | |
3053 | ||
3054 | path = btrfs_alloc_path(); | |
3055 | if (!path) | |
3056 | return -ENOMEM; | |
3057 | ||
3058 | trans = btrfs_start_transaction(root, 0); | |
3059 | if (IS_ERR(trans)) { | |
3060 | btrfs_free_path(path); | |
3061 | return PTR_ERR(trans); | |
3062 | } | |
3063 | ||
3064 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3065 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3066 | key.offset = 0; |
3067 | ||
3068 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3069 | sizeof(*item)); | |
3070 | if (ret) | |
3071 | goto out; | |
3072 | ||
3073 | leaf = path->nodes[0]; | |
3074 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3075 | ||
b159fa28 | 3076 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3077 | |
3078 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3079 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3080 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3081 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3082 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3083 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3084 | ||
3085 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3086 | ||
3087 | btrfs_mark_buffer_dirty(leaf); | |
3088 | out: | |
3089 | btrfs_free_path(path); | |
3a45bb20 | 3090 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3091 | if (err && !ret) |
3092 | ret = err; | |
3093 | return ret; | |
3094 | } | |
3095 | ||
6bccf3ab | 3096 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3097 | { |
6bccf3ab | 3098 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3099 | struct btrfs_trans_handle *trans; |
3100 | struct btrfs_path *path; | |
3101 | struct btrfs_key key; | |
3102 | int ret, err; | |
3103 | ||
3104 | path = btrfs_alloc_path(); | |
3105 | if (!path) | |
3106 | return -ENOMEM; | |
3107 | ||
3108 | trans = btrfs_start_transaction(root, 0); | |
3109 | if (IS_ERR(trans)) { | |
3110 | btrfs_free_path(path); | |
3111 | return PTR_ERR(trans); | |
3112 | } | |
3113 | ||
3114 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3115 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3116 | key.offset = 0; |
3117 | ||
3118 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3119 | if (ret < 0) | |
3120 | goto out; | |
3121 | if (ret > 0) { | |
3122 | ret = -ENOENT; | |
3123 | goto out; | |
3124 | } | |
3125 | ||
3126 | ret = btrfs_del_item(trans, root, path); | |
3127 | out: | |
3128 | btrfs_free_path(path); | |
3a45bb20 | 3129 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3130 | if (err && !ret) |
3131 | ret = err; | |
3132 | return ret; | |
3133 | } | |
3134 | ||
59641015 ID |
3135 | /* |
3136 | * This is a heuristic used to reduce the number of chunks balanced on | |
3137 | * resume after balance was interrupted. | |
3138 | */ | |
3139 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3140 | { | |
3141 | /* | |
3142 | * Turn on soft mode for chunk types that were being converted. | |
3143 | */ | |
3144 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3145 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3146 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3147 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3148 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3149 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3150 | ||
3151 | /* | |
3152 | * Turn on usage filter if is not already used. The idea is | |
3153 | * that chunks that we have already balanced should be | |
3154 | * reasonably full. Don't do it for chunks that are being | |
3155 | * converted - that will keep us from relocating unconverted | |
3156 | * (albeit full) chunks. | |
3157 | */ | |
3158 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3159 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3160 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3161 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3162 | bctl->data.usage = 90; | |
3163 | } | |
3164 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3165 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3166 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3167 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3168 | bctl->sys.usage = 90; | |
3169 | } | |
3170 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3171 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3172 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3173 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3174 | bctl->meta.usage = 90; | |
3175 | } | |
3176 | } | |
3177 | ||
c9e9f97b ID |
3178 | /* |
3179 | * Should be called with both balance and volume mutexes held to | |
3180 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3181 | * restriper. Same goes for unset_balance_control. | |
3182 | */ | |
3183 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3184 | { | |
3185 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3186 | ||
3187 | BUG_ON(fs_info->balance_ctl); | |
3188 | ||
3189 | spin_lock(&fs_info->balance_lock); | |
3190 | fs_info->balance_ctl = bctl; | |
3191 | spin_unlock(&fs_info->balance_lock); | |
3192 | } | |
3193 | ||
3194 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3195 | { | |
3196 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3197 | ||
3198 | BUG_ON(!fs_info->balance_ctl); | |
3199 | ||
3200 | spin_lock(&fs_info->balance_lock); | |
3201 | fs_info->balance_ctl = NULL; | |
3202 | spin_unlock(&fs_info->balance_lock); | |
3203 | ||
3204 | kfree(bctl); | |
3205 | } | |
3206 | ||
ed25e9b2 ID |
3207 | /* |
3208 | * Balance filters. Return 1 if chunk should be filtered out | |
3209 | * (should not be balanced). | |
3210 | */ | |
899c81ea | 3211 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3212 | struct btrfs_balance_args *bargs) |
3213 | { | |
899c81ea ID |
3214 | chunk_type = chunk_to_extended(chunk_type) & |
3215 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3216 | |
899c81ea | 3217 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3218 | return 0; |
3219 | ||
3220 | return 1; | |
3221 | } | |
3222 | ||
dba72cb3 | 3223 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3224 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3225 | { |
3226 | struct btrfs_block_group_cache *cache; | |
3227 | u64 chunk_used; | |
3228 | u64 user_thresh_min; | |
3229 | u64 user_thresh_max; | |
3230 | int ret = 1; | |
3231 | ||
3232 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3233 | chunk_used = btrfs_block_group_used(&cache->item); | |
3234 | ||
3235 | if (bargs->usage_min == 0) | |
3236 | user_thresh_min = 0; | |
3237 | else | |
3238 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3239 | bargs->usage_min); | |
3240 | ||
3241 | if (bargs->usage_max == 0) | |
3242 | user_thresh_max = 1; | |
3243 | else if (bargs->usage_max > 100) | |
3244 | user_thresh_max = cache->key.offset; | |
3245 | else | |
3246 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3247 | bargs->usage_max); | |
3248 | ||
3249 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3250 | ret = 0; | |
3251 | ||
3252 | btrfs_put_block_group(cache); | |
3253 | return ret; | |
3254 | } | |
3255 | ||
dba72cb3 | 3256 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3257 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3258 | { |
3259 | struct btrfs_block_group_cache *cache; | |
3260 | u64 chunk_used, user_thresh; | |
3261 | int ret = 1; | |
3262 | ||
3263 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3264 | chunk_used = btrfs_block_group_used(&cache->item); | |
3265 | ||
bc309467 | 3266 | if (bargs->usage_min == 0) |
3e39cea6 | 3267 | user_thresh = 1; |
a105bb88 ID |
3268 | else if (bargs->usage > 100) |
3269 | user_thresh = cache->key.offset; | |
3270 | else | |
3271 | user_thresh = div_factor_fine(cache->key.offset, | |
3272 | bargs->usage); | |
3273 | ||
5ce5b3c0 ID |
3274 | if (chunk_used < user_thresh) |
3275 | ret = 0; | |
3276 | ||
3277 | btrfs_put_block_group(cache); | |
3278 | return ret; | |
3279 | } | |
3280 | ||
409d404b ID |
3281 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3282 | struct btrfs_chunk *chunk, | |
3283 | struct btrfs_balance_args *bargs) | |
3284 | { | |
3285 | struct btrfs_stripe *stripe; | |
3286 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3287 | int i; | |
3288 | ||
3289 | for (i = 0; i < num_stripes; i++) { | |
3290 | stripe = btrfs_stripe_nr(chunk, i); | |
3291 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3292 | return 0; | |
3293 | } | |
3294 | ||
3295 | return 1; | |
3296 | } | |
3297 | ||
94e60d5a ID |
3298 | /* [pstart, pend) */ |
3299 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3300 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3301 | struct btrfs_balance_args *bargs) |
3302 | { | |
3303 | struct btrfs_stripe *stripe; | |
3304 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3305 | u64 stripe_offset; | |
3306 | u64 stripe_length; | |
3307 | int factor; | |
3308 | int i; | |
3309 | ||
3310 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3311 | return 0; | |
3312 | ||
3313 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3314 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3315 | factor = num_stripes / 2; | |
3316 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3317 | factor = num_stripes - 1; | |
3318 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3319 | factor = num_stripes - 2; | |
3320 | } else { | |
3321 | factor = num_stripes; | |
3322 | } | |
94e60d5a ID |
3323 | |
3324 | for (i = 0; i < num_stripes; i++) { | |
3325 | stripe = btrfs_stripe_nr(chunk, i); | |
3326 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3327 | continue; | |
3328 | ||
3329 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3330 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3331 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3332 | |
3333 | if (stripe_offset < bargs->pend && | |
3334 | stripe_offset + stripe_length > bargs->pstart) | |
3335 | return 0; | |
3336 | } | |
3337 | ||
3338 | return 1; | |
3339 | } | |
3340 | ||
ea67176a ID |
3341 | /* [vstart, vend) */ |
3342 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3343 | struct btrfs_chunk *chunk, | |
3344 | u64 chunk_offset, | |
3345 | struct btrfs_balance_args *bargs) | |
3346 | { | |
3347 | if (chunk_offset < bargs->vend && | |
3348 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3349 | /* at least part of the chunk is inside this vrange */ | |
3350 | return 0; | |
3351 | ||
3352 | return 1; | |
3353 | } | |
3354 | ||
dee32d0a GAP |
3355 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3356 | struct btrfs_chunk *chunk, | |
3357 | struct btrfs_balance_args *bargs) | |
3358 | { | |
3359 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3360 | ||
3361 | if (bargs->stripes_min <= num_stripes | |
3362 | && num_stripes <= bargs->stripes_max) | |
3363 | return 0; | |
3364 | ||
3365 | return 1; | |
3366 | } | |
3367 | ||
899c81ea | 3368 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3369 | struct btrfs_balance_args *bargs) |
3370 | { | |
3371 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3372 | return 0; | |
3373 | ||
899c81ea ID |
3374 | chunk_type = chunk_to_extended(chunk_type) & |
3375 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3376 | |
899c81ea | 3377 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3378 | return 1; |
3379 | ||
3380 | return 0; | |
3381 | } | |
3382 | ||
2ff7e61e | 3383 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3384 | struct extent_buffer *leaf, |
3385 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3386 | { | |
0b246afa | 3387 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3388 | struct btrfs_balance_args *bargs = NULL; |
3389 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3390 | ||
3391 | /* type filter */ | |
3392 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3393 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3394 | return 0; | |
3395 | } | |
3396 | ||
3397 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3398 | bargs = &bctl->data; | |
3399 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3400 | bargs = &bctl->sys; | |
3401 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3402 | bargs = &bctl->meta; | |
3403 | ||
ed25e9b2 ID |
3404 | /* profiles filter */ |
3405 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3406 | chunk_profiles_filter(chunk_type, bargs)) { | |
3407 | return 0; | |
5ce5b3c0 ID |
3408 | } |
3409 | ||
3410 | /* usage filter */ | |
3411 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3412 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3413 | return 0; |
bc309467 | 3414 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3415 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3416 | return 0; |
409d404b ID |
3417 | } |
3418 | ||
3419 | /* devid filter */ | |
3420 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3421 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3422 | return 0; | |
94e60d5a ID |
3423 | } |
3424 | ||
3425 | /* drange filter, makes sense only with devid filter */ | |
3426 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3427 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3428 | return 0; |
ea67176a ID |
3429 | } |
3430 | ||
3431 | /* vrange filter */ | |
3432 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3433 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3434 | return 0; | |
ed25e9b2 ID |
3435 | } |
3436 | ||
dee32d0a GAP |
3437 | /* stripes filter */ |
3438 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3439 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3440 | return 0; | |
3441 | } | |
3442 | ||
cfa4c961 ID |
3443 | /* soft profile changing mode */ |
3444 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3445 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3446 | return 0; | |
3447 | } | |
3448 | ||
7d824b6f DS |
3449 | /* |
3450 | * limited by count, must be the last filter | |
3451 | */ | |
3452 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3453 | if (bargs->limit == 0) | |
3454 | return 0; | |
3455 | else | |
3456 | bargs->limit--; | |
12907fc7 DS |
3457 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3458 | /* | |
3459 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3460 | * determined here because we do not have the global information |
12907fc7 DS |
3461 | * about the count of all chunks that satisfy the filters. |
3462 | */ | |
3463 | if (bargs->limit_max == 0) | |
3464 | return 0; | |
3465 | else | |
3466 | bargs->limit_max--; | |
7d824b6f DS |
3467 | } |
3468 | ||
f43ffb60 ID |
3469 | return 1; |
3470 | } | |
3471 | ||
c9e9f97b | 3472 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3473 | { |
19a39dce | 3474 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3475 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3476 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3477 | struct list_head *devices; | |
ec44a35c CM |
3478 | struct btrfs_device *device; |
3479 | u64 old_size; | |
3480 | u64 size_to_free; | |
12907fc7 | 3481 | u64 chunk_type; |
f43ffb60 | 3482 | struct btrfs_chunk *chunk; |
5a488b9d | 3483 | struct btrfs_path *path = NULL; |
ec44a35c | 3484 | struct btrfs_key key; |
ec44a35c | 3485 | struct btrfs_key found_key; |
c9e9f97b | 3486 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3487 | struct extent_buffer *leaf; |
3488 | int slot; | |
c9e9f97b ID |
3489 | int ret; |
3490 | int enospc_errors = 0; | |
19a39dce | 3491 | bool counting = true; |
12907fc7 | 3492 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3493 | u64 limit_data = bctl->data.limit; |
3494 | u64 limit_meta = bctl->meta.limit; | |
3495 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3496 | u32 count_data = 0; |
3497 | u32 count_meta = 0; | |
3498 | u32 count_sys = 0; | |
2c9fe835 | 3499 | int chunk_reserved = 0; |
cf25ce51 | 3500 | u64 bytes_used = 0; |
ec44a35c | 3501 | |
ec44a35c | 3502 | /* step one make some room on all the devices */ |
c9e9f97b | 3503 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3504 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3505 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3506 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3507 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
2b82032c | 3508 | if (!device->writeable || |
7cc8e58d MX |
3509 | btrfs_device_get_total_bytes(device) - |
3510 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3511 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3512 | continue; |
3513 | ||
3514 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3515 | if (ret == -ENOSPC) |
3516 | break; | |
5a488b9d LB |
3517 | if (ret) { |
3518 | /* btrfs_shrink_device never returns ret > 0 */ | |
3519 | WARN_ON(ret > 0); | |
3520 | goto error; | |
3521 | } | |
ec44a35c | 3522 | |
a22285a6 | 3523 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3524 | if (IS_ERR(trans)) { |
3525 | ret = PTR_ERR(trans); | |
3526 | btrfs_info_in_rcu(fs_info, | |
3527 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3528 | rcu_str_deref(device->name), ret, | |
3529 | old_size, old_size - size_to_free); | |
3530 | goto error; | |
3531 | } | |
ec44a35c CM |
3532 | |
3533 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3534 | if (ret) { |
3a45bb20 | 3535 | btrfs_end_transaction(trans); |
5a488b9d LB |
3536 | /* btrfs_grow_device never returns ret > 0 */ |
3537 | WARN_ON(ret > 0); | |
3538 | btrfs_info_in_rcu(fs_info, | |
3539 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3540 | rcu_str_deref(device->name), ret, | |
3541 | old_size, old_size - size_to_free); | |
3542 | goto error; | |
3543 | } | |
ec44a35c | 3544 | |
3a45bb20 | 3545 | btrfs_end_transaction(trans); |
ec44a35c CM |
3546 | } |
3547 | ||
3548 | /* step two, relocate all the chunks */ | |
3549 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3550 | if (!path) { |
3551 | ret = -ENOMEM; | |
3552 | goto error; | |
3553 | } | |
19a39dce ID |
3554 | |
3555 | /* zero out stat counters */ | |
3556 | spin_lock(&fs_info->balance_lock); | |
3557 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3558 | spin_unlock(&fs_info->balance_lock); | |
3559 | again: | |
7d824b6f | 3560 | if (!counting) { |
12907fc7 DS |
3561 | /* |
3562 | * The single value limit and min/max limits use the same bytes | |
3563 | * in the | |
3564 | */ | |
7d824b6f DS |
3565 | bctl->data.limit = limit_data; |
3566 | bctl->meta.limit = limit_meta; | |
3567 | bctl->sys.limit = limit_sys; | |
3568 | } | |
ec44a35c CM |
3569 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3570 | key.offset = (u64)-1; | |
3571 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3572 | ||
d397712b | 3573 | while (1) { |
19a39dce | 3574 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3575 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3576 | ret = -ECANCELED; |
3577 | goto error; | |
3578 | } | |
3579 | ||
67c5e7d4 | 3580 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3581 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3582 | if (ret < 0) { |
3583 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3584 | goto error; |
67c5e7d4 | 3585 | } |
ec44a35c CM |
3586 | |
3587 | /* | |
3588 | * this shouldn't happen, it means the last relocate | |
3589 | * failed | |
3590 | */ | |
3591 | if (ret == 0) | |
c9e9f97b | 3592 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3593 | |
3594 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3595 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3596 | if (ret) { |
67c5e7d4 | 3597 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3598 | ret = 0; |
ec44a35c | 3599 | break; |
c9e9f97b | 3600 | } |
7d9eb12c | 3601 | |
f43ffb60 ID |
3602 | leaf = path->nodes[0]; |
3603 | slot = path->slots[0]; | |
3604 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3605 | |
67c5e7d4 FM |
