]>
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 YZ |
136 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
137 | struct btrfs_root *root, | |
138 | struct btrfs_device *device); | |
139 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 140 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 141 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 142 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 143 | |
67a2c45e | 144 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 145 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
146 | struct list_head *btrfs_get_fs_uuids(void) |
147 | { | |
148 | return &fs_uuids; | |
149 | } | |
8a4b83cc | 150 | |
2208a378 ID |
151 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
152 | { | |
153 | struct btrfs_fs_devices *fs_devs; | |
154 | ||
78f2c9e6 | 155 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
156 | if (!fs_devs) |
157 | return ERR_PTR(-ENOMEM); | |
158 | ||
159 | mutex_init(&fs_devs->device_list_mutex); | |
160 | ||
161 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 162 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
163 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
164 | INIT_LIST_HEAD(&fs_devs->list); | |
165 | ||
166 | return fs_devs; | |
167 | } | |
168 | ||
169 | /** | |
170 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
171 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
172 | * generated. | |
173 | * | |
174 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
175 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
176 | * can be destroyed with kfree() right away. | |
177 | */ | |
178 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
179 | { | |
180 | struct btrfs_fs_devices *fs_devs; | |
181 | ||
182 | fs_devs = __alloc_fs_devices(); | |
183 | if (IS_ERR(fs_devs)) | |
184 | return fs_devs; | |
185 | ||
186 | if (fsid) | |
187 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
188 | else | |
189 | generate_random_uuid(fs_devs->fsid); | |
190 | ||
191 | return fs_devs; | |
192 | } | |
193 | ||
e4404d6e YZ |
194 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
195 | { | |
196 | struct btrfs_device *device; | |
197 | WARN_ON(fs_devices->opened); | |
198 | while (!list_empty(&fs_devices->devices)) { | |
199 | device = list_entry(fs_devices->devices.next, | |
200 | struct btrfs_device, dev_list); | |
201 | list_del(&device->dev_list); | |
606686ee | 202 | rcu_string_free(device->name); |
e4404d6e YZ |
203 | kfree(device); |
204 | } | |
205 | kfree(fs_devices); | |
206 | } | |
207 | ||
b8b8ff59 LC |
208 | static void btrfs_kobject_uevent(struct block_device *bdev, |
209 | enum kobject_action action) | |
210 | { | |
211 | int ret; | |
212 | ||
213 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
214 | if (ret) | |
efe120a0 | 215 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
216 | action, |
217 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
218 | &disk_to_dev(bdev->bd_disk)->kobj); | |
219 | } | |
220 | ||
143bede5 | 221 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
222 | { |
223 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 224 | |
2b82032c YZ |
225 | while (!list_empty(&fs_uuids)) { |
226 | fs_devices = list_entry(fs_uuids.next, | |
227 | struct btrfs_fs_devices, list); | |
228 | list_del(&fs_devices->list); | |
e4404d6e | 229 | free_fs_devices(fs_devices); |
8a4b83cc | 230 | } |
8a4b83cc CM |
231 | } |
232 | ||
12bd2fc0 ID |
233 | static struct btrfs_device *__alloc_device(void) |
234 | { | |
235 | struct btrfs_device *dev; | |
236 | ||
78f2c9e6 | 237 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
238 | if (!dev) |
239 | return ERR_PTR(-ENOMEM); | |
240 | ||
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); |
d0164adc MG |
251 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
252 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
12bd2fc0 ID |
253 | |
254 | return dev; | |
255 | } | |
256 | ||
a1b32a59 CM |
257 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
258 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
259 | { |
260 | struct btrfs_device *dev; | |
8a4b83cc | 261 | |
c6e30871 | 262 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 263 | if (dev->devid == devid && |
8f18cf13 | 264 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 265 | return dev; |
a443755f | 266 | } |
8a4b83cc CM |
267 | } |
268 | return NULL; | |
269 | } | |
270 | ||
a1b32a59 | 271 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 272 | { |
8a4b83cc CM |
273 | struct btrfs_fs_devices *fs_devices; |
274 | ||
c6e30871 | 275 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
276 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
277 | return fs_devices; | |
278 | } | |
279 | return NULL; | |
280 | } | |
281 | ||
beaf8ab3 SB |
282 | static int |
283 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
284 | int flush, struct block_device **bdev, | |
285 | struct buffer_head **bh) | |
286 | { | |
287 | int ret; | |
288 | ||
289 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
290 | ||
291 | if (IS_ERR(*bdev)) { | |
292 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
293 | goto error; |
294 | } | |
295 | ||
296 | if (flush) | |
297 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
298 | ret = set_blocksize(*bdev, 4096); | |
299 | if (ret) { | |
300 | blkdev_put(*bdev, flags); | |
301 | goto error; | |
302 | } | |
303 | invalidate_bdev(*bdev); | |
304 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
305 | if (IS_ERR(*bh)) { |
306 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
307 | blkdev_put(*bdev, flags); |
308 | goto error; | |
309 | } | |
310 | ||
311 | return 0; | |
312 | ||
313 | error: | |
314 | *bdev = NULL; | |
315 | *bh = NULL; | |
316 | return ret; | |
317 | } | |
318 | ||
ffbd517d CM |
319 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
320 | struct bio *head, struct bio *tail) | |
321 | { | |
322 | ||
323 | struct bio *old_head; | |
324 | ||
325 | old_head = pending_bios->head; | |
326 | pending_bios->head = head; | |
327 | if (pending_bios->tail) | |
328 | tail->bi_next = old_head; | |
329 | else | |
330 | pending_bios->tail = tail; | |
331 | } | |
332 | ||
8b712842 CM |
333 | /* |
334 | * we try to collect pending bios for a device so we don't get a large | |
335 | * number of procs sending bios down to the same device. This greatly | |
336 | * improves the schedulers ability to collect and merge the bios. | |
337 | * | |
338 | * But, it also turns into a long list of bios to process and that is sure | |
339 | * to eventually make the worker thread block. The solution here is to | |
340 | * make some progress and then put this work struct back at the end of | |
341 | * the list if the block device is congested. This way, multiple devices | |
342 | * can make progress from a single worker thread. | |
343 | */ | |
143bede5 | 344 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
345 | { |
346 | struct bio *pending; | |
347 | struct backing_dev_info *bdi; | |
b64a2851 | 348 | struct btrfs_fs_info *fs_info; |
ffbd517d | 349 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
350 | struct bio *tail; |
351 | struct bio *cur; | |
352 | int again = 0; | |
ffbd517d | 353 | unsigned long num_run; |
d644d8a1 | 354 | unsigned long batch_run = 0; |
b64a2851 | 355 | unsigned long limit; |
b765ead5 | 356 | unsigned long last_waited = 0; |
d84275c9 | 357 | int force_reg = 0; |
0e588859 | 358 | int sync_pending = 0; |
211588ad CM |
359 | struct blk_plug plug; |
360 | ||
361 | /* | |
362 | * this function runs all the bios we've collected for | |
363 | * a particular device. We don't want to wander off to | |
364 | * another device without first sending all of these down. | |
365 | * So, setup a plug here and finish it off before we return | |
366 | */ | |
367 | blk_start_plug(&plug); | |
8b712842 | 368 | |
bedf762b | 369 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
370 | fs_info = device->dev_root->fs_info; |
371 | limit = btrfs_async_submit_limit(fs_info); | |
372 | limit = limit * 2 / 3; | |
373 | ||
8b712842 CM |
374 | loop: |
375 | spin_lock(&device->io_lock); | |
376 | ||
a6837051 | 377 | loop_lock: |
d84275c9 | 378 | num_run = 0; |
ffbd517d | 379 | |
8b712842 CM |
380 | /* take all the bios off the list at once and process them |
381 | * later on (without the lock held). But, remember the | |
382 | * tail and other pointers so the bios can be properly reinserted | |
383 | * into the list if we hit congestion | |
384 | */ | |
d84275c9 | 385 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 386 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
387 | force_reg = 1; |
388 | } else { | |
ffbd517d | 389 | pending_bios = &device->pending_bios; |
d84275c9 CM |
390 | force_reg = 0; |
391 | } | |
ffbd517d CM |
392 | |
393 | pending = pending_bios->head; | |
394 | tail = pending_bios->tail; | |
8b712842 | 395 | WARN_ON(pending && !tail); |
8b712842 CM |
396 | |
397 | /* | |
398 | * if pending was null this time around, no bios need processing | |
399 | * at all and we can stop. Otherwise it'll loop back up again | |
400 | * and do an additional check so no bios are missed. | |
401 | * | |
402 | * device->running_pending is used to synchronize with the | |
403 | * schedule_bio code. | |
404 | */ | |
ffbd517d CM |
405 | if (device->pending_sync_bios.head == NULL && |
406 | device->pending_bios.head == NULL) { | |
8b712842 CM |
407 | again = 0; |
408 | device->running_pending = 0; | |
ffbd517d CM |
409 | } else { |
410 | again = 1; | |
411 | device->running_pending = 1; | |
8b712842 | 412 | } |
ffbd517d CM |
413 | |
414 | pending_bios->head = NULL; | |
415 | pending_bios->tail = NULL; | |
416 | ||
8b712842 CM |
417 | spin_unlock(&device->io_lock); |
418 | ||
d397712b | 419 | while (pending) { |
ffbd517d CM |
420 | |
421 | rmb(); | |
d84275c9 CM |
422 | /* we want to work on both lists, but do more bios on the |
423 | * sync list than the regular list | |
424 | */ | |
425 | if ((num_run > 32 && | |
426 | pending_bios != &device->pending_sync_bios && | |
427 | device->pending_sync_bios.head) || | |
428 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
429 | device->pending_bios.head)) { | |
ffbd517d CM |
430 | spin_lock(&device->io_lock); |
431 | requeue_list(pending_bios, pending, tail); | |
432 | goto loop_lock; | |
433 | } | |
434 | ||
8b712842 CM |
435 | cur = pending; |
436 | pending = pending->bi_next; | |
437 | cur->bi_next = NULL; | |
b64a2851 | 438 | |
ee863954 DS |
439 | /* |
440 | * atomic_dec_return implies a barrier for waitqueue_active | |
441 | */ | |
66657b31 | 442 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
443 | waitqueue_active(&fs_info->async_submit_wait)) |
444 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 445 | |
dac56212 | 446 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 447 | |
2ab1ba68 CM |
448 | /* |
449 | * if we're doing the sync list, record that our | |
450 | * plug has some sync requests on it | |
451 | * | |
452 | * If we're doing the regular list and there are | |
453 | * sync requests sitting around, unplug before | |
454 | * we add more | |
455 | */ | |
456 | if (pending_bios == &device->pending_sync_bios) { | |
457 | sync_pending = 1; | |
458 | } else if (sync_pending) { | |
459 | blk_finish_plug(&plug); | |
460 | blk_start_plug(&plug); | |
461 | sync_pending = 0; | |
462 | } | |
463 | ||
4e49ea4a | 464 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
465 | num_run++; |
466 | batch_run++; | |
853d8ec4 DS |
467 | |
468 | cond_resched(); | |
8b712842 CM |
469 | |
470 | /* | |
471 | * we made progress, there is more work to do and the bdi | |
472 | * is now congested. Back off and let other work structs | |
473 | * run instead | |
474 | */ | |
57fd5a5f | 475 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 476 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 477 | struct io_context *ioc; |
8b712842 | 478 | |
b765ead5 CM |
479 | ioc = current->io_context; |
480 | ||
481 | /* | |
482 | * the main goal here is that we don't want to | |
483 | * block if we're going to be able to submit | |
484 | * more requests without blocking. | |
485 | * | |
486 | * This code does two great things, it pokes into | |
487 | * the elevator code from a filesystem _and_ | |
488 | * it makes assumptions about how batching works. | |
489 | */ | |
490 | if (ioc && ioc->nr_batch_requests > 0 && | |
491 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
492 | (last_waited == 0 || | |
493 | ioc->last_waited == last_waited)) { | |
494 | /* | |
495 | * we want to go through our batch of | |
496 | * requests and stop. So, we copy out | |
497 | * the ioc->last_waited time and test | |
498 | * against it before looping | |
499 | */ | |
500 | last_waited = ioc->last_waited; | |
853d8ec4 | 501 | cond_resched(); |
b765ead5 CM |
502 | continue; |
503 | } | |
8b712842 | 504 | spin_lock(&device->io_lock); |
ffbd517d | 505 | requeue_list(pending_bios, pending, tail); |
a6837051 | 506 | device->running_pending = 1; |
8b712842 CM |
507 | |
508 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
509 | btrfs_queue_work(fs_info->submit_workers, |
510 | &device->work); | |
8b712842 CM |
511 | goto done; |
512 | } | |
d85c8a6f CM |
513 | /* unplug every 64 requests just for good measure */ |
514 | if (batch_run % 64 == 0) { | |
515 | blk_finish_plug(&plug); | |
516 | blk_start_plug(&plug); | |
517 | sync_pending = 0; | |
518 | } | |
8b712842 | 519 | } |
ffbd517d | 520 | |
51684082 CM |
521 | cond_resched(); |
522 | if (again) | |
523 | goto loop; | |
524 | ||
525 | spin_lock(&device->io_lock); | |
526 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
527 | goto loop_lock; | |
528 | spin_unlock(&device->io_lock); | |
529 | ||
8b712842 | 530 | done: |
211588ad | 531 | blk_finish_plug(&plug); |
8b712842 CM |
532 | } |
533 | ||
b2950863 | 534 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
535 | { |
536 | struct btrfs_device *device; | |
537 | ||
538 | device = container_of(work, struct btrfs_device, work); | |
539 | run_scheduled_bios(device); | |
540 | } | |
541 | ||
4fde46f0 AJ |
542 | |
543 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
544 | { | |
545 | struct btrfs_fs_devices *fs_devs; | |
546 | struct btrfs_device *dev; | |
547 | ||
548 | if (!cur_dev->name) | |
549 | return; | |
550 | ||
551 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
552 | int del = 1; | |
553 | ||
554 | if (fs_devs->opened) | |
555 | continue; | |
556 | if (fs_devs->seeding) | |
557 | continue; | |
558 | ||
559 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
560 | ||
561 | if (dev == cur_dev) | |
562 | continue; | |
563 | if (!dev->name) | |
564 | continue; | |
565 | ||
566 | /* | |
567 | * Todo: This won't be enough. What if the same device | |
568 | * comes back (with new uuid and) with its mapper path? | |
569 | * But for now, this does help as mostly an admin will | |
570 | * either use mapper or non mapper path throughout. | |
571 | */ | |
572 | rcu_read_lock(); | |
573 | del = strcmp(rcu_str_deref(dev->name), | |
574 | rcu_str_deref(cur_dev->name)); | |
575 | rcu_read_unlock(); | |
576 | if (!del) | |
577 | break; | |
578 | } | |
579 | ||
580 | if (!del) { | |
581 | /* delete the stale device */ | |
582 | if (fs_devs->num_devices == 1) { | |
583 | btrfs_sysfs_remove_fsid(fs_devs); | |
584 | list_del(&fs_devs->list); | |
585 | free_fs_devices(fs_devs); | |
586 | } else { | |
587 | fs_devs->num_devices--; | |
588 | list_del(&dev->dev_list); | |
589 | rcu_string_free(dev->name); | |
590 | kfree(dev); | |
591 | } | |
592 | break; | |
593 | } | |
594 | } | |
595 | } | |
596 | ||
60999ca4 DS |
597 | /* |
598 | * Add new device to list of registered devices | |
599 | * | |
600 | * Returns: | |
601 | * 1 - first time device is seen | |
602 | * 0 - device already known | |
603 | * < 0 - error | |
604 | */ | |
a1b32a59 | 605 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
606 | struct btrfs_super_block *disk_super, |
607 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
608 | { | |
609 | struct btrfs_device *device; | |
610 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 611 | struct rcu_string *name; |
60999ca4 | 612 | int ret = 0; |
8a4b83cc CM |
613 | u64 found_transid = btrfs_super_generation(disk_super); |
614 | ||
615 | fs_devices = find_fsid(disk_super->fsid); | |
616 | if (!fs_devices) { | |
2208a378 ID |
617 | fs_devices = alloc_fs_devices(disk_super->fsid); |
618 | if (IS_ERR(fs_devices)) | |
619 | return PTR_ERR(fs_devices); | |
620 | ||
8a4b83cc | 621 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 622 | |
8a4b83cc CM |
623 | device = NULL; |
624 | } else { | |
a443755f CM |
625 | device = __find_device(&fs_devices->devices, devid, |
626 | disk_super->dev_item.uuid); | |
8a4b83cc | 627 | } |
443f24fe | 628 | |
8a4b83cc | 629 | if (!device) { |
2b82032c YZ |
630 | if (fs_devices->opened) |
631 | return -EBUSY; | |
632 | ||
12bd2fc0 ID |
633 | device = btrfs_alloc_device(NULL, &devid, |
634 | disk_super->dev_item.uuid); | |
635 | if (IS_ERR(device)) { | |
8a4b83cc | 636 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 637 | return PTR_ERR(device); |
8a4b83cc | 638 | } |
606686ee JB |
639 | |
640 | name = rcu_string_strdup(path, GFP_NOFS); | |
641 | if (!name) { | |
8a4b83cc CM |
642 | kfree(device); |
643 | return -ENOMEM; | |
644 | } | |
606686ee | 645 | rcu_assign_pointer(device->name, name); |
90519d66 | 646 | |
e5e9a520 | 647 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 648 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 649 | fs_devices->num_devices++; |
e5e9a520 CM |
650 | mutex_unlock(&fs_devices->device_list_mutex); |
651 | ||
60999ca4 | 652 | ret = 1; |
2b82032c | 653 | device->fs_devices = fs_devices; |
606686ee | 654 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
655 | /* |
656 | * When FS is already mounted. | |
657 | * 1. If you are here and if the device->name is NULL that | |
658 | * means this device was missing at time of FS mount. | |
659 | * 2. If you are here and if the device->name is different | |
660 | * from 'path' that means either | |
661 | * a. The same device disappeared and reappeared with | |
662 | * different name. or | |
663 | * b. The missing-disk-which-was-replaced, has | |
664 | * reappeared now. | |
665 | * | |
666 | * We must allow 1 and 2a above. But 2b would be a spurious | |
667 | * and unintentional. | |
668 | * | |
669 | * Further in case of 1 and 2a above, the disk at 'path' | |
670 | * would have missed some transaction when it was away and | |
671 | * in case of 2a the stale bdev has to be updated as well. | |
672 | * 2b must not be allowed at all time. | |
673 | */ | |
674 | ||
675 | /* | |
0f23ae74 CM |
676 | * For now, we do allow update to btrfs_fs_device through the |
677 | * btrfs dev scan cli after FS has been mounted. We're still | |
678 | * tracking a problem where systems fail mount by subvolume id | |
679 | * when we reject replacement on a mounted FS. | |
b96de000 | 680 | */ |
0f23ae74 | 681 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
682 | /* |
683 | * That is if the FS is _not_ mounted and if you | |
684 | * are here, that means there is more than one | |
685 | * disk with same uuid and devid.We keep the one | |
686 | * with larger generation number or the last-in if | |
687 | * generation are equal. | |
688 | */ | |
0f23ae74 | 689 | return -EEXIST; |
77bdae4d | 690 | } |
b96de000 | 691 | |
606686ee | 692 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
693 | if (!name) |
694 | return -ENOMEM; | |
606686ee JB |
695 | rcu_string_free(device->name); |
696 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
697 | if (device->missing) { |
698 | fs_devices->missing_devices--; | |
699 | device->missing = 0; | |
700 | } | |
8a4b83cc CM |
701 | } |
702 | ||
77bdae4d AJ |
703 | /* |
704 | * Unmount does not free the btrfs_device struct but would zero | |
705 | * generation along with most of the other members. So just update | |
706 | * it back. We need it to pick the disk with largest generation | |
707 | * (as above). | |
708 | */ | |
709 | if (!fs_devices->opened) | |
710 | device->generation = found_transid; | |
711 | ||
4fde46f0 AJ |
712 | /* |
713 | * if there is new btrfs on an already registered device, | |
714 | * then remove the stale device entry. | |
715 | */ | |
02feae3c AJ |
716 | if (ret > 0) |
717 | btrfs_free_stale_device(device); | |
4fde46f0 | 718 | |
8a4b83cc | 719 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
720 | |
721 | return ret; | |
8a4b83cc CM |
722 | } |
723 | ||
e4404d6e YZ |
724 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
725 | { | |
726 | struct btrfs_fs_devices *fs_devices; | |
727 | struct btrfs_device *device; | |
728 | struct btrfs_device *orig_dev; | |
729 | ||
2208a378 ID |
730 | fs_devices = alloc_fs_devices(orig->fsid); |
731 | if (IS_ERR(fs_devices)) | |
732 | return fs_devices; | |
e4404d6e | 733 | |
adbbb863 | 734 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 735 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 736 | |
46224705 | 737 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 738 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
739 | struct rcu_string *name; |
740 | ||
12bd2fc0 ID |
741 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
742 | orig_dev->uuid); | |
743 | if (IS_ERR(device)) | |
e4404d6e YZ |
744 | goto error; |
745 | ||
606686ee JB |
746 | /* |
747 | * This is ok to do without rcu read locked because we hold the | |
748 | * uuid mutex so nothing we touch in here is going to disappear. | |
749 | */ | |
e755f780 | 750 | if (orig_dev->name) { |
78f2c9e6 DS |
751 | name = rcu_string_strdup(orig_dev->name->str, |
752 | GFP_KERNEL); | |
e755f780 AJ |
753 | if (!name) { |
754 | kfree(device); | |
755 | goto error; | |
756 | } | |
757 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 758 | } |
e4404d6e | 759 | |
e4404d6e YZ |
760 | list_add(&device->dev_list, &fs_devices->devices); |
761 | device->fs_devices = fs_devices; | |
762 | fs_devices->num_devices++; | |
763 | } | |
adbbb863 | 764 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
765 | return fs_devices; |
766 | error: | |
adbbb863 | 767 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
768 | free_fs_devices(fs_devices); |
769 | return ERR_PTR(-ENOMEM); | |
770 | } | |
771 | ||
9eaed21e | 772 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 773 | { |
c6e30871 | 774 | struct btrfs_device *device, *next; |
443f24fe | 775 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 776 | |
dfe25020 CM |
777 | mutex_lock(&uuid_mutex); |
778 | again: | |
46224705 | 779 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 780 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 781 | if (device->in_fs_metadata) { |
63a212ab | 782 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
783 | (!latest_dev || |
784 | device->generation > latest_dev->generation)) { | |
785 | latest_dev = device; | |
a6b0d5c8 | 786 | } |
2b82032c | 787 | continue; |
a6b0d5c8 | 788 | } |
2b82032c | 789 | |
8dabb742 SB |
790 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
791 | /* | |
792 | * In the first step, keep the device which has | |
793 | * the correct fsid and the devid that is used | |
794 | * for the dev_replace procedure. | |
795 | * In the second step, the dev_replace state is | |
796 | * read from the device tree and it is known | |
797 | * whether the procedure is really active or | |
798 | * not, which means whether this device is | |
799 | * used or whether it should be removed. | |
800 | */ | |
801 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
802 | continue; | |
803 | } | |
804 | } | |
2b82032c | 805 | if (device->bdev) { |
d4d77629 | 806 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
807 | device->bdev = NULL; |
808 | fs_devices->open_devices--; | |
809 | } | |
810 | if (device->writeable) { | |
811 | list_del_init(&device->dev_alloc_list); | |
812 | device->writeable = 0; | |
8dabb742 SB |
813 | if (!device->is_tgtdev_for_dev_replace) |
814 | fs_devices->rw_devices--; | |
2b82032c | 815 | } |
e4404d6e YZ |
816 | list_del_init(&device->dev_list); |
817 | fs_devices->num_devices--; | |
606686ee | 818 | rcu_string_free(device->name); |
e4404d6e | 819 | kfree(device); |
dfe25020 | 820 | } |
2b82032c YZ |
821 | |
822 | if (fs_devices->seed) { | |
823 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
824 | goto again; |
825 | } | |
826 | ||
443f24fe | 827 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 828 | |
dfe25020 | 829 | mutex_unlock(&uuid_mutex); |
dfe25020 | 830 | } |
a0af469b | 831 | |
1f78160c XG |
832 | static void __free_device(struct work_struct *work) |
833 | { | |
834 | struct btrfs_device *device; | |
835 | ||
836 | device = container_of(work, struct btrfs_device, rcu_work); | |
606686ee | 837 | rcu_string_free(device->name); |
1f78160c XG |
838 | kfree(device); |
839 | } | |
840 | ||
841 | static void free_device(struct rcu_head *head) | |
842 | { | |
843 | struct btrfs_device *device; | |
844 | ||
845 | device = container_of(head, struct btrfs_device, rcu); | |
846 | ||
847 | INIT_WORK(&device->rcu_work, __free_device); | |
848 | schedule_work(&device->rcu_work); | |
849 | } | |
850 | ||
14238819 AJ |
851 | static void btrfs_close_bdev(struct btrfs_device *device) |
852 | { | |
853 | if (device->bdev && device->writeable) { | |
854 | sync_blockdev(device->bdev); | |
855 | invalidate_bdev(device->bdev); | |
856 | } | |
857 | ||
858 | if (device->bdev) | |
859 | blkdev_put(device->bdev, device->mode); | |
860 | } | |
861 | ||
f448341a AJ |
862 | static void btrfs_close_one_device(struct btrfs_device *device) |
863 | { | |
864 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
865 | struct btrfs_device *new_device; | |
866 | struct rcu_string *name; | |
867 | ||
868 | if (device->bdev) | |
869 | fs_devices->open_devices--; | |
870 | ||
871 | if (device->writeable && | |
872 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
873 | list_del_init(&device->dev_alloc_list); | |
874 | fs_devices->rw_devices--; | |
875 | } | |
876 | ||
877 | if (device->missing) | |
878 | fs_devices->missing_devices--; | |