3606 | if (found_key.objectid != key.objectid) { |
3607 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3608 | break; |
67c5e7d4 | 3609 | } |
7d9eb12c | 3610 | |
f43ffb60 | 3611 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3612 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3613 | |
19a39dce ID |
3614 | if (!counting) { |
3615 | spin_lock(&fs_info->balance_lock); | |
3616 | bctl->stat.considered++; | |
3617 | spin_unlock(&fs_info->balance_lock); | |
3618 | } | |
3619 | ||
2ff7e61e | 3620 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3621 | found_key.offset); |
2c9fe835 | 3622 | |
b3b4aa74 | 3623 | btrfs_release_path(path); |
67c5e7d4 FM |
3624 | if (!ret) { |
3625 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3626 | goto loop; |
67c5e7d4 | 3627 | } |
f43ffb60 | 3628 | |
19a39dce | 3629 | if (counting) { |
67c5e7d4 | 3630 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3631 | spin_lock(&fs_info->balance_lock); |
3632 | bctl->stat.expected++; | |
3633 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3634 | |
3635 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3636 | count_data++; | |
3637 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3638 | count_sys++; | |
3639 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3640 | count_meta++; | |
3641 | ||
3642 | goto loop; | |
3643 | } | |
3644 | ||
3645 | /* | |
3646 | * Apply limit_min filter, no need to check if the LIMITS | |
3647 | * filter is used, limit_min is 0 by default | |
3648 | */ | |
3649 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3650 | count_data < bctl->data.limit_min) | |
3651 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3652 | count_meta < bctl->meta.limit_min) | |
3653 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3654 | count_sys < bctl->sys.limit_min)) { | |
3655 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3656 | goto loop; |
3657 | } | |
3658 | ||
cf25ce51 LB |
3659 | ASSERT(fs_info->data_sinfo); |
3660 | spin_lock(&fs_info->data_sinfo->lock); | |
3661 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3662 | spin_unlock(&fs_info->data_sinfo->lock); | |
3663 | ||
3664 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3665 | !chunk_reserved && !bytes_used) { | |
2c9fe835 ZL |
3666 | trans = btrfs_start_transaction(chunk_root, 0); |
3667 | if (IS_ERR(trans)) { | |
3668 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3669 | ret = PTR_ERR(trans); | |
3670 | goto error; | |
3671 | } | |
3672 | ||
2ff7e61e | 3673 | ret = btrfs_force_chunk_alloc(trans, fs_info, |
2c9fe835 | 3674 | BTRFS_BLOCK_GROUP_DATA); |
3a45bb20 | 3675 | btrfs_end_transaction(trans); |
2c9fe835 ZL |
3676 | if (ret < 0) { |
3677 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3678 | goto error; | |
3679 | } | |
2c9fe835 ZL |
3680 | chunk_reserved = 1; |
3681 | } | |
3682 | ||
5b4aacef | 3683 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3684 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3685 | if (ret && ret != -ENOSPC) |
3686 | goto error; | |
19a39dce | 3687 | if (ret == -ENOSPC) { |
c9e9f97b | 3688 | enospc_errors++; |
19a39dce ID |
3689 | } else { |
3690 | spin_lock(&fs_info->balance_lock); | |
3691 | bctl->stat.completed++; | |
3692 | spin_unlock(&fs_info->balance_lock); | |
3693 | } | |
f43ffb60 | 3694 | loop: |
795a3321 ID |
3695 | if (found_key.offset == 0) |
3696 | break; | |
ba1bf481 | 3697 | key.offset = found_key.offset - 1; |
ec44a35c | 3698 | } |
c9e9f97b | 3699 | |
19a39dce ID |
3700 | if (counting) { |
3701 | btrfs_release_path(path); | |
3702 | counting = false; | |
3703 | goto again; | |
3704 | } | |
ec44a35c CM |
3705 | error: |
3706 | btrfs_free_path(path); | |
c9e9f97b | 3707 | if (enospc_errors) { |
efe120a0 | 3708 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3709 | enospc_errors); |
c9e9f97b ID |
3710 | if (!ret) |
3711 | ret = -ENOSPC; | |
3712 | } | |
3713 | ||
ec44a35c CM |
3714 | return ret; |
3715 | } | |
3716 | ||
0c460c0d ID |
3717 | /** |
3718 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3719 | * @flags: profile to validate | |
3720 | * @extended: if true @flags is treated as an extended profile | |
3721 | */ | |
3722 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3723 | { | |
3724 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3725 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3726 | ||
3727 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3728 | ||
3729 | /* 1) check that all other bits are zeroed */ | |
3730 | if (flags & ~mask) | |
3731 | return 0; | |
3732 | ||
3733 | /* 2) see if profile is reduced */ | |
3734 | if (flags == 0) | |
3735 | return !extended; /* "0" is valid for usual profiles */ | |
3736 | ||
3737 | /* true if exactly one bit set */ | |
3738 | return (flags & (flags - 1)) == 0; | |
3739 | } | |
3740 | ||
837d5b6e ID |
3741 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3742 | { | |
a7e99c69 ID |
3743 | /* cancel requested || normal exit path */ |
3744 | return atomic_read(&fs_info->balance_cancel_req) || | |
3745 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3746 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3747 | } |
3748 | ||
c9e9f97b ID |
3749 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3750 | { | |
0940ebf6 ID |
3751 | int ret; |
3752 | ||
c9e9f97b | 3753 | unset_balance_control(fs_info); |
6bccf3ab | 3754 | ret = del_balance_item(fs_info); |
0f788c58 | 3755 | if (ret) |
34d97007 | 3756 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f | 3757 | |
171938e5 | 3758 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
c9e9f97b ID |
3759 | } |
3760 | ||
bdcd3c97 AM |
3761 | /* Non-zero return value signifies invalidity */ |
3762 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3763 | u64 allowed) | |
3764 | { | |
3765 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3766 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3767 | (bctl_arg->target & ~allowed))); | |
3768 | } | |
3769 | ||
c9e9f97b ID |
3770 | /* |
3771 | * Should be called with both balance and volume mutexes held | |
3772 | */ | |
3773 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3774 | struct btrfs_ioctl_balance_args *bargs) | |
3775 | { | |
3776 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
14506127 | 3777 | u64 meta_target, data_target; |
f43ffb60 | 3778 | u64 allowed; |
e4837f8f | 3779 | int mixed = 0; |
c9e9f97b | 3780 | int ret; |
8dabb742 | 3781 | u64 num_devices; |
de98ced9 | 3782 | unsigned seq; |
c9e9f97b | 3783 | |
837d5b6e | 3784 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3785 | atomic_read(&fs_info->balance_pause_req) || |
3786 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3787 | ret = -EINVAL; |
3788 | goto out; | |
3789 | } | |
3790 | ||
e4837f8f ID |
3791 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3792 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3793 | mixed = 1; | |
3794 | ||
f43ffb60 ID |
3795 | /* |
3796 | * In case of mixed groups both data and meta should be picked, | |
3797 | * and identical options should be given for both of them. | |
3798 | */ | |
e4837f8f ID |
3799 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3800 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3801 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3802 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3803 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3804 | btrfs_err(fs_info, |
3805 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3806 | ret = -EINVAL; |
3807 | goto out; | |
3808 | } | |
3809 | } | |
3810 | ||
8dabb742 | 3811 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3812 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3813 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3814 | BUG_ON(num_devices < 1); | |
3815 | num_devices--; | |
3816 | } | |
73beece9 | 3817 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
88be159c AH |
3818 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3819 | if (num_devices > 1) | |
e4d8ec0f | 3820 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3821 | if (num_devices > 2) |
3822 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3823 | if (num_devices > 3) | |
3824 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3825 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3826 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3827 | btrfs_err(fs_info, |
3828 | "unable to start balance with target data profile %llu", | |
3829 | bctl->data.target); | |
e4d8ec0f ID |
3830 | ret = -EINVAL; |
3831 | goto out; | |
3832 | } | |
bdcd3c97 | 3833 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3834 | btrfs_err(fs_info, |
5d163e0e JM |
3835 | "unable to start balance with target metadata profile %llu", |
3836 | bctl->meta.target); | |
e4d8ec0f ID |
3837 | ret = -EINVAL; |
3838 | goto out; | |
3839 | } | |
bdcd3c97 | 3840 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3841 | btrfs_err(fs_info, |
5d163e0e JM |
3842 | "unable to start balance with target system profile %llu", |
3843 | bctl->sys.target); | |
e4d8ec0f ID |
3844 | ret = -EINVAL; |
3845 | goto out; | |
3846 | } | |
3847 | ||
e4d8ec0f ID |
3848 | /* allow to reduce meta or sys integrity only if force set */ |
3849 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3850 | BTRFS_BLOCK_GROUP_RAID10 | |
3851 | BTRFS_BLOCK_GROUP_RAID5 | | |
3852 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3853 | do { |
3854 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3855 | ||
3856 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3857 | (fs_info->avail_system_alloc_bits & allowed) && | |
3858 | !(bctl->sys.target & allowed)) || | |
3859 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3860 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3861 | !(bctl->meta.target & allowed))) { | |
3862 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3863 | btrfs_info(fs_info, |
3864 | "force reducing metadata integrity"); | |
de98ced9 | 3865 | } else { |
5d163e0e JM |
3866 | btrfs_err(fs_info, |
3867 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3868 | ret = -EINVAL; |
3869 | goto out; | |
3870 | } | |
e4d8ec0f | 3871 | } |
de98ced9 | 3872 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3873 | |
14506127 AB |
3874 | /* if we're not converting, the target field is uninitialized */ |
3875 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3876 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3877 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3878 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3879 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3880 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 3881 | btrfs_warn(fs_info, |
5d163e0e | 3882 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
14506127 | 3883 | meta_target, data_target); |
ee592d07 ST |
3884 | } |
3885 | ||
6bccf3ab | 3886 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3887 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3888 | goto out; |
3889 | ||
59641015 ID |
3890 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3891 | BUG_ON(ret == -EEXIST); | |
3892 | set_balance_control(bctl); | |
3893 | } else { | |
3894 | BUG_ON(ret != -EEXIST); | |
3895 | spin_lock(&fs_info->balance_lock); | |
3896 | update_balance_args(bctl); | |
3897 | spin_unlock(&fs_info->balance_lock); | |
3898 | } | |
c9e9f97b | 3899 | |
837d5b6e | 3900 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3901 | mutex_unlock(&fs_info->balance_mutex); |
3902 | ||
3903 | ret = __btrfs_balance(fs_info); | |
3904 | ||
3905 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3906 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
3907 | |
3908 | if (bargs) { | |
3909 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3910 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3911 | } |
3912 | ||
3a01aa7a ID |
3913 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3914 | balance_need_close(fs_info)) { | |
3915 | __cancel_balance(fs_info); | |
3916 | } | |
3917 | ||
837d5b6e | 3918 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3919 | |
3920 | return ret; | |
3921 | out: | |
59641015 ID |
3922 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3923 | __cancel_balance(fs_info); | |
ed0fb78f | 3924 | else { |
59641015 | 3925 | kfree(bctl); |
171938e5 | 3926 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
ed0fb78f | 3927 | } |
59641015 ID |
3928 | return ret; |
3929 | } | |
3930 | ||
3931 | static int balance_kthread(void *data) | |
3932 | { | |
2b6ba629 | 3933 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3934 | int ret = 0; |
59641015 ID |
3935 | |
3936 | mutex_lock(&fs_info->volume_mutex); | |
3937 | mutex_lock(&fs_info->balance_mutex); | |
3938 | ||
2b6ba629 | 3939 | if (fs_info->balance_ctl) { |
efe120a0 | 3940 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3941 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3942 | } |
59641015 ID |
3943 | |
3944 | mutex_unlock(&fs_info->balance_mutex); | |
3945 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3946 | |
59641015 ID |
3947 | return ret; |
3948 | } | |
3949 | ||
2b6ba629 ID |
3950 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3951 | { | |
3952 | struct task_struct *tsk; | |
3953 | ||
3954 | spin_lock(&fs_info->balance_lock); | |
3955 | if (!fs_info->balance_ctl) { | |
3956 | spin_unlock(&fs_info->balance_lock); | |
3957 | return 0; | |
3958 | } | |
3959 | spin_unlock(&fs_info->balance_lock); | |
3960 | ||
3cdde224 | 3961 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 3962 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3963 | return 0; |
3964 | } | |
3965 | ||
3966 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3967 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3968 | } |
3969 | ||
68310a5e | 3970 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3971 | { |
59641015 ID |
3972 | struct btrfs_balance_control *bctl; |
3973 | struct btrfs_balance_item *item; | |
3974 | struct btrfs_disk_balance_args disk_bargs; | |
3975 | struct btrfs_path *path; | |
3976 | struct extent_buffer *leaf; | |
3977 | struct btrfs_key key; | |
3978 | int ret; | |
3979 | ||
3980 | path = btrfs_alloc_path(); | |
3981 | if (!path) | |
3982 | return -ENOMEM; | |
3983 | ||
59641015 | 3984 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3985 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
3986 | key.offset = 0; |
3987 | ||
68310a5e | 3988 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3989 | if (ret < 0) |
68310a5e | 3990 | goto out; |
59641015 ID |
3991 | if (ret > 0) { /* ret = -ENOENT; */ |
3992 | ret = 0; | |
68310a5e ID |
3993 | goto out; |
3994 | } | |
3995 | ||
3996 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3997 | if (!bctl) { | |
3998 | ret = -ENOMEM; | |
3999 | goto out; | |
59641015 ID |
4000 | } |
4001 | ||
4002 | leaf = path->nodes[0]; | |
4003 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4004 | ||
68310a5e ID |
4005 | bctl->fs_info = fs_info; |
4006 | bctl->flags = btrfs_balance_flags(leaf, item); | |
4007 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4008 | |
4009 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4010 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4011 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4012 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4013 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4014 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4015 | ||
171938e5 | 4016 | WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)); |
ed0fb78f | 4017 | |
68310a5e ID |
4018 | mutex_lock(&fs_info->volume_mutex); |
4019 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 4020 | |
68310a5e ID |
4021 | set_balance_control(bctl); |
4022 | ||
4023 | mutex_unlock(&fs_info->balance_mutex); | |
4024 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4025 | out: |
4026 | btrfs_free_path(path); | |
ec44a35c CM |
4027 | return ret; |
4028 | } | |
4029 | ||
837d5b6e ID |
4030 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4031 | { | |
4032 | int ret = 0; | |
4033 | ||
4034 | mutex_lock(&fs_info->balance_mutex); | |
4035 | if (!fs_info->balance_ctl) { | |
4036 | mutex_unlock(&fs_info->balance_mutex); | |
4037 | return -ENOTCONN; | |
4038 | } | |
4039 | ||
4040 | if (atomic_read(&fs_info->balance_running)) { | |
4041 | atomic_inc(&fs_info->balance_pause_req); | |
4042 | mutex_unlock(&fs_info->balance_mutex); | |
4043 | ||
4044 | wait_event(fs_info->balance_wait_q, | |
4045 | atomic_read(&fs_info->balance_running) == 0); | |
4046 | ||
4047 | mutex_lock(&fs_info->balance_mutex); | |
4048 | /* we are good with balance_ctl ripped off from under us */ | |
4049 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4050 | atomic_dec(&fs_info->balance_pause_req); | |
4051 | } else { | |
4052 | ret = -ENOTCONN; | |
4053 | } | |
4054 | ||
4055 | mutex_unlock(&fs_info->balance_mutex); | |
4056 | return ret; | |
4057 | } | |
4058 | ||
a7e99c69 ID |
4059 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4060 | { | |
e649e587 ID |
4061 | if (fs_info->sb->s_flags & MS_RDONLY) |
4062 | return -EROFS; | |
4063 | ||
a7e99c69 ID |
4064 | mutex_lock(&fs_info->balance_mutex); |
4065 | if (!fs_info->balance_ctl) { | |
4066 | mutex_unlock(&fs_info->balance_mutex); | |
4067 | return -ENOTCONN; | |
4068 | } | |
4069 | ||
4070 | atomic_inc(&fs_info->balance_cancel_req); | |
4071 | /* | |
4072 | * if we are running just wait and return, balance item is | |
4073 | * deleted in btrfs_balance in this case | |
4074 | */ | |
4075 | if (atomic_read(&fs_info->balance_running)) { | |
4076 | mutex_unlock(&fs_info->balance_mutex); | |
4077 | wait_event(fs_info->balance_wait_q, | |
4078 | atomic_read(&fs_info->balance_running) == 0); | |
4079 | mutex_lock(&fs_info->balance_mutex); | |
4080 | } else { | |
4081 | /* __cancel_balance needs volume_mutex */ | |
4082 | mutex_unlock(&fs_info->balance_mutex); | |
4083 | mutex_lock(&fs_info->volume_mutex); | |
4084 | mutex_lock(&fs_info->balance_mutex); | |
4085 | ||
4086 | if (fs_info->balance_ctl) | |
4087 | __cancel_balance(fs_info); | |
4088 | ||
4089 | mutex_unlock(&fs_info->volume_mutex); | |
4090 | } | |
4091 | ||
4092 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4093 | atomic_dec(&fs_info->balance_cancel_req); | |
4094 | mutex_unlock(&fs_info->balance_mutex); | |
4095 | return 0; | |
4096 | } | |
4097 | ||
803b2f54 SB |
4098 | static int btrfs_uuid_scan_kthread(void *data) |
4099 | { | |
4100 | struct btrfs_fs_info *fs_info = data; | |
4101 | struct btrfs_root *root = fs_info->tree_root; | |
4102 | struct btrfs_key key; | |
803b2f54 SB |
4103 | struct btrfs_path *path = NULL; |
4104 | int ret = 0; | |
4105 | struct extent_buffer *eb; | |
4106 | int slot; | |
4107 | struct btrfs_root_item root_item; | |
4108 | u32 item_size; | |
f45388f3 | 4109 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4110 | |
4111 | path = btrfs_alloc_path(); | |
4112 | if (!path) { | |
4113 | ret = -ENOMEM; | |
4114 | goto out; | |
4115 | } | |
4116 | ||
4117 | key.objectid = 0; | |
4118 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4119 | key.offset = 0; | |
4120 | ||
803b2f54 | 4121 | while (1) { |
6174d3cb | 4122 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4123 | if (ret) { |
4124 | if (ret > 0) | |
4125 | ret = 0; | |
4126 | break; | |
4127 | } | |
4128 | ||
4129 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4130 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4131 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4132 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4133 | goto skip; | |
4134 | ||
4135 | eb = path->nodes[0]; | |
4136 | slot = path->slots[0]; | |
4137 | item_size = btrfs_item_size_nr(eb, slot); | |
4138 | if (item_size < sizeof(root_item)) | |
4139 | goto skip; | |
4140 | ||
803b2f54 SB |
4141 | read_extent_buffer(eb, &root_item, |
4142 | btrfs_item_ptr_offset(eb, slot), | |
4143 | (int)sizeof(root_item)); | |
4144 | if (btrfs_root_refs(&root_item) == 0) | |
4145 | goto skip; | |
f45388f3 FDBM |
4146 | |
4147 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4148 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4149 | if (trans) | |
4150 | goto update_tree; | |
4151 | ||
4152 | btrfs_release_path(path); | |
803b2f54 SB |
4153 | /* |
4154 | * 1 - subvol uuid item | |
4155 | * 1 - received_subvol uuid item | |
4156 | */ | |
4157 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4158 | if (IS_ERR(trans)) { | |
4159 | ret = PTR_ERR(trans); | |
4160 | break; | |
4161 | } | |
f45388f3 FDBM |
4162 | continue; |
4163 | } else { | |
4164 | goto skip; | |
4165 | } | |
4166 | update_tree: | |
4167 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
6bccf3ab | 4168 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4169 | root_item.uuid, |
4170 | BTRFS_UUID_KEY_SUBVOL, | |
4171 | key.objectid); | |
4172 | if (ret < 0) { | |
efe120a0 | 4173 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4174 | ret); |
803b2f54 SB |
4175 | break; |
4176 | } | |
4177 | } | |
4178 | ||
4179 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
6bccf3ab | 4180 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4181 | root_item.received_uuid, |
4182 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4183 | key.objectid); | |
4184 | if (ret < 0) { | |
efe120a0 | 4185 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4186 | ret); |
803b2f54 SB |
4187 | break; |
4188 | } | |
4189 | } | |
4190 | ||
f45388f3 | 4191 | skip: |
803b2f54 | 4192 | if (trans) { |
3a45bb20 | 4193 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4194 | trans = NULL; |
803b2f54 SB |
4195 | if (ret) |
4196 | break; | |
4197 | } | |
4198 | ||
803b2f54 SB |
4199 | btrfs_release_path(path); |
4200 | if (key.offset < (u64)-1) { | |
4201 | key.offset++; | |
4202 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4203 | key.offset = 0; | |
4204 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4205 | } else if (key.objectid < (u64)-1) { | |
4206 | key.offset = 0; | |
4207 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4208 | key.objectid++; | |
4209 | } else { | |
4210 | break; | |
4211 | } | |
4212 | cond_resched(); | |
4213 | } | |
4214 | ||
4215 | out: | |
4216 | btrfs_free_path(path); | |
f45388f3 | 4217 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4218 | btrfs_end_transaction(trans); |
803b2f54 | 4219 | if (ret) |
efe120a0 | 4220 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4221 | else |
afcdd129 | 4222 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4223 | up(&fs_info->uuid_tree_rescan_sem); |
4224 | return 0; | |
4225 | } | |
4226 | ||
70f80175 SB |
4227 | /* |
4228 | * Callback for btrfs_uuid_tree_iterate(). | |
4229 | * returns: | |
4230 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4231 | * < 0 if an error occurred. |
70f80175 SB |
4232 | * > 0 if the check failed, which means the caller shall remove the entry. |
4233 | */ | |
4234 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4235 | u8 *uuid, u8 type, u64 subid) | |
4236 | { | |
4237 | struct btrfs_key key; | |
4238 | int ret = 0; | |
4239 | struct btrfs_root *subvol_root; | |
4240 | ||
4241 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4242 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4243 | goto out; | |
4244 | ||
4245 | key.objectid = subid; | |
4246 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4247 | key.offset = (u64)-1; | |
4248 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4249 | if (IS_ERR(subvol_root)) { | |
4250 | ret = PTR_ERR(subvol_root); | |
4251 | if (ret == -ENOENT) | |
4252 | ret = 1; | |
4253 | goto out; | |
4254 | } | |
4255 | ||
4256 | switch (type) { | |
4257 | case BTRFS_UUID_KEY_SUBVOL: | |
4258 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4259 | ret = 1; | |
4260 | break; | |
4261 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4262 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4263 | BTRFS_UUID_SIZE)) | |
4264 | ret = 1; | |
4265 | break; | |
4266 | } | |
4267 | ||
4268 | out: | |
4269 | return ret; | |
4270 | } | |
4271 | ||
4272 | static int btrfs_uuid_rescan_kthread(void *data) | |
4273 | { | |
4274 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4275 | int ret; | |
4276 | ||
4277 | /* | |
4278 | * 1st step is to iterate through the existing UUID tree and | |
4279 | * to delete all entries that contain outdated data. | |
4280 | * 2nd step is to add all missing entries to the UUID tree. | |
4281 | */ | |
4282 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4283 | if (ret < 0) { | |
efe120a0 | 4284 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4285 | up(&fs_info->uuid_tree_rescan_sem); |
4286 | return ret; | |
4287 | } | |
4288 | return btrfs_uuid_scan_kthread(data); | |
4289 | } | |
4290 | ||
f7a81ea4 SB |
4291 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4292 | { | |
4293 | struct btrfs_trans_handle *trans; | |
4294 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4295 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4296 | struct task_struct *task; |
4297 | int ret; | |
f7a81ea4 SB |
4298 | |
4299 | /* | |
4300 | * 1 - root node | |
4301 | * 1 - root item | |
4302 | */ | |
4303 | trans = btrfs_start_transaction(tree_root, 2); | |
4304 | if (IS_ERR(trans)) | |
4305 | return PTR_ERR(trans); | |
4306 | ||
4307 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4308 | BTRFS_UUID_TREE_OBJECTID); | |
4309 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4310 | ret = PTR_ERR(uuid_root); |
66642832 | 4311 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4312 | btrfs_end_transaction(trans); |
6d13f549 | 4313 | return ret; |
f7a81ea4 SB |
4314 | } |
4315 | ||
4316 | fs_info->uuid_root = uuid_root; | |
4317 | ||
3a45bb20 | 4318 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4319 | if (ret) |
4320 | return ret; | |
4321 | ||
4322 | down(&fs_info->uuid_tree_rescan_sem); | |
4323 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4324 | if (IS_ERR(task)) { | |
70f80175 | 4325 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4326 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4327 | up(&fs_info->uuid_tree_rescan_sem); |
4328 | return PTR_ERR(task); | |
4329 | } | |
4330 | ||
4331 | return 0; | |
f7a81ea4 | 4332 | } |
803b2f54 | 4333 | |
70f80175 SB |
4334 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4335 | { | |
4336 | struct task_struct *task; | |
4337 | ||
4338 | down(&fs_info->uuid_tree_rescan_sem); | |
4339 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4340 | if (IS_ERR(task)) { | |
4341 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4342 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4343 | up(&fs_info->uuid_tree_rescan_sem); |
4344 | return PTR_ERR(task); | |
4345 | } | |
4346 | ||
4347 | return 0; | |
4348 | } | |
4349 | ||
8f18cf13 CM |
4350 | /* |
4351 | * shrinking a device means finding all of the device extents past | |
4352 | * the new size, and then following the back refs to the chunks. | |
4353 | * The chunk relocation code actually frees the device extent | |
4354 | */ | |
4355 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4356 | { | |
0b246afa JM |
4357 | struct btrfs_fs_info *fs_info = device->fs_info; |
4358 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4359 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4360 | struct btrfs_dev_extent *dev_extent = NULL; |
4361 | struct btrfs_path *path; | |
4362 | u64 length; | |
8f18cf13 CM |
4363 | u64 chunk_offset; |
4364 | int ret; | |
4365 | int slot; | |
ba1bf481 JB |
4366 | int failed = 0; |
4367 | bool retried = false; | |
53e489bc | 4368 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4369 | struct extent_buffer *l; |
4370 | struct btrfs_key key; | |
0b246afa | 4371 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4372 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4373 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4374 | u64 diff; |
4375 | ||
4376 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4377 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4378 | |
63a212ab SB |
4379 | if (device->is_tgtdev_for_dev_replace) |
4380 | return -EINVAL; | |
4381 | ||
8f18cf13 CM |
4382 | path = btrfs_alloc_path(); |
4383 | if (!path) | |
4384 | return -ENOMEM; | |
4385 | ||
e4058b54 | 4386 | path->reada = READA_FORWARD; |
8f18cf13 | 4387 | |
34441361 | 4388 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4389 | |
7cc8e58d | 4390 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4391 | if (device->writeable) { |
2b82032c | 4392 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4393 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4394 | } |
34441361 | 4395 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4396 | |
ba1bf481 | 4397 | again: |
8f18cf13 CM |
4398 | key.objectid = device->devid; |
4399 | key.offset = (u64)-1; | |
4400 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4401 | ||
213e64da | 4402 | do { |
0b246afa | 4403 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4404 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4405 | if (ret < 0) { |
0b246afa | 4406 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4407 | goto done; |
67c5e7d4 | 4408 | } |
8f18cf13 CM |
4409 | |
4410 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4411 | if (ret) |
0b246afa | 4412 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4413 | if (ret < 0) |
4414 | goto done; | |
4415 | if (ret) { | |
4416 | ret = 0; | |
b3b4aa74 | 4417 | btrfs_release_path(path); |
bf1fb512 | 4418 | break; |
8f18cf13 CM |
4419 | } |
4420 | ||
4421 | l = path->nodes[0]; | |
4422 | slot = path->slots[0]; | |
4423 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4424 | ||
ba1bf481 | 4425 | if (key.objectid != device->devid) { |
0b246afa | 4426 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4427 | btrfs_release_path(path); |
bf1fb512 | 4428 | break; |
ba1bf481 | 4429 | } |
8f18cf13 CM |
4430 | |
4431 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4432 | length = btrfs_dev_extent_length(l, dev_extent); | |
4433 | ||
ba1bf481 | 4434 | if (key.offset + length <= new_size) { |
0b246afa | 4435 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4436 | btrfs_release_path(path); |
d6397bae | 4437 | break; |
ba1bf481 | 4438 | } |
8f18cf13 | 4439 | |
8f18cf13 | 4440 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4441 | btrfs_release_path(path); |
8f18cf13 | 4442 | |
0b246afa JM |
4443 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4444 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4445 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4446 | goto done; |
ba1bf481 JB |
4447 | if (ret == -ENOSPC) |
4448 | failed++; | |
213e64da | 4449 | } while (key.offset-- > 0); |
ba1bf481 JB |
4450 | |
4451 | if (failed && !retried) { | |
4452 | failed = 0; | |
4453 | retried = true; | |
4454 | goto again; | |
4455 | } else if (failed && retried) { | |
4456 | ret = -ENOSPC; | |
ba1bf481 | 4457 | goto done; |
8f18cf13 CM |
4458 | } |
4459 | ||
d6397bae | 4460 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4461 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4462 | if (IS_ERR(trans)) { |
4463 | ret = PTR_ERR(trans); | |
4464 | goto done; | |
4465 | } | |
4466 | ||
34441361 | 4467 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4468 | |
4469 | /* | |
4470 | * We checked in the above loop all device extents that were already in | |
4471 | * the device tree. However before we have updated the device's | |
4472 | * total_bytes to the new size, we might have had chunk allocations that | |
4473 | * have not complete yet (new block groups attached to transaction | |
4474 | * handles), and therefore their device extents were not yet in the | |
4475 | * device tree and we missed them in the loop above. So if we have any | |
4476 | * pending chunk using a device extent that overlaps the device range | |
4477 | * that we can not use anymore, commit the current transaction and | |
4478 | * repeat the search on the device tree - this way we guarantee we will | |
4479 | * not have chunks using device extents that end beyond 'new_size'. | |
4480 | */ | |
4481 | if (!checked_pending_chunks) { | |
4482 | u64 start = new_size; | |
4483 | u64 len = old_size - new_size; | |
4484 | ||
499f377f JM |
4485 | if (contains_pending_extent(trans->transaction, device, |
4486 | &start, len)) { | |
34441361 | 4487 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4488 | checked_pending_chunks = true; |
4489 | failed = 0; | |
4490 | retried = false; | |
3a45bb20 | 4491 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4492 | if (ret) |
4493 | goto done; | |
4494 | goto again; | |
4495 | } | |
4496 | } | |
4497 | ||
7cc8e58d | 4498 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4499 | if (list_empty(&device->resized_list)) |
4500 | list_add_tail(&device->resized_list, | |
0b246afa | 4501 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4502 | |
d6397bae | 4503 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4504 | btrfs_set_super_total_bytes(super_copy, |
4505 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4506 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4507 | |
4508 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4509 | ret = btrfs_update_device(trans, device); | |
3a45bb20 | 4510 | btrfs_end_transaction(trans); |
8f18cf13 CM |
4511 | done: |
4512 | btrfs_free_path(path); | |
53e489bc | 4513 | if (ret) { |
34441361 | 4514 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4515 | btrfs_device_set_total_bytes(device, old_size); |
4516 | if (device->writeable) | |
4517 | device->fs_devices->total_rw_bytes += diff; | |
a5ed45f8 | 4518 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4519 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4520 | } |
8f18cf13 CM |
4521 | return ret; |
4522 | } | |
4523 | ||
2ff7e61e | 4524 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4525 | struct btrfs_key *key, |
4526 | struct btrfs_chunk *chunk, int item_size) | |
4527 | { | |
0b246afa | 4528 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4529 | struct btrfs_disk_key disk_key; |
4530 | u32 array_size; | |
4531 | u8 *ptr; | |
4532 | ||
34441361 | 4533 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4534 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4535 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4536 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4537 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4538 | return -EFBIG; |
fe48a5c0 | 4539 | } |
0b86a832 CM |
4540 | |
4541 | ptr = super_copy->sys_chunk_array + array_size; | |
4542 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4543 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4544 | ptr += sizeof(disk_key); | |
4545 | memcpy(ptr, chunk, item_size); | |
4546 | item_size += sizeof(disk_key); | |
4547 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4548 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4549 | |
0b86a832 CM |
4550 | return 0; |
4551 | } | |
4552 | ||
73c5de00 AJ |
4553 | /* |
4554 | * sort the devices in descending order by max_avail, total_avail | |
4555 | */ | |
4556 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4557 | { |
73c5de00 AJ |
4558 | const struct btrfs_device_info *di_a = a; |
4559 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4560 | |
73c5de00 | 4561 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4562 | return -1; |
73c5de00 | 4563 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4564 | return 1; |
73c5de00 AJ |
4565 | if (di_a->total_avail > di_b->total_avail) |
4566 | return -1; | |
4567 | if (di_a->total_avail < di_b->total_avail) | |
4568 | return 1; | |
4569 | return 0; | |
b2117a39 | 4570 | } |
0b86a832 | 4571 | |
53b381b3 DW |
4572 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4573 | { | |
ffe2d203 | 4574 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4575 | return; |
4576 | ||
ceda0864 | 4577 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4578 | } |
4579 | ||
da17066c | 4580 | #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \ |
23f8f9b7 GH |
4581 | - sizeof(struct btrfs_chunk)) \ |
4582 | / sizeof(struct btrfs_stripe) + 1) | |
4583 | ||
4584 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4585 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4586 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4587 | / sizeof(struct btrfs_stripe) + 1) | |
4588 | ||
73c5de00 | 4589 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4590 | u64 start, u64 type) |
b2117a39 | 4591 | { |
2ff7e61e | 4592 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4593 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4594 | struct btrfs_device *device; |
73c5de00 AJ |
4595 | struct map_lookup *map = NULL; |
4596 | struct extent_map_tree *em_tree; | |
4597 | struct extent_map *em; | |
4598 | struct btrfs_device_info *devices_info = NULL; | |
4599 | u64 total_avail; | |
4600 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4601 | int data_stripes; /* number of stripes that count for |
4602 | block group size */ | |
73c5de00 AJ |
4603 | int sub_stripes; /* sub_stripes info for map */ |
4604 | int dev_stripes; /* stripes per dev */ | |
4605 | int devs_max; /* max devs to use */ | |
4606 | int devs_min; /* min devs needed */ | |
4607 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4608 | int ncopies; /* how many copies to data has */ | |
4609 | int ret; | |
4610 | u64 max_stripe_size; | |
4611 | u64 max_chunk_size; | |
4612 | u64 stripe_size; | |
4613 | u64 num_bytes; | |
4614 | int ndevs; | |
4615 | int i; | |
4616 | int j; | |
31e50229 | 4617 | int index; |
593060d7 | 4618 | |
0c460c0d | 4619 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4620 | |
73c5de00 AJ |
4621 | if (list_empty(&fs_devices->alloc_list)) |
4622 | return -ENOSPC; | |
b2117a39 | 4623 | |
31e50229 | 4624 | index = __get_raid_index(type); |
73c5de00 | 4625 | |
31e50229 LB |
4626 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4627 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4628 | devs_max = btrfs_raid_array[index].devs_max; | |
4629 | devs_min = btrfs_raid_array[index].devs_min; | |
4630 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4631 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4632 | |
9b3f68b9 | 4633 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4634 | max_stripe_size = SZ_1G; |
73c5de00 | 4635 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 GH |
4636 | if (!devs_max) |
4637 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4638 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4639 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4640 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4641 | max_stripe_size = SZ_1G; | |
1100373f | 4642 | else |
ee22184b | 4643 | max_stripe_size = SZ_256M; |
73c5de00 | 4644 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4645 | if (!devs_max) |
4646 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4647 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4648 | max_stripe_size = SZ_32M; |
73c5de00 | 4649 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4650 | if (!devs_max) |
4651 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4652 | } else { |
351fd353 | 4653 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4654 | type); |
4655 | BUG_ON(1); | |
9b3f68b9 CM |
4656 | } |
4657 | ||
2b82032c YZ |
4658 | /* we don't want a chunk larger than 10% of writeable space */ |
4659 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4660 | max_chunk_size); | |
9b3f68b9 | 4661 | |