879 | ||
14238819 | 880 | btrfs_close_bdev(device); |
e2bf6e89 | 881 | |
f448341a AJ |
882 | new_device = btrfs_alloc_device(NULL, &device->devid, |
883 | device->uuid); | |
884 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
885 | ||
886 | /* Safe because we are under uuid_mutex */ | |
887 | if (device->name) { | |
888 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
889 | BUG_ON(!name); /* -ENOMEM */ | |
890 | rcu_assign_pointer(new_device->name, name); | |
891 | } | |
892 | ||
893 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
894 | new_device->fs_devices = device->fs_devices; | |
895 | ||
896 | call_rcu(&device->rcu, free_device); | |
897 | } | |
898 | ||
2b82032c | 899 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 900 | { |
2037a093 | 901 | struct btrfs_device *device, *tmp; |
e4404d6e | 902 | |
2b82032c YZ |
903 | if (--fs_devices->opened > 0) |
904 | return 0; | |
8a4b83cc | 905 | |
c9513edb | 906 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 907 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
f190aa47 | 908 | btrfs_close_one_device(device); |
8a4b83cc | 909 | } |
c9513edb XG |
910 | mutex_unlock(&fs_devices->device_list_mutex); |
911 | ||
e4404d6e YZ |
912 | WARN_ON(fs_devices->open_devices); |
913 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
914 | fs_devices->opened = 0; |
915 | fs_devices->seeding = 0; | |
2b82032c | 916 | |
8a4b83cc CM |
917 | return 0; |
918 | } | |
919 | ||
2b82032c YZ |
920 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
921 | { | |
e4404d6e | 922 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
923 | int ret; |
924 | ||
925 | mutex_lock(&uuid_mutex); | |
926 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
927 | if (!fs_devices->opened) { |
928 | seed_devices = fs_devices->seed; | |
929 | fs_devices->seed = NULL; | |
930 | } | |
2b82032c | 931 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
932 | |
933 | while (seed_devices) { | |
934 | fs_devices = seed_devices; | |
935 | seed_devices = fs_devices->seed; | |
936 | __btrfs_close_devices(fs_devices); | |
937 | free_fs_devices(fs_devices); | |
938 | } | |
bc178622 ES |
939 | /* |
940 | * Wait for rcu kworkers under __btrfs_close_devices | |
941 | * to finish all blkdev_puts so device is really | |
942 | * free when umount is done. | |
943 | */ | |
944 | rcu_barrier(); | |
2b82032c YZ |
945 | return ret; |
946 | } | |
947 | ||
e4404d6e YZ |
948 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
949 | fmode_t flags, void *holder) | |
8a4b83cc | 950 | { |
d5e2003c | 951 | struct request_queue *q; |
8a4b83cc CM |
952 | struct block_device *bdev; |
953 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 954 | struct btrfs_device *device; |
443f24fe | 955 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
956 | struct buffer_head *bh; |
957 | struct btrfs_super_block *disk_super; | |
a0af469b | 958 | u64 devid; |
2b82032c | 959 | int seeding = 1; |
a0af469b | 960 | int ret = 0; |
8a4b83cc | 961 | |
d4d77629 TH |
962 | flags |= FMODE_EXCL; |
963 | ||
c6e30871 | 964 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
965 | if (device->bdev) |
966 | continue; | |
dfe25020 CM |
967 | if (!device->name) |
968 | continue; | |
969 | ||
f63e0cca ES |
970 | /* Just open everything we can; ignore failures here */ |
971 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
972 | &bdev, &bh)) | |
beaf8ab3 | 973 | continue; |
a0af469b CM |
974 | |
975 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 976 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
977 | if (devid != device->devid) |
978 | goto error_brelse; | |
979 | ||
2b82032c YZ |
980 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
981 | BTRFS_UUID_SIZE)) | |
982 | goto error_brelse; | |
983 | ||
984 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
985 | if (!latest_dev || |
986 | device->generation > latest_dev->generation) | |
987 | latest_dev = device; | |
a0af469b | 988 | |
2b82032c YZ |
989 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
990 | device->writeable = 0; | |
991 | } else { | |
992 | device->writeable = !bdev_read_only(bdev); | |
993 | seeding = 0; | |
994 | } | |
995 | ||
d5e2003c | 996 | q = bdev_get_queue(bdev); |
90180da4 | 997 | if (blk_queue_discard(q)) |
d5e2003c | 998 | device->can_discard = 1; |
d5e2003c | 999 | |
8a4b83cc | 1000 | device->bdev = bdev; |
dfe25020 | 1001 | device->in_fs_metadata = 0; |
15916de8 CM |
1002 | device->mode = flags; |
1003 | ||
c289811c CM |
1004 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1005 | fs_devices->rotating = 1; | |
1006 | ||
a0af469b | 1007 | fs_devices->open_devices++; |
55e50e45 ID |
1008 | if (device->writeable && |
1009 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
1010 | fs_devices->rw_devices++; |
1011 | list_add(&device->dev_alloc_list, | |
1012 | &fs_devices->alloc_list); | |
1013 | } | |
4f6c9328 | 1014 | brelse(bh); |
a0af469b | 1015 | continue; |
a061fc8d | 1016 | |
a0af469b CM |
1017 | error_brelse: |
1018 | brelse(bh); | |
d4d77629 | 1019 | blkdev_put(bdev, flags); |
a0af469b | 1020 | continue; |
8a4b83cc | 1021 | } |
a0af469b | 1022 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1023 | ret = -EINVAL; |
a0af469b CM |
1024 | goto out; |
1025 | } | |
2b82032c YZ |
1026 | fs_devices->seeding = seeding; |
1027 | fs_devices->opened = 1; | |
443f24fe | 1028 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1029 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1030 | out: |
2b82032c YZ |
1031 | return ret; |
1032 | } | |
1033 | ||
1034 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 1035 | fmode_t flags, void *holder) |
2b82032c YZ |
1036 | { |
1037 | int ret; | |
1038 | ||
1039 | mutex_lock(&uuid_mutex); | |
1040 | if (fs_devices->opened) { | |
e4404d6e YZ |
1041 | fs_devices->opened++; |
1042 | ret = 0; | |
2b82032c | 1043 | } else { |
15916de8 | 1044 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1045 | } |
8a4b83cc | 1046 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1047 | return ret; |
1048 | } | |
1049 | ||
6cf86a00 AJ |
1050 | void btrfs_release_disk_super(struct page *page) |
1051 | { | |
1052 | kunmap(page); | |
1053 | put_page(page); | |
1054 | } | |
1055 | ||
1056 | int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, | |
1057 | struct page **page, struct btrfs_super_block **disk_super) | |
1058 | { | |
1059 | void *p; | |
1060 | pgoff_t index; | |
1061 | ||
1062 | /* make sure our super fits in the device */ | |
1063 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1064 | return 1; | |
1065 | ||
1066 | /* make sure our super fits in the page */ | |
1067 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1068 | return 1; | |
1069 | ||
1070 | /* make sure our super doesn't straddle pages on disk */ | |
1071 | index = bytenr >> PAGE_SHIFT; | |
1072 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1073 | return 1; | |
1074 | ||
1075 | /* pull in the page with our super */ | |
1076 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1077 | index, GFP_KERNEL); | |
1078 | ||
1079 | if (IS_ERR_OR_NULL(*page)) | |
1080 | return 1; | |
1081 | ||
1082 | p = kmap(*page); | |
1083 | ||
1084 | /* align our pointer to the offset of the super block */ | |
1085 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1086 | ||
1087 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1088 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1089 | btrfs_release_disk_super(*page); | |
1090 | return 1; | |
1091 | } | |
1092 | ||
1093 | if ((*disk_super)->label[0] && | |
1094 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1095 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1096 | ||
1097 | return 0; | |
1098 | } | |
1099 | ||
6f60cbd3 DS |
1100 | /* |
1101 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1102 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1103 | * is read via pagecache | |
1104 | */ | |
97288f2c | 1105 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1106 | struct btrfs_fs_devices **fs_devices_ret) |
1107 | { | |
1108 | struct btrfs_super_block *disk_super; | |
1109 | struct block_device *bdev; | |
6f60cbd3 | 1110 | struct page *page; |
6f60cbd3 | 1111 | int ret = -EINVAL; |
8a4b83cc | 1112 | u64 devid; |
f2984462 | 1113 | u64 transid; |
02db0844 | 1114 | u64 total_devices; |
6f60cbd3 | 1115 | u64 bytenr; |
8a4b83cc | 1116 | |
6f60cbd3 DS |
1117 | /* |
1118 | * we would like to check all the supers, but that would make | |
1119 | * a btrfs mount succeed after a mkfs from a different FS. | |
1120 | * So, we need to add a special mount option to scan for | |
1121 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1122 | */ | |
1123 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1124 | flags |= FMODE_EXCL; |
10f6327b | 1125 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1126 | |
1127 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1128 | if (IS_ERR(bdev)) { |
1129 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1130 | goto error; |
6f60cbd3 DS |
1131 | } |
1132 | ||
6cf86a00 | 1133 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) |
6f60cbd3 DS |
1134 | goto error_bdev_put; |
1135 | ||
a343832f | 1136 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1137 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1138 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1139 | |
8a4b83cc | 1140 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1141 | if (ret > 0) { |
1142 | if (disk_super->label[0]) { | |
62e85577 | 1143 | pr_info("BTRFS: device label %s ", disk_super->label); |
60999ca4 | 1144 | } else { |
62e85577 | 1145 | pr_info("BTRFS: device fsid %pU ", disk_super->fsid); |
60999ca4 DS |
1146 | } |
1147 | ||
62e85577 | 1148 | pr_cont("devid %llu transid %llu %s\n", devid, transid, path); |
60999ca4 DS |
1149 | ret = 0; |
1150 | } | |
02db0844 JB |
1151 | if (!ret && fs_devices_ret) |
1152 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 | 1153 | |
6cf86a00 | 1154 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1155 | |
1156 | error_bdev_put: | |
d4d77629 | 1157 | blkdev_put(bdev, flags); |
8a4b83cc | 1158 | error: |
beaf8ab3 | 1159 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1160 | return ret; |
1161 | } | |
0b86a832 | 1162 | |
6d07bcec MX |
1163 | /* helper to account the used device space in the range */ |
1164 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1165 | u64 end, u64 *length) | |
1166 | { | |
1167 | struct btrfs_key key; | |
1168 | struct btrfs_root *root = device->dev_root; | |
1169 | struct btrfs_dev_extent *dev_extent; | |
1170 | struct btrfs_path *path; | |
1171 | u64 extent_end; | |
1172 | int ret; | |
1173 | int slot; | |
1174 | struct extent_buffer *l; | |
1175 | ||
1176 | *length = 0; | |
1177 | ||
63a212ab | 1178 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1179 | return 0; |
1180 | ||
1181 | path = btrfs_alloc_path(); | |
1182 | if (!path) | |
1183 | return -ENOMEM; | |
e4058b54 | 1184 | path->reada = READA_FORWARD; |
6d07bcec MX |
1185 | |
1186 | key.objectid = device->devid; | |
1187 | key.offset = start; | |
1188 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1189 | ||
1190 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1191 | if (ret < 0) | |
1192 | goto out; | |
1193 | if (ret > 0) { | |
1194 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1195 | if (ret < 0) | |
1196 | goto out; | |
1197 | } | |
1198 | ||
1199 | while (1) { | |
1200 | l = path->nodes[0]; | |
1201 | slot = path->slots[0]; | |
1202 | if (slot >= btrfs_header_nritems(l)) { | |
1203 | ret = btrfs_next_leaf(root, path); | |
1204 | if (ret == 0) | |
1205 | continue; | |
1206 | if (ret < 0) | |
1207 | goto out; | |
1208 | ||
1209 | break; | |
1210 | } | |
1211 | btrfs_item_key_to_cpu(l, &key, slot); | |
1212 | ||
1213 | if (key.objectid < device->devid) | |
1214 | goto next; | |
1215 | ||
1216 | if (key.objectid > device->devid) | |
1217 | break; | |
1218 | ||
962a298f | 1219 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1220 | goto next; |
1221 | ||
1222 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1223 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1224 | dev_extent); | |
1225 | if (key.offset <= start && extent_end > end) { | |
1226 | *length = end - start + 1; | |
1227 | break; | |
1228 | } else if (key.offset <= start && extent_end > start) | |
1229 | *length += extent_end - start; | |
1230 | else if (key.offset > start && extent_end <= end) | |
1231 | *length += extent_end - key.offset; | |
1232 | else if (key.offset > start && key.offset <= end) { | |
1233 | *length += end - key.offset + 1; | |
1234 | break; | |
1235 | } else if (key.offset > end) | |
1236 | break; | |
1237 | ||
1238 | next: | |
1239 | path->slots[0]++; | |
1240 | } | |
1241 | ret = 0; | |
1242 | out: | |
1243 | btrfs_free_path(path); | |
1244 | return ret; | |
1245 | } | |
1246 | ||
499f377f | 1247 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1248 | struct btrfs_device *device, |
1249 | u64 *start, u64 len) | |
1250 | { | |
499f377f | 1251 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; |
6df9a95e | 1252 | struct extent_map *em; |
499f377f | 1253 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1254 | int ret = 0; |
1b984508 | 1255 | u64 physical_start = *start; |
6df9a95e | 1256 | |
499f377f JM |
1257 | if (transaction) |
1258 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1259 | again: |
1260 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1261 | struct map_lookup *map; |
1262 | int i; | |
1263 | ||
95617d69 | 1264 | map = em->map_lookup; |
6df9a95e | 1265 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1266 | u64 end; |
1267 | ||
6df9a95e JB |
1268 | if (map->stripes[i].dev != device) |
1269 | continue; | |
1b984508 | 1270 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1271 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1272 | physical_start) |
6df9a95e | 1273 | continue; |
c152b63e FM |
1274 | /* |
1275 | * Make sure that while processing the pinned list we do | |
1276 | * not override our *start with a lower value, because | |
1277 | * we can have pinned chunks that fall within this | |
1278 | * device hole and that have lower physical addresses | |
1279 | * than the pending chunks we processed before. If we | |
1280 | * do not take this special care we can end up getting | |
1281 | * 2 pending chunks that start at the same physical | |
1282 | * device offsets because the end offset of a pinned | |
1283 | * chunk can be equal to the start offset of some | |
1284 | * pending chunk. | |
1285 | */ | |
1286 | end = map->stripes[i].physical + em->orig_block_len; | |
1287 | if (end > *start) { | |
1288 | *start = end; | |
1289 | ret = 1; | |
1290 | } | |
6df9a95e JB |
1291 | } |
1292 | } | |
499f377f JM |
1293 | if (search_list != &fs_info->pinned_chunks) { |
1294 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1295 | goto again; |
1296 | } | |
6df9a95e JB |
1297 | |
1298 | return ret; | |
1299 | } | |
1300 | ||
1301 | ||
0b86a832 | 1302 | /* |
499f377f JM |
1303 | * find_free_dev_extent_start - find free space in the specified device |
1304 | * @device: the device which we search the free space in | |
1305 | * @num_bytes: the size of the free space that we need | |
1306 | * @search_start: the position from which to begin the search | |
1307 | * @start: store the start of the free space. | |
1308 | * @len: the size of the free space. that we find, or the size | |
1309 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1310 | * |
0b86a832 CM |
1311 | * this uses a pretty simple search, the expectation is that it is |
1312 | * called very infrequently and that a given device has a small number | |
1313 | * of extents | |
7bfc837d MX |
1314 | * |
1315 | * @start is used to store the start of the free space if we find. But if we | |
1316 | * don't find suitable free space, it will be used to store the start position | |
1317 | * of the max free space. | |
1318 | * | |
1319 | * @len is used to store the size of the free space that we find. | |
1320 | * But if we don't find suitable free space, it is used to store the size of | |
1321 | * the max free space. | |
0b86a832 | 1322 | */ |
499f377f JM |
1323 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1324 | struct btrfs_device *device, u64 num_bytes, | |
1325 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 CM |
1326 | { |
1327 | struct btrfs_key key; | |
1328 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1329 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1330 | struct btrfs_path *path; |
7bfc837d MX |
1331 | u64 hole_size; |
1332 | u64 max_hole_start; | |
1333 | u64 max_hole_size; | |
1334 | u64 extent_end; | |
0b86a832 CM |
1335 | u64 search_end = device->total_bytes; |
1336 | int ret; | |
7bfc837d | 1337 | int slot; |
0b86a832 | 1338 | struct extent_buffer *l; |
8cdc7c5b FM |
1339 | u64 min_search_start; |
1340 | ||
1341 | /* | |
1342 | * We don't want to overwrite the superblock on the drive nor any area | |
1343 | * used by the boot loader (grub for example), so we make sure to start | |
1344 | * at an offset of at least 1MB. | |
1345 | */ | |
1346 | min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024); | |
1347 | search_start = max(search_start, min_search_start); | |
0b86a832 | 1348 | |
6df9a95e JB |
1349 | path = btrfs_alloc_path(); |
1350 | if (!path) | |
1351 | return -ENOMEM; | |
f2ab7618 | 1352 | |
7bfc837d MX |
1353 | max_hole_start = search_start; |
1354 | max_hole_size = 0; | |
1355 | ||
f2ab7618 | 1356 | again: |
63a212ab | 1357 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1358 | ret = -ENOSPC; |
6df9a95e | 1359 | goto out; |
7bfc837d MX |
1360 | } |
1361 | ||
e4058b54 | 1362 | path->reada = READA_FORWARD; |
6df9a95e JB |
1363 | path->search_commit_root = 1; |
1364 | path->skip_locking = 1; | |
7bfc837d | 1365 | |
0b86a832 CM |
1366 | key.objectid = device->devid; |
1367 | key.offset = search_start; | |
1368 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1369 | |
125ccb0a | 1370 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1371 | if (ret < 0) |
7bfc837d | 1372 | goto out; |
1fcbac58 YZ |
1373 | if (ret > 0) { |
1374 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1375 | if (ret < 0) | |
7bfc837d | 1376 | goto out; |
1fcbac58 | 1377 | } |
7bfc837d | 1378 | |
0b86a832 CM |
1379 | while (1) { |
1380 | l = path->nodes[0]; | |
1381 | slot = path->slots[0]; | |
1382 | if (slot >= btrfs_header_nritems(l)) { | |
1383 | ret = btrfs_next_leaf(root, path); | |
1384 | if (ret == 0) | |
1385 | continue; | |
1386 | if (ret < 0) | |
7bfc837d MX |
1387 | goto out; |
1388 | ||
1389 | break; | |
0b86a832 CM |
1390 | } |
1391 | btrfs_item_key_to_cpu(l, &key, slot); | |
1392 | ||
1393 | if (key.objectid < device->devid) | |
1394 | goto next; | |
1395 | ||
1396 | if (key.objectid > device->devid) | |
7bfc837d | 1397 | break; |
0b86a832 | 1398 | |
962a298f | 1399 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1400 | goto next; |
9779b72f | 1401 | |
7bfc837d MX |
1402 | if (key.offset > search_start) { |
1403 | hole_size = key.offset - search_start; | |
9779b72f | 1404 | |
6df9a95e JB |
1405 | /* |
1406 | * Have to check before we set max_hole_start, otherwise | |
1407 | * we could end up sending back this offset anyway. | |
1408 | */ | |
499f377f | 1409 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1410 | &search_start, |
1b984508 FL |
1411 | hole_size)) { |
1412 | if (key.offset >= search_start) { | |
1413 | hole_size = key.offset - search_start; | |
1414 | } else { | |
1415 | WARN_ON_ONCE(1); | |
1416 | hole_size = 0; | |
1417 | } | |
1418 | } | |
6df9a95e | 1419 | |
7bfc837d MX |
1420 | if (hole_size > max_hole_size) { |
1421 | max_hole_start = search_start; | |
1422 | max_hole_size = hole_size; | |
1423 | } | |
9779b72f | 1424 | |
7bfc837d MX |
1425 | /* |
1426 | * If this free space is greater than which we need, | |
1427 | * it must be the max free space that we have found | |
1428 | * until now, so max_hole_start must point to the start | |
1429 | * of this free space and the length of this free space | |
1430 | * is stored in max_hole_size. Thus, we return | |
1431 | * max_hole_start and max_hole_size and go back to the | |
1432 | * caller. | |
1433 | */ | |
1434 | if (hole_size >= num_bytes) { | |
1435 | ret = 0; | |
1436 | goto out; | |
0b86a832 CM |
1437 | } |
1438 | } | |
0b86a832 | 1439 | |
0b86a832 | 1440 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1441 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1442 | dev_extent); | |
1443 | if (extent_end > search_start) | |
1444 | search_start = extent_end; | |
0b86a832 CM |
1445 | next: |
1446 | path->slots[0]++; | |
1447 | cond_resched(); | |
1448 | } | |
0b86a832 | 1449 | |
38c01b96 | 1450 | /* |
1451 | * At this point, search_start should be the end of | |
1452 | * allocated dev extents, and when shrinking the device, | |
1453 | * search_end may be smaller than search_start. | |
1454 | */ | |
f2ab7618 | 1455 | if (search_end > search_start) { |
38c01b96 | 1456 | hole_size = search_end - search_start; |
1457 | ||
499f377f | 1458 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1459 | hole_size)) { |
1460 | btrfs_release_path(path); | |
1461 | goto again; | |
1462 | } | |
0b86a832 | 1463 | |
f2ab7618 ZL |
1464 | if (hole_size > max_hole_size) { |
1465 | max_hole_start = search_start; | |
1466 | max_hole_size = hole_size; | |
1467 | } | |
6df9a95e JB |
1468 | } |
1469 | ||
7bfc837d | 1470 | /* See above. */ |
f2ab7618 | 1471 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1472 | ret = -ENOSPC; |
1473 | else | |
1474 | ret = 0; | |
1475 | ||
1476 | out: | |
2b82032c | 1477 | btrfs_free_path(path); |
7bfc837d | 1478 | *start = max_hole_start; |
b2117a39 | 1479 | if (len) |
7bfc837d | 1480 | *len = max_hole_size; |
0b86a832 CM |
1481 | return ret; |
1482 | } | |
1483 | ||
499f377f JM |
1484 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1485 | struct btrfs_device *device, u64 num_bytes, | |
1486 | u64 *start, u64 *len) | |
1487 | { | |
499f377f | 1488 | /* FIXME use last free of some kind */ |
499f377f | 1489 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1490 | num_bytes, 0, start, len); |
499f377f JM |
1491 | } |
1492 | ||
b2950863 | 1493 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1494 | struct btrfs_device *device, |
2196d6e8 | 1495 | u64 start, u64 *dev_extent_len) |
8f18cf13 CM |
1496 | { |
1497 | int ret; | |
1498 | struct btrfs_path *path; | |
1499 | struct btrfs_root *root = device->dev_root; | |
1500 | struct btrfs_key key; | |
a061fc8d CM |
1501 | struct btrfs_key found_key; |
1502 | struct extent_buffer *leaf = NULL; | |
1503 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1504 | |
1505 | path = btrfs_alloc_path(); | |
1506 | if (!path) | |
1507 | return -ENOMEM; | |
1508 | ||
1509 | key.objectid = device->devid; | |
1510 | key.offset = start; | |
1511 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1512 | again: |
8f18cf13 | 1513 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1514 | if (ret > 0) { |
1515 | ret = btrfs_previous_item(root, path, key.objectid, | |
1516 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1517 | if (ret) |
1518 | goto out; | |
a061fc8d CM |
1519 | leaf = path->nodes[0]; |
1520 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1521 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1522 | struct btrfs_dev_extent); | |
1523 | BUG_ON(found_key.offset > start || found_key.offset + | |
1524 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1525 | key = found_key; |
1526 | btrfs_release_path(path); | |
1527 | goto again; | |
a061fc8d CM |
1528 | } else if (ret == 0) { |
1529 | leaf = path->nodes[0]; | |
1530 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1531 | struct btrfs_dev_extent); | |
79787eaa | 1532 | } else { |
34d97007 | 1533 | btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed"); |
79787eaa | 1534 | goto out; |
a061fc8d | 1535 | } |
8f18cf13 | 1536 | |
2196d6e8 MX |
1537 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1538 | ||
8f18cf13 | 1539 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1540 | if (ret) { |
34d97007 | 1541 | btrfs_handle_fs_error(root->fs_info, ret, |
79787eaa | 1542 | "Failed to remove dev extent item"); |
13212b54 | 1543 | } else { |
3204d33c | 1544 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1545 | } |
b0b802d7 | 1546 | out: |
8f18cf13 CM |
1547 | btrfs_free_path(path); |
1548 | return ret; | |
1549 | } | |
1550 | ||
48a3b636 ES |
1551 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1552 | struct btrfs_device *device, | |
1553 | u64 chunk_tree, u64 chunk_objectid, | |
1554 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1555 | { |
1556 | int ret; | |
1557 | struct btrfs_path *path; | |
1558 | struct btrfs_root *root = device->dev_root; | |
1559 | struct btrfs_dev_extent *extent; | |
1560 | struct extent_buffer *leaf; | |
1561 | struct btrfs_key key; | |
1562 | ||
dfe25020 | 1563 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1564 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1565 | path = btrfs_alloc_path(); |
1566 | if (!path) | |
1567 | return -ENOMEM; | |
1568 | ||
0b86a832 | 1569 | key.objectid = device->devid; |
2b82032c | 1570 | key.offset = start; |
0b86a832 CM |
1571 | key.type = BTRFS_DEV_EXTENT_KEY; |
1572 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1573 | sizeof(*extent)); | |
2cdcecbc MF |
1574 | if (ret) |
1575 | goto out; | |
0b86a832 CM |
1576 | |
1577 | leaf = path->nodes[0]; | |
1578 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1579 | struct btrfs_dev_extent); | |
e17cade2 CM |
1580 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1581 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1582 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1583 | ||
1584 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1585 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1586 | |
0b86a832 CM |
1587 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1588 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1589 | out: |
0b86a832 CM |
1590 | btrfs_free_path(path); |