31e818fe | 4662 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4663 | GFP_NOFS); |
4664 | if (!devices_info) | |
4665 | return -ENOMEM; | |
0cad8a11 | 4666 | |
9f680ce0 | 4667 | /* |
73c5de00 AJ |
4668 | * in the first pass through the devices list, we gather information |
4669 | * about the available holes on each device. | |
9f680ce0 | 4670 | */ |
73c5de00 | 4671 | ndevs = 0; |
ebcc9301 | 4672 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4673 | u64 max_avail; |
4674 | u64 dev_offset; | |
b2117a39 | 4675 | |
73c5de00 | 4676 | if (!device->writeable) { |
31b1a2bd | 4677 | WARN(1, KERN_ERR |
efe120a0 | 4678 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4679 | continue; |
4680 | } | |
b2117a39 | 4681 | |
63a212ab SB |
4682 | if (!device->in_fs_metadata || |
4683 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4684 | continue; |
b2117a39 | 4685 | |
73c5de00 AJ |
4686 | if (device->total_bytes > device->bytes_used) |
4687 | total_avail = device->total_bytes - device->bytes_used; | |
4688 | else | |
4689 | total_avail = 0; | |
38c01b96 | 4690 | |
4691 | /* If there is no space on this device, skip it. */ | |
4692 | if (total_avail == 0) | |
4693 | continue; | |
b2117a39 | 4694 | |
6df9a95e | 4695 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4696 | max_stripe_size * dev_stripes, |
4697 | &dev_offset, &max_avail); | |
4698 | if (ret && ret != -ENOSPC) | |
4699 | goto error; | |
b2117a39 | 4700 | |
73c5de00 AJ |
4701 | if (ret == 0) |
4702 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4703 | |
73c5de00 AJ |
4704 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4705 | continue; | |
b2117a39 | 4706 | |
063d006f ES |
4707 | if (ndevs == fs_devices->rw_devices) { |
4708 | WARN(1, "%s: found more than %llu devices\n", | |
4709 | __func__, fs_devices->rw_devices); | |
4710 | break; | |
4711 | } | |
73c5de00 AJ |
4712 | devices_info[ndevs].dev_offset = dev_offset; |
4713 | devices_info[ndevs].max_avail = max_avail; | |
4714 | devices_info[ndevs].total_avail = total_avail; | |
4715 | devices_info[ndevs].dev = device; | |
4716 | ++ndevs; | |
4717 | } | |
b2117a39 | 4718 | |
73c5de00 AJ |
4719 | /* |
4720 | * now sort the devices by hole size / available space | |
4721 | */ | |
4722 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4723 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4724 | |
73c5de00 | 4725 | /* round down to number of usable stripes */ |
e5600fd6 | 4726 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4727 | |
73c5de00 AJ |
4728 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4729 | ret = -ENOSPC; | |
4730 | goto error; | |
b2117a39 | 4731 | } |
9f680ce0 | 4732 | |
f148ef4d NB |
4733 | ndevs = min(ndevs, devs_max); |
4734 | ||
73c5de00 AJ |
4735 | /* |
4736 | * the primary goal is to maximize the number of stripes, so use as many | |
4737 | * devices as possible, even if the stripes are not maximum sized. | |
4738 | */ | |
4739 | stripe_size = devices_info[ndevs-1].max_avail; | |
4740 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4741 | |
53b381b3 DW |
4742 | /* |
4743 | * this will have to be fixed for RAID1 and RAID10 over | |
4744 | * more drives | |
4745 | */ | |
4746 | data_stripes = num_stripes / ncopies; | |
4747 | ||
500ceed8 | 4748 | if (type & BTRFS_BLOCK_GROUP_RAID5) |
53b381b3 | 4749 | data_stripes = num_stripes - 1; |
500ceed8 NB |
4750 | |
4751 | if (type & BTRFS_BLOCK_GROUP_RAID6) | |
53b381b3 | 4752 | data_stripes = num_stripes - 2; |
86db2578 CM |
4753 | |
4754 | /* | |
4755 | * Use the number of data stripes to figure out how big this chunk | |
4756 | * is really going to be in terms of logical address space, | |
4757 | * and compare that answer with the max chunk size | |
4758 | */ | |
4759 | if (stripe_size * data_stripes > max_chunk_size) { | |
4760 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4761 | |
4762 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4763 | |
4764 | /* bump the answer up to a 16MB boundary */ | |
4765 | stripe_size = (stripe_size + mask) & ~mask; | |
4766 | ||
4767 | /* but don't go higher than the limits we found | |
4768 | * while searching for free extents | |
4769 | */ | |
4770 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4771 | stripe_size = devices_info[ndevs-1].max_avail; | |
4772 | } | |
4773 | ||
b8b93add | 4774 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4775 | |
4776 | /* align to BTRFS_STRIPE_LEN */ | |
500ceed8 | 4777 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4778 | |
4779 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4780 | if (!map) { | |
4781 | ret = -ENOMEM; | |
4782 | goto error; | |
4783 | } | |
4784 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4785 | |
73c5de00 AJ |
4786 | for (i = 0; i < ndevs; ++i) { |
4787 | for (j = 0; j < dev_stripes; ++j) { | |
4788 | int s = i * dev_stripes + j; | |
4789 | map->stripes[s].dev = devices_info[i].dev; | |
4790 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4791 | j * stripe_size; | |
6324fbf3 | 4792 | } |
6324fbf3 | 4793 | } |
da17066c | 4794 | map->sector_size = info->sectorsize; |
500ceed8 NB |
4795 | map->stripe_len = BTRFS_STRIPE_LEN; |
4796 | map->io_align = BTRFS_STRIPE_LEN; | |
4797 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4798 | map->type = type; |
2b82032c | 4799 | map->sub_stripes = sub_stripes; |
0b86a832 | 4800 | |
53b381b3 | 4801 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4802 | |
6bccf3ab | 4803 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4804 | |
172ddd60 | 4805 | em = alloc_extent_map(); |
2b82032c | 4806 | if (!em) { |
298a8f9c | 4807 | kfree(map); |
b2117a39 MX |
4808 | ret = -ENOMEM; |
4809 | goto error; | |
593060d7 | 4810 | } |
298a8f9c | 4811 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4812 | em->map_lookup = map; |
2b82032c | 4813 | em->start = start; |
73c5de00 | 4814 | em->len = num_bytes; |
2b82032c YZ |
4815 | em->block_start = 0; |
4816 | em->block_len = em->len; | |
6df9a95e | 4817 | em->orig_block_len = stripe_size; |
593060d7 | 4818 | |
0b246afa | 4819 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4820 | write_lock(&em_tree->lock); |
09a2a8f9 | 4821 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4822 | if (!ret) { |
4823 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
490b54d6 | 4824 | refcount_inc(&em->refs); |
6df9a95e | 4825 | } |
890871be | 4826 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4827 | if (ret) { |
4828 | free_extent_map(em); | |
1dd4602f | 4829 | goto error; |
0f5d42b2 | 4830 | } |
0b86a832 | 4831 | |
2ff7e61e | 4832 | ret = btrfs_make_block_group(trans, info, 0, type, |
04487488 JB |
4833 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, |
4834 | start, num_bytes); | |
6df9a95e JB |
4835 | if (ret) |
4836 | goto error_del_extent; | |
2b82032c | 4837 | |
7cc8e58d MX |
4838 | for (i = 0; i < map->num_stripes; i++) { |
4839 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4840 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4841 | } | |
43530c46 | 4842 | |
a5ed45f8 | 4843 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4844 | |
0f5d42b2 | 4845 | free_extent_map(em); |
0b246afa | 4846 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4847 | |
b2117a39 | 4848 | kfree(devices_info); |
2b82032c | 4849 | return 0; |
b2117a39 | 4850 | |
6df9a95e | 4851 | error_del_extent: |
0f5d42b2 JB |
4852 | write_lock(&em_tree->lock); |
4853 | remove_extent_mapping(em_tree, em); | |
4854 | write_unlock(&em_tree->lock); | |
4855 | ||
4856 | /* One for our allocation */ | |
4857 | free_extent_map(em); | |
4858 | /* One for the tree reference */ | |
4859 | free_extent_map(em); | |
495e64f4 FM |
4860 | /* One for the pending_chunks list reference */ |
4861 | free_extent_map(em); | |
b2117a39 | 4862 | error: |
b2117a39 MX |
4863 | kfree(devices_info); |
4864 | return ret; | |
2b82032c YZ |
4865 | } |
4866 | ||
6df9a95e | 4867 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
6bccf3ab | 4868 | struct btrfs_fs_info *fs_info, |
6df9a95e | 4869 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4870 | { |
6bccf3ab JM |
4871 | struct btrfs_root *extent_root = fs_info->extent_root; |
4872 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4873 | struct btrfs_key key; |
2b82032c YZ |
4874 | struct btrfs_device *device; |
4875 | struct btrfs_chunk *chunk; | |
4876 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4877 | struct extent_map *em; |
4878 | struct map_lookup *map; | |
4879 | size_t item_size; | |
4880 | u64 dev_offset; | |
4881 | u64 stripe_size; | |
4882 | int i = 0; | |
140e639f | 4883 | int ret = 0; |
2b82032c | 4884 | |
592d92ee LB |
4885 | em = get_chunk_map(fs_info, chunk_offset, chunk_size); |
4886 | if (IS_ERR(em)) | |
4887 | return PTR_ERR(em); | |
6df9a95e | 4888 | |
95617d69 | 4889 | map = em->map_lookup; |
6df9a95e JB |
4890 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4891 | stripe_size = em->orig_block_len; | |
4892 | ||
2b82032c | 4893 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4894 | if (!chunk) { |
4895 | ret = -ENOMEM; | |
4896 | goto out; | |
4897 | } | |
4898 | ||
50460e37 FM |
4899 | /* |
4900 | * Take the device list mutex to prevent races with the final phase of | |
4901 | * a device replace operation that replaces the device object associated | |
4902 | * with the map's stripes, because the device object's id can change | |
4903 | * at any time during that final phase of the device replace operation | |
4904 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4905 | */ | |
0b246afa | 4906 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
4907 | for (i = 0; i < map->num_stripes; i++) { |
4908 | device = map->stripes[i].dev; | |
4909 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4910 | |
0b86a832 | 4911 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4912 | if (ret) |
50460e37 | 4913 | break; |
6df9a95e JB |
4914 | ret = btrfs_alloc_dev_extent(trans, device, |
4915 | chunk_root->root_key.objectid, | |
4916 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4917 | chunk_offset, dev_offset, | |
4918 | stripe_size); | |
4919 | if (ret) | |
50460e37 FM |
4920 | break; |
4921 | } | |
4922 | if (ret) { | |
0b246afa | 4923 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 4924 | goto out; |
2b82032c YZ |
4925 | } |
4926 | ||
2b82032c | 4927 | stripe = &chunk->stripe; |
6df9a95e JB |
4928 | for (i = 0; i < map->num_stripes; i++) { |
4929 | device = map->stripes[i].dev; | |
4930 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4931 | |
e17cade2 CM |
4932 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4933 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4934 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4935 | stripe++; |
0b86a832 | 4936 | } |
0b246afa | 4937 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4938 | |
2b82032c | 4939 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4940 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4941 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4942 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4943 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4944 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4945 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 4946 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 4947 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4948 | |
2b82032c YZ |
4949 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4950 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4951 | key.offset = chunk_offset; | |
0b86a832 | 4952 | |
2b82032c | 4953 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4954 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4955 | /* | |
4956 | * TODO: Cleanup of inserted chunk root in case of | |
4957 | * failure. | |
4958 | */ | |
2ff7e61e | 4959 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 4960 | } |
1abe9b8a | 4961 | |
6df9a95e | 4962 | out: |
0b86a832 | 4963 | kfree(chunk); |
6df9a95e | 4964 | free_extent_map(em); |
4ed1d16e | 4965 | return ret; |
2b82032c | 4966 | } |
0b86a832 | 4967 | |
2b82032c YZ |
4968 | /* |
4969 | * Chunk allocation falls into two parts. The first part does works | |
4970 | * that make the new allocated chunk useable, but not do any operation | |
4971 | * that modifies the chunk tree. The second part does the works that | |
4972 | * require modifying the chunk tree. This division is important for the | |
4973 | * bootstrap process of adding storage to a seed btrfs. | |
4974 | */ | |
4975 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4976 | struct btrfs_fs_info *fs_info, u64 type) |
2b82032c YZ |
4977 | { |
4978 | u64 chunk_offset; | |
2b82032c | 4979 | |
0b246afa JM |
4980 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
4981 | chunk_offset = find_next_chunk(fs_info); | |
72b468c8 | 4982 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
4983 | } |
4984 | ||
d397712b | 4985 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 4986 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
4987 | { |
4988 | u64 chunk_offset; | |
4989 | u64 sys_chunk_offset; | |
2b82032c | 4990 | u64 alloc_profile; |
2b82032c YZ |
4991 | int ret; |
4992 | ||
6df9a95e | 4993 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 4994 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 4995 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
4996 | if (ret) |
4997 | return ret; | |
2b82032c | 4998 | |
0b246afa | 4999 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5000 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5001 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5002 | return ret; |
2b82032c YZ |
5003 | } |
5004 | ||
d20983b4 MX |
5005 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5006 | { | |
5007 | int max_errors; | |
5008 | ||
5009 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5010 | BTRFS_BLOCK_GROUP_RAID10 | | |
5011 | BTRFS_BLOCK_GROUP_RAID5 | | |
5012 | BTRFS_BLOCK_GROUP_DUP)) { | |
5013 | max_errors = 1; | |
5014 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5015 | max_errors = 2; | |
5016 | } else { | |
5017 | max_errors = 0; | |
005d6427 | 5018 | } |
2b82032c | 5019 | |
d20983b4 | 5020 | return max_errors; |
2b82032c YZ |
5021 | } |
5022 | ||
2ff7e61e | 5023 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5024 | { |
5025 | struct extent_map *em; | |
5026 | struct map_lookup *map; | |
2b82032c | 5027 | int readonly = 0; |
d20983b4 | 5028 | int miss_ndevs = 0; |
2b82032c YZ |
5029 | int i; |
5030 | ||
592d92ee LB |
5031 | em = get_chunk_map(fs_info, chunk_offset, 1); |
5032 | if (IS_ERR(em)) | |
2b82032c YZ |
5033 | return 1; |
5034 | ||
95617d69 | 5035 | map = em->map_lookup; |
2b82032c | 5036 | for (i = 0; i < map->num_stripes; i++) { |
d20983b4 MX |
5037 | if (map->stripes[i].dev->missing) { |
5038 | miss_ndevs++; | |
5039 | continue; | |
5040 | } | |
5041 | ||
2b82032c YZ |
5042 | if (!map->stripes[i].dev->writeable) { |
5043 | readonly = 1; | |
d20983b4 | 5044 | goto end; |
2b82032c YZ |
5045 | } |
5046 | } | |
d20983b4 MX |
5047 | |
5048 | /* | |
5049 | * If the number of missing devices is larger than max errors, | |
5050 | * we can not write the data into that chunk successfully, so | |
5051 | * set it readonly. | |
5052 | */ | |
5053 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5054 | readonly = 1; | |
5055 | end: | |
0b86a832 | 5056 | free_extent_map(em); |
2b82032c | 5057 | return readonly; |
0b86a832 CM |
5058 | } |
5059 | ||
5060 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5061 | { | |
a8067e02 | 5062 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5063 | } |
5064 | ||
5065 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5066 | { | |
5067 | struct extent_map *em; | |
5068 | ||
d397712b | 5069 | while (1) { |
890871be | 5070 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5071 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5072 | if (em) | |
5073 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5074 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5075 | if (!em) |
5076 | break; | |
0b86a832 CM |
5077 | /* once for us */ |
5078 | free_extent_map(em); | |
5079 | /* once for the tree */ | |
5080 | free_extent_map(em); | |
5081 | } | |
5082 | } | |
5083 | ||
5d964051 | 5084 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5085 | { |
5086 | struct extent_map *em; | |
5087 | struct map_lookup *map; | |
f188591e CM |
5088 | int ret; |
5089 | ||
592d92ee LB |
5090 | em = get_chunk_map(fs_info, logical, len); |
5091 | if (IS_ERR(em)) | |
5092 | /* | |
5093 | * We could return errors for these cases, but that could get | |
5094 | * ugly and we'd probably do the same thing which is just not do | |
5095 | * anything else and exit, so return 1 so the callers don't try | |
5096 | * to use other copies. | |
5097 | */ | |
fb7669b5 | 5098 | return 1; |
fb7669b5 | 5099 | |
95617d69 | 5100 | map = em->map_lookup; |
f188591e CM |
5101 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5102 | ret = map->num_stripes; | |
321aecc6 CM |
5103 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5104 | ret = map->sub_stripes; | |
53b381b3 DW |
5105 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5106 | ret = 2; | |
5107 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5108 | ret = 3; | |
f188591e CM |
5109 | else |
5110 | ret = 1; | |
5111 | free_extent_map(em); | |
ad6d620e | 5112 | |
73beece9 | 5113 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
6fad823f LB |
5114 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5115 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5116 | ret++; |
73beece9 | 5117 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5118 | |
f188591e CM |
5119 | return ret; |
5120 | } | |
5121 | ||
2ff7e61e | 5122 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5123 | u64 logical) |
5124 | { | |
5125 | struct extent_map *em; | |
5126 | struct map_lookup *map; | |
0b246afa | 5127 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5128 | |
592d92ee | 5129 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5130 | |
69f03f13 NB |
5131 | if (!WARN_ON(IS_ERR(em))) { |
5132 | map = em->map_lookup; | |
5133 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5134 | len = map->stripe_len * nr_data_stripes(map); | |
5135 | free_extent_map(em); | |
5136 | } | |
53b381b3 DW |
5137 | return len; |
5138 | } | |
5139 | ||
e4ff5fb5 | 5140 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5141 | { |
5142 | struct extent_map *em; | |
5143 | struct map_lookup *map; | |
53b381b3 DW |
5144 | int ret = 0; |
5145 | ||
592d92ee | 5146 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5147 | |
69f03f13 NB |
5148 | if(!WARN_ON(IS_ERR(em))) { |
5149 | map = em->map_lookup; | |
5150 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5151 | ret = 1; | |
5152 | free_extent_map(em); | |
5153 | } | |
53b381b3 DW |
5154 | return ret; |
5155 | } | |
5156 | ||
30d9861f SB |
5157 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5158 | struct map_lookup *map, int first, int num, | |
5159 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5160 | { |
5161 | int i; | |
30d9861f SB |
5162 | int tolerance; |
5163 | struct btrfs_device *srcdev; | |
5164 | ||
5165 | if (dev_replace_is_ongoing && | |
5166 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5167 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5168 | srcdev = fs_info->dev_replace.srcdev; | |
5169 | else | |
5170 | srcdev = NULL; | |
5171 | ||
5172 | /* | |
5173 | * try to avoid the drive that is the source drive for a | |
5174 | * dev-replace procedure, only choose it if no other non-missing | |
5175 | * mirror is available | |
5176 | */ | |
5177 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5178 | if (map->stripes[optimal].dev->bdev && | |