1591 | return ret; | |
1592 | } | |
1593 | ||
6df9a95e | 1594 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1595 | { |
6df9a95e JB |
1596 | struct extent_map_tree *em_tree; |
1597 | struct extent_map *em; | |
1598 | struct rb_node *n; | |
1599 | u64 ret = 0; | |
0b86a832 | 1600 | |
6df9a95e JB |
1601 | em_tree = &fs_info->mapping_tree.map_tree; |
1602 | read_lock(&em_tree->lock); | |
1603 | n = rb_last(&em_tree->map); | |
1604 | if (n) { | |
1605 | em = rb_entry(n, struct extent_map, rb_node); | |
1606 | ret = em->start + em->len; | |
0b86a832 | 1607 | } |
6df9a95e JB |
1608 | read_unlock(&em_tree->lock); |
1609 | ||
0b86a832 CM |
1610 | return ret; |
1611 | } | |
1612 | ||
53f10659 ID |
1613 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1614 | u64 *devid_ret) | |
0b86a832 CM |
1615 | { |
1616 | int ret; | |
1617 | struct btrfs_key key; | |
1618 | struct btrfs_key found_key; | |
2b82032c YZ |
1619 | struct btrfs_path *path; |
1620 | ||
2b82032c YZ |
1621 | path = btrfs_alloc_path(); |
1622 | if (!path) | |
1623 | return -ENOMEM; | |
0b86a832 CM |
1624 | |
1625 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1626 | key.type = BTRFS_DEV_ITEM_KEY; | |
1627 | key.offset = (u64)-1; | |
1628 | ||
53f10659 | 1629 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1630 | if (ret < 0) |
1631 | goto error; | |
1632 | ||
79787eaa | 1633 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1634 | |
53f10659 ID |
1635 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1636 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1637 | BTRFS_DEV_ITEM_KEY); |
1638 | if (ret) { | |
53f10659 | 1639 | *devid_ret = 1; |
0b86a832 CM |
1640 | } else { |
1641 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1642 | path->slots[0]); | |
53f10659 | 1643 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1644 | } |
1645 | ret = 0; | |
1646 | error: | |
2b82032c | 1647 | btrfs_free_path(path); |
0b86a832 CM |
1648 | return ret; |
1649 | } | |
1650 | ||
1651 | /* | |
1652 | * the device information is stored in the chunk root | |
1653 | * the btrfs_device struct should be fully filled in | |
1654 | */ | |
48a3b636 ES |
1655 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1656 | struct btrfs_root *root, | |
1657 | struct btrfs_device *device) | |
0b86a832 CM |
1658 | { |
1659 | int ret; | |
1660 | struct btrfs_path *path; | |
1661 | struct btrfs_dev_item *dev_item; | |
1662 | struct extent_buffer *leaf; | |
1663 | struct btrfs_key key; | |
1664 | unsigned long ptr; | |
0b86a832 CM |
1665 | |
1666 | root = root->fs_info->chunk_root; | |
1667 | ||
1668 | path = btrfs_alloc_path(); | |
1669 | if (!path) | |
1670 | return -ENOMEM; | |
1671 | ||
0b86a832 CM |
1672 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1673 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1674 | key.offset = device->devid; |
0b86a832 CM |
1675 | |
1676 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1677 | sizeof(*dev_item)); |
0b86a832 CM |
1678 | if (ret) |
1679 | goto out; | |
1680 | ||
1681 | leaf = path->nodes[0]; | |
1682 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1683 | ||
1684 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1685 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1686 | btrfs_set_device_type(leaf, dev_item, device->type); |
1687 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1688 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1689 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1690 | btrfs_set_device_total_bytes(leaf, dev_item, |
1691 | btrfs_device_get_disk_total_bytes(device)); | |
1692 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1693 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1694 | btrfs_set_device_group(leaf, dev_item, 0); |
1695 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1696 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1697 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1698 | |
410ba3a2 | 1699 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1700 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1701 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1702 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1703 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1704 | |
2b82032c | 1705 | ret = 0; |
0b86a832 CM |
1706 | out: |
1707 | btrfs_free_path(path); | |
1708 | return ret; | |
1709 | } | |
8f18cf13 | 1710 | |
5a1972bd QW |
1711 | /* |
1712 | * Function to update ctime/mtime for a given device path. | |
1713 | * Mainly used for ctime/mtime based probe like libblkid. | |
1714 | */ | |
1715 | static void update_dev_time(char *path_name) | |
1716 | { | |
1717 | struct file *filp; | |
1718 | ||
1719 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1720 | if (IS_ERR(filp)) |
5a1972bd QW |
1721 | return; |
1722 | file_update_time(filp); | |
1723 | filp_close(filp, NULL); | |
5a1972bd QW |
1724 | } |
1725 | ||
a061fc8d CM |
1726 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1727 | struct btrfs_device *device) | |
1728 | { | |
1729 | int ret; | |
1730 | struct btrfs_path *path; | |
a061fc8d | 1731 | struct btrfs_key key; |
a061fc8d CM |
1732 | struct btrfs_trans_handle *trans; |
1733 | ||
1734 | root = root->fs_info->chunk_root; | |
1735 | ||
1736 | path = btrfs_alloc_path(); | |
1737 | if (!path) | |
1738 | return -ENOMEM; | |
1739 | ||
a22285a6 | 1740 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1741 | if (IS_ERR(trans)) { |
1742 | btrfs_free_path(path); | |
1743 | return PTR_ERR(trans); | |
1744 | } | |
a061fc8d CM |
1745 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1746 | key.type = BTRFS_DEV_ITEM_KEY; | |
1747 | key.offset = device->devid; | |
1748 | ||
1749 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1750 | if (ret < 0) | |
1751 | goto out; | |
1752 | ||
1753 | if (ret > 0) { | |
1754 | ret = -ENOENT; | |
1755 | goto out; | |
1756 | } | |
1757 | ||
1758 | ret = btrfs_del_item(trans, root, path); | |
1759 | if (ret) | |
1760 | goto out; | |
a061fc8d CM |
1761 | out: |
1762 | btrfs_free_path(path); | |
1763 | btrfs_commit_transaction(trans, root); | |
1764 | return ret; | |
1765 | } | |
1766 | ||
3cc31a0d DS |
1767 | /* |
1768 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1769 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1770 | * replace. | |
1771 | */ | |
1772 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1773 | u64 num_devices) | |
a061fc8d | 1774 | { |
a061fc8d | 1775 | u64 all_avail; |
de98ced9 | 1776 | unsigned seq; |
418775a2 | 1777 | int i; |
a061fc8d | 1778 | |
de98ced9 | 1779 | do { |
bd45ffbc | 1780 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1781 | |
bd45ffbc AJ |
1782 | all_avail = fs_info->avail_data_alloc_bits | |
1783 | fs_info->avail_system_alloc_bits | | |
1784 | fs_info->avail_metadata_alloc_bits; | |
1785 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1786 | |
418775a2 DS |
1787 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1788 | if (!(all_avail & btrfs_raid_group[i])) | |
1789 | continue; | |
a061fc8d | 1790 | |
418775a2 DS |
1791 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1792 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1793 | |
418775a2 DS |
1794 | if (ret) |
1795 | return ret; | |
1796 | } | |
53b381b3 DW |
1797 | } |
1798 | ||
bd45ffbc | 1799 | return 0; |
f1fa7f26 AJ |
1800 | } |
1801 | ||
88acff64 AJ |
1802 | struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs, |
1803 | struct btrfs_device *device) | |
a061fc8d | 1804 | { |
2b82032c | 1805 | struct btrfs_device *next_device; |
88acff64 AJ |
1806 | |
1807 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1808 | if (next_device != device && | |
1809 | !next_device->missing && next_device->bdev) | |
1810 | return next_device; | |
1811 | } | |
1812 | ||
1813 | return NULL; | |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1818 | * and replace it with the provided or the next active device, in the context | |
1819 | * where this function called, there should be always be another device (or | |
1820 | * this_dev) which is active. | |
1821 | */ | |
1822 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1823 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1824 | { | |
1825 | struct btrfs_device *next_device; | |
1826 | ||
1827 | if (this_dev) | |
1828 | next_device = this_dev; | |
1829 | else | |
1830 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1831 | device); | |
1832 | ASSERT(next_device); | |
1833 | ||
1834 | if (fs_info->sb->s_bdev && | |
1835 | (fs_info->sb->s_bdev == device->bdev)) | |
1836 | fs_info->sb->s_bdev = next_device->bdev; | |
1837 | ||
1838 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1839 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1840 | } | |
1841 | ||
6b526ed7 | 1842 | int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid) |
f1fa7f26 AJ |
1843 | { |
1844 | struct btrfs_device *device; | |
1f78160c | 1845 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1846 | u64 num_devices; |
a061fc8d | 1847 | int ret = 0; |
1f78160c | 1848 | bool clear_super = false; |
a061fc8d | 1849 | |
a061fc8d CM |
1850 | mutex_lock(&uuid_mutex); |
1851 | ||
8dabb742 | 1852 | num_devices = root->fs_info->fs_devices->num_devices; |
73beece9 | 1853 | btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0); |
8dabb742 SB |
1854 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { |
1855 | WARN_ON(num_devices < 1); | |
1856 | num_devices--; | |
1857 | } | |
73beece9 | 1858 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0); |
8dabb742 | 1859 | |
3cc31a0d | 1860 | ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1); |
f1fa7f26 | 1861 | if (ret) |
a061fc8d | 1862 | goto out; |
a061fc8d | 1863 | |
5c5c0df0 | 1864 | ret = btrfs_find_device_by_devspec(root, devid, device_path, |
24fc572f AJ |
1865 | &device); |
1866 | if (ret) | |
53b381b3 | 1867 | goto out; |
dfe25020 | 1868 | |
63a212ab | 1869 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1870 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1871 | goto out; |
63a212ab SB |
1872 | } |
1873 | ||
2b82032c | 1874 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1875 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1876 | goto out; |
2b82032c YZ |
1877 | } |
1878 | ||
1879 | if (device->writeable) { | |
0c1daee0 | 1880 | lock_chunks(root); |
2b82032c | 1881 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1882 | device->fs_devices->rw_devices--; |
0c1daee0 | 1883 | unlock_chunks(root); |
1f78160c | 1884 | clear_super = true; |
dfe25020 | 1885 | } |
a061fc8d | 1886 | |
d7901554 | 1887 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1888 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1889 | mutex_lock(&uuid_mutex); |
a061fc8d | 1890 | if (ret) |
9b3517e9 | 1891 | goto error_undo; |
a061fc8d | 1892 | |
63a212ab SB |
1893 | /* |
1894 | * TODO: the superblock still includes this device in its num_devices | |
1895 | * counter although write_all_supers() is not locked out. This | |
1896 | * could give a filesystem state which requires a degraded mount. | |
1897 | */ | |
a061fc8d CM |
1898 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1899 | if (ret) | |
9b3517e9 | 1900 | goto error_undo; |
a061fc8d | 1901 | |
2b82032c | 1902 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1903 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1904 | |
1905 | /* | |
1906 | * the device list mutex makes sure that we don't change | |
1907 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1908 | * the device supers. Whoever is writing all supers, should |
1909 | * lock the device list mutex before getting the number of | |
1910 | * devices in the super block (super_copy). Conversely, | |
1911 | * whoever updates the number of devices in the super block | |
1912 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1913 | */ |
1f78160c XG |
1914 | |
1915 | cur_devices = device->fs_devices; | |
e5e9a520 | 1916 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1917 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1918 | |
e4404d6e | 1919 | device->fs_devices->num_devices--; |
02db0844 | 1920 | device->fs_devices->total_devices--; |
2b82032c | 1921 | |
cd02dca5 | 1922 | if (device->missing) |
3a7d55c8 | 1923 | device->fs_devices->missing_devices--; |
cd02dca5 | 1924 | |
88acff64 | 1925 | btrfs_assign_next_active_device(root->fs_info, device, NULL); |
2b82032c | 1926 | |
0bfaa9c5 | 1927 | if (device->bdev) { |
e4404d6e | 1928 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1929 | /* remove sysfs entry */ |
32576040 | 1930 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); |
0bfaa9c5 | 1931 | } |
99994cde | 1932 | |
6c41761f DS |
1933 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1934 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
d7306801 | 1935 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2b82032c | 1936 | |
cea67ab9 JM |
1937 | /* |
1938 | * at this point, the device is zero sized and detached from | |
1939 | * the devices list. All that's left is to zero out the old | |
1940 | * supers and free the device. | |
1941 | */ | |
1942 | if (device->writeable) | |
1943 | btrfs_scratch_superblocks(device->bdev, device->name->str); | |
1944 | ||
1945 | btrfs_close_bdev(device); | |
1946 | call_rcu(&device->rcu, free_device); | |
1947 | ||
1f78160c | 1948 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1949 | struct btrfs_fs_devices *fs_devices; |
1950 | fs_devices = root->fs_info->fs_devices; | |
1951 | while (fs_devices) { | |
8321cf25 RS |
1952 | if (fs_devices->seed == cur_devices) { |
1953 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1954 | break; |
8321cf25 | 1955 | } |
e4404d6e | 1956 | fs_devices = fs_devices->seed; |
2b82032c | 1957 | } |
1f78160c | 1958 | cur_devices->seed = NULL; |
1f78160c | 1959 | __btrfs_close_devices(cur_devices); |
1f78160c | 1960 | free_fs_devices(cur_devices); |
2b82032c YZ |
1961 | } |
1962 | ||
5af3e8cc SB |
1963 | root->fs_info->num_tolerated_disk_barrier_failures = |
1964 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1965 | ||
a061fc8d CM |
1966 | out: |
1967 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1968 | return ret; |
24fc572f | 1969 | |
9b3517e9 ID |
1970 | error_undo: |
1971 | if (device->writeable) { | |
0c1daee0 | 1972 | lock_chunks(root); |
9b3517e9 ID |
1973 | list_add(&device->dev_alloc_list, |
1974 | &root->fs_info->fs_devices->alloc_list); | |
c3929c36 | 1975 | device->fs_devices->rw_devices++; |
0c1daee0 | 1976 | unlock_chunks(root); |
9b3517e9 | 1977 | } |
24fc572f | 1978 | goto out; |
a061fc8d CM |
1979 | } |
1980 | ||
084b6e7c QW |
1981 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1982 | struct btrfs_device *srcdev) | |
e93c89c1 | 1983 | { |
d51908ce AJ |
1984 | struct btrfs_fs_devices *fs_devices; |
1985 | ||
e93c89c1 | 1986 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 1987 | |
25e8e911 AJ |
1988 | /* |
1989 | * in case of fs with no seed, srcdev->fs_devices will point | |
1990 | * to fs_devices of fs_info. However when the dev being replaced is | |
1991 | * a seed dev it will point to the seed's local fs_devices. In short | |
1992 | * srcdev will have its correct fs_devices in both the cases. | |
1993 | */ | |
1994 | fs_devices = srcdev->fs_devices; | |
d51908ce | 1995 | |
e93c89c1 SB |
1996 | list_del_rcu(&srcdev->dev_list); |
1997 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 1998 | fs_devices->num_devices--; |
82372bc8 | 1999 | if (srcdev->missing) |
d51908ce | 2000 | fs_devices->missing_devices--; |
e93c89c1 | 2001 | |
48b3b9d4 | 2002 | if (srcdev->writeable) |
82372bc8 | 2003 | fs_devices->rw_devices--; |
1357272f | 2004 | |
82372bc8 | 2005 | if (srcdev->bdev) |
d51908ce | 2006 | fs_devices->open_devices--; |
084b6e7c QW |
2007 | } |
2008 | ||
2009 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2010 | struct btrfs_device *srcdev) | |
2011 | { | |
2012 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2013 | |
48b3b9d4 AJ |
2014 | if (srcdev->writeable) { |
2015 | /* zero out the old super if it is writable */ | |
2016 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2017 | } | |
14238819 AJ |
2018 | |
2019 | btrfs_close_bdev(srcdev); | |
2020 | ||
e93c89c1 | 2021 | call_rcu(&srcdev->rcu, free_device); |
94d5f0c2 AJ |
2022 | |
2023 | /* | |
2024 | * unless fs_devices is seed fs, num_devices shouldn't go | |
2025 | * zero | |
2026 | */ | |
2027 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
2028 | ||
2029 | /* if this is no devs we rather delete the fs_devices */ | |
2030 | if (!fs_devices->num_devices) { | |
2031 | struct btrfs_fs_devices *tmp_fs_devices; | |
2032 | ||
2033 | tmp_fs_devices = fs_info->fs_devices; | |
2034 | while (tmp_fs_devices) { | |
2035 | if (tmp_fs_devices->seed == fs_devices) { | |
2036 | tmp_fs_devices->seed = fs_devices->seed; | |
2037 | break; | |
2038 | } | |
2039 | tmp_fs_devices = tmp_fs_devices->seed; | |
2040 | } | |
2041 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2042 | __btrfs_close_devices(fs_devices); |
2043 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2044 | } |
e93c89c1 SB |
2045 | } |
2046 | ||
2047 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2048 | struct btrfs_device *tgtdev) | |
2049 | { | |
67a2c45e | 2050 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2051 | WARN_ON(!tgtdev); |
2052 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2053 | |
32576040 | 2054 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2055 | |
779bf3fe | 2056 | if (tgtdev->bdev) |
e93c89c1 | 2057 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2058 | |
e93c89c1 | 2059 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2060 | |
88acff64 | 2061 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2062 | |
e93c89c1 | 2063 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2064 | |
2065 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2066 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2067 | |
2068 | /* | |
2069 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2070 | * may lead to a call to btrfs_show_devname() which will try | |
2071 | * to hold device_list_mutex. And here this device | |
2072 | * is already out of device list, so we don't have to hold | |
2073 | * the device_list_mutex lock. | |
2074 | */ | |
2075 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2076 | |
2077 | btrfs_close_bdev(tgtdev); | |
779bf3fe | 2078 | call_rcu(&tgtdev->rcu, free_device); |
e93c89c1 SB |
2079 | } |
2080 | ||
48a3b636 ES |
2081 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
2082 | struct btrfs_device **device) | |
7ba15b7d SB |
2083 | { |
2084 | int ret = 0; | |
2085 | struct btrfs_super_block *disk_super; | |
2086 | u64 devid; | |
2087 | u8 *dev_uuid; | |
2088 | struct block_device *bdev; | |
2089 | struct buffer_head *bh; | |
2090 | ||
2091 | *device = NULL; | |
2092 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
2093 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
2094 | if (ret) | |
2095 | return ret; | |
2096 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2097 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2098 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 2099 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
2100 | disk_super->fsid); |
2101 | brelse(bh); | |
2102 | if (!*device) | |
2103 | ret = -ENOENT; | |
2104 | blkdev_put(bdev, FMODE_READ); | |
2105 | return ret; | |
2106 | } | |
2107 | ||
2108 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
2109 | char *device_path, | |
2110 | struct btrfs_device **device) | |
2111 | { | |
2112 | *device = NULL; | |
2113 | if (strcmp(device_path, "missing") == 0) { | |
2114 | struct list_head *devices; | |
2115 | struct btrfs_device *tmp; | |
2116 | ||
2117 | devices = &root->fs_info->fs_devices->devices; | |
2118 | /* | |
2119 | * It is safe to read the devices since the volume_mutex | |
2120 | * is held by the caller. | |
2121 | */ | |
2122 | list_for_each_entry(tmp, devices, dev_list) { | |
2123 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2124 | *device = tmp; | |
2125 | break; | |
2126 | } | |
2127 | } | |
2128 | ||
d74a6259 AJ |
2129 | if (!*device) |
2130 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2131 | |
2132 | return 0; | |
2133 | } else { | |
2134 | return btrfs_find_device_by_path(root, device_path, device); | |
2135 | } | |
2136 | } | |
2137 | ||
5c5c0df0 DS |
2138 | /* |
2139 | * Lookup a device given by device id, or the path if the id is 0. | |
2140 | */ | |
2141 | int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid, | |
2142 | char *devpath, | |
24e0474b AJ |
2143 | struct btrfs_device **device) |
2144 | { | |
2145 | int ret; | |
2146 | ||
5c5c0df0 | 2147 | if (devid) { |
24e0474b | 2148 | ret = 0; |
5c5c0df0 | 2149 | *device = btrfs_find_device(root->fs_info, devid, NULL, |
24e0474b AJ |
2150 | NULL); |
2151 | if (!*device) | |
2152 | ret = -ENOENT; | |
2153 | } else { | |
5c5c0df0 | 2154 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2155 | return -EINVAL; |
2156 | ||
5c5c0df0 | 2157 | ret = btrfs_find_device_missing_or_by_path(root, devpath, |
24e0474b AJ |
2158 | device); |
2159 | } | |
2160 | return ret; | |
2161 | } | |
2162 | ||
2b82032c YZ |
2163 | /* |
2164 | * does all the dirty work required for changing file system's UUID. | |
2165 | */ | |
125ccb0a | 2166 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
2167 | { |
2168 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
2169 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 2170 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 2171 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
2172 | struct btrfs_device *device; |
2173 | u64 super_flags; | |
2174 | ||
2175 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2176 | if (!fs_devices->seeding) |
2b82032c YZ |
2177 | return -EINVAL; |
2178 | ||
2208a378 ID |
2179 | seed_devices = __alloc_fs_devices(); |
2180 | if (IS_ERR(seed_devices)) | |
2181 | return PTR_ERR(seed_devices); | |
2b82032c | 2182 | |
e4404d6e YZ |
2183 | old_devices = clone_fs_devices(fs_devices); |
2184 | if (IS_ERR(old_devices)) { | |
2185 | kfree(seed_devices); | |
2186 | return PTR_ERR(old_devices); | |
2b82032c | 2187 | } |
e4404d6e | 2188 | |
2b82032c YZ |
2189 | list_add(&old_devices->list, &fs_uuids); |
2190 | ||
e4404d6e YZ |
2191 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2192 | seed_devices->opened = 1; | |
2193 | INIT_LIST_HEAD(&seed_devices->devices); | |
2194 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2195 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
2196 | |
2197 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
2198 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2199 | synchronize_rcu); | |
2196d6e8 MX |
2200 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2201 | device->fs_devices = seed_devices; | |
c9513edb | 2202 | |
2196d6e8 | 2203 | lock_chunks(root); |
e4404d6e | 2204 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
2196d6e8 | 2205 | unlock_chunks(root); |
e4404d6e | 2206 | |
2b82032c YZ |
2207 | fs_devices->seeding = 0; |
2208 | fs_devices->num_devices = 0; | |
2209 | fs_devices->open_devices = 0; | |
69611ac8 | 2210 | fs_devices->missing_devices = 0; |
69611ac8 | 2211 | fs_devices->rotating = 0; |
e4404d6e | 2212 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2213 | |
2214 | generate_random_uuid(fs_devices->fsid); | |
2215 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2216 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
f7171750 FDBM |
2217 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2218 | ||
2b82032c YZ |
2219 | super_flags = btrfs_super_flags(disk_super) & |
2220 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2221 | btrfs_set_super_flags(disk_super, super_flags); | |
2222 | ||
2223 | return 0; | |
2224 | } | |
2225 | ||
2226 | /* | |
01327610 | 2227 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2228 | */ |
2229 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2230 | struct btrfs_root *root) | |
2231 | { | |
2232 | struct btrfs_path *path; | |
2233 | struct extent_buffer *leaf; | |
2234 | struct btrfs_dev_item *dev_item; | |
2235 | struct btrfs_device *device; | |
2236 | struct btrfs_key key; | |
2237 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2238 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2239 | u64 devid; | |
2240 | int ret; | |
2241 | ||
2242 | path = btrfs_alloc_path(); | |
2243 | if (!path) | |
2244 | return -ENOMEM; | |
2245 | ||
2246 | root = root->fs_info->chunk_root; | |
2247 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2248 | key.offset = 0; | |
2249 | key.type = BTRFS_DEV_ITEM_KEY; | |
2250 | ||
2251 | while (1) { | |
2252 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2253 | if (ret < 0) | |
2254 | goto error; | |
2255 | ||
2256 | leaf = path->nodes[0]; | |
2257 | next_slot: | |
2258 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2259 | ret = btrfs_next_leaf(root, path); | |
2260 | if (ret > 0) | |
2261 | break; | |
2262 | if (ret < 0) | |
2263 | goto error; | |
2264 | leaf = path->nodes[0]; | |
2265 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2266 | btrfs_release_path(path); |
2b82032c YZ |
2267 | continue; |
2268 | } | |
2269 | ||
2270 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2271 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2272 | key.type != BTRFS_DEV_ITEM_KEY) | |
2273 | break; | |
2274 | ||
2275 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2276 | struct btrfs_dev_item); | |
2277 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2278 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2279 | BTRFS_UUID_SIZE); |
1473b24e | 2280 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2281 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
2282 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2283 | fs_uuid); | |
79787eaa | 2284 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2285 | |
2286 | if (device->fs_devices->seeding) { | |
2287 | btrfs_set_device_generation(leaf, dev_item, | |
2288 | device->generation); | |
2289 | btrfs_mark_buffer_dirty(leaf); | |
2290 | } | |
2291 | ||