5179 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5180 | return optimal; | |
5181 | for (i = first; i < first + num; i++) { | |
5182 | if (map->stripes[i].dev->bdev && | |
5183 | (tolerance || map->stripes[i].dev != srcdev)) | |
5184 | return i; | |
5185 | } | |
dfe25020 | 5186 | } |
30d9861f | 5187 | |
dfe25020 CM |
5188 | /* we couldn't find one that doesn't fail. Just return something |
5189 | * and the io error handling code will clean up eventually | |
5190 | */ | |
5191 | return optimal; | |
5192 | } | |
5193 | ||
53b381b3 DW |
5194 | static inline int parity_smaller(u64 a, u64 b) |
5195 | { | |
5196 | return a > b; | |
5197 | } | |
5198 | ||
5199 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5200 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5201 | { |
5202 | struct btrfs_bio_stripe s; | |
5203 | int i; | |
5204 | u64 l; | |
5205 | int again = 1; | |
5206 | ||
5207 | while (again) { | |
5208 | again = 0; | |
cc7539ed | 5209 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5210 | if (parity_smaller(bbio->raid_map[i], |
5211 | bbio->raid_map[i+1])) { | |
53b381b3 | 5212 | s = bbio->stripes[i]; |
8e5cfb55 | 5213 | l = bbio->raid_map[i]; |
53b381b3 | 5214 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5215 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5216 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5217 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5218 | |
53b381b3 DW |
5219 | again = 1; |
5220 | } | |
5221 | } | |
5222 | } | |
5223 | } | |
5224 | ||
6e9606d2 ZL |
5225 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5226 | { | |
5227 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5228 | /* the size of the btrfs_bio */ |
6e9606d2 | 5229 | sizeof(struct btrfs_bio) + |
e57cf21e | 5230 | /* plus the variable array for the stripes */ |
6e9606d2 | 5231 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5232 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5233 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5234 | /* |
5235 | * plus the raid_map, which includes both the tgt dev | |
5236 | * and the stripes | |
5237 | */ | |
5238 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5239 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5240 | |
5241 | atomic_set(&bbio->error, 0); | |
140475ae | 5242 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5243 | |
5244 | return bbio; | |
5245 | } | |
5246 | ||
5247 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5248 | { | |
140475ae ER |
5249 | WARN_ON(!refcount_read(&bbio->refs)); |
5250 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5251 | } |
5252 | ||
5253 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5254 | { | |
5255 | if (!bbio) | |
5256 | return; | |
140475ae | 5257 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5258 | kfree(bbio); |
5259 | } | |
5260 | ||
0b3d4cd3 LB |
5261 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5262 | /* | |
5263 | * Please note that, discard won't be sent to target device of device | |
5264 | * replace. | |
5265 | */ | |
5266 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5267 | u64 logical, u64 length, | |
5268 | struct btrfs_bio **bbio_ret) | |
5269 | { | |
5270 | struct extent_map *em; | |
5271 | struct map_lookup *map; | |
5272 | struct btrfs_bio *bbio; | |
5273 | u64 offset; | |
5274 | u64 stripe_nr; | |
5275 | u64 stripe_nr_end; | |
5276 | u64 stripe_end_offset; | |
5277 | u64 stripe_cnt; | |
5278 | u64 stripe_len; | |
5279 | u64 stripe_offset; | |
5280 | u64 num_stripes; | |
5281 | u32 stripe_index; | |
5282 | u32 factor = 0; | |
5283 | u32 sub_stripes = 0; | |
5284 | u64 stripes_per_dev = 0; | |
5285 | u32 remaining_stripes = 0; | |
5286 | u32 last_stripe = 0; | |
5287 | int ret = 0; | |
5288 | int i; | |
5289 | ||
5290 | /* discard always return a bbio */ | |
5291 | ASSERT(bbio_ret); | |
5292 | ||
5293 | em = get_chunk_map(fs_info, logical, length); | |
5294 | if (IS_ERR(em)) | |
5295 | return PTR_ERR(em); | |
5296 | ||
5297 | map = em->map_lookup; | |
5298 | /* we don't discard raid56 yet */ | |
5299 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5300 | ret = -EOPNOTSUPP; | |
5301 | goto out; | |
5302 | } | |
5303 | ||
5304 | offset = logical - em->start; | |
5305 | length = min_t(u64, em->len - offset, length); | |
5306 | ||
5307 | stripe_len = map->stripe_len; | |
5308 | /* | |
5309 | * stripe_nr counts the total number of stripes we have to stride | |
5310 | * to get to this block | |
5311 | */ | |
5312 | stripe_nr = div64_u64(offset, stripe_len); | |
5313 | ||
5314 | /* stripe_offset is the offset of this block in its stripe */ | |
5315 | stripe_offset = offset - stripe_nr * stripe_len; | |
5316 | ||
5317 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5318 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5319 | stripe_cnt = stripe_nr_end - stripe_nr; |
5320 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5321 | (offset + length); | |
5322 | /* | |
5323 | * after this, stripe_nr is the number of stripes on this | |
5324 | * device we have to walk to find the data, and stripe_index is | |
5325 | * the number of our device in the stripe array | |
5326 | */ | |
5327 | num_stripes = 1; | |
5328 | stripe_index = 0; | |
5329 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5330 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5331 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5332 | sub_stripes = 1; | |
5333 | else | |
5334 | sub_stripes = map->sub_stripes; | |
5335 | ||
5336 | factor = map->num_stripes / sub_stripes; | |
5337 | num_stripes = min_t(u64, map->num_stripes, | |
5338 | sub_stripes * stripe_cnt); | |
5339 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5340 | stripe_index *= sub_stripes; | |
5341 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5342 | &remaining_stripes); | |
5343 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5344 | last_stripe *= sub_stripes; | |
5345 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5346 | BTRFS_BLOCK_GROUP_DUP)) { | |
5347 | num_stripes = map->num_stripes; | |
5348 | } else { | |
5349 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5350 | &stripe_index); | |
5351 | } | |
5352 | ||
5353 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5354 | if (!bbio) { | |
5355 | ret = -ENOMEM; | |
5356 | goto out; | |
5357 | } | |
5358 | ||
5359 | for (i = 0; i < num_stripes; i++) { | |
5360 | bbio->stripes[i].physical = | |
5361 | map->stripes[stripe_index].physical + | |
5362 | stripe_offset + stripe_nr * map->stripe_len; | |
5363 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5364 | ||
5365 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5366 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5367 | bbio->stripes[i].length = stripes_per_dev * | |
5368 | map->stripe_len; | |
5369 | ||
5370 | if (i / sub_stripes < remaining_stripes) | |
5371 | bbio->stripes[i].length += | |
5372 | map->stripe_len; | |
5373 | ||
5374 | /* | |
5375 | * Special for the first stripe and | |
5376 | * the last stripe: | |
5377 | * | |
5378 | * |-------|...|-------| | |
5379 | * |----------| | |
5380 | * off end_off | |
5381 | */ | |
5382 | if (i < sub_stripes) | |
5383 | bbio->stripes[i].length -= | |
5384 | stripe_offset; | |
5385 | ||
5386 | if (stripe_index >= last_stripe && | |
5387 | stripe_index <= (last_stripe + | |
5388 | sub_stripes - 1)) | |
5389 | bbio->stripes[i].length -= | |
5390 | stripe_end_offset; | |
5391 | ||
5392 | if (i == sub_stripes - 1) | |
5393 | stripe_offset = 0; | |
5394 | } else { | |
5395 | bbio->stripes[i].length = length; | |
5396 | } | |
5397 | ||
5398 | stripe_index++; | |
5399 | if (stripe_index == map->num_stripes) { | |
5400 | stripe_index = 0; | |
5401 | stripe_nr++; | |
5402 | } | |
5403 | } | |
5404 | ||
5405 | *bbio_ret = bbio; | |
5406 | bbio->map_type = map->type; | |
5407 | bbio->num_stripes = num_stripes; | |
5408 | out: | |
5409 | free_extent_map(em); | |
5410 | return ret; | |
5411 | } | |
5412 | ||
5ab56090 LB |
5413 | /* |
5414 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5415 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5416 | * left cursor can also be read from the target drive. | |
5417 | * | |
5418 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5419 | * array of stripes. | |
5420 | * For READ, it also needs to be supported using the same mirror number. | |
5421 | * | |
5422 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5423 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5424 | * case. | |
5425 | */ | |
5426 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5427 | u64 logical, u64 length, | |
5428 | u64 srcdev_devid, int *mirror_num, | |
5429 | u64 *physical) | |
5430 | { | |
5431 | struct btrfs_bio *bbio = NULL; | |
5432 | int num_stripes; | |
5433 | int index_srcdev = 0; | |
5434 | int found = 0; | |
5435 | u64 physical_of_found = 0; | |
5436 | int i; | |
5437 | int ret = 0; | |
5438 | ||
5439 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5440 | logical, &length, &bbio, 0, 0); | |
5441 | if (ret) { | |
5442 | ASSERT(bbio == NULL); | |
5443 | return ret; | |
5444 | } | |
5445 | ||
5446 | num_stripes = bbio->num_stripes; | |
5447 | if (*mirror_num > num_stripes) { | |
5448 | /* | |
5449 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5450 | * that means that the requested area is not left of the left | |
5451 | * cursor | |
5452 | */ | |
5453 | btrfs_put_bbio(bbio); | |
5454 | return -EIO; | |
5455 | } | |
5456 | ||
5457 | /* | |
5458 | * process the rest of the function using the mirror_num of the source | |
5459 | * drive. Therefore look it up first. At the end, patch the device | |
5460 | * pointer to the one of the target drive. | |
5461 | */ | |
5462 | for (i = 0; i < num_stripes; i++) { | |
5463 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5464 | continue; | |
5465 | ||
5466 | /* | |
5467 | * In case of DUP, in order to keep it simple, only add the | |
5468 | * mirror with the lowest physical address | |
5469 | */ | |
5470 | if (found && | |
5471 | physical_of_found <= bbio->stripes[i].physical) | |
5472 | continue; | |
5473 | ||
5474 | index_srcdev = i; | |
5475 | found = 1; | |
5476 | physical_of_found = bbio->stripes[i].physical; | |
5477 | } | |
5478 | ||
5479 | btrfs_put_bbio(bbio); | |
5480 | ||
5481 | ASSERT(found); | |
5482 | if (!found) | |
5483 | return -EIO; | |
5484 | ||
5485 | *mirror_num = index_srcdev + 1; | |
5486 | *physical = physical_of_found; | |
5487 | return ret; | |
5488 | } | |
5489 | ||
73c0f228 LB |
5490 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5491 | struct btrfs_bio **bbio_ret, | |
5492 | struct btrfs_dev_replace *dev_replace, | |
5493 | int *num_stripes_ret, int *max_errors_ret) | |
5494 | { | |
5495 | struct btrfs_bio *bbio = *bbio_ret; | |
5496 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5497 | int tgtdev_indexes = 0; | |
5498 | int num_stripes = *num_stripes_ret; | |
5499 | int max_errors = *max_errors_ret; | |
5500 | int i; | |
5501 | ||
5502 | if (op == BTRFS_MAP_WRITE) { | |
5503 | int index_where_to_add; | |
5504 | ||
5505 | /* | |
5506 | * duplicate the write operations while the dev replace | |
5507 | * procedure is running. Since the copying of the old disk to | |
5508 | * the new disk takes place at run time while the filesystem is | |
5509 | * mounted writable, the regular write operations to the old | |
5510 | * disk have to be duplicated to go to the new disk as well. | |
5511 | * | |
5512 | * Note that device->missing is handled by the caller, and that | |
5513 | * the write to the old disk is already set up in the stripes | |
5514 | * array. | |
5515 | */ | |
5516 | index_where_to_add = num_stripes; | |
5517 | for (i = 0; i < num_stripes; i++) { | |
5518 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5519 | /* write to new disk, too */ | |
5520 | struct btrfs_bio_stripe *new = | |
5521 | bbio->stripes + index_where_to_add; | |
5522 | struct btrfs_bio_stripe *old = | |
5523 | bbio->stripes + i; | |
5524 | ||
5525 | new->physical = old->physical; | |
5526 | new->length = old->length; | |
5527 | new->dev = dev_replace->tgtdev; | |
5528 | bbio->tgtdev_map[i] = index_where_to_add; | |
5529 | index_where_to_add++; | |
5530 | max_errors++; | |
5531 | tgtdev_indexes++; | |
5532 | } | |
5533 | } | |
5534 | num_stripes = index_where_to_add; | |
5535 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5536 | int index_srcdev = 0; | |
5537 | int found = 0; | |
5538 | u64 physical_of_found = 0; | |
5539 | ||
5540 | /* | |
5541 | * During the dev-replace procedure, the target drive can also | |
5542 | * be used to read data in case it is needed to repair a corrupt | |
5543 | * block elsewhere. This is possible if the requested area is | |
5544 | * left of the left cursor. In this area, the target drive is a | |
5545 | * full copy of the source drive. | |
5546 | */ | |
5547 | for (i = 0; i < num_stripes; i++) { | |
5548 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5549 | /* | |
5550 | * In case of DUP, in order to keep it simple, | |
5551 | * only add the mirror with the lowest physical | |
5552 | * address | |
5553 | */ | |
5554 | if (found && | |
5555 | physical_of_found <= | |
5556 | bbio->stripes[i].physical) | |
5557 | continue; | |
5558 | index_srcdev = i; | |
5559 | found = 1; | |
5560 | physical_of_found = bbio->stripes[i].physical; | |
5561 | } | |
5562 | } | |
5563 | if (found) { | |
5564 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5565 | bbio->stripes + num_stripes; | |
5566 | ||
5567 | tgtdev_stripe->physical = physical_of_found; | |
5568 | tgtdev_stripe->length = | |
5569 | bbio->stripes[index_srcdev].length; | |
5570 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5571 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5572 | ||
5573 | tgtdev_indexes++; | |
5574 | num_stripes++; | |
5575 | } | |
5576 | } | |
5577 | ||
5578 | *num_stripes_ret = num_stripes; | |
5579 | *max_errors_ret = max_errors; | |
5580 | bbio->num_tgtdevs = tgtdev_indexes; | |
5581 | *bbio_ret = bbio; | |
5582 | } | |
5583 | ||
2b19a1fe LB |
5584 | static bool need_full_stripe(enum btrfs_map_op op) |
5585 | { | |
5586 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5587 | } | |
5588 | ||
cf8cddd3 CH |
5589 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5590 | enum btrfs_map_op op, | |
f2d8d74d | 5591 | u64 logical, u64 *length, |
a1d3c478 | 5592 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5593 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5594 | { |
5595 | struct extent_map *em; | |
5596 | struct map_lookup *map; | |
0b86a832 | 5597 | u64 offset; |
593060d7 CM |
5598 | u64 stripe_offset; |
5599 | u64 stripe_nr; | |
53b381b3 | 5600 | u64 stripe_len; |
9d644a62 | 5601 | u32 stripe_index; |
cea9e445 | 5602 | int i; |
de11cc12 | 5603 | int ret = 0; |
f2d8d74d | 5604 | int num_stripes; |
a236aed1 | 5605 | int max_errors = 0; |
2c8cdd6e | 5606 | int tgtdev_indexes = 0; |
a1d3c478 | 5607 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5608 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5609 | int dev_replace_is_ongoing = 0; | |
5610 | int num_alloc_stripes; | |
ad6d620e SB |
5611 | int patch_the_first_stripe_for_dev_replace = 0; |
5612 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5613 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5614 | |
0b3d4cd3 LB |
5615 | if (op == BTRFS_MAP_DISCARD) |
5616 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5617 | *length, bbio_ret); | |
5618 | ||
592d92ee LB |
5619 | em = get_chunk_map(fs_info, logical, *length); |
5620 | if (IS_ERR(em)) | |
5621 | return PTR_ERR(em); | |
0b86a832 | 5622 | |
95617d69 | 5623 | map = em->map_lookup; |
0b86a832 | 5624 | offset = logical - em->start; |
593060d7 | 5625 | |
53b381b3 | 5626 | stripe_len = map->stripe_len; |
593060d7 CM |
5627 | stripe_nr = offset; |
5628 | /* | |
5629 | * stripe_nr counts the total number of stripes we have to stride | |
5630 | * to get to this block | |
5631 | */ | |
47c5713f | 5632 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5633 | |
53b381b3 | 5634 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5635 | if (offset < stripe_offset) { |
5d163e0e JM |
5636 | btrfs_crit(fs_info, |
5637 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5638 | stripe_offset, offset, em->start, logical, |
5639 | stripe_len); | |
5640 | free_extent_map(em); | |
5641 | return -EINVAL; | |
5642 | } | |
593060d7 CM |
5643 | |
5644 | /* stripe_offset is the offset of this block in its stripe*/ | |
5645 | stripe_offset = offset - stripe_offset; | |
5646 | ||
53b381b3 | 5647 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5648 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5649 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5650 | raid56_full_stripe_start = offset; | |
5651 | ||
5652 | /* allow a write of a full stripe, but make sure we don't | |
5653 | * allow straddling of stripes | |
5654 | */ | |
47c5713f DS |
5655 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5656 | full_stripe_len); | |
53b381b3 DW |
5657 | raid56_full_stripe_start *= full_stripe_len; |
5658 | } | |
5659 | ||
0b3d4cd3 | 5660 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5661 | u64 max_len; |
5662 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5663 | For other RAID types and for RAID[56] reads, just allow a single | |
5664 | stripe (on a single disk). */ | |
ffe2d203 | 5665 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5666 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5667 | max_len = stripe_len * nr_data_stripes(map) - |
5668 | (offset - raid56_full_stripe_start); | |
5669 | } else { | |
5670 | /* we limit the length of each bio to what fits in a stripe */ | |
5671 | max_len = stripe_len - stripe_offset; | |
5672 | } | |
5673 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5674 | } else { |
5675 | *length = em->len - offset; | |
5676 | } | |
f2d8d74d | 5677 | |
53b381b3 DW |
5678 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5679 | it cares about is the length */ | |
a1d3c478 | 5680 | if (!bbio_ret) |
cea9e445 CM |
5681 | goto out; |
5682 | ||
73beece9 | 5683 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5684 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5685 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5686 | btrfs_dev_replace_unlock(dev_replace, 0); |
5687 | else | |
5688 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5689 | |
ad6d620e | 5690 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5691 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5692 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5693 | dev_replace->srcdev->devid, | |
5694 | &mirror_num, | |
5695 | &physical_to_patch_in_first_stripe); | |
5696 | if (ret) | |
ad6d620e | 5697 | goto out; |
5ab56090 LB |
5698 | else |
5699 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5700 | } else if (mirror_num > map->num_stripes) { |
5701 | mirror_num = 0; | |
5702 | } | |
5703 | ||
f2d8d74d | 5704 | num_stripes = 1; |
cea9e445 | 5705 | stripe_index = 0; |
fce3bb9a | 5706 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5707 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5708 | &stripe_index); | |
0b3d4cd3 | 5709 | if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS) |
28e1cc7d | 5710 | mirror_num = 1; |
fce3bb9a | 5711 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
0b3d4cd3 | 5712 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
f2d8d74d | 5713 | num_stripes = map->num_stripes; |
2fff734f | 5714 | else if (mirror_num) |
f188591e | 5715 | stripe_index = mirror_num - 1; |
dfe25020 | 5716 | else { |
30d9861f | 5717 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5718 | map->num_stripes, |
30d9861f SB |
5719 | current->pid % map->num_stripes, |
5720 | dev_replace_is_ongoing); | |
a1d3c478 | 5721 | mirror_num = stripe_index + 1; |