2292 | path->slots[0]++; | |
2293 | goto next_slot; | |
2294 | } | |
2295 | ret = 0; | |
2296 | error: | |
2297 | btrfs_free_path(path); | |
2298 | return ret; | |
2299 | } | |
2300 | ||
788f20eb CM |
2301 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
2302 | { | |
d5e2003c | 2303 | struct request_queue *q; |
788f20eb CM |
2304 | struct btrfs_trans_handle *trans; |
2305 | struct btrfs_device *device; | |
2306 | struct block_device *bdev; | |
788f20eb | 2307 | struct list_head *devices; |
2b82032c | 2308 | struct super_block *sb = root->fs_info->sb; |
606686ee | 2309 | struct rcu_string *name; |
3c1dbdf5 | 2310 | u64 tmp; |
2b82032c | 2311 | int seeding_dev = 0; |
788f20eb CM |
2312 | int ret = 0; |
2313 | ||
2b82032c | 2314 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 2315 | return -EROFS; |
788f20eb | 2316 | |
a5d16333 | 2317 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 2318 | root->fs_info->bdev_holder); |
7f59203a JB |
2319 | if (IS_ERR(bdev)) |
2320 | return PTR_ERR(bdev); | |
a2135011 | 2321 | |
2b82032c YZ |
2322 | if (root->fs_info->fs_devices->seeding) { |
2323 | seeding_dev = 1; | |
2324 | down_write(&sb->s_umount); | |
2325 | mutex_lock(&uuid_mutex); | |
2326 | } | |
2327 | ||
8c8bee1d | 2328 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2329 | |
788f20eb | 2330 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
2331 | |
2332 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 2333 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2334 | if (device->bdev == bdev) { |
2335 | ret = -EEXIST; | |
d25628bd LB |
2336 | mutex_unlock( |
2337 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2338 | goto error; |
788f20eb CM |
2339 | } |
2340 | } | |
d25628bd | 2341 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2342 | |
12bd2fc0 ID |
2343 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
2344 | if (IS_ERR(device)) { | |
788f20eb | 2345 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2346 | ret = PTR_ERR(device); |
2b82032c | 2347 | goto error; |
788f20eb CM |
2348 | } |
2349 | ||
78f2c9e6 | 2350 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2351 | if (!name) { |
788f20eb | 2352 | kfree(device); |
2b82032c YZ |
2353 | ret = -ENOMEM; |
2354 | goto error; | |
788f20eb | 2355 | } |
606686ee | 2356 | rcu_assign_pointer(device->name, name); |
2b82032c | 2357 | |
a22285a6 | 2358 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2359 | if (IS_ERR(trans)) { |
606686ee | 2360 | rcu_string_free(device->name); |
98d5dc13 TI |
2361 | kfree(device); |
2362 | ret = PTR_ERR(trans); | |
2363 | goto error; | |
2364 | } | |
2365 | ||
d5e2003c JB |
2366 | q = bdev_get_queue(bdev); |
2367 | if (blk_queue_discard(q)) | |
2368 | device->can_discard = 1; | |
2b82032c | 2369 | device->writeable = 1; |
2b82032c | 2370 | device->generation = trans->transid; |
788f20eb CM |
2371 | device->io_width = root->sectorsize; |
2372 | device->io_align = root->sectorsize; | |
2373 | device->sector_size = root->sectorsize; | |
2374 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2375 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2376 | device->commit_total_bytes = device->total_bytes; |
788f20eb CM |
2377 | device->dev_root = root->fs_info->dev_root; |
2378 | device->bdev = bdev; | |
dfe25020 | 2379 | device->in_fs_metadata = 1; |
63a212ab | 2380 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2381 | device->mode = FMODE_EXCL; |
27087f37 | 2382 | device->dev_stats_valid = 1; |
2b82032c | 2383 | set_blocksize(device->bdev, 4096); |
788f20eb | 2384 | |
2b82032c YZ |
2385 | if (seeding_dev) { |
2386 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2387 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2388 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2389 | } |
788f20eb | 2390 | |
2b82032c | 2391 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2392 | |
e5e9a520 | 2393 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
2196d6e8 | 2394 | lock_chunks(root); |
1f78160c | 2395 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2396 | list_add(&device->dev_alloc_list, |
2397 | &root->fs_info->fs_devices->alloc_list); | |
2398 | root->fs_info->fs_devices->num_devices++; | |
2399 | root->fs_info->fs_devices->open_devices++; | |
2400 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2401 | root->fs_info->fs_devices->total_devices++; |
2b82032c | 2402 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2403 | |
2bf64758 JB |
2404 | spin_lock(&root->fs_info->free_chunk_lock); |
2405 | root->fs_info->free_chunk_space += device->total_bytes; | |
2406 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2407 | ||
c289811c CM |
2408 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2409 | root->fs_info->fs_devices->rotating = 1; | |
2410 | ||
3c1dbdf5 | 2411 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); |
6c41761f | 2412 | btrfs_set_super_total_bytes(root->fs_info->super_copy, |
3c1dbdf5 | 2413 | tmp + device->total_bytes); |
788f20eb | 2414 | |
3c1dbdf5 | 2415 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); |
6c41761f | 2416 | btrfs_set_super_num_devices(root->fs_info->super_copy, |
3c1dbdf5 | 2417 | tmp + 1); |
0d39376a AJ |
2418 | |
2419 | /* add sysfs device entry */ | |
e3bd6973 | 2420 | btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device); |
0d39376a | 2421 | |
2196d6e8 MX |
2422 | /* |
2423 | * we've got more storage, clear any full flags on the space | |
2424 | * infos | |
2425 | */ | |
2426 | btrfs_clear_space_info_full(root->fs_info); | |
2427 | ||
2428 | unlock_chunks(root); | |
e5e9a520 | 2429 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2430 | |
2b82032c | 2431 | if (seeding_dev) { |
2196d6e8 | 2432 | lock_chunks(root); |
2b82032c | 2433 | ret = init_first_rw_device(trans, root, device); |
2196d6e8 | 2434 | unlock_chunks(root); |
005d6427 | 2435 | if (ret) { |
66642832 | 2436 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2437 | goto error_trans; |
005d6427 | 2438 | } |
2196d6e8 MX |
2439 | } |
2440 | ||
2441 | ret = btrfs_add_device(trans, root, device); | |
2442 | if (ret) { | |
66642832 | 2443 | btrfs_abort_transaction(trans, ret); |
2196d6e8 MX |
2444 | goto error_trans; |
2445 | } | |
2446 | ||
2447 | if (seeding_dev) { | |
2448 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2449 | ||
2b82032c | 2450 | ret = btrfs_finish_sprout(trans, root); |
005d6427 | 2451 | if (ret) { |
66642832 | 2452 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2453 | goto error_trans; |
005d6427 | 2454 | } |
b2373f25 AJ |
2455 | |
2456 | /* Sprouting would change fsid of the mounted root, | |
2457 | * so rename the fsid on the sysfs | |
2458 | */ | |
2459 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2460 | root->fs_info->fsid); | |
c1b7e474 | 2461 | if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj, |
2e7910d6 | 2462 | fsid_buf)) |
f14d104d DS |
2463 | btrfs_warn(root->fs_info, |
2464 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2465 | } |
2466 | ||
5af3e8cc SB |
2467 | root->fs_info->num_tolerated_disk_barrier_failures = |
2468 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2469 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2470 | |
2b82032c YZ |
2471 | if (seeding_dev) { |
2472 | mutex_unlock(&uuid_mutex); | |
2473 | up_write(&sb->s_umount); | |
788f20eb | 2474 | |
79787eaa JM |
2475 | if (ret) /* transaction commit */ |
2476 | return ret; | |
2477 | ||
2b82032c | 2478 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa | 2479 | if (ret < 0) |
34d97007 | 2480 | btrfs_handle_fs_error(root->fs_info, ret, |
5d163e0e | 2481 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2482 | trans = btrfs_attach_transaction(root); |
2483 | if (IS_ERR(trans)) { | |
2484 | if (PTR_ERR(trans) == -ENOENT) | |
2485 | return 0; | |
2486 | return PTR_ERR(trans); | |
2487 | } | |
2488 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2489 | } |
c9e9f97b | 2490 | |
5a1972bd QW |
2491 | /* Update ctime/mtime for libblkid */ |
2492 | update_dev_time(device_path); | |
2b82032c | 2493 | return ret; |
79787eaa JM |
2494 | |
2495 | error_trans: | |
79787eaa | 2496 | btrfs_end_transaction(trans, root); |
606686ee | 2497 | rcu_string_free(device->name); |
32576040 | 2498 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); |
79787eaa | 2499 | kfree(device); |
2b82032c | 2500 | error: |
e525fd89 | 2501 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2502 | if (seeding_dev) { |
2503 | mutex_unlock(&uuid_mutex); | |
2504 | up_write(&sb->s_umount); | |
2505 | } | |
c9e9f97b | 2506 | return ret; |
788f20eb CM |
2507 | } |
2508 | ||
e93c89c1 | 2509 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2510 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2511 | struct btrfs_device **device_out) |
2512 | { | |
2513 | struct request_queue *q; | |
2514 | struct btrfs_device *device; | |
2515 | struct block_device *bdev; | |
2516 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2517 | struct list_head *devices; | |
2518 | struct rcu_string *name; | |
12bd2fc0 | 2519 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2520 | int ret = 0; |
2521 | ||
2522 | *device_out = NULL; | |
1c43366d MX |
2523 | if (fs_info->fs_devices->seeding) { |
2524 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2525 | return -EINVAL; |
1c43366d | 2526 | } |
e93c89c1 SB |
2527 | |
2528 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2529 | fs_info->bdev_holder); | |
1c43366d MX |
2530 | if (IS_ERR(bdev)) { |
2531 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2532 | return PTR_ERR(bdev); |
1c43366d | 2533 | } |
e93c89c1 SB |
2534 | |
2535 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2536 | ||
2537 | devices = &fs_info->fs_devices->devices; | |
2538 | list_for_each_entry(device, devices, dev_list) { | |
2539 | if (device->bdev == bdev) { | |
5d163e0e JM |
2540 | btrfs_err(fs_info, |
2541 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2542 | ret = -EEXIST; |
2543 | goto error; | |
2544 | } | |
2545 | } | |
2546 | ||
1c43366d | 2547 | |
7cc8e58d MX |
2548 | if (i_size_read(bdev->bd_inode) < |
2549 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2550 | btrfs_err(fs_info, |
2551 | "target device is smaller than source device!"); | |
1c43366d MX |
2552 | ret = -EINVAL; |
2553 | goto error; | |
2554 | } | |
2555 | ||
2556 | ||
12bd2fc0 ID |
2557 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2558 | if (IS_ERR(device)) { | |
2559 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2560 | goto error; |
2561 | } | |
2562 | ||
2563 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2564 | if (!name) { | |
2565 | kfree(device); | |
2566 | ret = -ENOMEM; | |
2567 | goto error; | |
2568 | } | |
2569 | rcu_assign_pointer(device->name, name); | |
2570 | ||
2571 | q = bdev_get_queue(bdev); | |
2572 | if (blk_queue_discard(q)) | |
2573 | device->can_discard = 1; | |
2574 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2575 | device->writeable = 1; | |
e93c89c1 SB |
2576 | device->generation = 0; |
2577 | device->io_width = root->sectorsize; | |
2578 | device->io_align = root->sectorsize; | |
2579 | device->sector_size = root->sectorsize; | |
7cc8e58d MX |
2580 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2581 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2582 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2583 | ASSERT(list_empty(&srcdev->resized_list)); |
2584 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2585 | device->commit_bytes_used = device->bytes_used; |
e93c89c1 SB |
2586 | device->dev_root = fs_info->dev_root; |
2587 | device->bdev = bdev; | |
2588 | device->in_fs_metadata = 1; | |
2589 | device->is_tgtdev_for_dev_replace = 1; | |
2590 | device->mode = FMODE_EXCL; | |
27087f37 | 2591 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2592 | set_blocksize(device->bdev, 4096); |
2593 | device->fs_devices = fs_info->fs_devices; | |
2594 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2595 | fs_info->fs_devices->num_devices++; | |
2596 | fs_info->fs_devices->open_devices++; | |
e93c89c1 SB |
2597 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2598 | ||
2599 | *device_out = device; | |
2600 | return ret; | |
2601 | ||
2602 | error: | |
2603 | blkdev_put(bdev, FMODE_EXCL); | |
2604 | return ret; | |
2605 | } | |
2606 | ||
2607 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2608 | struct btrfs_device *tgtdev) | |
2609 | { | |
2610 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2611 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2612 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2613 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2614 | tgtdev->dev_root = fs_info->dev_root; | |
2615 | tgtdev->in_fs_metadata = 1; | |
2616 | } | |
2617 | ||
d397712b CM |
2618 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2619 | struct btrfs_device *device) | |
0b86a832 CM |
2620 | { |
2621 | int ret; | |
2622 | struct btrfs_path *path; | |
2623 | struct btrfs_root *root; | |
2624 | struct btrfs_dev_item *dev_item; | |
2625 | struct extent_buffer *leaf; | |
2626 | struct btrfs_key key; | |
2627 | ||
2628 | root = device->dev_root->fs_info->chunk_root; | |
2629 | ||
2630 | path = btrfs_alloc_path(); | |
2631 | if (!path) | |
2632 | return -ENOMEM; | |
2633 | ||
2634 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2635 | key.type = BTRFS_DEV_ITEM_KEY; | |
2636 | key.offset = device->devid; | |
2637 | ||
2638 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2639 | if (ret < 0) | |
2640 | goto out; | |
2641 | ||
2642 | if (ret > 0) { | |
2643 | ret = -ENOENT; | |
2644 | goto out; | |
2645 | } | |
2646 | ||
2647 | leaf = path->nodes[0]; | |
2648 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2649 | ||
2650 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2651 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2652 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2653 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2654 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2655 | btrfs_set_device_total_bytes(leaf, dev_item, |
2656 | btrfs_device_get_disk_total_bytes(device)); | |
2657 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2658 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2659 | btrfs_mark_buffer_dirty(leaf); |
2660 | ||
2661 | out: | |
2662 | btrfs_free_path(path); | |
2663 | return ret; | |
2664 | } | |
2665 | ||
2196d6e8 | 2666 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2667 | struct btrfs_device *device, u64 new_size) |
2668 | { | |
2669 | struct btrfs_super_block *super_copy = | |
6c41761f | 2670 | device->dev_root->fs_info->super_copy; |
935e5cc9 | 2671 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2672 | u64 old_total; |
2673 | u64 diff; | |
8f18cf13 | 2674 | |
2b82032c YZ |
2675 | if (!device->writeable) |
2676 | return -EACCES; | |
2196d6e8 MX |
2677 | |
2678 | lock_chunks(device->dev_root); | |
2679 | old_total = btrfs_super_total_bytes(super_copy); | |
2680 | diff = new_size - device->total_bytes; | |
2681 | ||
63a212ab | 2682 | if (new_size <= device->total_bytes || |
2196d6e8 MX |
2683 | device->is_tgtdev_for_dev_replace) { |
2684 | unlock_chunks(device->dev_root); | |
2b82032c | 2685 | return -EINVAL; |
2196d6e8 | 2686 | } |
2b82032c | 2687 | |
935e5cc9 | 2688 | fs_devices = device->dev_root->fs_info->fs_devices; |
2b82032c | 2689 | |
8f18cf13 | 2690 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2691 | device->fs_devices->total_rw_bytes += diff; |
2692 | ||
7cc8e58d MX |
2693 | btrfs_device_set_total_bytes(device, new_size); |
2694 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4184ea7f | 2695 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
935e5cc9 MX |
2696 | if (list_empty(&device->resized_list)) |
2697 | list_add_tail(&device->resized_list, | |
2698 | &fs_devices->resized_devices); | |
2196d6e8 | 2699 | unlock_chunks(device->dev_root); |
4184ea7f | 2700 | |
8f18cf13 CM |
2701 | return btrfs_update_device(trans, device); |
2702 | } | |
2703 | ||
2704 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
a688a04a | 2705 | struct btrfs_root *root, u64 chunk_objectid, |
8f18cf13 CM |
2706 | u64 chunk_offset) |
2707 | { | |
2708 | int ret; | |
2709 | struct btrfs_path *path; | |
2710 | struct btrfs_key key; | |
2711 | ||
2712 | root = root->fs_info->chunk_root; | |
2713 | path = btrfs_alloc_path(); | |
2714 | if (!path) | |
2715 | return -ENOMEM; | |
2716 | ||
2717 | key.objectid = chunk_objectid; | |
2718 | key.offset = chunk_offset; | |
2719 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2720 | ||
2721 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2722 | if (ret < 0) |
2723 | goto out; | |
2724 | else if (ret > 0) { /* Logic error or corruption */ | |
34d97007 | 2725 | btrfs_handle_fs_error(root->fs_info, -ENOENT, |
79787eaa JM |
2726 | "Failed lookup while freeing chunk."); |
2727 | ret = -ENOENT; | |
2728 | goto out; | |
2729 | } | |
8f18cf13 CM |
2730 | |
2731 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2732 | if (ret < 0) |
34d97007 | 2733 | btrfs_handle_fs_error(root->fs_info, ret, |
79787eaa JM |
2734 | "Failed to delete chunk item."); |
2735 | out: | |
8f18cf13 | 2736 | btrfs_free_path(path); |
65a246c5 | 2737 | return ret; |
8f18cf13 CM |
2738 | } |
2739 | ||
b2950863 | 2740 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2741 | chunk_offset) |
2742 | { | |
6c41761f | 2743 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2744 | struct btrfs_disk_key *disk_key; |
2745 | struct btrfs_chunk *chunk; | |
2746 | u8 *ptr; | |
2747 | int ret = 0; | |
2748 | u32 num_stripes; | |
2749 | u32 array_size; | |
2750 | u32 len = 0; | |
2751 | u32 cur; | |
2752 | struct btrfs_key key; | |
2753 | ||
2196d6e8 | 2754 | lock_chunks(root); |
8f18cf13 CM |
2755 | array_size = btrfs_super_sys_array_size(super_copy); |
2756 | ||
2757 | ptr = super_copy->sys_chunk_array; | |
2758 | cur = 0; | |
2759 | ||
2760 | while (cur < array_size) { | |
2761 | disk_key = (struct btrfs_disk_key *)ptr; | |
2762 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2763 | ||
2764 | len = sizeof(*disk_key); | |
2765 | ||
2766 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2767 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2768 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2769 | len += btrfs_chunk_item_size(num_stripes); | |
2770 | } else { | |
2771 | ret = -EIO; | |
2772 | break; | |
2773 | } | |
2774 | if (key.objectid == chunk_objectid && | |
2775 | key.offset == chunk_offset) { | |
2776 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2777 | array_size -= len; | |
2778 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2779 | } else { | |
2780 | ptr += len; | |
2781 | cur += len; | |
2782 | } | |
2783 | } | |
2196d6e8 | 2784 | unlock_chunks(root); |
8f18cf13 CM |
2785 | return ret; |
2786 | } | |
2787 | ||
47ab2a6c JB |
2788 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
2789 | struct btrfs_root *root, u64 chunk_offset) | |
8f18cf13 CM |
2790 | { |
2791 | struct extent_map_tree *em_tree; | |
8f18cf13 | 2792 | struct extent_map *em; |
47ab2a6c | 2793 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
8f18cf13 | 2794 | struct map_lookup *map; |
2196d6e8 | 2795 | u64 dev_extent_len = 0; |
47ab2a6c | 2796 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2797 | int i, ret = 0; |
57ba4cb8 | 2798 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
8f18cf13 | 2799 | |
47ab2a6c | 2800 | /* Just in case */ |
8f18cf13 | 2801 | root = root->fs_info->chunk_root; |
8f18cf13 CM |
2802 | em_tree = &root->fs_info->mapping_tree.map_tree; |
2803 | ||
890871be | 2804 | read_lock(&em_tree->lock); |
8f18cf13 | 2805 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2806 | read_unlock(&em_tree->lock); |
8f18cf13 | 2807 | |
47ab2a6c JB |
2808 | if (!em || em->start > chunk_offset || |
2809 | em->start + em->len < chunk_offset) { | |
2810 | /* | |
2811 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2812 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2813 | * do anything we still error out. |
2814 | */ | |
2815 | ASSERT(0); | |
2816 | if (em) | |
2817 | free_extent_map(em); | |
2818 | return -EINVAL; | |
2819 | } | |
95617d69 | 2820 | map = em->map_lookup; |
39c2d7fa | 2821 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 2822 | check_system_chunk(trans, extent_root, map->type); |
39c2d7fa | 2823 | unlock_chunks(root->fs_info->chunk_root); |
8f18cf13 | 2824 | |
57ba4cb8 FM |
2825 | /* |
2826 | * Take the device list mutex to prevent races with the final phase of | |
2827 | * a device replace operation that replaces the device object associated | |
2828 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2829 | */ | |
2830 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2831 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2832 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2833 | ret = btrfs_free_dev_extent(trans, device, |
2834 | map->stripes[i].physical, | |
2835 | &dev_extent_len); | |
47ab2a6c | 2836 | if (ret) { |
57ba4cb8 | 2837 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2838 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2839 | goto out; |
2840 | } | |
a061fc8d | 2841 | |
2196d6e8 MX |
2842 | if (device->bytes_used > 0) { |
2843 | lock_chunks(root); | |
2844 | btrfs_device_set_bytes_used(device, | |
2845 | device->bytes_used - dev_extent_len); | |
2846 | spin_lock(&root->fs_info->free_chunk_lock); | |
2847 | root->fs_info->free_chunk_space += dev_extent_len; | |
2848 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2849 | btrfs_clear_space_info_full(root->fs_info); | |
2850 | unlock_chunks(root); | |
2851 | } | |
a061fc8d | 2852 | |
dfe25020 CM |
2853 | if (map->stripes[i].dev) { |
2854 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2855 | if (ret) { |
57ba4cb8 | 2856 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2857 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2858 | goto out; |
2859 | } | |
dfe25020 | 2860 | } |
8f18cf13 | 2861 | } |
57ba4cb8 FM |
2862 | mutex_unlock(&fs_devices->device_list_mutex); |
2863 | ||
a688a04a | 2864 | ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset); |
47ab2a6c | 2865 | if (ret) { |
66642832 | 2866 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2867 | goto out; |
2868 | } | |
8f18cf13 | 2869 | |
1abe9b8a | 2870 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2871 | ||
8f18cf13 CM |
2872 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2873 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
47ab2a6c | 2874 | if (ret) { |
66642832 | 2875 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2876 | goto out; |
2877 | } | |
8f18cf13 CM |
2878 | } |
2879 | ||
04216820 | 2880 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em); |
47ab2a6c | 2881 | if (ret) { |
66642832 | 2882 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2883 | goto out; |
2884 | } | |
2b82032c | 2885 | |
47ab2a6c | 2886 | out: |
2b82032c YZ |
2887 | /* once for us */ |
2888 | free_extent_map(em); | |
47ab2a6c JB |
2889 | return ret; |
2890 | } | |
2b82032c | 2891 | |
dc2ee4e2 | 2892 | static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset) |
47ab2a6c JB |
2893 | { |
2894 | struct btrfs_root *extent_root; | |
47ab2a6c | 2895 | int ret; |
5d8eb6fe | 2896 | struct btrfs_block_group_cache *block_group; |
2b82032c | 2897 | |
47ab2a6c JB |
2898 | root = root->fs_info->chunk_root; |
2899 | extent_root = root->fs_info->extent_root; | |
2900 | ||
67c5e7d4 FM |
2901 | /* |
2902 | * Prevent races with automatic removal of unused block groups. | |
2903 | * After we relocate and before we remove the chunk with offset | |
2904 | * chunk_offset, automatic removal of the block group can kick in, | |
2905 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2906 | * | |
2907 | * Make sure to acquire this mutex before doing a tree search (dev | |
2908 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2909 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2910 | * we release the path used to search the chunk/dev tree and before | |
2911 | * the current task acquires this mutex and calls us. | |
2912 | */ | |
2913 | ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex)); | |
2914 | ||
47ab2a6c JB |
2915 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2916 | if (ret) | |
2917 | return -ENOSPC; | |
2918 | ||
2919 | /* step one, relocate all the extents inside this chunk */ | |
55e3a601 | 2920 | btrfs_scrub_pause(root); |
47ab2a6c | 2921 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
55e3a601 | 2922 | btrfs_scrub_continue(root); |
47ab2a6c JB |
2923 | if (ret) |
2924 | return ret; | |
2925 | ||
47ab2a6c | 2926 | /* |
5d8eb6fe NA |
2927 | * step two, flag the chunk as removed and let |
2928 | * btrfs_delete_unused_bgs() remove it. | |
47ab2a6c | 2929 | */ |
5d8eb6fe NA |
2930 | block_group = btrfs_lookup_block_group(root->fs_info, chunk_offset); |
2931 | spin_lock(&block_group->lock); | |
2932 | block_group->removed = 1; | |
2933 | spin_unlock(&block_group->lock); | |
2934 | btrfs_put_block_group(block_group); | |
2935 | ||
2936 | return 0; | |
2b82032c YZ |
2937 | } |
2938 | ||
2939 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2940 | { | |
2941 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2942 | struct btrfs_path *path; | |
2943 | struct extent_buffer *leaf; | |
2944 | struct btrfs_chunk *chunk; | |
2945 | struct btrfs_key key; | |
2946 | struct btrfs_key found_key; | |
2b82032c | 2947 | u64 chunk_type; |
ba1bf481 JB |
2948 | bool retried = false; |
2949 | int failed = 0; | |
2b82032c YZ |
2950 | int ret; |
2951 | ||
2952 | path = btrfs_alloc_path(); | |
2953 | if (!path) | |
2954 | return -ENOMEM; | |
2955 | ||
ba1bf481 | 2956 | again: |
2b82032c YZ |
2957 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2958 | key.offset = (u64)-1; | |
2959 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2960 | ||
2961 | while (1) { | |
67c5e7d4 | 2962 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
2b82032c | 2963 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
2964 | if (ret < 0) { |
2965 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c | 2966 | goto error; |