dfe25020 | 5722 | } |
2fff734f | 5723 | |
611f0e00 | 5724 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
0b3d4cd3 | 5725 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) { |
f2d8d74d | 5726 | num_stripes = map->num_stripes; |
a1d3c478 | 5727 | } else if (mirror_num) { |
f188591e | 5728 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5729 | } else { |
5730 | mirror_num = 1; | |
5731 | } | |
2fff734f | 5732 | |
321aecc6 | 5733 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5734 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5735 | |
47c5713f | 5736 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5737 | stripe_index *= map->sub_stripes; |
5738 | ||
cf8cddd3 | 5739 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
f2d8d74d | 5740 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5741 | else if (mirror_num) |
5742 | stripe_index += mirror_num - 1; | |
dfe25020 | 5743 | else { |
3e74317a | 5744 | int old_stripe_index = stripe_index; |
30d9861f SB |
5745 | stripe_index = find_live_mirror(fs_info, map, |
5746 | stripe_index, | |
dfe25020 | 5747 | map->sub_stripes, stripe_index + |
30d9861f SB |
5748 | current->pid % map->sub_stripes, |
5749 | dev_replace_is_ongoing); | |
3e74317a | 5750 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5751 | } |
53b381b3 | 5752 | |
ffe2d203 | 5753 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5754 | if (need_raid_map && |
cf8cddd3 | 5755 | (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS || |
af8e2d1d | 5756 | mirror_num > 1)) { |
53b381b3 | 5757 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5758 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5759 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5760 | |
5761 | /* RAID[56] write or recovery. Return all stripes */ | |
5762 | num_stripes = map->num_stripes; | |
5763 | max_errors = nr_parity_stripes(map); | |
5764 | ||
53b381b3 DW |
5765 | *length = map->stripe_len; |
5766 | stripe_index = 0; | |
5767 | stripe_offset = 0; | |
5768 | } else { | |
5769 | /* | |
5770 | * Mirror #0 or #1 means the original data block. | |
5771 | * Mirror #2 is RAID5 parity block. | |
5772 | * Mirror #3 is RAID6 Q block. | |
5773 | */ | |
47c5713f DS |
5774 | stripe_nr = div_u64_rem(stripe_nr, |
5775 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5776 | if (mirror_num > 1) |
5777 | stripe_index = nr_data_stripes(map) + | |
5778 | mirror_num - 2; | |
5779 | ||
5780 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5781 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5782 | &stripe_index); | |
0b3d4cd3 LB |
5783 | if ((op != BTRFS_MAP_WRITE && |
5784 | op != BTRFS_MAP_GET_READ_MIRRORS) && | |
5785 | mirror_num <= 1) | |
28e1cc7d | 5786 | mirror_num = 1; |
53b381b3 | 5787 | } |
8790d502 CM |
5788 | } else { |
5789 | /* | |
47c5713f DS |
5790 | * after this, stripe_nr is the number of stripes on this |
5791 | * device we have to walk to find the data, and stripe_index is | |
5792 | * the number of our device in the stripe array | |
8790d502 | 5793 | */ |
47c5713f DS |
5794 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5795 | &stripe_index); | |
a1d3c478 | 5796 | mirror_num = stripe_index + 1; |
8790d502 | 5797 | } |
e042d1ec | 5798 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5799 | btrfs_crit(fs_info, |
5800 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5801 | stripe_index, map->num_stripes); |
5802 | ret = -EINVAL; | |
5803 | goto out; | |
5804 | } | |
cea9e445 | 5805 | |
472262f3 | 5806 | num_alloc_stripes = num_stripes; |
6fad823f | 5807 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5808 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5809 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5810 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5811 | num_alloc_stripes++; |
2c8cdd6e | 5812 | tgtdev_indexes = num_stripes; |
ad6d620e | 5813 | } |
2c8cdd6e | 5814 | |
6e9606d2 | 5815 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5816 | if (!bbio) { |
5817 | ret = -ENOMEM; | |
5818 | goto out; | |
5819 | } | |
6fad823f | 5820 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5821 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5822 | |
8e5cfb55 | 5823 | /* build raid_map */ |
2b19a1fe LB |
5824 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5825 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5826 | u64 tmp; |
9d644a62 | 5827 | unsigned rot; |
8e5cfb55 ZL |
5828 | |
5829 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5830 | sizeof(struct btrfs_bio_stripe) * | |
5831 | num_alloc_stripes + | |
5832 | sizeof(int) * tgtdev_indexes); | |
5833 | ||
5834 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5835 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5836 | |
5837 | /* Fill in the logical address of each stripe */ | |
5838 | tmp = stripe_nr * nr_data_stripes(map); | |
5839 | for (i = 0; i < nr_data_stripes(map); i++) | |
5840 | bbio->raid_map[(i+rot) % num_stripes] = | |
5841 | em->start + (tmp + i) * map->stripe_len; | |
5842 | ||
5843 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5844 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5845 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5846 | RAID6_Q_STRIPE; | |
5847 | } | |
5848 | ||
b89203f7 | 5849 | |
0b3d4cd3 LB |
5850 | for (i = 0; i < num_stripes; i++) { |
5851 | bbio->stripes[i].physical = | |
5852 | map->stripes[stripe_index].physical + | |
5853 | stripe_offset + | |
5854 | stripe_nr * map->stripe_len; | |
5855 | bbio->stripes[i].dev = | |
5856 | map->stripes[stripe_index].dev; | |
5857 | stripe_index++; | |
593060d7 | 5858 | } |
de11cc12 | 5859 | |
2b19a1fe | 5860 | if (need_full_stripe(op)) |
d20983b4 | 5861 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5862 | |
8e5cfb55 ZL |
5863 | if (bbio->raid_map) |
5864 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5865 | |
73c0f228 | 5866 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5867 | need_full_stripe(op)) { |
73c0f228 LB |
5868 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5869 | &max_errors); | |
472262f3 SB |
5870 | } |
5871 | ||
de11cc12 | 5872 | *bbio_ret = bbio; |
10f11900 | 5873 | bbio->map_type = map->type; |
de11cc12 LZ |
5874 | bbio->num_stripes = num_stripes; |
5875 | bbio->max_errors = max_errors; | |
5876 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5877 | |
5878 | /* | |
5879 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5880 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5881 | * available as a mirror | |
5882 | */ | |
5883 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5884 | WARN_ON(num_stripes > 1); | |
5885 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5886 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5887 | bbio->mirror_num = map->num_stripes + 1; | |
5888 | } | |
cea9e445 | 5889 | out: |
73beece9 LB |
5890 | if (dev_replace_is_ongoing) { |
5891 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5892 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5893 | } | |
0b86a832 | 5894 | free_extent_map(em); |
de11cc12 | 5895 | return ret; |
0b86a832 CM |
5896 | } |
5897 | ||
cf8cddd3 | 5898 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 5899 | u64 logical, u64 *length, |
a1d3c478 | 5900 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5901 | { |
b3d3fa51 | 5902 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5903 | mirror_num, 0); |
f2d8d74d CM |
5904 | } |
5905 | ||
af8e2d1d | 5906 | /* For Scrub/replace */ |
cf8cddd3 | 5907 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 5908 | u64 logical, u64 *length, |
825ad4c9 | 5909 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 5910 | { |
825ad4c9 | 5911 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
5912 | } |
5913 | ||
ab8d0fc4 | 5914 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
5915 | u64 chunk_start, u64 physical, u64 devid, |
5916 | u64 **logical, int *naddrs, int *stripe_len) | |
5917 | { | |
a512bbf8 YZ |
5918 | struct extent_map *em; |
5919 | struct map_lookup *map; | |
5920 | u64 *buf; | |
5921 | u64 bytenr; | |
5922 | u64 length; | |
5923 | u64 stripe_nr; | |
53b381b3 | 5924 | u64 rmap_len; |
a512bbf8 YZ |
5925 | int i, j, nr = 0; |
5926 | ||
592d92ee LB |
5927 | em = get_chunk_map(fs_info, chunk_start, 1); |
5928 | if (IS_ERR(em)) | |
835d974f | 5929 | return -EIO; |
835d974f | 5930 | |
95617d69 | 5931 | map = em->map_lookup; |
a512bbf8 | 5932 | length = em->len; |
53b381b3 DW |
5933 | rmap_len = map->stripe_len; |
5934 | ||
a512bbf8 | 5935 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5936 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5937 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5938 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5939 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5940 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5941 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5942 | } | |
a512bbf8 | 5943 | |
31e818fe | 5944 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5945 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5946 | |
5947 | for (i = 0; i < map->num_stripes; i++) { | |
5948 | if (devid && map->stripes[i].dev->devid != devid) | |
5949 | continue; | |
5950 | if (map->stripes[i].physical > physical || | |
5951 | map->stripes[i].physical + length <= physical) | |
5952 | continue; | |
5953 | ||
5954 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 5955 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5956 | |
5957 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5958 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5959 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5960 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5961 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5962 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5963 | * Alternatively, just use rmap_len below instead of | |
5964 | * map->stripe_len */ | |
5965 | ||
5966 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5967 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5968 | for (j = 0; j < nr; j++) { |
5969 | if (buf[j] == bytenr) | |
5970 | break; | |
5971 | } | |
934d375b CM |
5972 | if (j == nr) { |
5973 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5974 | buf[nr++] = bytenr; |
934d375b | 5975 | } |
a512bbf8 YZ |
5976 | } |
5977 | ||
a512bbf8 YZ |
5978 | *logical = buf; |
5979 | *naddrs = nr; | |
53b381b3 | 5980 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5981 | |
5982 | free_extent_map(em); | |
5983 | return 0; | |
f2d8d74d CM |
5984 | } |
5985 | ||
4246a0b6 | 5986 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 5987 | { |
326e1dbb MS |
5988 | bio->bi_private = bbio->private; |
5989 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 5990 | bio_endio(bio); |
326e1dbb | 5991 | |
6e9606d2 | 5992 | btrfs_put_bbio(bbio); |
8408c716 MX |
5993 | } |
5994 | ||
4246a0b6 | 5995 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 5996 | { |
9be3395b | 5997 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 5998 | int is_orig_bio = 0; |
8790d502 | 5999 | |
4e4cbee9 | 6000 | if (bio->bi_status) { |
a1d3c478 | 6001 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6002 | if (bio->bi_status == BLK_STS_IOERR || |
6003 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6004 | unsigned int stripe_index = |
9be3395b | 6005 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6006 | struct btrfs_device *dev; |
442a4f63 SB |
6007 | |
6008 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6009 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6010 | if (dev->bdev) { |
37226b21 | 6011 | if (bio_op(bio) == REQ_OP_WRITE) |
597a60fa SB |
6012 | btrfs_dev_stat_inc(dev, |
6013 | BTRFS_DEV_STAT_WRITE_ERRS); | |
6014 | else | |
6015 | btrfs_dev_stat_inc(dev, | |
6016 | BTRFS_DEV_STAT_READ_ERRS); | |
70fd7614 | 6017 | if (bio->bi_opf & REQ_PREFLUSH) |
597a60fa SB |
6018 | btrfs_dev_stat_inc(dev, |
6019 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
6020 | btrfs_dev_stat_print_on_error(dev); | |
6021 | } | |
442a4f63 SB |
6022 | } |
6023 | } | |
8790d502 | 6024 | |
a1d3c478 | 6025 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6026 | is_orig_bio = 1; |
6027 | ||
c404e0dc MX |
6028 | btrfs_bio_counter_dec(bbio->fs_info); |
6029 | ||
a1d3c478 | 6030 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6031 | if (!is_orig_bio) { |
6032 | bio_put(bio); | |
a1d3c478 | 6033 | bio = bbio->orig_bio; |
7d2b4daa | 6034 | } |
c7b22bb1 | 6035 | |
9be3395b | 6036 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6037 | /* only send an error to the higher layers if it is |
53b381b3 | 6038 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6039 | */ |
a1d3c478 | 6040 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6041 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6042 | } else { |
1259ab75 CM |
6043 | /* |
6044 | * this bio is actually up to date, we didn't | |
6045 | * go over the max number of errors | |
6046 | */ | |
4e4cbee9 | 6047 | bio->bi_status = 0; |
1259ab75 | 6048 | } |
c55f1396 | 6049 | |
4246a0b6 | 6050 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6051 | } else if (!is_orig_bio) { |
8790d502 CM |
6052 | bio_put(bio); |
6053 | } | |
8790d502 CM |
6054 | } |
6055 | ||
8b712842 CM |
6056 | /* |
6057 | * see run_scheduled_bios for a description of why bios are collected for | |
6058 | * async submit. | |
6059 | * | |
6060 | * This will add one bio to the pending list for a device and make sure | |
6061 | * the work struct is scheduled. | |
6062 | */ | |
2ff7e61e | 6063 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6064 | struct bio *bio) |
8b712842 | 6065 | { |
0b246afa | 6066 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6067 | int should_queue = 1; |
ffbd517d | 6068 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6069 | |
53b381b3 | 6070 | if (device->missing || !device->bdev) { |
4246a0b6 | 6071 | bio_io_error(bio); |
53b381b3 DW |
6072 | return; |
6073 | } | |
6074 | ||
8b712842 | 6075 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6076 | if (bio_op(bio) == REQ_OP_READ) { |
492bb6de | 6077 | bio_get(bio); |
4e49ea4a | 6078 | btrfsic_submit_bio(bio); |
492bb6de | 6079 | bio_put(bio); |
143bede5 | 6080 | return; |
8b712842 CM |
6081 | } |
6082 | ||
6083 | /* | |
0986fe9e | 6084 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
6085 | * higher layers. Otherwise, the async bio makes it appear we have |
6086 | * made progress against dirty pages when we've really just put it | |
6087 | * on a queue for later | |
6088 | */ | |
0b246afa | 6089 | atomic_inc(&fs_info->nr_async_bios); |
492bb6de | 6090 | WARN_ON(bio->bi_next); |
8b712842 | 6091 | bio->bi_next = NULL; |
8b712842 CM |
6092 | |
6093 | spin_lock(&device->io_lock); | |
67f055c7 | 6094 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6095 | pending_bios = &device->pending_sync_bios; |
6096 | else | |
6097 | pending_bios = &device->pending_bios; | |
8b712842 | 6098 | |
ffbd517d CM |
6099 | if (pending_bios->tail) |
6100 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6101 | |
ffbd517d CM |
6102 | pending_bios->tail = bio; |
6103 | if (!pending_bios->head) | |
6104 | pending_bios->head = bio; | |
8b712842 CM |
6105 | if (device->running_pending) |
6106 | should_queue = 0; | |
6107 | ||
6108 | spin_unlock(&device->io_lock); | |
6109 | ||
6110 | if (should_queue) | |
0b246afa | 6111 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6112 | } |
6113 | ||
2ff7e61e JM |
6114 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6115 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6116 | { |
6117 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6118 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6119 | |
6120 | bio->bi_private = bbio; | |
9be3395b | 6121 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6122 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6123 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6124 | #ifdef DEBUG |
6125 | { | |
6126 | struct rcu_string *name; | |
6127 | ||
6128 | rcu_read_lock(); | |
6129 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6130 | btrfs_debug(fs_info, |
6131 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6132 | bio_op(bio), bio->bi_opf, | |
6133 | (u64)bio->bi_iter.bi_sector, | |
6134 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6135 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6136 | rcu_read_unlock(); |
6137 | } | |
6138 | #endif | |
6139 | bio->bi_bdev = dev->bdev; | |
c404e0dc | 6140 | |
2ff7e61e | 6141 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6142 | |
de1ee92a | 6143 | if (async) |
2ff7e61e | 6144 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6145 | else |
4e49ea4a | 6146 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6147 | } |
6148 | ||
de1ee92a JB |
6149 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6150 | { | |
6151 | atomic_inc(&bbio->error); | |
6152 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6153 | /* Should be the original bio. */ |
8408c716 MX |
6154 | WARN_ON(bio != bbio->orig_bio); |
6155 | ||
9be3395b | 6156 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6157 | bio->bi_iter.bi_sector = logical >> 9; |
4e4cbee9 | 6158 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 6159 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6160 | } |
6161 | } | |
6162 | ||
2ff7e61e | 6163 | int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
8b712842 | 6164 | int mirror_num, int async_submit) |
0b86a832 | 6165 | { |
0b86a832 | 6166 | struct btrfs_device *dev; |
8790d502 | 6167 | struct bio *first_bio = bio; |
4f024f37 | 6168 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6169 | u64 length = 0; |
6170 | u64 map_length; | |
0b86a832 | 6171 | int ret; |
08da757d ZL |
6172 | int dev_nr; |
6173 | int total_devs; | |
a1d3c478 | 6174 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6175 | |
4f024f37 | 6176 | length = bio->bi_iter.bi_size; |
0b86a832 | 6177 | map_length = length; |
cea9e445 | 6178 | |
0b246afa JM |
6179 | btrfs_bio_counter_inc_blocked(fs_info); |
6180 | ret = __btrfs_map_block(fs_info, bio_op(bio), logical, | |
37226b21 | 6181 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6182 | if (ret) { |
0b246afa | 6183 | btrfs_bio_counter_dec(fs_info); |
79787eaa | 6184 | return ret; |
c404e0dc | 6185 | } |
cea9e445 | 6186 | |
a1d3c478 | 6187 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6188 | bbio->orig_bio = first_bio; |
6189 | bbio->private = first_bio->bi_private; | |
6190 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6191 | bbio->fs_info = fs_info; |
53b381b3 DW |
6192 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6193 | ||
ad1ba2a0 | 6194 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6195 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6196 | /* In this case, map_length has been set to the length of |
6197 | a single stripe; not the whole write */ | |
37226b21 | 6198 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6199 | ret = raid56_parity_write(fs_info, bio, bbio, |
6200 | map_length); | |
53b381b3 | 6201 | } else { |
2ff7e61e JM |
6202 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6203 | map_length, mirror_num, 1); | |
53b381b3 | 6204 | } |
4245215d | 6205 | |
0b246afa | 6206 | btrfs_bio_counter_dec(fs_info); |
c404e0dc | 6207 | return ret; |
53b381b3 DW |
6208 | } |
6209 | ||
cea9e445 | 6210 | if (map_length < length) { |
0b246afa | 6211 | btrfs_crit(fs_info, |
5d163e0e JM |
6212 | "mapping failed logical %llu bio len %llu len %llu", |
6213 | logical, length, map_length); | |
cea9e445 CM |
6214 | BUG(); |
6215 | } | |
a1d3c478 | 6216 | |
08da757d | 6217 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6218 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6219 | if (!dev || !dev->bdev || |