67c5e7d4 | 2967 | } |
79787eaa | 2968 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2969 | |
2970 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2971 | key.type); | |
67c5e7d4 FM |
2972 | if (ret) |
2973 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c YZ |
2974 | if (ret < 0) |
2975 | goto error; | |
2976 | if (ret > 0) | |
2977 | break; | |
1a40e23b | 2978 | |
2b82032c YZ |
2979 | leaf = path->nodes[0]; |
2980 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2981 | |
2b82032c YZ |
2982 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2983 | struct btrfs_chunk); | |
2984 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2985 | btrfs_release_path(path); |
8f18cf13 | 2986 | |
2b82032c | 2987 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
a688a04a | 2988 | ret = btrfs_relocate_chunk(chunk_root, |
2b82032c | 2989 | found_key.offset); |
ba1bf481 JB |
2990 | if (ret == -ENOSPC) |
2991 | failed++; | |
14586651 HS |
2992 | else |
2993 | BUG_ON(ret); | |
2b82032c | 2994 | } |
67c5e7d4 | 2995 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 2996 | |
2b82032c YZ |
2997 | if (found_key.offset == 0) |
2998 | break; | |
2999 | key.offset = found_key.offset - 1; | |
3000 | } | |
3001 | ret = 0; | |
ba1bf481 JB |
3002 | if (failed && !retried) { |
3003 | failed = 0; | |
3004 | retried = true; | |
3005 | goto again; | |
fae7f21c | 3006 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3007 | ret = -ENOSPC; |
3008 | } | |
2b82032c YZ |
3009 | error: |
3010 | btrfs_free_path(path); | |
3011 | return ret; | |
8f18cf13 CM |
3012 | } |
3013 | ||
0940ebf6 ID |
3014 | static int insert_balance_item(struct btrfs_root *root, |
3015 | struct btrfs_balance_control *bctl) | |
3016 | { | |
3017 | struct btrfs_trans_handle *trans; | |
3018 | struct btrfs_balance_item *item; | |
3019 | struct btrfs_disk_balance_args disk_bargs; | |
3020 | struct btrfs_path *path; | |
3021 | struct extent_buffer *leaf; | |
3022 | struct btrfs_key key; | |
3023 | int ret, err; | |
3024 | ||
3025 | path = btrfs_alloc_path(); | |
3026 | if (!path) | |
3027 | return -ENOMEM; | |
3028 | ||
3029 | trans = btrfs_start_transaction(root, 0); | |
3030 | if (IS_ERR(trans)) { | |
3031 | btrfs_free_path(path); | |
3032 | return PTR_ERR(trans); | |
3033 | } | |
3034 | ||
3035 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3036 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3037 | key.offset = 0; |
3038 | ||
3039 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3040 | sizeof(*item)); | |
3041 | if (ret) | |
3042 | goto out; | |
3043 | ||
3044 | leaf = path->nodes[0]; | |
3045 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3046 | ||
3047 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
3048 | ||
3049 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3050 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3051 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3052 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3053 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3054 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3055 | ||
3056 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3057 | ||
3058 | btrfs_mark_buffer_dirty(leaf); | |
3059 | out: | |
3060 | btrfs_free_path(path); | |
3061 | err = btrfs_commit_transaction(trans, root); | |
3062 | if (err && !ret) | |
3063 | ret = err; | |
3064 | return ret; | |
3065 | } | |
3066 | ||
3067 | static int del_balance_item(struct btrfs_root *root) | |
3068 | { | |
3069 | struct btrfs_trans_handle *trans; | |
3070 | struct btrfs_path *path; | |
3071 | struct btrfs_key key; | |
3072 | int ret, err; | |
3073 | ||
3074 | path = btrfs_alloc_path(); | |
3075 | if (!path) | |
3076 | return -ENOMEM; | |
3077 | ||
3078 | trans = btrfs_start_transaction(root, 0); | |
3079 | if (IS_ERR(trans)) { | |
3080 | btrfs_free_path(path); | |
3081 | return PTR_ERR(trans); | |
3082 | } | |
3083 | ||
3084 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3085 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3086 | key.offset = 0; |
3087 | ||
3088 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3089 | if (ret < 0) | |
3090 | goto out; | |
3091 | if (ret > 0) { | |
3092 | ret = -ENOENT; | |
3093 | goto out; | |
3094 | } | |
3095 | ||
3096 | ret = btrfs_del_item(trans, root, path); | |
3097 | out: | |
3098 | btrfs_free_path(path); | |
3099 | err = btrfs_commit_transaction(trans, root); | |
3100 | if (err && !ret) | |
3101 | ret = err; | |
3102 | return ret; | |
3103 | } | |
3104 | ||
59641015 ID |
3105 | /* |
3106 | * This is a heuristic used to reduce the number of chunks balanced on | |
3107 | * resume after balance was interrupted. | |
3108 | */ | |
3109 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3110 | { | |
3111 | /* | |
3112 | * Turn on soft mode for chunk types that were being converted. | |
3113 | */ | |
3114 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3115 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3116 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3117 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3118 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3119 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3120 | ||
3121 | /* | |
3122 | * Turn on usage filter if is not already used. The idea is | |
3123 | * that chunks that we have already balanced should be | |
3124 | * reasonably full. Don't do it for chunks that are being | |
3125 | * converted - that will keep us from relocating unconverted | |
3126 | * (albeit full) chunks. | |
3127 | */ | |
3128 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3129 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3130 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3131 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3132 | bctl->data.usage = 90; | |
3133 | } | |
3134 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3135 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3136 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3137 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3138 | bctl->sys.usage = 90; | |
3139 | } | |
3140 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3141 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3142 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3143 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3144 | bctl->meta.usage = 90; | |
3145 | } | |
3146 | } | |
3147 | ||
c9e9f97b ID |
3148 | /* |
3149 | * Should be called with both balance and volume mutexes held to | |
3150 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3151 | * restriper. Same goes for unset_balance_control. | |
3152 | */ | |
3153 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3154 | { | |
3155 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3156 | ||
3157 | BUG_ON(fs_info->balance_ctl); | |
3158 | ||
3159 | spin_lock(&fs_info->balance_lock); | |
3160 | fs_info->balance_ctl = bctl; | |
3161 | spin_unlock(&fs_info->balance_lock); | |
3162 | } | |
3163 | ||
3164 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3165 | { | |
3166 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3167 | ||
3168 | BUG_ON(!fs_info->balance_ctl); | |
3169 | ||
3170 | spin_lock(&fs_info->balance_lock); | |
3171 | fs_info->balance_ctl = NULL; | |
3172 | spin_unlock(&fs_info->balance_lock); | |
3173 | ||
3174 | kfree(bctl); | |
3175 | } | |
3176 | ||
ed25e9b2 ID |
3177 | /* |
3178 | * Balance filters. Return 1 if chunk should be filtered out | |
3179 | * (should not be balanced). | |
3180 | */ | |
899c81ea | 3181 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3182 | struct btrfs_balance_args *bargs) |
3183 | { | |
899c81ea ID |
3184 | chunk_type = chunk_to_extended(chunk_type) & |
3185 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3186 | |
899c81ea | 3187 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3188 | return 0; |
3189 | ||
3190 | return 1; | |
3191 | } | |
3192 | ||
dba72cb3 | 3193 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3194 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3195 | { |
3196 | struct btrfs_block_group_cache *cache; | |
3197 | u64 chunk_used; | |
3198 | u64 user_thresh_min; | |
3199 | u64 user_thresh_max; | |
3200 | int ret = 1; | |
3201 | ||
3202 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3203 | chunk_used = btrfs_block_group_used(&cache->item); | |
3204 | ||
3205 | if (bargs->usage_min == 0) | |
3206 | user_thresh_min = 0; | |
3207 | else | |
3208 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3209 | bargs->usage_min); | |
3210 | ||
3211 | if (bargs->usage_max == 0) | |
3212 | user_thresh_max = 1; | |
3213 | else if (bargs->usage_max > 100) | |
3214 | user_thresh_max = cache->key.offset; | |
3215 | else | |
3216 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3217 | bargs->usage_max); | |
3218 | ||
3219 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3220 | ret = 0; | |
3221 | ||
3222 | btrfs_put_block_group(cache); | |
3223 | return ret; | |
3224 | } | |
3225 | ||
dba72cb3 | 3226 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3227 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3228 | { |
3229 | struct btrfs_block_group_cache *cache; | |
3230 | u64 chunk_used, user_thresh; | |
3231 | int ret = 1; | |
3232 | ||
3233 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3234 | chunk_used = btrfs_block_group_used(&cache->item); | |
3235 | ||
bc309467 | 3236 | if (bargs->usage_min == 0) |
3e39cea6 | 3237 | user_thresh = 1; |
a105bb88 ID |
3238 | else if (bargs->usage > 100) |
3239 | user_thresh = cache->key.offset; | |
3240 | else | |
3241 | user_thresh = div_factor_fine(cache->key.offset, | |
3242 | bargs->usage); | |
3243 | ||
5ce5b3c0 ID |
3244 | if (chunk_used < user_thresh) |
3245 | ret = 0; | |
3246 | ||
3247 | btrfs_put_block_group(cache); | |
3248 | return ret; | |
3249 | } | |
3250 | ||
409d404b ID |
3251 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3252 | struct btrfs_chunk *chunk, | |
3253 | struct btrfs_balance_args *bargs) | |
3254 | { | |
3255 | struct btrfs_stripe *stripe; | |
3256 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3257 | int i; | |
3258 | ||
3259 | for (i = 0; i < num_stripes; i++) { | |
3260 | stripe = btrfs_stripe_nr(chunk, i); | |
3261 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3262 | return 0; | |
3263 | } | |
3264 | ||
3265 | return 1; | |
3266 | } | |
3267 | ||
94e60d5a ID |
3268 | /* [pstart, pend) */ |
3269 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3270 | struct btrfs_chunk *chunk, | |
3271 | u64 chunk_offset, | |
3272 | struct btrfs_balance_args *bargs) | |
3273 | { | |
3274 | struct btrfs_stripe *stripe; | |
3275 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3276 | u64 stripe_offset; | |
3277 | u64 stripe_length; | |
3278 | int factor; | |
3279 | int i; | |
3280 | ||
3281 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3282 | return 0; | |
3283 | ||
3284 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3285 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3286 | factor = num_stripes / 2; | |
3287 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3288 | factor = num_stripes - 1; | |
3289 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3290 | factor = num_stripes - 2; | |
3291 | } else { | |
3292 | factor = num_stripes; | |
3293 | } | |
94e60d5a ID |
3294 | |
3295 | for (i = 0; i < num_stripes; i++) { | |
3296 | stripe = btrfs_stripe_nr(chunk, i); | |
3297 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3298 | continue; | |
3299 | ||
3300 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3301 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3302 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3303 | |
3304 | if (stripe_offset < bargs->pend && | |
3305 | stripe_offset + stripe_length > bargs->pstart) | |
3306 | return 0; | |
3307 | } | |
3308 | ||
3309 | return 1; | |
3310 | } | |
3311 | ||
ea67176a ID |
3312 | /* [vstart, vend) */ |
3313 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3314 | struct btrfs_chunk *chunk, | |
3315 | u64 chunk_offset, | |
3316 | struct btrfs_balance_args *bargs) | |
3317 | { | |
3318 | if (chunk_offset < bargs->vend && | |
3319 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3320 | /* at least part of the chunk is inside this vrange */ | |
3321 | return 0; | |
3322 | ||
3323 | return 1; | |
3324 | } | |
3325 | ||
dee32d0a GAP |
3326 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3327 | struct btrfs_chunk *chunk, | |
3328 | struct btrfs_balance_args *bargs) | |
3329 | { | |
3330 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3331 | ||
3332 | if (bargs->stripes_min <= num_stripes | |
3333 | && num_stripes <= bargs->stripes_max) | |
3334 | return 0; | |
3335 | ||
3336 | return 1; | |
3337 | } | |
3338 | ||
899c81ea | 3339 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3340 | struct btrfs_balance_args *bargs) |
3341 | { | |
3342 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3343 | return 0; | |
3344 | ||
899c81ea ID |
3345 | chunk_type = chunk_to_extended(chunk_type) & |
3346 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3347 | |
899c81ea | 3348 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3349 | return 1; |
3350 | ||
3351 | return 0; | |
3352 | } | |
3353 | ||
f43ffb60 ID |
3354 | static int should_balance_chunk(struct btrfs_root *root, |
3355 | struct extent_buffer *leaf, | |
3356 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3357 | { | |
3358 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3359 | struct btrfs_balance_args *bargs = NULL; | |
3360 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3361 | ||
3362 | /* type filter */ | |
3363 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3364 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3365 | return 0; | |
3366 | } | |
3367 | ||
3368 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3369 | bargs = &bctl->data; | |
3370 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3371 | bargs = &bctl->sys; | |
3372 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3373 | bargs = &bctl->meta; | |
3374 | ||
ed25e9b2 ID |
3375 | /* profiles filter */ |
3376 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3377 | chunk_profiles_filter(chunk_type, bargs)) { | |
3378 | return 0; | |
5ce5b3c0 ID |
3379 | } |
3380 | ||
3381 | /* usage filter */ | |
3382 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3383 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3384 | return 0; | |
bc309467 DS |
3385 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
3386 | chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3387 | return 0; | |
409d404b ID |
3388 | } |
3389 | ||
3390 | /* devid filter */ | |
3391 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3392 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3393 | return 0; | |
94e60d5a ID |
3394 | } |
3395 | ||
3396 | /* drange filter, makes sense only with devid filter */ | |
3397 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3398 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3399 | return 0; | |
ea67176a ID |
3400 | } |
3401 | ||
3402 | /* vrange filter */ | |
3403 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3404 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3405 | return 0; | |
ed25e9b2 ID |
3406 | } |
3407 | ||
dee32d0a GAP |
3408 | /* stripes filter */ |
3409 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3410 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3411 | return 0; | |
3412 | } | |
3413 | ||
cfa4c961 ID |
3414 | /* soft profile changing mode */ |
3415 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3416 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3417 | return 0; | |
3418 | } | |
3419 | ||
7d824b6f DS |
3420 | /* |
3421 | * limited by count, must be the last filter | |
3422 | */ | |
3423 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3424 | if (bargs->limit == 0) | |
3425 | return 0; | |
3426 | else | |
3427 | bargs->limit--; | |
12907fc7 DS |
3428 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3429 | /* | |
3430 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3431 | * determined here because we do not have the global information |
12907fc7 DS |
3432 | * about the count of all chunks that satisfy the filters. |
3433 | */ | |
3434 | if (bargs->limit_max == 0) | |
3435 | return 0; | |
3436 | else | |
3437 | bargs->limit_max--; | |
7d824b6f DS |
3438 | } |
3439 | ||
f43ffb60 ID |
3440 | return 1; |
3441 | } | |
3442 | ||
c9e9f97b | 3443 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3444 | { |
19a39dce | 3445 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3446 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3447 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3448 | struct list_head *devices; | |
ec44a35c CM |
3449 | struct btrfs_device *device; |
3450 | u64 old_size; | |
3451 | u64 size_to_free; | |
12907fc7 | 3452 | u64 chunk_type; |
f43ffb60 | 3453 | struct btrfs_chunk *chunk; |
5a488b9d | 3454 | struct btrfs_path *path = NULL; |
ec44a35c | 3455 | struct btrfs_key key; |
ec44a35c | 3456 | struct btrfs_key found_key; |
c9e9f97b | 3457 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3458 | struct extent_buffer *leaf; |
3459 | int slot; | |
c9e9f97b ID |
3460 | int ret; |
3461 | int enospc_errors = 0; | |
19a39dce | 3462 | bool counting = true; |
12907fc7 | 3463 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3464 | u64 limit_data = bctl->data.limit; |
3465 | u64 limit_meta = bctl->meta.limit; | |
3466 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3467 | u32 count_data = 0; |
3468 | u32 count_meta = 0; | |
3469 | u32 count_sys = 0; | |
2c9fe835 | 3470 | int chunk_reserved = 0; |
cf25ce51 | 3471 | u64 bytes_used = 0; |
ec44a35c | 3472 | |
ec44a35c | 3473 | /* step one make some room on all the devices */ |
c9e9f97b | 3474 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3475 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3476 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3477 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3478 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
2b82032c | 3479 | if (!device->writeable || |
7cc8e58d MX |
3480 | btrfs_device_get_total_bytes(device) - |
3481 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3482 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3483 | continue; |
3484 | ||
3485 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3486 | if (ret == -ENOSPC) |
3487 | break; | |
5a488b9d LB |
3488 | if (ret) { |
3489 | /* btrfs_shrink_device never returns ret > 0 */ | |
3490 | WARN_ON(ret > 0); | |
3491 | goto error; | |
3492 | } | |
ec44a35c | 3493 | |
a22285a6 | 3494 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3495 | if (IS_ERR(trans)) { |
3496 | ret = PTR_ERR(trans); | |
3497 | btrfs_info_in_rcu(fs_info, | |
3498 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3499 | rcu_str_deref(device->name), ret, | |
3500 | old_size, old_size - size_to_free); | |
3501 | goto error; | |
3502 | } | |
ec44a35c CM |
3503 | |
3504 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d LB |
3505 | if (ret) { |
3506 | btrfs_end_transaction(trans, dev_root); | |
3507 | /* btrfs_grow_device never returns ret > 0 */ | |
3508 | WARN_ON(ret > 0); | |
3509 | btrfs_info_in_rcu(fs_info, | |
3510 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3511 | rcu_str_deref(device->name), ret, | |
3512 | old_size, old_size - size_to_free); | |
3513 | goto error; | |
3514 | } | |
ec44a35c CM |
3515 | |
3516 | btrfs_end_transaction(trans, dev_root); | |
3517 | } | |
3518 | ||
3519 | /* step two, relocate all the chunks */ | |
3520 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3521 | if (!path) { |
3522 | ret = -ENOMEM; | |
3523 | goto error; | |
3524 | } | |
19a39dce ID |
3525 | |
3526 | /* zero out stat counters */ | |
3527 | spin_lock(&fs_info->balance_lock); | |
3528 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3529 | spin_unlock(&fs_info->balance_lock); | |
3530 | again: | |
7d824b6f | 3531 | if (!counting) { |
12907fc7 DS |
3532 | /* |
3533 | * The single value limit and min/max limits use the same bytes | |
3534 | * in the | |
3535 | */ | |
7d824b6f DS |
3536 | bctl->data.limit = limit_data; |
3537 | bctl->meta.limit = limit_meta; | |
3538 | bctl->sys.limit = limit_sys; | |
3539 | } | |
ec44a35c CM |
3540 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3541 | key.offset = (u64)-1; | |
3542 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3543 | ||
d397712b | 3544 | while (1) { |
19a39dce | 3545 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3546 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3547 | ret = -ECANCELED; |
3548 | goto error; | |
3549 | } | |
3550 | ||
67c5e7d4 | 3551 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3552 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3553 | if (ret < 0) { |
3554 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3555 | goto error; |
67c5e7d4 | 3556 | } |
ec44a35c CM |
3557 | |
3558 | /* | |
3559 | * this shouldn't happen, it means the last relocate | |
3560 | * failed | |
3561 | */ | |
3562 | if (ret == 0) | |
c9e9f97b | 3563 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3564 | |
3565 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3566 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3567 | if (ret) { |
67c5e7d4 | 3568 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3569 | ret = 0; |
ec44a35c | 3570 | break; |
c9e9f97b | 3571 | } |
7d9eb12c | 3572 | |
f43ffb60 ID |
3573 | leaf = path->nodes[0]; |
3574 | slot = path->slots[0]; | |
3575 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3576 | |
67c5e7d4 FM |
3577 | if (found_key.objectid != key.objectid) { |
3578 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3579 | break; |
67c5e7d4 | 3580 | } |
7d9eb12c | 3581 | |
f43ffb60 | 3582 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3583 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3584 | |
19a39dce ID |
3585 | if (!counting) { |
3586 | spin_lock(&fs_info->balance_lock); | |
3587 | bctl->stat.considered++; | |
3588 | spin_unlock(&fs_info->balance_lock); | |
3589 | } | |
3590 | ||
f43ffb60 ID |
3591 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3592 | found_key.offset); | |
2c9fe835 | 3593 | |
b3b4aa74 | 3594 | btrfs_release_path(path); |
67c5e7d4 FM |
3595 | if (!ret) { |
3596 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3597 | goto loop; |
67c5e7d4 | 3598 | } |
f43ffb60 | 3599 | |
19a39dce | 3600 | if (counting) { |
67c5e7d4 | 3601 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3602 | spin_lock(&fs_info->balance_lock); |
3603 | bctl->stat.expected++; | |
3604 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3605 | |
3606 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3607 | count_data++; | |
3608 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3609 | count_sys++; | |
3610 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3611 | count_meta++; | |
3612 | ||
3613 | goto loop; | |
3614 | } | |
3615 | ||
3616 | /* | |
3617 | * Apply limit_min filter, no need to check if the LIMITS | |
3618 | * filter is used, limit_min is 0 by default | |
3619 | */ | |
3620 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3621 | count_data < bctl->data.limit_min) | |
3622 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3623 | count_meta < bctl->meta.limit_min) | |
3624 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3625 | count_sys < bctl->sys.limit_min)) { | |
3626 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3627 | goto loop; |
3628 | } | |
3629 | ||
cf25ce51 LB |
3630 | ASSERT(fs_info->data_sinfo); |
3631 | spin_lock(&fs_info->data_sinfo->lock); | |
3632 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3633 | spin_unlock(&fs_info->data_sinfo->lock); | |
3634 | ||
3635 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3636 | !chunk_reserved && !bytes_used) { | |
2c9fe835 ZL |
3637 | trans = btrfs_start_transaction(chunk_root, 0); |
3638 | if (IS_ERR(trans)) { | |
3639 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3640 | ret = PTR_ERR(trans); | |
3641 | goto error; | |
3642 | } | |
3643 | ||
3644 | ret = btrfs_force_chunk_alloc(trans, chunk_root, | |
3645 | BTRFS_BLOCK_GROUP_DATA); | |
8a7d656f | 3646 | btrfs_end_transaction(trans, chunk_root); |
2c9fe835 ZL |
3647 | if (ret < 0) { |
3648 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3649 | goto error; | |
3650 | } | |
2c9fe835 ZL |
3651 | chunk_reserved = 1; |
3652 | } | |
3653 | ||
ec44a35c | 3654 | ret = btrfs_relocate_chunk(chunk_root, |
ec44a35c | 3655 | found_key.offset); |
67c5e7d4 | 3656 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3657 | if (ret && ret != -ENOSPC) |
3658 | goto error; | |
19a39dce | 3659 | if (ret == -ENOSPC) { |
c9e9f97b | 3660 | enospc_errors++; |
19a39dce ID |
3661 | } else { |
3662 | spin_lock(&fs_info->balance_lock); | |
3663 | bctl->stat.completed++; | |
3664 | spin_unlock(&fs_info->balance_lock); | |
3665 | } | |
f43ffb60 | 3666 | loop: |
795a3321 ID |
3667 | if (found_key.offset == 0) |
3668 | break; | |
ba1bf481 | 3669 | key.offset = found_key.offset - 1; |
ec44a35c | 3670 | } |
c9e9f97b | 3671 | |
19a39dce ID |
3672 | if (counting) { |
3673 | btrfs_release_path(path); | |
3674 | counting = false; | |
3675 | goto again; | |
3676 | } | |
ec44a35c CM |
3677 | error: |
3678 | btrfs_free_path(path); | |
c9e9f97b | 3679 | if (enospc_errors) { |
efe120a0 | 3680 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3681 | enospc_errors); |
c9e9f97b ID |
3682 | if (!ret) |
3683 | ret = -ENOSPC; | |
3684 | } | |
3685 | ||
ec44a35c CM |
3686 | return ret; |
3687 | } | |
3688 | ||
0c460c0d ID |
3689 | /** |
3690 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3691 | * @flags: profile to validate | |
3692 | * @extended: if true @flags is treated as an extended profile | |
3693 | */ | |
3694 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3695 | { | |
3696 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3697 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3698 | ||
3699 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3700 | ||
3701 | /* 1) check that all other bits are zeroed */ | |
3702 | if (flags & ~mask) | |
3703 | return 0; | |
3704 | ||
3705 | /* 2) see if profile is reduced */ | |
3706 | if (flags == 0) | |
3707 | return !extended; /* "0" is valid for usual profiles */ | |
3708 | ||
3709 | /* true if exactly one bit set */ | |
3710 | return (flags & (flags - 1)) == 0; | |
3711 | } | |
3712 | ||
837d5b6e ID |
3713 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3714 | { | |
a7e99c69 ID |
3715 | /* cancel requested || normal exit path */ |
3716 | return atomic_read(&fs_info->balance_cancel_req) || | |
3717 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3718 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3719 | } |
3720 | ||
c9e9f97b ID |
3721 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3722 | { | |
0940ebf6 ID |
3723 | int ret; |
3724 | ||
c9e9f97b | 3725 | unset_balance_control(fs_info); |
0940ebf6 | 3726 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 | 3727 | if (ret) |
34d97007 | 3728 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f ID |
3729 | |
3730 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3731 | } |
3732 | ||
bdcd3c97 AM |
3733 | /* Non-zero return value signifies invalidity */ |
3734 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3735 | u64 allowed) | |
3736 | { | |
3737 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3738 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3739 | (bctl_arg->target & ~allowed))); | |
3740 | } | |
3741 | ||
c9e9f97b ID |
3742 | /* |
3743 | * Should be called with both balance and volume mutexes held | |
3744 | */ | |
3745 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3746 | struct btrfs_ioctl_balance_args *bargs) | |
3747 | { | |
3748 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3749 | u64 allowed; |
e4837f8f | 3750 | int mixed = 0; |
c9e9f97b | 3751 | int ret; |
8dabb742 | 3752 | u64 num_devices; |
de98ced9 | 3753 | unsigned seq; |
c9e9f97b | 3754 | |
837d5b6e | 3755 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3756 | atomic_read(&fs_info->balance_pause_req) || |
3757 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3758 | ret = -EINVAL; |
3759 | goto out; | |
3760 | } | |
3761 | ||
e4837f8f ID |
3762 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3763 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3764 | mixed = 1; | |
3765 | ||
f43ffb60 ID |
3766 | /* |
3767 | * In case of mixed groups both data and meta should be picked, | |
3768 | * and identical options should be given for both of them. | |
3769 | */ | |
e4837f8f ID |
3770 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3771 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3772 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3773 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3774 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3775 | btrfs_err(fs_info, |
3776 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3777 | ret = -EINVAL; |
3778 | goto out; | |
3779 | } | |
3780 | } | |
3781 | ||
8dabb742 | 3782 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3783 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3784 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3785 | BUG_ON(num_devices < 1); | |
3786 | num_devices--; | |
3787 | } | |
73beece9 | 3788 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
88be159c AH |
3789 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3790 | if (num_devices > 1) | |
e4d8ec0f | 3791 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3792 | if (num_devices > 2) |
3793 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3794 | if (num_devices > 3) | |
3795 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3796 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3797 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3798 | btrfs_err(fs_info, |
3799 | "unable to start balance with target data profile %llu", | |
3800 | bctl->data.target); | |
e4d8ec0f ID |
3801 | ret = -EINVAL; |
3802 | goto out; | |
3803 | } | |
bdcd3c97 | 3804 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3805 | btrfs_err(fs_info, |
5d163e0e JM |
3806 | "unable to start balance with target metadata profile %llu", |
3807 | bctl->meta.target); | |
e4d8ec0f ID |
3808 | ret = -EINVAL; |
3809 | goto out; | |
3810 | } | |
bdcd3c97 | 3811 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3812 | btrfs_err(fs_info, |
5d163e0e JM |
3813 | "unable to start balance with target system profile %llu", |
3814 | bctl->sys.target); | |
e4d8ec0f ID |
3815 | ret = -EINVAL; |
3816 | goto out; | |
3817 | } | |
3818 | ||
e4d8ec0f ID |
3819 | /* allow to reduce meta or sys integrity only if force set */ |
3820 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3821 | BTRFS_BLOCK_GROUP_RAID10 | |
3822 | BTRFS_BLOCK_GROUP_RAID5 | | |
3823 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3824 | do { |
3825 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3826 | ||
3827 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3828 | (fs_info->avail_system_alloc_bits & allowed) && | |
3829 | !(bctl->sys.target & allowed)) || | |
3830 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3831 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3832 | !(bctl->meta.target & allowed))) { | |
3833 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3834 | btrfs_info(fs_info, |
3835 | "force reducing metadata integrity"); | |
de98ced9 | 3836 | } else { |
5d163e0e JM |
3837 | btrfs_err(fs_info, |
3838 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3839 | ret = -EINVAL; |
3840 | goto out; | |
3841 | } | |
e4d8ec0f | 3842 | } |
de98ced9 | 3843 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3844 | |
ee592d07 ST |
3845 | if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) < |
3846 | btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) { | |
3847 | btrfs_warn(fs_info, | |
5d163e0e JM |
3848 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
3849 | bctl->meta.target, bctl->data.target); | |
ee592d07 ST |
3850 | } |
3851 | ||
5af3e8cc | 3852 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
943c6e99 ZL |
3853 | fs_info->num_tolerated_disk_barrier_failures = min( |
3854 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3855 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3856 | bctl->sys.target)); | |
5af3e8cc SB |
3857 | } |
3858 | ||
0940ebf6 | 3859 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3860 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3861 | goto out; |
3862 | ||
59641015 ID |
3863 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3864 | BUG_ON(ret == -EEXIST); | |
3865 | set_balance_control(bctl); | |
3866 | } else { | |
3867 | BUG_ON(ret != -EEXIST); | |
3868 | spin_lock(&fs_info->balance_lock); | |
3869 | update_balance_args(bctl); | |
3870 | spin_unlock(&fs_info->balance_lock); | |
3871 | } | |
c9e9f97b | 3872 | |
837d5b6e | 3873 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3874 | mutex_unlock(&fs_info->balance_mutex); |
3875 | ||
3876 | ret = __btrfs_balance(fs_info); | |
3877 | ||
3878 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3879 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3880 | |
bf023ecf ID |
3881 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3882 | fs_info->num_tolerated_disk_barrier_failures = | |
3883 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3884 | } | |
3885 | ||
c9e9f97b ID |
3886 | if (bargs) { |
3887 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3888 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3889 | } |
3890 | ||
3a01aa7a ID |
3891 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3892 | balance_need_close(fs_info)) { | |
3893 | __cancel_balance(fs_info); | |
3894 | } | |
3895 | ||
837d5b6e | 3896 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3897 | |
3898 | return ret; | |
3899 | out: | |
59641015 ID |
3900 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3901 | __cancel_balance(fs_info); | |
ed0fb78f | 3902 | else { |
59641015 | 3903 | kfree(bctl); |
ed0fb78f ID |
3904 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3905 | } | |
59641015 ID |
3906 | return ret; |
3907 | } | |
3908 | ||
3909 | static int balance_kthread(void *data) | |
3910 | { | |
2b6ba629 | 3911 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3912 | int ret = 0; |
59641015 ID |
3913 | |
3914 | mutex_lock(&fs_info->volume_mutex); | |
3915 | mutex_lock(&fs_info->balance_mutex); | |
3916 | ||
2b6ba629 | 3917 | if (fs_info->balance_ctl) { |
efe120a0 | 3918 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3919 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3920 | } |
59641015 ID |
3921 | |
3922 | mutex_unlock(&fs_info->balance_mutex); | |
3923 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3924 | |
59641015 ID |
3925 | return ret; |
3926 | } | |
3927 | ||
2b6ba629 ID |
3928 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3929 | { | |
3930 | struct task_struct *tsk; | |
3931 | ||
3932 | spin_lock(&fs_info->balance_lock); | |
3933 | if (!fs_info->balance_ctl) { | |
3934 | spin_unlock(&fs_info->balance_lock); | |
3935 | return 0; | |
3936 | } | |
3937 | spin_unlock(&fs_info->balance_lock); | |
3938 | ||
3cdde224 | 3939 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 3940 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3941 | return 0; |
3942 | } | |
3943 | ||
3944 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3945 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3946 | } |
3947 | ||
68310a5e | 3948 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3949 | { |
59641015 ID |
3950 | struct btrfs_balance_control *bctl; |
3951 | struct btrfs_balance_item *item; | |
3952 | struct btrfs_disk_balance_args disk_bargs; | |
3953 | struct btrfs_path *path; | |
3954 | struct extent_buffer *leaf; | |
3955 | struct btrfs_key key; | |
3956 | int ret; | |
3957 | ||
3958 | path = btrfs_alloc_path(); | |
3959 | if (!path) | |
3960 | return -ENOMEM; | |
3961 | ||
59641015 | 3962 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3963 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
3964 | key.offset = 0; |
3965 | ||
68310a5e | 3966 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3967 | if (ret < 0) |
68310a5e | 3968 | goto out; |
59641015 ID |
3969 | if (ret > 0) { /* ret = -ENOENT; */ |
3970 | ret = 0; | |
68310a5e ID |
3971 | goto out; |
3972 | } | |
3973 | ||
3974 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3975 | if (!bctl) { | |
3976 | ret = -ENOMEM; | |
3977 | goto out; | |
59641015 ID |
3978 | } |
3979 | ||
3980 | leaf = path->nodes[0]; | |
3981 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3982 | ||
68310a5e ID |
3983 | bctl->fs_info = fs_info; |
3984 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3985 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3986 | |
3987 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3988 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3989 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3990 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3991 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3992 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3993 | ||
ed0fb78f ID |
3994 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3995 | ||
68310a5e ID |
3996 | mutex_lock(&fs_info->volume_mutex); |
3997 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3998 | |
68310a5e ID |
3999 | set_balance_control(bctl); |
4000 | ||
4001 | mutex_unlock(&fs_info->balance_mutex); | |
4002 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4003 | out: |
4004 | btrfs_free_path(path); | |
ec44a35c CM |
4005 | return ret; |
4006 | } | |
4007 | ||
837d5b6e ID |
4008 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4009 | { | |
4010 | int ret = 0; | |
4011 | ||
4012 | mutex_lock(&fs_info->balance_mutex); | |
4013 | if (!fs_info->balance_ctl) { | |
4014 | mutex_unlock(&fs_info->balance_mutex); | |
4015 | return -ENOTCONN; | |
4016 | } | |
4017 | ||
4018 | if (atomic_read(&fs_info->balance_running)) { | |
4019 | atomic_inc(&fs_info->balance_pause_req); | |
4020 | mutex_unlock(&fs_info->balance_mutex); | |
4021 | ||
4022 | wait_event(fs_info->balance_wait_q, | |
4023 | atomic_read(&fs_info->balance_running) == 0); | |
4024 | ||
4025 | mutex_lock(&fs_info->balance_mutex); | |
4026 | /* we are good with balance_ctl ripped off from under us */ | |
4027 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4028 | atomic_dec(&fs_info->balance_pause_req); | |
4029 | } else { | |
4030 | ret = -ENOTCONN; | |
4031 | } | |
4032 | ||
4033 | mutex_unlock(&fs_info->balance_mutex); | |
4034 | return ret; | |
4035 | } | |
4036 | ||
a7e99c69 ID |
4037 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4038 | { | |
e649e587 ID |
4039 | if (fs_info->sb->s_flags & MS_RDONLY) |
4040 | return -EROFS; | |
4041 | ||
a7e99c69 ID |
4042 | mutex_lock(&fs_info->balance_mutex); |
4043 | if (!fs_info->balance_ctl) { | |
4044 | mutex_unlock(&fs_info->balance_mutex); | |
4045 | return -ENOTCONN; | |
4046 | } | |
4047 | ||
4048 | atomic_inc(&fs_info->balance_cancel_req); | |
4049 | /* | |
4050 | * if we are running just wait and return, balance item is | |
4051 | * deleted in btrfs_balance in this case | |
4052 | */ | |
4053 | if (atomic_read(&fs_info->balance_running)) { | |
4054 | mutex_unlock(&fs_info->balance_mutex); | |
4055 | wait_event(fs_info->balance_wait_q, | |
4056 | atomic_read(&fs_info->balance_running) == 0); | |
4057 | mutex_lock(&fs_info->balance_mutex); | |
4058 | } else { | |
4059 | /* __cancel_balance needs volume_mutex */ | |
4060 | mutex_unlock(&fs_info->balance_mutex); | |
4061 | mutex_lock(&fs_info->volume_mutex); | |
4062 | mutex_lock(&fs_info->balance_mutex); | |
4063 | ||
4064 | if (fs_info->balance_ctl) | |
4065 | __cancel_balance(fs_info); | |
4066 | ||
4067 | mutex_unlock(&fs_info->volume_mutex); | |
4068 | } | |
4069 | ||
4070 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4071 | atomic_dec(&fs_info->balance_cancel_req); | |
4072 | mutex_unlock(&fs_info->balance_mutex); | |
4073 | return 0; | |
4074 | } | |
4075 | ||
803b2f54 SB |
4076 | static int btrfs_uuid_scan_kthread(void *data) |
4077 | { | |
4078 | struct btrfs_fs_info *fs_info = data; | |
4079 | struct btrfs_root *root = fs_info->tree_root; | |
4080 | struct btrfs_key key; | |
4081 | struct btrfs_key max_key; | |
4082 | struct btrfs_path *path = NULL; | |
4083 | int ret = 0; | |
4084 | struct extent_buffer *eb; | |
4085 | int slot; | |
4086 | struct btrfs_root_item root_item; | |
4087 | u32 item_size; | |
f45388f3 | 4088 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4089 | |
4090 | path = btrfs_alloc_path(); | |
4091 | if (!path) { | |
4092 | ret = -ENOMEM; | |
4093 | goto out; | |
4094 | } | |
4095 | ||
4096 | key.objectid = 0; | |
4097 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4098 | key.offset = 0; | |
4099 | ||
4100 | max_key.objectid = (u64)-1; | |
4101 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
4102 | max_key.offset = (u64)-1; | |
4103 | ||
803b2f54 | 4104 | while (1) { |
6174d3cb | 4105 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4106 | if (ret) { |
4107 | if (ret > 0) | |
4108 | ret = 0; | |
4109 | break; | |
4110 | } | |
4111 | ||
4112 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4113 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4114 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4115 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4116 | goto skip; | |
4117 | ||
4118 | eb = path->nodes[0]; | |
4119 | slot = path->slots[0]; | |
4120 | item_size = btrfs_item_size_nr(eb, slot); | |
4121 | if (item_size < sizeof(root_item)) | |
4122 | goto skip; | |
4123 | ||
803b2f54 SB |
4124 | read_extent_buffer(eb, &root_item, |
4125 | btrfs_item_ptr_offset(eb, slot), | |
4126 | (int)sizeof(root_item)); | |
4127 | if (btrfs_root_refs(&root_item) == 0) | |
4128 | goto skip; | |
f45388f3 FDBM |
4129 | |
4130 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4131 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4132 | if (trans) | |
4133 | goto update_tree; | |
4134 | ||
4135 | btrfs_release_path(path); | |
803b2f54 SB |
4136 | /* |
4137 | * 1 - subvol uuid item | |
4138 | * 1 - received_subvol uuid item | |
4139 | */ | |
4140 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4141 | if (IS_ERR(trans)) { | |
4142 | ret = PTR_ERR(trans); | |
4143 | break; | |
4144 | } | |
f45388f3 FDBM |
4145 | continue; |
4146 | } else { | |
4147 | goto skip; | |
4148 | } | |
4149 | update_tree: | |
4150 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
803b2f54 SB |
4151 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
4152 | root_item.uuid, | |
4153 | BTRFS_UUID_KEY_SUBVOL, | |
4154 | key.objectid); | |
4155 | if (ret < 0) { | |
efe120a0 | 4156 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4157 | ret); |
803b2f54 SB |
4158 | break; |
4159 | } | |
4160 | } | |
4161 | ||
4162 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
803b2f54 SB |
4163 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
4164 | root_item.received_uuid, | |
4165 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4166 | key.objectid); | |
4167 | if (ret < 0) { | |
efe120a0 | 4168 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4169 | ret); |
803b2f54 SB |
4170 | break; |
4171 | } | |
4172 | } | |
4173 | ||
f45388f3 | 4174 | skip: |
803b2f54 SB |
4175 | if (trans) { |
4176 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 4177 | trans = NULL; |
803b2f54 SB |
4178 | if (ret) |
4179 | break; | |
4180 | } | |
4181 | ||
803b2f54 SB |
4182 | btrfs_release_path(path); |
4183 | if (key.offset < (u64)-1) { | |
4184 | key.offset++; | |
4185 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4186 | key.offset = 0; | |
4187 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4188 | } else if (key.objectid < (u64)-1) { | |
4189 | key.offset = 0; | |
4190 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4191 | key.objectid++; | |
4192 | } else { | |
4193 | break; | |
4194 | } | |
4195 | cond_resched(); | |
4196 | } | |
4197 | ||
4198 | out: | |
4199 | btrfs_free_path(path); | |
f45388f3 FDBM |
4200 | if (trans && !IS_ERR(trans)) |
4201 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 4202 | if (ret) |
efe120a0 | 4203 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4204 | else |
afcdd129 | 4205 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4206 | up(&fs_info->uuid_tree_rescan_sem); |
4207 | return 0; | |
4208 | } | |
4209 | ||
70f80175 SB |
4210 | /* |
4211 | * Callback for btrfs_uuid_tree_iterate(). | |
4212 | * returns: | |
4213 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4214 | * < 0 if an error occurred. |
70f80175 SB |
4215 | * > 0 if the check failed, which means the caller shall remove the entry. |
4216 | */ | |
4217 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4218 | u8 *uuid, u8 type, u64 subid) | |
4219 | { | |
4220 | struct btrfs_key key; | |
4221 | int ret = 0; | |
4222 | struct btrfs_root *subvol_root; | |
4223 | ||
4224 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4225 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4226 | goto out; | |
4227 | ||
4228 | key.objectid = subid; | |
4229 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4230 | key.offset = (u64)-1; | |
4231 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4232 | if (IS_ERR(subvol_root)) { | |
4233 | ret = PTR_ERR(subvol_root); | |
4234 | if (ret == -ENOENT) | |
4235 | ret = 1; | |
4236 | goto out; | |
4237 | } | |
4238 | ||
4239 | switch (type) { | |
4240 | case BTRFS_UUID_KEY_SUBVOL: | |
4241 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4242 | ret = 1; | |
4243 | break; | |
4244 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4245 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4246 | BTRFS_UUID_SIZE)) | |
4247 | ret = 1; | |
4248 | break; | |
4249 | } | |
4250 | ||
4251 | out: | |
4252 | return ret; | |
4253 | } | |
4254 | ||
4255 | static int btrfs_uuid_rescan_kthread(void *data) | |
4256 | { | |
4257 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4258 | int ret; | |
4259 | ||
4260 | /* | |
4261 | * 1st step is to iterate through the existing UUID tree and | |
4262 | * to delete all entries that contain outdated data. | |
4263 | * 2nd step is to add all missing entries to the UUID tree. | |
4264 | */ | |
4265 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4266 | if (ret < 0) { | |
efe120a0 | 4267 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4268 | up(&fs_info->uuid_tree_rescan_sem); |
4269 | return ret; | |
4270 | } | |
4271 | return btrfs_uuid_scan_kthread(data); | |
4272 | } | |
4273 | ||
f7a81ea4 SB |
4274 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4275 | { | |
4276 | struct btrfs_trans_handle *trans; | |
4277 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4278 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4279 | struct task_struct *task; |
4280 | int ret; | |
f7a81ea4 SB |
4281 | |
4282 | /* | |
4283 | * 1 - root node | |
4284 | * 1 - root item | |
4285 | */ | |
4286 | trans = btrfs_start_transaction(tree_root, 2); | |
4287 | if (IS_ERR(trans)) | |
4288 | return PTR_ERR(trans); | |
4289 | ||
4290 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4291 | BTRFS_UUID_TREE_OBJECTID); | |
4292 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4293 | ret = PTR_ERR(uuid_root); |
66642832 | 4294 | btrfs_abort_transaction(trans, ret); |
65d4f4c1 | 4295 | btrfs_end_transaction(trans, tree_root); |
6d13f549 | 4296 | return ret; |
f7a81ea4 SB |
4297 | } |
4298 | ||
4299 | fs_info->uuid_root = uuid_root; | |
4300 | ||
803b2f54 SB |
4301 | ret = btrfs_commit_transaction(trans, tree_root); |
4302 | if (ret) | |
4303 | return ret; | |
4304 | ||
4305 | down(&fs_info->uuid_tree_rescan_sem); | |
4306 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4307 | if (IS_ERR(task)) { | |
70f80175 | 4308 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4309 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4310 | up(&fs_info->uuid_tree_rescan_sem); |
4311 | return PTR_ERR(task); | |
4312 | } | |
4313 | ||
4314 | return 0; | |
f7a81ea4 | 4315 | } |
803b2f54 | 4316 | |
70f80175 SB |
4317 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4318 | { | |
4319 | struct task_struct *task; | |
4320 | ||
4321 | down(&fs_info->uuid_tree_rescan_sem); | |
4322 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4323 | if (IS_ERR(task)) { | |
4324 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4325 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4326 | up(&fs_info->uuid_tree_rescan_sem); |
4327 | return PTR_ERR(task); | |
4328 | } | |
4329 | ||
4330 | return 0; | |
4331 | } | |
4332 | ||
8f18cf13 CM |
4333 | /* |
4334 | * shrinking a device means finding all of the device extents past | |
4335 | * the new size, and then following the back refs to the chunks. | |
4336 | * The chunk relocation code actually frees the device extent | |
4337 | */ | |
4338 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4339 | { | |
4340 | struct btrfs_trans_handle *trans; | |
4341 | struct btrfs_root *root = device->dev_root; | |
4342 | struct btrfs_dev_extent *dev_extent = NULL; | |
4343 | struct btrfs_path *path; | |
4344 | u64 length; | |
8f18cf13 CM |
4345 | u64 chunk_offset; |
4346 | int ret; | |
4347 | int slot; | |
ba1bf481 JB |
4348 | int failed = 0; |
4349 | bool retried = false; | |
53e489bc | 4350 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4351 | struct extent_buffer *l; |
4352 | struct btrfs_key key; | |
6c41761f | 4353 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 4354 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4355 | u64 old_size = btrfs_device_get_total_bytes(device); |
4356 | u64 diff = old_size - new_size; | |
8f18cf13 | 4357 | |
63a212ab SB |
4358 | if (device->is_tgtdev_for_dev_replace) |
4359 | return -EINVAL; | |
4360 | ||
8f18cf13 CM |
4361 | path = btrfs_alloc_path(); |
4362 | if (!path) | |
4363 | return -ENOMEM; | |
4364 | ||
e4058b54 | 4365 | path->reada = READA_FORWARD; |
8f18cf13 | 4366 | |
7d9eb12c CM |
4367 | lock_chunks(root); |
4368 | ||
7cc8e58d | 4369 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4370 | if (device->writeable) { |
2b82032c | 4371 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
4372 | spin_lock(&root->fs_info->free_chunk_lock); |
4373 | root->fs_info->free_chunk_space -= diff; | |
4374 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4375 | } | |
7d9eb12c | 4376 | unlock_chunks(root); |
8f18cf13 | 4377 | |
ba1bf481 | 4378 | again: |
8f18cf13 CM |
4379 | key.objectid = device->devid; |
4380 | key.offset = (u64)-1; | |
4381 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4382 | ||
213e64da | 4383 | do { |
67c5e7d4 | 4384 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4385 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 FM |
4386 | if (ret < 0) { |
4387 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 | 4388 | goto done; |
67c5e7d4 | 4389 | } |
8f18cf13 CM |
4390 | |
4391 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 FM |
4392 | if (ret) |
4393 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 CM |
4394 | if (ret < 0) |
4395 | goto done; | |
4396 | if (ret) { | |
4397 | ret = 0; | |
b3b4aa74 | 4398 | btrfs_release_path(path); |
bf1fb512 | 4399 | break; |
8f18cf13 CM |
4400 | } |
4401 | ||
4402 | l = path->nodes[0]; | |
4403 | slot = path->slots[0]; | |
4404 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4405 | ||
ba1bf481 | 4406 | if (key.objectid != device->devid) { |
67c5e7d4 | 4407 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4408 | btrfs_release_path(path); |
bf1fb512 | 4409 | break; |
ba1bf481 | 4410 | } |
8f18cf13 CM |
4411 | |
4412 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4413 | length = btrfs_dev_extent_length(l, dev_extent); | |
4414 | ||
ba1bf481 | 4415 | if (key.offset + length <= new_size) { |
67c5e7d4 | 4416 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4417 | btrfs_release_path(path); |
d6397bae | 4418 | break; |
ba1bf481 | 4419 | } |