a967efb3 | 6220 | (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) { |
de1ee92a | 6221 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6222 | continue; |
6223 | } | |
6224 | ||
3aa8e074 | 6225 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6226 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6227 | else |
a1d3c478 | 6228 | bio = first_bio; |
de1ee92a | 6229 | |
2ff7e61e JM |
6230 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6231 | dev_nr, async_submit); | |
8790d502 | 6232 | } |
0b246afa | 6233 | btrfs_bio_counter_dec(fs_info); |
0b86a832 CM |
6234 | return 0; |
6235 | } | |
6236 | ||
aa1b8cd4 | 6237 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6238 | u8 *uuid, u8 *fsid) |
0b86a832 | 6239 | { |
2b82032c YZ |
6240 | struct btrfs_device *device; |
6241 | struct btrfs_fs_devices *cur_devices; | |
6242 | ||
aa1b8cd4 | 6243 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6244 | while (cur_devices) { |
6245 | if (!fsid || | |
6246 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
35c70103 | 6247 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6248 | if (device) |
6249 | return device; | |
6250 | } | |
6251 | cur_devices = cur_devices->seed; | |
6252 | } | |
6253 | return NULL; | |
0b86a832 CM |
6254 | } |
6255 | ||
2ff7e61e | 6256 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6257 | u64 devid, u8 *dev_uuid) |
6258 | { | |
6259 | struct btrfs_device *device; | |
dfe25020 | 6260 | |
12bd2fc0 ID |
6261 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6262 | if (IS_ERR(device)) | |
7cbd8a83 | 6263 | return NULL; |
12bd2fc0 ID |
6264 | |
6265 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6266 | device->fs_devices = fs_devices; |
dfe25020 | 6267 | fs_devices->num_devices++; |
12bd2fc0 ID |
6268 | |
6269 | device->missing = 1; | |
cd02dca5 | 6270 | fs_devices->missing_devices++; |
12bd2fc0 | 6271 | |
dfe25020 CM |
6272 | return device; |
6273 | } | |
6274 | ||
12bd2fc0 ID |
6275 | /** |
6276 | * btrfs_alloc_device - allocate struct btrfs_device | |
6277 | * @fs_info: used only for generating a new devid, can be NULL if | |
6278 | * devid is provided (i.e. @devid != NULL). | |
6279 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6280 | * is generated. | |
6281 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6282 | * is generated. | |
6283 | * | |
6284 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6285 | * on error. Returned struct is not linked onto any lists and can be | |
6286 | * destroyed with kfree() right away. | |
6287 | */ | |
6288 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6289 | const u64 *devid, | |
6290 | const u8 *uuid) | |
6291 | { | |
6292 | struct btrfs_device *dev; | |
6293 | u64 tmp; | |
6294 | ||
fae7f21c | 6295 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6296 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6297 | |
6298 | dev = __alloc_device(); | |
6299 | if (IS_ERR(dev)) | |
6300 | return dev; | |
6301 | ||
6302 | if (devid) | |
6303 | tmp = *devid; | |
6304 | else { | |
6305 | int ret; | |
6306 | ||
6307 | ret = find_next_devid(fs_info, &tmp); | |
6308 | if (ret) { | |
6309 | kfree(dev); | |
6310 | return ERR_PTR(ret); | |
6311 | } | |
6312 | } | |
6313 | dev->devid = tmp; | |
6314 | ||
6315 | if (uuid) | |
6316 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6317 | else | |
6318 | generate_random_uuid(dev->uuid); | |
6319 | ||
9e0af237 LB |
6320 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6321 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6322 | |
6323 | return dev; | |
6324 | } | |
6325 | ||
e06cd3dd | 6326 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6327 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6328 | struct extent_buffer *leaf, |
6329 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6330 | { |
0b86a832 | 6331 | u64 length; |
f04b772b | 6332 | u64 stripe_len; |
e06cd3dd LB |
6333 | u16 num_stripes; |
6334 | u16 sub_stripes; | |
6335 | u64 type; | |
0b86a832 | 6336 | |
e17cade2 | 6337 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6338 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6339 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6340 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6341 | type = btrfs_chunk_type(leaf, chunk); | |
6342 | ||
f04b772b | 6343 | if (!num_stripes) { |
0b246afa | 6344 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6345 | num_stripes); |
6346 | return -EIO; | |
6347 | } | |
0b246afa JM |
6348 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6349 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6350 | return -EIO; |
6351 | } | |
0b246afa JM |
6352 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6353 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6354 | btrfs_chunk_sector_size(leaf, chunk)); |
6355 | return -EIO; | |
6356 | } | |
0b246afa JM |
6357 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6358 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6359 | return -EIO; |
6360 | } | |
3d8da678 | 6361 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6362 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6363 | stripe_len); |
6364 | return -EIO; | |
6365 | } | |
6366 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6367 | type) { |
0b246afa | 6368 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6369 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6370 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6371 | btrfs_chunk_type(leaf, chunk)); | |
6372 | return -EIO; | |
6373 | } | |
e06cd3dd LB |
6374 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6375 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6376 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6377 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6378 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6379 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6380 | num_stripes != 1)) { | |
0b246afa | 6381 | btrfs_err(fs_info, |
e06cd3dd LB |
6382 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6383 | num_stripes, sub_stripes, | |
6384 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6385 | return -EIO; | |
6386 | } | |
6387 | ||
6388 | return 0; | |
6389 | } | |
6390 | ||
2ff7e61e | 6391 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6392 | struct extent_buffer *leaf, |
6393 | struct btrfs_chunk *chunk) | |
6394 | { | |
0b246afa | 6395 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6396 | struct map_lookup *map; |
6397 | struct extent_map *em; | |
6398 | u64 logical; | |
6399 | u64 length; | |
e06cd3dd LB |
6400 | u64 devid; |
6401 | u8 uuid[BTRFS_UUID_SIZE]; | |
6402 | int num_stripes; | |
6403 | int ret; | |
6404 | int i; | |
6405 | ||
6406 | logical = key->offset; | |
6407 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6408 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6409 | ||
2ff7e61e | 6410 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6411 | if (ret) |
6412 | return ret; | |
a061fc8d | 6413 | |
890871be | 6414 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6415 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6416 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6417 | |
6418 | /* already mapped? */ | |
6419 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6420 | free_extent_map(em); | |
0b86a832 CM |
6421 | return 0; |
6422 | } else if (em) { | |
6423 | free_extent_map(em); | |
6424 | } | |
0b86a832 | 6425 | |
172ddd60 | 6426 | em = alloc_extent_map(); |
0b86a832 CM |
6427 | if (!em) |
6428 | return -ENOMEM; | |
593060d7 | 6429 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6430 | if (!map) { |
6431 | free_extent_map(em); | |
6432 | return -ENOMEM; | |
6433 | } | |
6434 | ||
298a8f9c | 6435 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6436 | em->map_lookup = map; |
0b86a832 CM |
6437 | em->start = logical; |
6438 | em->len = length; | |
70c8a91c | 6439 | em->orig_start = 0; |
0b86a832 | 6440 | em->block_start = 0; |
c8b97818 | 6441 | em->block_len = em->len; |
0b86a832 | 6442 | |
593060d7 CM |
6443 | map->num_stripes = num_stripes; |
6444 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6445 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6446 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6447 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6448 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6449 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6450 | for (i = 0; i < num_stripes; i++) { |
6451 | map->stripes[i].physical = | |
6452 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6453 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6454 | read_extent_buffer(leaf, uuid, (unsigned long) |
6455 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6456 | BTRFS_UUID_SIZE); | |
0b246afa | 6457 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6458 | uuid, NULL); |
3cdde224 | 6459 | if (!map->stripes[i].dev && |
0b246afa | 6460 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6461 | free_extent_map(em); |
c5502451 | 6462 | btrfs_report_missing_device(fs_info, devid, uuid); |
593060d7 CM |
6463 | return -EIO; |
6464 | } | |
dfe25020 CM |
6465 | if (!map->stripes[i].dev) { |
6466 | map->stripes[i].dev = | |
2ff7e61e JM |
6467 | add_missing_dev(fs_info->fs_devices, devid, |
6468 | uuid); | |
dfe25020 | 6469 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6470 | free_extent_map(em); |
6471 | return -EIO; | |
6472 | } | |
c5502451 | 6473 | btrfs_report_missing_device(fs_info, devid, uuid); |
dfe25020 CM |
6474 | } |
6475 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6476 | } |
6477 | ||
890871be | 6478 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6479 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6480 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6481 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6482 | free_extent_map(em); |
6483 | ||
6484 | return 0; | |
6485 | } | |
6486 | ||
143bede5 | 6487 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6488 | struct btrfs_dev_item *dev_item, |
6489 | struct btrfs_device *device) | |
6490 | { | |
6491 | unsigned long ptr; | |
0b86a832 CM |
6492 | |
6493 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6494 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6495 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6496 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6497 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6498 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6499 | device->type = btrfs_device_type(leaf, dev_item); |
6500 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6501 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6502 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6503 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6504 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6505 | |
410ba3a2 | 6506 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6507 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6508 | } |
6509 | ||
2ff7e61e | 6510 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6511 | u8 *fsid) |
2b82032c YZ |
6512 | { |
6513 | struct btrfs_fs_devices *fs_devices; | |
6514 | int ret; | |
6515 | ||
b367e47f | 6516 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2dfeca9b | 6517 | ASSERT(fsid); |
2b82032c | 6518 | |
0b246afa | 6519 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6520 | while (fs_devices) { |
5f375835 MX |
6521 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6522 | return fs_devices; | |
6523 | ||
2b82032c YZ |
6524 | fs_devices = fs_devices->seed; |
6525 | } | |
6526 | ||
6527 | fs_devices = find_fsid(fsid); | |
6528 | if (!fs_devices) { | |
0b246afa | 6529 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6530 | return ERR_PTR(-ENOENT); |
6531 | ||
6532 | fs_devices = alloc_fs_devices(fsid); | |
6533 | if (IS_ERR(fs_devices)) | |
6534 | return fs_devices; | |
6535 | ||
6536 | fs_devices->seeding = 1; | |
6537 | fs_devices->opened = 1; | |
6538 | return fs_devices; | |
2b82032c | 6539 | } |
e4404d6e YZ |
6540 | |
6541 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6542 | if (IS_ERR(fs_devices)) |
6543 | return fs_devices; | |
2b82032c | 6544 | |
97288f2c | 6545 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
0b246afa | 6546 | fs_info->bdev_holder); |
48d28232 JL |
6547 | if (ret) { |
6548 | free_fs_devices(fs_devices); | |
5f375835 | 6549 | fs_devices = ERR_PTR(ret); |
2b82032c | 6550 | goto out; |
48d28232 | 6551 | } |
2b82032c YZ |
6552 | |
6553 | if (!fs_devices->seeding) { | |
6554 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6555 | free_fs_devices(fs_devices); |
5f375835 | 6556 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6557 | goto out; |
6558 | } | |
6559 | ||
0b246afa JM |
6560 | fs_devices->seed = fs_info->fs_devices->seed; |
6561 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6562 | out: |
5f375835 | 6563 | return fs_devices; |
2b82032c YZ |
6564 | } |
6565 | ||
2ff7e61e | 6566 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6567 | struct extent_buffer *leaf, |
6568 | struct btrfs_dev_item *dev_item) | |
6569 | { | |
0b246afa | 6570 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6571 | struct btrfs_device *device; |
6572 | u64 devid; | |
6573 | int ret; | |
2b82032c | 6574 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6575 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6576 | ||
0b86a832 | 6577 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6578 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6579 | BTRFS_UUID_SIZE); |
1473b24e | 6580 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6581 | BTRFS_UUID_SIZE); |
6582 | ||
0b246afa | 6583 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) { |
2ff7e61e | 6584 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6585 | if (IS_ERR(fs_devices)) |
6586 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6587 | } |
6588 | ||
0b246afa | 6589 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6590 | if (!device) { |
c5502451 QW |
6591 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6592 | btrfs_report_missing_device(fs_info, devid, dev_uuid); | |
2b82032c | 6593 | return -EIO; |
c5502451 | 6594 | } |
2b82032c | 6595 | |
2ff7e61e | 6596 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
5f375835 MX |
6597 | if (!device) |
6598 | return -ENOMEM; | |
c5502451 | 6599 | btrfs_report_missing_device(fs_info, devid, dev_uuid); |
5f375835 | 6600 | } else { |
c5502451 QW |
6601 | if (!device->bdev) { |
6602 | btrfs_report_missing_device(fs_info, devid, dev_uuid); | |
6603 | if (!btrfs_test_opt(fs_info, DEGRADED)) | |
6604 | return -EIO; | |
6605 | } | |
5f375835 MX |
6606 | |
6607 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6608 | /* |
6609 | * this happens when a device that was properly setup | |
6610 | * in the device info lists suddenly goes bad. | |
6611 | * device->bdev is NULL, and so we have to set | |
6612 | * device->missing to one here | |
6613 | */ | |
5f375835 | 6614 | device->fs_devices->missing_devices++; |
cd02dca5 | 6615 | device->missing = 1; |
2b82032c | 6616 | } |
5f375835 MX |
6617 | |
6618 | /* Move the device to its own fs_devices */ | |
6619 | if (device->fs_devices != fs_devices) { | |
6620 | ASSERT(device->missing); | |
6621 | ||
6622 | list_move(&device->dev_list, &fs_devices->devices); | |
6623 | device->fs_devices->num_devices--; | |
6624 | fs_devices->num_devices++; | |
6625 | ||
6626 | device->fs_devices->missing_devices--; | |
6627 | fs_devices->missing_devices++; | |
6628 | ||
6629 | device->fs_devices = fs_devices; | |
6630 | } | |
2b82032c YZ |
6631 | } |
6632 | ||
0b246afa | 6633 | if (device->fs_devices != fs_info->fs_devices) { |
2b82032c YZ |
6634 | BUG_ON(device->writeable); |
6635 | if (device->generation != | |
6636 | btrfs_device_generation(leaf, dev_item)) | |
6637 | return -EINVAL; | |
6324fbf3 | 6638 | } |
0b86a832 CM |
6639 | |
6640 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6641 | device->in_fs_metadata = 1; |
63a212ab | 6642 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6643 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6644 | atomic64_add(device->total_bytes - device->bytes_used, |
6645 | &fs_info->free_chunk_space); | |
2bf64758 | 6646 | } |
0b86a832 | 6647 | ret = 0; |
0b86a832 CM |
6648 | return ret; |
6649 | } | |
6650 | ||
6bccf3ab | 6651 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6652 | { |
6bccf3ab | 6653 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6654 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6655 | struct extent_buffer *sb; |
0b86a832 | 6656 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6657 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6658 | u8 *array_ptr; |
6659 | unsigned long sb_array_offset; | |
84eed90f | 6660 | int ret = 0; |
0b86a832 CM |
6661 | u32 num_stripes; |
6662 | u32 array_size; | |
6663 | u32 len = 0; | |
1ffb22cf | 6664 | u32 cur_offset; |
e06cd3dd | 6665 | u64 type; |
84eed90f | 6666 | struct btrfs_key key; |
0b86a832 | 6667 | |
0b246afa | 6668 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6669 | /* |
6670 | * This will create extent buffer of nodesize, superblock size is | |
6671 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6672 | * overallocate but we can keep it as-is, only the first page is used. | |
6673 | */ | |
2ff7e61e | 6674 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6675 | if (IS_ERR(sb)) |
6676 | return PTR_ERR(sb); | |
4db8c528 | 6677 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6678 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6679 | /* |
01327610 | 6680 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6681 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6682 | * pages up-to-date when the page is larger: extent does not cover the |
6683 | * whole page and consequently check_page_uptodate does not find all | |
6684 | * the page's extents up-to-date (the hole beyond sb), | |
6685 | * write_extent_buffer then triggers a WARN_ON. | |
6686 | * | |
6687 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6688 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6689 | * to silence the warning eg. on PowerPC 64. | |
6690 | */ | |
09cbfeaf | 6691 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6692 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6693 | |
a061fc8d | 6694 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6695 | array_size = btrfs_super_sys_array_size(super_copy); |
6696 | ||
1ffb22cf DS |
6697 | array_ptr = super_copy->sys_chunk_array; |
6698 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6699 | cur_offset = 0; | |
0b86a832 | 6700 | |
1ffb22cf DS |
6701 | while (cur_offset < array_size) { |
6702 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6703 | len = sizeof(*disk_key); |
6704 | if (cur_offset + len > array_size) | |
6705 | goto out_short_read; | |
6706 | ||
0b86a832 CM |
6707 | btrfs_disk_key_to_cpu(&key, disk_key); |
6708 | ||
1ffb22cf DS |
6709 | array_ptr += len; |
6710 | sb_array_offset += len; | |
6711 | cur_offset += len; | |
0b86a832 | 6712 | |
0d81ba5d | 6713 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6714 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6715 | /* |
6716 | * At least one btrfs_chunk with one stripe must be | |
6717 | * present, exact stripe count check comes afterwards | |
6718 | */ | |
6719 | len = btrfs_chunk_item_size(1); | |
6720 | if (cur_offset + len > array_size) | |
6721 | goto out_short_read; | |
6722 | ||
6723 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6724 | if (!num_stripes) { |
ab8d0fc4 JM |
6725 | btrfs_err(fs_info, |
6726 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6727 | num_stripes, cur_offset); |
6728 | ret = -EIO; | |
6729 | break; | |
6730 | } | |
6731 | ||
e06cd3dd LB |
6732 | type = btrfs_chunk_type(sb, chunk); |
6733 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6734 | btrfs_err(fs_info, |
e06cd3dd LB |
6735 | "invalid chunk type %llu in sys_array at offset %u", |
6736 | type, cur_offset); | |
6737 | ret = -EIO; | |
6738 | break; | |
6739 | } | |
6740 | ||
e3540eab DS |
6741 | len = btrfs_chunk_item_size(num_stripes); |
6742 | if (cur_offset + len > array_size) | |
6743 | goto out_short_read; | |
6744 | ||
2ff7e61e | 6745 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6746 | if (ret) |
6747 | break; | |
0b86a832 | 6748 | } else { |
ab8d0fc4 JM |
6749 | btrfs_err(fs_info, |
6750 | "unexpected item type %u in sys_array at offset %u", | |
6751 | (u32)key.type, cur_offset); | |
84eed90f CM |
6752 | ret = -EIO; |
6753 | break; | |
0b86a832 | 6754 | } |
1ffb22cf DS |
6755 | array_ptr += len; |
6756 | sb_array_offset += len; | |
6757 | cur_offset += len; | |
0b86a832 | 6758 | } |
d865177a | 6759 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6760 | free_extent_buffer_stale(sb); |
84eed90f | 6761 | return ret; |
e3540eab DS |
6762 | |
6763 | out_short_read: | |
ab8d0fc4 | 6764 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6765 | len, cur_offset); |
d865177a | 6766 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6767 | free_extent_buffer_stale(sb); |
e3540eab | 6768 | return -EIO; |
0b86a832 CM |
6769 | } |
6770 | ||
c5502451 QW |
6771 | void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, u64 devid, |
6772 | u8 *uuid) | |
6773 | { | |
6774 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", devid, uuid); | |
6775 | } | |
6776 | ||
21634a19 QW |
6777 | /* |
6778 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6779 | * | |
6780 | * Return true if all chunks meet the minimal RW mount requirements. | |
6781 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6782 | */ | |
6783 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info) | |
6784 | { | |
6785 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
6786 | struct extent_map *em; | |
6787 | u64 next_start = 0; | |
6788 | bool ret = true; | |
6789 | ||
6790 | read_lock(&map_tree->map_tree.lock); | |
6791 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
6792 | read_unlock(&map_tree->map_tree.lock); | |
6793 | /* No chunk at all? Return false anyway */ | |
6794 | if (!em) { | |
6795 | ret = false; | |
6796 | goto out; | |
6797 | } | |
6798 | while (em) { | |
6799 | struct map_lookup *map; | |
6800 | int missing = 0; | |
6801 | int max_tolerated; | |
6802 | int i; | |
6803 | ||
6804 | map = em->map_lookup; | |
6805 | max_tolerated = | |
6806 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6807 | map->type); | |
6808 | for (i = 0; i < map->num_stripes; i++) { | |
6809 | struct btrfs_device *dev = map->stripes[i].dev; | |
6810 | ||
6811 | if (!dev || !dev->bdev || dev->missing || | |
6812 | dev->last_flush_error) | |
6813 | missing++; | |
6814 | } | |
6815 | if (missing > max_tolerated) { | |
6816 | btrfs_warn(fs_info, | |
6817 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", | |
6818 | em->start, missing, max_tolerated); | |
6819 | free_extent_map(em); | |
6820 | ret = false; | |
6821 | goto out; | |
6822 | } | |
6823 | next_start = extent_map_end(em); | |
6824 | free_extent_map(em); | |
6825 | ||
6826 | read_lock(&map_tree->map_tree.lock); | |
6827 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
6828 | (u64)(-1) - next_start); | |
6829 | read_unlock(&map_tree->map_tree.lock); | |
6830 | } | |
6831 | out: | |
6832 | return ret; | |
6833 | } | |
6834 | ||
5b4aacef | 6835 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6836 | { |
5b4aacef | 6837 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6838 | struct btrfs_path *path; |
6839 | struct extent_buffer *leaf; | |
6840 | struct btrfs_key key; | |
6841 | struct btrfs_key found_key; | |
6842 | int ret; | |
6843 | int slot; | |
99e3ecfc | 6844 | u64 total_dev = 0; |
0b86a832 | 6845 | |
0b86a832 CM |
6846 | path = btrfs_alloc_path(); |
6847 | if (!path) | |
6848 | return -ENOMEM; | |
6849 | ||
b367e47f | 6850 | mutex_lock(&uuid_mutex); |
34441361 | 6851 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6852 | |
395927a9 FDBM |
6853 | /* |
6854 | * Read all device items, and then all the chunk items. All | |
6855 | * device items are found before any chunk item (their object id | |
6856 | * is smaller than the lowest possible object id for a chunk | |
6857 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6858 | */ |
6859 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6860 | key.offset = 0; | |
6861 | key.type = 0; | |
0b86a832 | 6862 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6863 | if (ret < 0) |
6864 | goto error; | |
d397712b | 6865 | while (1) { |
0b86a832 CM |
6866 | leaf = path->nodes[0]; |
6867 | slot = path->slots[0]; | |
6868 | if (slot >= btrfs_header_nritems(leaf)) { | |
6869 | ret = btrfs_next_leaf(root, path); | |
6870 | if (ret == 0) | |
6871 | continue; | |
6872 | if (ret < 0) | |
6873 | goto error; | |
6874 | break; | |
6875 | } | |
6876 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6877 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6878 | struct btrfs_dev_item *dev_item; | |
6879 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6880 | struct btrfs_dev_item); |
2ff7e61e | 6881 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
6882 | if (ret) |
6883 | goto error; | |
99e3ecfc | 6884 | total_dev++; |
0b86a832 CM |
6885 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6886 | struct btrfs_chunk *chunk; | |
6887 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 6888 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
6889 | if (ret) |
6890 | goto error; | |
0b86a832 CM |
6891 | } |
6892 | path->slots[0]++; | |
6893 | } | |
99e3ecfc LB |
6894 | |
6895 | /* | |
6896 | * After loading chunk tree, we've got all device information, | |
6897 | * do another round of validation checks. | |
6898 | */ | |
0b246afa JM |
6899 | if (total_dev != fs_info->fs_devices->total_devices) { |
6900 | btrfs_err(fs_info, | |
99e3ecfc | 6901 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 6902 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
6903 | total_dev); |
6904 | ret = -EINVAL; | |
6905 | goto error; | |
6906 | } | |
0b246afa JM |
6907 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
6908 | fs_info->fs_devices->total_rw_bytes) { | |
6909 | btrfs_err(fs_info, | |
99e3ecfc | 6910 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
6911 | btrfs_super_total_bytes(fs_info->super_copy), |
6912 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
6913 | ret = -EINVAL; |
6914 | goto error; | |
6915 | } | |
0b86a832 CM |
6916 | ret = 0; |
6917 | error: | |
34441361 | 6918 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
6919 | mutex_unlock(&uuid_mutex); |
6920 | ||
2b82032c | 6921 | btrfs_free_path(path); |
0b86a832 CM |
6922 | return ret; |
6923 | } | |
442a4f63 | 6924 | |
cb517eab MX |
6925 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6926 | { | |
6927 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6928 | struct btrfs_device *device; | |
6929 | ||
29cc83f6 LB |
6930 | while (fs_devices) { |
6931 | mutex_lock(&fs_devices->device_list_mutex); | |
6932 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 6933 | device->fs_info = fs_info; |
29cc83f6 LB |
6934 | mutex_unlock(&fs_devices->device_list_mutex); |
6935 | ||
6936 | fs_devices = fs_devices->seed; | |
6937 | } | |
cb517eab MX |
6938 | } |
6939 | ||
733f4fbb SB |
6940 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6941 | { | |
6942 | int i; | |
6943 | ||
6944 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6945 | btrfs_dev_stat_reset(dev, i); | |
6946 | } | |
6947 | ||
6948 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6949 | { | |
6950 | struct btrfs_key key; | |
6951 | struct btrfs_key found_key; | |
6952 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6953 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6954 | struct extent_buffer *eb; | |
6955 | int slot; | |
6956 | int ret = 0; | |
6957 | struct btrfs_device *device; | |
6958 | struct btrfs_path *path = NULL; | |
6959 | int i; | |
6960 | ||
6961 | path = btrfs_alloc_path(); | |
6962 | if (!path) { | |
6963 | ret = -ENOMEM; | |
6964 | goto out; | |
6965 | } | |
6966 | ||
6967 | mutex_lock(&fs_devices->device_list_mutex); | |
6968 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6969 | int item_size; | |
6970 | struct btrfs_dev_stats_item *ptr; | |
6971 | ||
242e2956 DS |
6972 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6973 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6974 | key.offset = device->devid; |
6975 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6976 | if (ret) { | |
733f4fbb SB |
6977 | __btrfs_reset_dev_stats(device); |
6978 | device->dev_stats_valid = 1; | |
6979 | btrfs_release_path(path); | |
6980 | continue; | |
6981 | } | |
6982 | slot = path->slots[0]; | |
6983 | eb = path->nodes[0]; | |
6984 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6985 | item_size = btrfs_item_size_nr(eb, slot); | |
6986 | ||
6987 | ptr = btrfs_item_ptr(eb, slot, | |
6988 | struct btrfs_dev_stats_item); | |
6989 | ||
6990 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6991 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6992 | btrfs_dev_stat_set(device, i, | |
6993 | btrfs_dev_stats_value(eb, ptr, i)); | |
6994 | else | |
6995 | btrfs_dev_stat_reset(device, i); | |
6996 | } | |
6997 | ||
6998 | device->dev_stats_valid = 1; | |
6999 | btrfs_dev_stat_print_on_load(device); | |
7000 | btrfs_release_path(path); | |
7001 | } | |
7002 | mutex_unlock(&fs_devices->device_list_mutex); | |
7003 | ||
7004 | out: | |
7005 | btrfs_free_path(path); | |
7006 | return ret < 0 ? ret : 0; | |
7007 | } | |
7008 | ||
7009 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6bccf3ab | 7010 | struct btrfs_fs_info *fs_info, |
733f4fbb SB |
7011 | struct btrfs_device *device) |
7012 | { | |
6bccf3ab | 7013 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7014 | struct btrfs_path *path; |
7015 | struct btrfs_key key; | |
7016 | struct extent_buffer *eb; | |
7017 | struct btrfs_dev_stats_item *ptr; | |
7018 | int ret; | |
7019 | int i; | |
7020 | ||
242e2956 DS |
7021 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7022 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7023 | key.offset = device->devid; |
7024 | ||
7025 | path = btrfs_alloc_path(); | |
fa252992 DS |
7026 | if (!path) |
7027 | return -ENOMEM; | |
733f4fbb SB |
7028 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7029 | if (ret < 0) { | |
0b246afa | 7030 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7031 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7032 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7033 | goto out; |
7034 | } | |
7035 | ||
7036 | if (ret == 0 && | |
7037 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7038 | /* need to delete old one and insert a new one */ | |
7039 | ret = btrfs_del_item(trans, dev_root, path); | |
7040 | if (ret != 0) { | |
0b246afa | 7041 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7042 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7043 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7044 | goto out; |
7045 | } | |
7046 | ret = 1; | |
7047 | } | |
7048 | ||
7049 | if (ret == 1) { | |
7050 | /* need to insert a new item */ | |
7051 | btrfs_release_path(path); | |
7052 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7053 | &key, sizeof(*ptr)); | |
7054 | if (ret < 0) { | |
0b246afa | 7055 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7056 | "insert dev_stats item for device %s failed %d", |
7057 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7058 | goto out; |
7059 | } | |
7060 | } | |
7061 | ||
7062 | eb = path->nodes[0]; | |
7063 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7064 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7065 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7066 | btrfs_dev_stat_read(device, i)); | |
7067 | btrfs_mark_buffer_dirty(eb); | |
7068 | ||
7069 | out: | |
7070 | btrfs_free_path(path); | |
7071 | return ret; | |
7072 | } | |
7073 | ||
7074 | /* | |
7075 | * called from commit_transaction. Writes all changed device stats to disk. | |
7076 | */ | |
7077 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7078 | struct btrfs_fs_info *fs_info) | |
7079 | { | |
733f4fbb SB |
7080 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7081 | struct btrfs_device *device; | |
addc3fa7 | 7082 | int stats_cnt; |
733f4fbb SB |
7083 | int ret = 0; |
7084 | ||
7085 | mutex_lock(&fs_devices->device_list_mutex); | |
7086 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 7087 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
7088 | continue; |
7089 | ||
addc3fa7 | 7090 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
6bccf3ab | 7091 | ret = update_dev_stat_item(trans, fs_info, device); |
733f4fbb | 7092 | if (!ret) |
addc3fa7 | 7093 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7094 | } |
7095 | mutex_unlock(&fs_devices->device_list_mutex); | |
7096 | ||
7097 | return ret; | |
7098 | } | |
7099 | ||
442a4f63 SB |
7100 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7101 | { | |
7102 | btrfs_dev_stat_inc(dev, index); | |
7103 | btrfs_dev_stat_print_on_error(dev); | |
7104 | } | |
7105 | ||
48a3b636 | 7106 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7107 | { |
733f4fbb SB |
7108 | if (!dev->dev_stats_valid) |
7109 | return; | |
fb456252 | 7110 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7111 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7112 | rcu_str_deref(dev->name), |
442a4f63 SB |
7113 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7114 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7115 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7116 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7117 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7118 | } |
c11d2c23 | 7119 | |
733f4fbb SB |
7120 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7121 | { | |
a98cdb85 SB |
7122 | int i; |
7123 | ||
7124 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7125 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7126 | break; | |
7127 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7128 | return; /* all values == 0, suppress message */ | |
7129 | ||
fb456252 | 7130 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7131 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7132 | rcu_str_deref(dev->name), |
733f4fbb SB |
7133 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7134 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7135 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7136 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7137 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7138 | } | |
7139 | ||
2ff7e61e | 7140 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7141 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7142 | { |
7143 | struct btrfs_device *dev; | |
0b246afa | 7144 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7145 | int i; |
7146 | ||
7147 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7148 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7149 | mutex_unlock(&fs_devices->device_list_mutex); |
7150 | ||
7151 | if (!dev) { | |
0b246afa | 7152 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7153 | return -ENODEV; |
733f4fbb | 7154 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7155 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7156 | return -ENODEV; |
b27f7c0c | 7157 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7158 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7159 | if (stats->nr_items > i) | |
7160 | stats->values[i] = | |
7161 | btrfs_dev_stat_read_and_reset(dev, i); | |
7162 | else | |
7163 | btrfs_dev_stat_reset(dev, i); | |
7164 | } | |
7165 | } else { | |
7166 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7167 | if (stats->nr_items > i) | |
7168 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7169 | } | |
7170 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7171 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7172 | return 0; | |
7173 | } | |
a8a6dab7 | 7174 | |
da353f6b | 7175 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7176 | { |
7177 | struct buffer_head *bh; | |
7178 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7179 | int copy_num; |
a8a6dab7 | 7180 | |
12b1c263 AJ |
7181 | if (!bdev) |
7182 | return; | |
a8a6dab7 | 7183 | |
12b1c263 AJ |
7184 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7185 | copy_num++) { | |
a8a6dab7 | 7186 | |
12b1c263 AJ |
7187 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7188 | continue; | |
7189 | ||
7190 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7191 | ||
7192 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7193 | set_buffer_dirty(bh); | |
7194 | sync_dirty_buffer(bh); | |
7195 | brelse(bh); | |
7196 | } | |
7197 | ||
7198 | /* Notify udev that device has changed */ | |
7199 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7200 | ||
7201 | /* Update ctime/mtime for device path for libblkid */ | |
7202 | update_dev_time(device_path); | |
a8a6dab7 | 7203 | } |
935e5cc9 MX |
7204 | |
7205 | /* | |
7206 | * Update the size of all devices, which is used for writing out the | |
7207 | * super blocks. | |
7208 | */ | |
7209 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7210 | { | |
7211 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7212 | struct btrfs_device *curr, *next; | |
7213 | ||
7214 | if (list_empty(&fs_devices->resized_devices)) | |
7215 | return; | |
7216 | ||
7217 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7218 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7219 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7220 | resized_list) { | |
7221 | list_del_init(&curr->resized_list); | |
7222 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7223 | } | |
34441361 | 7224 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7225 | mutex_unlock(&fs_devices->device_list_mutex); |
7226 | } | |
ce7213c7 MX |
7227 | |
7228 | /* Must be invoked during the transaction commit */ | |
2ff7e61e | 7229 | void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info, |
ce7213c7 MX |
7230 | struct btrfs_transaction *transaction) |
7231 | { | |
7232 | struct extent_map *em; | |
7233 | struct map_lookup *map; | |
7234 | struct btrfs_device *dev; | |
7235 | int i; | |
7236 | ||
7237 | if (list_empty(&transaction->pending_chunks)) | |
7238 | return; | |
7239 | ||
7240 | /* In order to kick the device replace finish process */ | |
34441361 | 7241 | mutex_lock(&fs_info->chunk_mutex); |
ce7213c7 | 7242 | list_for_each_entry(em, &transaction->pending_chunks, list) { |
95617d69 | 7243 | map = em->map_lookup; |
ce7213c7 MX |
7244 | |
7245 | for (i = 0; i < map->num_stripes; i++) { | |
7246 | dev = map->stripes[i].dev; | |
7247 | dev->commit_bytes_used = dev->bytes_used; | |
7248 | } | |
7249 | } | |
34441361 | 7250 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7251 | } |
5a13f430 AJ |
7252 | |
7253 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7254 | { | |
7255 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7256 | while (fs_devices) { | |
7257 | fs_devices->fs_info = fs_info; | |
7258 | fs_devices = fs_devices->seed; | |
7259 | } | |
7260 | } | |
7261 | ||
7262 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7263 | { | |
7264 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7265 | while (fs_devices) { | |
7266 | fs_devices->fs_info = NULL; | |
7267 | fs_devices = fs_devices->seed; | |
7268 | } | |
7269 | } |