8f18cf13 | 4420 | |
8f18cf13 | 4421 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4422 | btrfs_release_path(path); |
8f18cf13 | 4423 | |
dc2ee4e2 | 4424 | ret = btrfs_relocate_chunk(root, chunk_offset); |
67c5e7d4 | 4425 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
ba1bf481 | 4426 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4427 | goto done; |
ba1bf481 JB |
4428 | if (ret == -ENOSPC) |
4429 | failed++; | |
213e64da | 4430 | } while (key.offset-- > 0); |
ba1bf481 JB |
4431 | |
4432 | if (failed && !retried) { | |
4433 | failed = 0; | |
4434 | retried = true; | |
4435 | goto again; | |
4436 | } else if (failed && retried) { | |
4437 | ret = -ENOSPC; | |
ba1bf481 | 4438 | goto done; |
8f18cf13 CM |
4439 | } |
4440 | ||
d6397bae | 4441 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4442 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4443 | if (IS_ERR(trans)) { |
4444 | ret = PTR_ERR(trans); | |
4445 | goto done; | |
4446 | } | |
4447 | ||
d6397bae | 4448 | lock_chunks(root); |
53e489bc FM |
4449 | |
4450 | /* | |
4451 | * We checked in the above loop all device extents that were already in | |
4452 | * the device tree. However before we have updated the device's | |
4453 | * total_bytes to the new size, we might have had chunk allocations that | |
4454 | * have not complete yet (new block groups attached to transaction | |
4455 | * handles), and therefore their device extents were not yet in the | |
4456 | * device tree and we missed them in the loop above. So if we have any | |
4457 | * pending chunk using a device extent that overlaps the device range | |
4458 | * that we can not use anymore, commit the current transaction and | |
4459 | * repeat the search on the device tree - this way we guarantee we will | |
4460 | * not have chunks using device extents that end beyond 'new_size'. | |
4461 | */ | |
4462 | if (!checked_pending_chunks) { | |
4463 | u64 start = new_size; | |
4464 | u64 len = old_size - new_size; | |
4465 | ||
499f377f JM |
4466 | if (contains_pending_extent(trans->transaction, device, |
4467 | &start, len)) { | |
53e489bc FM |
4468 | unlock_chunks(root); |
4469 | checked_pending_chunks = true; | |
4470 | failed = 0; | |
4471 | retried = false; | |
4472 | ret = btrfs_commit_transaction(trans, root); | |
4473 | if (ret) | |
4474 | goto done; | |
4475 | goto again; | |
4476 | } | |
4477 | } | |
4478 | ||
7cc8e58d | 4479 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4480 | if (list_empty(&device->resized_list)) |
4481 | list_add_tail(&device->resized_list, | |
4482 | &root->fs_info->fs_devices->resized_devices); | |
d6397bae | 4483 | |
d6397bae CB |
4484 | WARN_ON(diff > old_total); |
4485 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4486 | unlock_chunks(root); | |
2196d6e8 MX |
4487 | |
4488 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4489 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4490 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4491 | done: |
4492 | btrfs_free_path(path); | |
53e489bc FM |
4493 | if (ret) { |
4494 | lock_chunks(root); | |
4495 | btrfs_device_set_total_bytes(device, old_size); | |
4496 | if (device->writeable) | |
4497 | device->fs_devices->total_rw_bytes += diff; | |
4498 | spin_lock(&root->fs_info->free_chunk_lock); | |
4499 | root->fs_info->free_chunk_space += diff; | |
4500 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4501 | unlock_chunks(root); | |
4502 | } | |
8f18cf13 CM |
4503 | return ret; |
4504 | } | |
4505 | ||
125ccb0a | 4506 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4507 | struct btrfs_key *key, |
4508 | struct btrfs_chunk *chunk, int item_size) | |
4509 | { | |
6c41761f | 4510 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
4511 | struct btrfs_disk_key disk_key; |
4512 | u32 array_size; | |
4513 | u8 *ptr; | |
4514 | ||
fe48a5c0 | 4515 | lock_chunks(root); |
0b86a832 | 4516 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4517 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 MX |
4518 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
4519 | unlock_chunks(root); | |
0b86a832 | 4520 | return -EFBIG; |
fe48a5c0 | 4521 | } |
0b86a832 CM |
4522 | |
4523 | ptr = super_copy->sys_chunk_array + array_size; | |
4524 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4525 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4526 | ptr += sizeof(disk_key); | |
4527 | memcpy(ptr, chunk, item_size); | |
4528 | item_size += sizeof(disk_key); | |
4529 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 MX |
4530 | unlock_chunks(root); |
4531 | ||
0b86a832 CM |
4532 | return 0; |
4533 | } | |
4534 | ||
73c5de00 AJ |
4535 | /* |
4536 | * sort the devices in descending order by max_avail, total_avail | |
4537 | */ | |
4538 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4539 | { |
73c5de00 AJ |
4540 | const struct btrfs_device_info *di_a = a; |
4541 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4542 | |
73c5de00 | 4543 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4544 | return -1; |
73c5de00 | 4545 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4546 | return 1; |
73c5de00 AJ |
4547 | if (di_a->total_avail > di_b->total_avail) |
4548 | return -1; | |
4549 | if (di_a->total_avail < di_b->total_avail) | |
4550 | return 1; | |
4551 | return 0; | |
b2117a39 | 4552 | } |
0b86a832 | 4553 | |
53b381b3 DW |
4554 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4555 | { | |
4556 | /* TODO allow them to set a preferred stripe size */ | |
ee22184b | 4557 | return SZ_64K; |
53b381b3 DW |
4558 | } |
4559 | ||
4560 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4561 | { | |
ffe2d203 | 4562 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4563 | return; |
4564 | ||
ceda0864 | 4565 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4566 | } |
4567 | ||
14a1e067 | 4568 | #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \ |
23f8f9b7 GH |
4569 | - sizeof(struct btrfs_chunk)) \ |
4570 | / sizeof(struct btrfs_stripe) + 1) | |
4571 | ||
4572 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4573 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4574 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4575 | / sizeof(struct btrfs_stripe) + 1) | |
4576 | ||
73c5de00 | 4577 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4578 | struct btrfs_root *extent_root, u64 start, |
4579 | u64 type) | |
b2117a39 | 4580 | { |
73c5de00 AJ |
4581 | struct btrfs_fs_info *info = extent_root->fs_info; |
4582 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4583 | struct list_head *cur; | |
4584 | struct map_lookup *map = NULL; | |
4585 | struct extent_map_tree *em_tree; | |
4586 | struct extent_map *em; | |
4587 | struct btrfs_device_info *devices_info = NULL; | |
4588 | u64 total_avail; | |
4589 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4590 | int data_stripes; /* number of stripes that count for |
4591 | block group size */ | |
73c5de00 AJ |
4592 | int sub_stripes; /* sub_stripes info for map */ |
4593 | int dev_stripes; /* stripes per dev */ | |
4594 | int devs_max; /* max devs to use */ | |
4595 | int devs_min; /* min devs needed */ | |
4596 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4597 | int ncopies; /* how many copies to data has */ | |
4598 | int ret; | |
4599 | u64 max_stripe_size; | |
4600 | u64 max_chunk_size; | |
4601 | u64 stripe_size; | |
4602 | u64 num_bytes; | |
53b381b3 | 4603 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4604 | int ndevs; |
4605 | int i; | |
4606 | int j; | |
31e50229 | 4607 | int index; |
593060d7 | 4608 | |
0c460c0d | 4609 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4610 | |
73c5de00 AJ |
4611 | if (list_empty(&fs_devices->alloc_list)) |
4612 | return -ENOSPC; | |
b2117a39 | 4613 | |
31e50229 | 4614 | index = __get_raid_index(type); |
73c5de00 | 4615 | |
31e50229 LB |
4616 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4617 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4618 | devs_max = btrfs_raid_array[index].devs_max; | |
4619 | devs_min = btrfs_raid_array[index].devs_min; | |
4620 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4621 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4622 | |
9b3f68b9 | 4623 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4624 | max_stripe_size = SZ_1G; |
73c5de00 | 4625 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 GH |
4626 | if (!devs_max) |
4627 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4628 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4629 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4630 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4631 | max_stripe_size = SZ_1G; | |
1100373f | 4632 | else |
ee22184b | 4633 | max_stripe_size = SZ_256M; |
73c5de00 | 4634 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4635 | if (!devs_max) |
4636 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4637 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4638 | max_stripe_size = SZ_32M; |
73c5de00 | 4639 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4640 | if (!devs_max) |
4641 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4642 | } else { |
351fd353 | 4643 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4644 | type); |
4645 | BUG_ON(1); | |
9b3f68b9 CM |
4646 | } |
4647 | ||
2b82032c YZ |
4648 | /* we don't want a chunk larger than 10% of writeable space */ |
4649 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4650 | max_chunk_size); | |
9b3f68b9 | 4651 | |
31e818fe | 4652 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4653 | GFP_NOFS); |
4654 | if (!devices_info) | |
4655 | return -ENOMEM; | |
0cad8a11 | 4656 | |
73c5de00 | 4657 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4658 | |
9f680ce0 | 4659 | /* |
73c5de00 AJ |
4660 | * in the first pass through the devices list, we gather information |
4661 | * about the available holes on each device. | |
9f680ce0 | 4662 | */ |
73c5de00 AJ |
4663 | ndevs = 0; |
4664 | while (cur != &fs_devices->alloc_list) { | |
4665 | struct btrfs_device *device; | |
4666 | u64 max_avail; | |
4667 | u64 dev_offset; | |
b2117a39 | 4668 | |
73c5de00 | 4669 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4670 | |
73c5de00 | 4671 | cur = cur->next; |
b2117a39 | 4672 | |
73c5de00 | 4673 | if (!device->writeable) { |
31b1a2bd | 4674 | WARN(1, KERN_ERR |
efe120a0 | 4675 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4676 | continue; |
4677 | } | |
b2117a39 | 4678 | |
63a212ab SB |
4679 | if (!device->in_fs_metadata || |
4680 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4681 | continue; |
b2117a39 | 4682 | |
73c5de00 AJ |
4683 | if (device->total_bytes > device->bytes_used) |
4684 | total_avail = device->total_bytes - device->bytes_used; | |
4685 | else | |
4686 | total_avail = 0; | |
38c01b96 | 4687 | |
4688 | /* If there is no space on this device, skip it. */ | |
4689 | if (total_avail == 0) | |
4690 | continue; | |
b2117a39 | 4691 | |
6df9a95e | 4692 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4693 | max_stripe_size * dev_stripes, |
4694 | &dev_offset, &max_avail); | |
4695 | if (ret && ret != -ENOSPC) | |
4696 | goto error; | |
b2117a39 | 4697 | |
73c5de00 AJ |
4698 | if (ret == 0) |
4699 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4700 | |
73c5de00 AJ |
4701 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4702 | continue; | |
b2117a39 | 4703 | |
063d006f ES |
4704 | if (ndevs == fs_devices->rw_devices) { |
4705 | WARN(1, "%s: found more than %llu devices\n", | |
4706 | __func__, fs_devices->rw_devices); | |
4707 | break; | |
4708 | } | |
73c5de00 AJ |
4709 | devices_info[ndevs].dev_offset = dev_offset; |
4710 | devices_info[ndevs].max_avail = max_avail; | |
4711 | devices_info[ndevs].total_avail = total_avail; | |
4712 | devices_info[ndevs].dev = device; | |
4713 | ++ndevs; | |
4714 | } | |
b2117a39 | 4715 | |
73c5de00 AJ |
4716 | /* |
4717 | * now sort the devices by hole size / available space | |
4718 | */ | |
4719 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4720 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4721 | |
73c5de00 AJ |
4722 | /* round down to number of usable stripes */ |
4723 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4724 | |
73c5de00 AJ |
4725 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4726 | ret = -ENOSPC; | |
4727 | goto error; | |
b2117a39 | 4728 | } |
9f680ce0 | 4729 | |
73c5de00 AJ |
4730 | if (devs_max && ndevs > devs_max) |
4731 | ndevs = devs_max; | |
4732 | /* | |
4733 | * the primary goal is to maximize the number of stripes, so use as many | |
4734 | * devices as possible, even if the stripes are not maximum sized. | |
4735 | */ | |
4736 | stripe_size = devices_info[ndevs-1].max_avail; | |
4737 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4738 | |
53b381b3 DW |
4739 | /* |
4740 | * this will have to be fixed for RAID1 and RAID10 over | |
4741 | * more drives | |
4742 | */ | |
4743 | data_stripes = num_stripes / ncopies; | |
4744 | ||
53b381b3 DW |
4745 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4746 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
b7f67055 | 4747 | extent_root->stripesize); |
53b381b3 DW |
4748 | data_stripes = num_stripes - 1; |
4749 | } | |
4750 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4751 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
b7f67055 | 4752 | extent_root->stripesize); |
53b381b3 DW |
4753 | data_stripes = num_stripes - 2; |
4754 | } | |
86db2578 CM |
4755 | |
4756 | /* | |
4757 | * Use the number of data stripes to figure out how big this chunk | |
4758 | * is really going to be in terms of logical address space, | |
4759 | * and compare that answer with the max chunk size | |
4760 | */ | |
4761 | if (stripe_size * data_stripes > max_chunk_size) { | |
4762 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4763 | |
4764 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4765 | |
4766 | /* bump the answer up to a 16MB boundary */ | |
4767 | stripe_size = (stripe_size + mask) & ~mask; | |
4768 | ||
4769 | /* but don't go higher than the limits we found | |
4770 | * while searching for free extents | |
4771 | */ | |
4772 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4773 | stripe_size = devices_info[ndevs-1].max_avail; | |
4774 | } | |
4775 | ||
b8b93add | 4776 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4777 | |
4778 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4779 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4780 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4781 | |
4782 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4783 | if (!map) { | |
4784 | ret = -ENOMEM; | |
4785 | goto error; | |
4786 | } | |
4787 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4788 | |
73c5de00 AJ |
4789 | for (i = 0; i < ndevs; ++i) { |
4790 | for (j = 0; j < dev_stripes; ++j) { | |
4791 | int s = i * dev_stripes + j; | |
4792 | map->stripes[s].dev = devices_info[i].dev; | |
4793 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4794 | j * stripe_size; | |
6324fbf3 | 4795 | } |
6324fbf3 | 4796 | } |
2b82032c | 4797 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4798 | map->stripe_len = raid_stripe_len; |
4799 | map->io_align = raid_stripe_len; | |
4800 | map->io_width = raid_stripe_len; | |
2b82032c | 4801 | map->type = type; |
2b82032c | 4802 | map->sub_stripes = sub_stripes; |
0b86a832 | 4803 | |
53b381b3 | 4804 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4805 | |
73c5de00 | 4806 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4807 | |
172ddd60 | 4808 | em = alloc_extent_map(); |
2b82032c | 4809 | if (!em) { |
298a8f9c | 4810 | kfree(map); |
b2117a39 MX |
4811 | ret = -ENOMEM; |
4812 | goto error; | |
593060d7 | 4813 | } |
298a8f9c | 4814 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4815 | em->map_lookup = map; |
2b82032c | 4816 | em->start = start; |
73c5de00 | 4817 | em->len = num_bytes; |
2b82032c YZ |
4818 | em->block_start = 0; |
4819 | em->block_len = em->len; | |
6df9a95e | 4820 | em->orig_block_len = stripe_size; |
593060d7 | 4821 | |
2b82032c | 4822 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4823 | write_lock(&em_tree->lock); |
09a2a8f9 | 4824 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4825 | if (!ret) { |
4826 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4827 | atomic_inc(&em->refs); | |
4828 | } | |
890871be | 4829 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4830 | if (ret) { |
4831 | free_extent_map(em); | |
1dd4602f | 4832 | goto error; |
0f5d42b2 | 4833 | } |
0b86a832 | 4834 | |
04487488 JB |
4835 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4836 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4837 | start, num_bytes); | |
6df9a95e JB |
4838 | if (ret) |
4839 | goto error_del_extent; | |
2b82032c | 4840 | |
7cc8e58d MX |
4841 | for (i = 0; i < map->num_stripes; i++) { |
4842 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4843 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4844 | } | |
43530c46 | 4845 | |
1c116187 MX |
4846 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4847 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4848 | map->num_stripes); | |
4849 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4850 | ||
0f5d42b2 | 4851 | free_extent_map(em); |
53b381b3 DW |
4852 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4853 | ||
b2117a39 | 4854 | kfree(devices_info); |
2b82032c | 4855 | return 0; |
b2117a39 | 4856 | |
6df9a95e | 4857 | error_del_extent: |
0f5d42b2 JB |
4858 | write_lock(&em_tree->lock); |
4859 | remove_extent_mapping(em_tree, em); | |
4860 | write_unlock(&em_tree->lock); | |
4861 | ||
4862 | /* One for our allocation */ | |
4863 | free_extent_map(em); | |
4864 | /* One for the tree reference */ | |
4865 | free_extent_map(em); | |
495e64f4 FM |
4866 | /* One for the pending_chunks list reference */ |
4867 | free_extent_map(em); | |
b2117a39 | 4868 | error: |
b2117a39 MX |
4869 | kfree(devices_info); |
4870 | return ret; | |
2b82032c YZ |
4871 | } |
4872 | ||
6df9a95e | 4873 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4874 | struct btrfs_root *extent_root, |
6df9a95e | 4875 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4876 | { |
2b82032c YZ |
4877 | struct btrfs_key key; |
4878 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4879 | struct btrfs_device *device; | |
4880 | struct btrfs_chunk *chunk; | |
4881 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4882 | struct extent_map_tree *em_tree; |
4883 | struct extent_map *em; | |
4884 | struct map_lookup *map; | |
4885 | size_t item_size; | |
4886 | u64 dev_offset; | |
4887 | u64 stripe_size; | |
4888 | int i = 0; | |
140e639f | 4889 | int ret = 0; |
2b82032c | 4890 | |
6df9a95e JB |
4891 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4892 | read_lock(&em_tree->lock); | |
4893 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4894 | read_unlock(&em_tree->lock); | |
4895 | ||
4896 | if (!em) { | |
5d163e0e JM |
4897 | btrfs_crit(extent_root->fs_info, |
4898 | "unable to find logical %Lu len %Lu", | |
4899 | chunk_offset, chunk_size); | |
6df9a95e JB |
4900 | return -EINVAL; |
4901 | } | |
4902 | ||
4903 | if (em->start != chunk_offset || em->len != chunk_size) { | |
5d163e0e JM |
4904 | btrfs_crit(extent_root->fs_info, |
4905 | "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu", | |
4906 | chunk_offset, chunk_size, em->start, em->len); | |
6df9a95e JB |
4907 | free_extent_map(em); |
4908 | return -EINVAL; | |
4909 | } | |
4910 | ||
95617d69 | 4911 | map = em->map_lookup; |
6df9a95e JB |
4912 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4913 | stripe_size = em->orig_block_len; | |
4914 | ||
2b82032c | 4915 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4916 | if (!chunk) { |
4917 | ret = -ENOMEM; | |
4918 | goto out; | |
4919 | } | |
4920 | ||
50460e37 FM |
4921 | /* |
4922 | * Take the device list mutex to prevent races with the final phase of | |
4923 | * a device replace operation that replaces the device object associated | |
4924 | * with the map's stripes, because the device object's id can change | |
4925 | * at any time during that final phase of the device replace operation | |
4926 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4927 | */ | |
4928 | mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
6df9a95e JB |
4929 | for (i = 0; i < map->num_stripes; i++) { |
4930 | device = map->stripes[i].dev; | |
4931 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4932 | |
0b86a832 | 4933 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4934 | if (ret) |
50460e37 | 4935 | break; |
6df9a95e JB |
4936 | ret = btrfs_alloc_dev_extent(trans, device, |
4937 | chunk_root->root_key.objectid, | |
4938 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4939 | chunk_offset, dev_offset, | |
4940 | stripe_size); | |
4941 | if (ret) | |
50460e37 FM |
4942 | break; |
4943 | } | |
4944 | if (ret) { | |
4945 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
4946 | goto out; | |
2b82032c YZ |
4947 | } |
4948 | ||
2b82032c | 4949 | stripe = &chunk->stripe; |
6df9a95e JB |
4950 | for (i = 0; i < map->num_stripes; i++) { |
4951 | device = map->stripes[i].dev; | |
4952 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4953 | |
e17cade2 CM |
4954 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4955 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4956 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4957 | stripe++; |
0b86a832 | 4958 | } |
50460e37 | 4959 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4960 | |
2b82032c | 4961 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4962 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4963 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4964 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4965 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4966 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4967 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4968 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4969 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4970 | |
2b82032c YZ |
4971 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4972 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4973 | key.offset = chunk_offset; | |
0b86a832 | 4974 | |
2b82032c | 4975 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4976 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4977 | /* | |
4978 | * TODO: Cleanup of inserted chunk root in case of | |
4979 | * failure. | |
4980 | */ | |
125ccb0a | 4981 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4982 | item_size); |
8f18cf13 | 4983 | } |
1abe9b8a | 4984 | |
6df9a95e | 4985 | out: |
0b86a832 | 4986 | kfree(chunk); |
6df9a95e | 4987 | free_extent_map(em); |
4ed1d16e | 4988 | return ret; |
2b82032c | 4989 | } |
0b86a832 | 4990 | |
2b82032c YZ |
4991 | /* |
4992 | * Chunk allocation falls into two parts. The first part does works | |
4993 | * that make the new allocated chunk useable, but not do any operation | |
4994 | * that modifies the chunk tree. The second part does the works that | |
4995 | * require modifying the chunk tree. This division is important for the | |
4996 | * bootstrap process of adding storage to a seed btrfs. | |
4997 | */ | |
4998 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4999 | struct btrfs_root *extent_root, u64 type) | |
5000 | { | |
5001 | u64 chunk_offset; | |
2b82032c | 5002 | |
a9629596 | 5003 | ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex)); |
6df9a95e JB |
5004 | chunk_offset = find_next_chunk(extent_root->fs_info); |
5005 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
5006 | } |
5007 | ||
d397712b | 5008 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
5009 | struct btrfs_root *root, |
5010 | struct btrfs_device *device) | |
5011 | { | |
5012 | u64 chunk_offset; | |
5013 | u64 sys_chunk_offset; | |
2b82032c | 5014 | u64 alloc_profile; |
2b82032c YZ |
5015 | struct btrfs_fs_info *fs_info = root->fs_info; |
5016 | struct btrfs_root *extent_root = fs_info->extent_root; | |
5017 | int ret; | |
5018 | ||
6df9a95e | 5019 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 5020 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
5021 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
5022 | alloc_profile); | |
79787eaa JM |
5023 | if (ret) |
5024 | return ret; | |
2b82032c | 5025 | |
6df9a95e | 5026 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 5027 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
5028 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
5029 | alloc_profile); | |
79787eaa | 5030 | return ret; |
2b82032c YZ |
5031 | } |
5032 | ||
d20983b4 MX |
5033 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5034 | { | |
5035 | int max_errors; | |
5036 | ||
5037 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5038 | BTRFS_BLOCK_GROUP_RAID10 | | |
5039 | BTRFS_BLOCK_GROUP_RAID5 | | |
5040 | BTRFS_BLOCK_GROUP_DUP)) { | |
5041 | max_errors = 1; | |
5042 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5043 | max_errors = 2; | |
5044 | } else { | |
5045 | max_errors = 0; | |
005d6427 | 5046 | } |
2b82032c | 5047 | |
d20983b4 | 5048 | return max_errors; |
2b82032c YZ |
5049 | } |
5050 | ||
5051 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
5052 | { | |
5053 | struct extent_map *em; | |
5054 | struct map_lookup *map; | |
5055 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5056 | int readonly = 0; | |
d20983b4 | 5057 | int miss_ndevs = 0; |
2b82032c YZ |
5058 | int i; |
5059 | ||
890871be | 5060 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 5061 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 5062 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
5063 | if (!em) |
5064 | return 1; | |
5065 | ||
95617d69 | 5066 | map = em->map_lookup; |
2b82032c | 5067 | for (i = 0; i < map->num_stripes; i++) { |
d20983b4 MX |
5068 | if (map->stripes[i].dev->missing) { |
5069 | miss_ndevs++; | |
5070 | continue; | |
5071 | } | |
5072 | ||
2b82032c YZ |
5073 | if (!map->stripes[i].dev->writeable) { |
5074 | readonly = 1; | |
d20983b4 | 5075 | goto end; |
2b82032c YZ |
5076 | } |
5077 | } | |
d20983b4 MX |
5078 | |
5079 | /* | |
5080 | * If the number of missing devices is larger than max errors, | |
5081 | * we can not write the data into that chunk successfully, so | |
5082 | * set it readonly. | |
5083 | */ | |
5084 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5085 | readonly = 1; | |
5086 | end: | |
0b86a832 | 5087 | free_extent_map(em); |
2b82032c | 5088 | return readonly; |
0b86a832 CM |
5089 | } |
5090 | ||
5091 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5092 | { | |
a8067e02 | 5093 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5094 | } |
5095 | ||
5096 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5097 | { | |
5098 | struct extent_map *em; | |
5099 | ||
d397712b | 5100 | while (1) { |
890871be | 5101 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5102 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5103 | if (em) | |
5104 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5105 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5106 | if (!em) |
5107 | break; | |
0b86a832 CM |
5108 | /* once for us */ |
5109 | free_extent_map(em); | |
5110 | /* once for the tree */ | |
5111 | free_extent_map(em); | |
5112 | } | |
5113 | } | |
5114 | ||
5d964051 | 5115 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 5116 | { |
5d964051 | 5117 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
5118 | struct extent_map *em; |
5119 | struct map_lookup *map; | |
5120 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5121 | int ret; | |
5122 | ||
890871be | 5123 | read_lock(&em_tree->lock); |
f188591e | 5124 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 5125 | read_unlock(&em_tree->lock); |
f188591e | 5126 | |
fb7669b5 JB |
5127 | /* |
5128 | * We could return errors for these cases, but that could get ugly and | |
5129 | * we'd probably do the same thing which is just not do anything else | |
5130 | * and exit, so return 1 so the callers don't try to use other copies. | |
5131 | */ | |
5132 | if (!em) { | |
351fd353 | 5133 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
5134 | logical+len); |
5135 | return 1; | |
5136 | } | |
5137 | ||
5138 | if (em->start > logical || em->start + em->len < logical) { | |
5d163e0e JM |
5139 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu", |
5140 | logical, logical+len, em->start, | |
5141 | em->start + em->len); | |
7d3d1744 | 5142 | free_extent_map(em); |
fb7669b5 JB |
5143 | return 1; |
5144 | } | |
5145 | ||
95617d69 | 5146 | map = em->map_lookup; |
f188591e CM |
5147 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5148 | ret = map->num_stripes; | |
321aecc6 CM |
5149 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5150 | ret = map->sub_stripes; | |
53b381b3 DW |
5151 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5152 | ret = 2; | |
5153 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5154 | ret = 3; | |
f188591e CM |
5155 | else |
5156 | ret = 1; | |
5157 | free_extent_map(em); | |
ad6d620e | 5158 | |
73beece9 | 5159 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
ad6d620e SB |
5160 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) |
5161 | ret++; | |
73beece9 | 5162 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5163 | |
f188591e CM |
5164 | return ret; |
5165 | } | |
5166 | ||
53b381b3 DW |
5167 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
5168 | struct btrfs_mapping_tree *map_tree, | |
5169 | u64 logical) | |
5170 | { | |
5171 | struct extent_map *em; | |
5172 | struct map_lookup *map; | |
5173 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5174 | unsigned long len = root->sectorsize; | |
5175 | ||
5176 | read_lock(&em_tree->lock); | |
5177 | em = lookup_extent_mapping(em_tree, logical, len); | |
5178 | read_unlock(&em_tree->lock); | |
5179 | BUG_ON(!em); | |
5180 | ||
5181 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
95617d69 | 5182 | map = em->map_lookup; |
ffe2d203 | 5183 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 5184 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
5185 | free_extent_map(em); |
5186 | return len; | |
5187 | } | |
5188 | ||
5189 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
5190 | u64 logical, u64 len, int mirror_num) | |
5191 | { | |
5192 | struct extent_map *em; | |
5193 | struct map_lookup *map; | |
5194 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5195 | int ret = 0; | |
5196 | ||
5197 | read_lock(&em_tree->lock); | |
5198 | em = lookup_extent_mapping(em_tree, logical, len); | |
5199 | read_unlock(&em_tree->lock); | |
5200 | BUG_ON(!em); | |
5201 | ||
5202 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
95617d69 | 5203 | map = em->map_lookup; |
ffe2d203 | 5204 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
5205 | ret = 1; |
5206 | free_extent_map(em); | |
5207 | return ret; | |
5208 | } | |
5209 | ||
30d9861f SB |
5210 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5211 | struct map_lookup *map, int first, int num, | |
5212 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5213 | { |
5214 | int i; | |
30d9861f SB |
5215 | int tolerance; |
5216 | struct btrfs_device *srcdev; | |
5217 | ||
5218 | if (dev_replace_is_ongoing && | |
5219 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5220 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5221 | srcdev = fs_info->dev_replace.srcdev; | |
5222 | else | |
5223 | srcdev = NULL; | |
5224 | ||
5225 | /* | |
5226 | * try to avoid the drive that is the source drive for a | |
5227 | * dev-replace procedure, only choose it if no other non-missing | |
5228 | * mirror is available | |
5229 | */ | |
5230 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5231 | if (map->stripes[optimal].dev->bdev && | |
5232 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5233 | return optimal; | |
5234 | for (i = first; i < first + num; i++) { | |
5235 | if (map->stripes[i].dev->bdev && | |
5236 | (tolerance || map->stripes[i].dev != srcdev)) | |
5237 | return i; | |
5238 | } | |
dfe25020 | 5239 | } |
30d9861f | 5240 | |
dfe25020 CM |
5241 | /* we couldn't find one that doesn't fail. Just return something |
5242 | * and the io error handling code will clean up eventually | |
5243 | */ | |
5244 | return optimal; | |
5245 | } | |
5246 | ||
53b381b3 DW |
5247 | static inline int parity_smaller(u64 a, u64 b) |
5248 | { | |
5249 | return a > b; | |
5250 | } | |
5251 | ||
5252 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5253 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5254 | { |
5255 | struct btrfs_bio_stripe s; | |
5256 | int i; | |
5257 | u64 l; | |
5258 | int again = 1; | |
5259 | ||
5260 | while (again) { | |
5261 | again = 0; | |
cc7539ed | 5262 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5263 | if (parity_smaller(bbio->raid_map[i], |
5264 | bbio->raid_map[i+1])) { | |
53b381b3 | 5265 | s = bbio->stripes[i]; |
8e5cfb55 | 5266 | l = bbio->raid_map[i]; |
53b381b3 | 5267 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5268 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5269 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5270 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5271 | |
53b381b3 DW |
5272 | again = 1; |
5273 | } | |
5274 | } | |
5275 | } | |
5276 | } | |
5277 | ||
6e9606d2 ZL |
5278 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5279 | { | |
5280 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5281 | /* the size of the btrfs_bio */ |
6e9606d2 | 5282 | sizeof(struct btrfs_bio) + |
e57cf21e | 5283 | /* plus the variable array for the stripes */ |
6e9606d2 | 5284 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5285 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5286 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5287 | /* |
5288 | * plus the raid_map, which includes both the tgt dev | |
5289 | * and the stripes | |
5290 | */ | |
5291 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5292 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5293 | |
5294 | atomic_set(&bbio->error, 0); | |
5295 | atomic_set(&bbio->refs, 1); | |
5296 | ||
5297 | return bbio; | |
5298 | } | |
5299 | ||
5300 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5301 | { | |
5302 | WARN_ON(!atomic_read(&bbio->refs)); | |
5303 | atomic_inc(&bbio->refs); | |
5304 | } | |
5305 | ||
5306 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5307 | { | |
5308 | if (!bbio) | |
5309 | return; | |
5310 | if (atomic_dec_and_test(&bbio->refs)) | |
5311 | kfree(bbio); | |
5312 | } | |
5313 | ||
b3d3fa51 | 5314 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int op, |
f2d8d74d | 5315 | u64 logical, u64 *length, |
a1d3c478 | 5316 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5317 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5318 | { |
5319 | struct extent_map *em; | |
5320 | struct map_lookup *map; | |
3ec706c8 | 5321 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
5322 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5323 | u64 offset; | |
593060d7 | 5324 | u64 stripe_offset; |
fce3bb9a | 5325 | u64 stripe_end_offset; |
593060d7 | 5326 | u64 stripe_nr; |
fce3bb9a LD |
5327 | u64 stripe_nr_orig; |
5328 | u64 stripe_nr_end; | |
53b381b3 | 5329 | u64 stripe_len; |
9d644a62 | 5330 | u32 stripe_index; |
cea9e445 | 5331 | int i; |
de11cc12 | 5332 | int ret = 0; |
f2d8d74d | 5333 | int num_stripes; |
a236aed1 | 5334 | int max_errors = 0; |
2c8cdd6e | 5335 | int tgtdev_indexes = 0; |
a1d3c478 | 5336 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5337 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5338 | int dev_replace_is_ongoing = 0; | |
5339 | int num_alloc_stripes; | |
ad6d620e SB |
5340 | int patch_the_first_stripe_for_dev_replace = 0; |
5341 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5342 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5343 | |
890871be | 5344 | read_lock(&em_tree->lock); |
0b86a832 | 5345 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 5346 | read_unlock(&em_tree->lock); |
f2d8d74d | 5347 | |
3b951516 | 5348 | if (!em) { |
c2cf52eb | 5349 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 5350 | logical, *length); |
9bb91873 JB |
5351 | return -EINVAL; |
5352 | } | |
5353 | ||
5354 | if (em->start > logical || em->start + em->len < logical) { | |
5d163e0e JM |
5355 | btrfs_crit(fs_info, |
5356 | "found a bad mapping, wanted %Lu, found %Lu-%Lu", | |
5357 | logical, em->start, em->start + em->len); | |
7d3d1744 | 5358 | free_extent_map(em); |
9bb91873 | 5359 | return -EINVAL; |
3b951516 | 5360 | } |
0b86a832 | 5361 | |
95617d69 | 5362 | map = em->map_lookup; |
0b86a832 | 5363 | offset = logical - em->start; |
593060d7 | 5364 | |
53b381b3 | 5365 | stripe_len = map->stripe_len; |
593060d7 CM |
5366 | stripe_nr = offset; |
5367 | /* | |
5368 | * stripe_nr counts the total number of stripes we have to stride | |
5369 | * to get to this block | |
5370 | */ | |
47c5713f | 5371 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5372 | |
53b381b3 | 5373 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5374 | if (offset < stripe_offset) { |
5d163e0e JM |
5375 | btrfs_crit(fs_info, |
5376 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5377 | stripe_offset, offset, em->start, logical, |
5378 | stripe_len); | |
5379 | free_extent_map(em); | |
5380 | return -EINVAL; | |
5381 | } | |
593060d7 CM |
5382 | |
5383 | /* stripe_offset is the offset of this block in its stripe*/ | |
5384 | stripe_offset = offset - stripe_offset; | |
5385 | ||
53b381b3 | 5386 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5387 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5388 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5389 | raid56_full_stripe_start = offset; | |
5390 | ||
5391 | /* allow a write of a full stripe, but make sure we don't | |
5392 | * allow straddling of stripes | |
5393 | */ | |
47c5713f DS |
5394 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5395 | full_stripe_len); | |
53b381b3 DW |
5396 | raid56_full_stripe_start *= full_stripe_len; |
5397 | } | |
5398 | ||
b3d3fa51 | 5399 | if (op == REQ_OP_DISCARD) { |
53b381b3 | 5400 | /* we don't discard raid56 yet */ |
ffe2d203 | 5401 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5402 | ret = -EOPNOTSUPP; |
5403 | goto out; | |
5404 | } | |
fce3bb9a | 5405 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
5406 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
5407 | u64 max_len; | |
5408 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5409 | For other RAID types and for RAID[56] reads, just allow a single | |
5410 | stripe (on a single disk). */ | |
ffe2d203 | 5411 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
b3d3fa51 | 5412 | (op == REQ_OP_WRITE)) { |
53b381b3 DW |
5413 | max_len = stripe_len * nr_data_stripes(map) - |
5414 | (offset - raid56_full_stripe_start); | |
5415 | } else { | |
5416 | /* we limit the length of each bio to what fits in a stripe */ | |
5417 | max_len = stripe_len - stripe_offset; | |
5418 | } | |
5419 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5420 | } else { |
5421 | *length = em->len - offset; | |
5422 | } | |
f2d8d74d | 5423 | |
53b381b3 DW |
5424 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5425 | it cares about is the length */ | |
a1d3c478 | 5426 | if (!bbio_ret) |
cea9e445 CM |
5427 | goto out; |
5428 | ||
73beece9 | 5429 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5430 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5431 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5432 | btrfs_dev_replace_unlock(dev_replace, 0); |
5433 | else | |
5434 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5435 | |
ad6d620e | 5436 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
b3d3fa51 MC |
5437 | op != REQ_OP_WRITE && op != REQ_OP_DISCARD && |
5438 | op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) { | |
ad6d620e SB |
5439 | /* |
5440 | * in dev-replace case, for repair case (that's the only | |
5441 | * case where the mirror is selected explicitly when | |
5442 | * calling btrfs_map_block), blocks left of the left cursor | |
5443 | * can also be read from the target drive. | |
5444 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5445 | * the last one to the array of stripes. For READ, it also | |
5446 | * needs to be supported using the same mirror number. | |
5447 | * If the requested block is not left of the left cursor, | |
5448 | * EIO is returned. This can happen because btrfs_num_copies() | |
5449 | * returns one more in the dev-replace case. | |
5450 | */ | |
5451 | u64 tmp_length = *length; | |
5452 | struct btrfs_bio *tmp_bbio = NULL; | |
5453 | int tmp_num_stripes; | |
5454 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5455 | int index_srcdev = 0; | |
5456 | int found = 0; | |
5457 | u64 physical_of_found = 0; | |
5458 | ||
5459 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
8e5cfb55 | 5460 | logical, &tmp_length, &tmp_bbio, 0, 0); |
ad6d620e SB |
5461 | if (ret) { |
5462 | WARN_ON(tmp_bbio != NULL); | |
5463 | goto out; | |
5464 | } | |
5465 | ||
5466 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5467 | if (mirror_num > tmp_num_stripes) { | |
5468 | /* | |
5469 | * REQ_GET_READ_MIRRORS does not contain this | |
5470 | * mirror, that means that the requested area | |
5471 | * is not left of the left cursor | |
5472 | */ | |
5473 | ret = -EIO; | |
6e9606d2 | 5474 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5475 | goto out; |
5476 | } | |
5477 | ||
5478 | /* | |
5479 | * process the rest of the function using the mirror_num | |
5480 | * of the source drive. Therefore look it up first. | |
5481 | * At the end, patch the device pointer to the one of the | |
5482 | * target drive. | |
5483 | */ | |
5484 | for (i = 0; i < tmp_num_stripes; i++) { | |
94a97dfe ZL |
5485 | if (tmp_bbio->stripes[i].dev->devid != srcdev_devid) |
5486 | continue; | |
5487 | ||
5488 | /* | |
5489 | * In case of DUP, in order to keep it simple, only add | |
5490 | * the mirror with the lowest physical address | |
5491 | */ | |
5492 | if (found && | |
5493 | physical_of_found <= tmp_bbio->stripes[i].physical) | |
5494 | continue; | |
5495 | ||
5496 | index_srcdev = i; | |
5497 | found = 1; | |
5498 | physical_of_found = tmp_bbio->stripes[i].physical; | |
ad6d620e SB |
5499 | } |
5500 | ||
94a97dfe ZL |
5501 | btrfs_put_bbio(tmp_bbio); |
5502 | ||
5503 | if (!found) { | |
ad6d620e SB |
5504 | WARN_ON(1); |
5505 | ret = -EIO; | |
ad6d620e SB |
5506 | goto out; |
5507 | } | |
5508 | ||
94a97dfe ZL |
5509 | mirror_num = index_srcdev + 1; |
5510 | patch_the_first_stripe_for_dev_replace = 1; | |
5511 | physical_to_patch_in_first_stripe = physical_of_found; | |
ad6d620e SB |
5512 | } else if (mirror_num > map->num_stripes) { |
5513 | mirror_num = 0; | |
5514 | } | |
5515 | ||
f2d8d74d | 5516 | num_stripes = 1; |
cea9e445 | 5517 | stripe_index = 0; |
fce3bb9a | 5518 | stripe_nr_orig = stripe_nr; |
fda2832f | 5519 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
b8b93add | 5520 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); |
fce3bb9a LD |
5521 | stripe_end_offset = stripe_nr_end * map->stripe_len - |
5522 | (offset + *length); | |
53b381b3 | 5523 | |
fce3bb9a | 5524 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
b3d3fa51 | 5525 | if (op == REQ_OP_DISCARD) |
fce3bb9a LD |
5526 | num_stripes = min_t(u64, map->num_stripes, |
5527 | stripe_nr_end - stripe_nr_orig); | |
47c5713f DS |
5528 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5529 | &stripe_index); | |
b3d3fa51 MC |
5530 | if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD && |
5531 | op != REQ_GET_READ_MIRRORS) | |
28e1cc7d | 5532 | mirror_num = 1; |
fce3bb9a | 5533 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
b3d3fa51 MC |
5534 | if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD || |
5535 | op == REQ_GET_READ_MIRRORS) | |
f2d8d74d | 5536 | num_stripes = map->num_stripes; |
2fff734f | 5537 | else if (mirror_num) |
f188591e | 5538 | stripe_index = mirror_num - 1; |
dfe25020 | 5539 | else { |
30d9861f | 5540 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5541 | map->num_stripes, |
30d9861f SB |
5542 | current->pid % map->num_stripes, |
5543 | dev_replace_is_ongoing); | |
a1d3c478 | 5544 | mirror_num = stripe_index + 1; |
dfe25020 | 5545 | } |
2fff734f | 5546 | |
611f0e00 | 5547 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
df5c82a8 | 5548 | if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD || |
b3d3fa51 | 5549 | op == REQ_GET_READ_MIRRORS) { |
f2d8d74d | 5550 | num_stripes = map->num_stripes; |
a1d3c478 | 5551 | } else if (mirror_num) { |
f188591e | 5552 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5553 | } else { |
5554 | mirror_num = 1; | |
5555 | } | |
2fff734f | 5556 | |
321aecc6 | 5557 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5558 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5559 | |
47c5713f | 5560 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5561 | stripe_index *= map->sub_stripes; |
5562 | ||
b3d3fa51 | 5563 | if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) |
f2d8d74d | 5564 | num_stripes = map->sub_stripes; |
b3d3fa51 | 5565 | else if (op == REQ_OP_DISCARD) |
fce3bb9a LD |
5566 | num_stripes = min_t(u64, map->sub_stripes * |
5567 | (stripe_nr_end - stripe_nr_orig), | |
5568 | map->num_stripes); | |
321aecc6 CM |
5569 | else if (mirror_num) |
5570 | stripe_index += mirror_num - 1; | |
dfe25020 | 5571 | else { |
3e74317a | 5572 | int old_stripe_index = stripe_index; |
30d9861f SB |
5573 | stripe_index = find_live_mirror(fs_info, map, |
5574 | stripe_index, | |
dfe25020 | 5575 | map->sub_stripes, stripe_index + |
30d9861f SB |
5576 | current->pid % map->sub_stripes, |
5577 | dev_replace_is_ongoing); | |
3e74317a | 5578 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5579 | } |
53b381b3 | 5580 | |
ffe2d203 | 5581 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5582 | if (need_raid_map && |
b3d3fa51 | 5583 | (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS || |
af8e2d1d | 5584 | mirror_num > 1)) { |
53b381b3 | 5585 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5586 | stripe_nr = div_u64(raid56_full_stripe_start, |
5587 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5588 | |
5589 | /* RAID[56] write or recovery. Return all stripes */ | |
5590 | num_stripes = map->num_stripes; | |
5591 | max_errors = nr_parity_stripes(map); | |
5592 | ||
53b381b3 DW |
5593 | *length = map->stripe_len; |
5594 | stripe_index = 0; | |
5595 | stripe_offset = 0; | |
5596 | } else { | |
5597 | /* | |
5598 | * Mirror #0 or #1 means the original data block. | |
5599 | * Mirror #2 is RAID5 parity block. | |
5600 | * Mirror #3 is RAID6 Q block. | |
5601 | */ | |
47c5713f DS |
5602 | stripe_nr = div_u64_rem(stripe_nr, |
5603 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5604 | if (mirror_num > 1) |
5605 | stripe_index = nr_data_stripes(map) + | |
5606 | mirror_num - 2; | |
5607 | ||
5608 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5609 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5610 | &stripe_index); | |
b3d3fa51 MC |
5611 | if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD && |
5612 | op != REQ_GET_READ_MIRRORS) && mirror_num <= 1) | |
28e1cc7d | 5613 | mirror_num = 1; |
53b381b3 | 5614 | } |
8790d502 CM |
5615 | } else { |
5616 | /* | |
47c5713f DS |
5617 | * after this, stripe_nr is the number of stripes on this |
5618 | * device we have to walk to find the data, and stripe_index is | |
5619 | * the number of our device in the stripe array | |
8790d502 | 5620 | */ |
47c5713f DS |
5621 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5622 | &stripe_index); | |
a1d3c478 | 5623 | mirror_num = stripe_index + 1; |
8790d502 | 5624 | } |
e042d1ec | 5625 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5626 | btrfs_crit(fs_info, |
5627 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5628 | stripe_index, map->num_stripes); |
5629 | ret = -EINVAL; | |
5630 | goto out; | |
5631 | } | |
cea9e445 | 5632 | |
472262f3 | 5633 | num_alloc_stripes = num_stripes; |
ad6d620e | 5634 | if (dev_replace_is_ongoing) { |
b3d3fa51 | 5635 | if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) |
ad6d620e | 5636 | num_alloc_stripes <<= 1; |
b3d3fa51 | 5637 | if (op == REQ_GET_READ_MIRRORS) |
ad6d620e | 5638 | num_alloc_stripes++; |
2c8cdd6e | 5639 | tgtdev_indexes = num_stripes; |
ad6d620e | 5640 | } |
2c8cdd6e | 5641 | |
6e9606d2 | 5642 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5643 | if (!bbio) { |
5644 | ret = -ENOMEM; | |
5645 | goto out; | |
5646 | } | |
2c8cdd6e MX |
5647 | if (dev_replace_is_ongoing) |
5648 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
de11cc12 | 5649 | |
8e5cfb55 | 5650 | /* build raid_map */ |
ffe2d203 | 5651 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
b3d3fa51 MC |
5652 | need_raid_map && |
5653 | ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) || | |
8e5cfb55 ZL |
5654 | mirror_num > 1)) { |
5655 | u64 tmp; | |
9d644a62 | 5656 | unsigned rot; |
8e5cfb55 ZL |
5657 | |
5658 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5659 | sizeof(struct btrfs_bio_stripe) * | |
5660 | num_alloc_stripes + | |
5661 | sizeof(int) * tgtdev_indexes); | |
5662 | ||
5663 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5664 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5665 | |
5666 | /* Fill in the logical address of each stripe */ | |
5667 | tmp = stripe_nr * nr_data_stripes(map); | |
5668 | for (i = 0; i < nr_data_stripes(map); i++) | |
5669 | bbio->raid_map[(i+rot) % num_stripes] = | |
5670 | em->start + (tmp + i) * map->stripe_len; | |
5671 | ||
5672 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5673 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5674 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5675 | RAID6_Q_STRIPE; | |
5676 | } | |
5677 | ||
b3d3fa51 | 5678 | if (op == REQ_OP_DISCARD) { |
9d644a62 DS |
5679 | u32 factor = 0; |
5680 | u32 sub_stripes = 0; | |
ec9ef7a1 LZ |
5681 | u64 stripes_per_dev = 0; |
5682 | u32 remaining_stripes = 0; | |
b89203f7 | 5683 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5684 | |
5685 | if (map->type & | |
5686 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5687 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5688 | sub_stripes = 1; | |
5689 | else | |
5690 | sub_stripes = map->sub_stripes; | |
5691 | ||
5692 | factor = map->num_stripes / sub_stripes; | |
5693 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5694 | stripe_nr_orig, | |
5695 | factor, | |
5696 | &remaining_stripes); | |
b89203f7 LB |
5697 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5698 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5699 | } |
5700 | ||
fce3bb9a | 5701 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5702 | bbio->stripes[i].physical = |
f2d8d74d CM |
5703 | map->stripes[stripe_index].physical + |
5704 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5705 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5706 | |
ec9ef7a1 LZ |
5707 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5708 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5709 | bbio->stripes[i].length = stripes_per_dev * | |
5710 | map->stripe_len; | |
b89203f7 | 5711 | |
ec9ef7a1 LZ |
5712 | if (i / sub_stripes < remaining_stripes) |
5713 | bbio->stripes[i].length += | |
5714 | map->stripe_len; | |
b89203f7 LB |
5715 | |
5716 | /* | |
5717 | * Special for the first stripe and | |
5718 | * the last stripe: | |
5719 | * | |
5720 | * |-------|...|-------| | |
5721 | * |----------| | |
5722 | * off end_off | |
5723 | */ | |
ec9ef7a1 | 5724 | if (i < sub_stripes) |
a1d3c478 | 5725 | bbio->stripes[i].length -= |
fce3bb9a | 5726 | stripe_offset; |
b89203f7 LB |
5727 | |
5728 | if (stripe_index >= last_stripe && | |
5729 | stripe_index <= (last_stripe + | |
5730 | sub_stripes - 1)) | |
a1d3c478 | 5731 | bbio->stripes[i].length -= |
fce3bb9a | 5732 | stripe_end_offset; |
b89203f7 | 5733 | |
ec9ef7a1 LZ |
5734 | if (i == sub_stripes - 1) |
5735 | stripe_offset = 0; | |
fce3bb9a | 5736 | } else |
a1d3c478 | 5737 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5738 | |
5739 | stripe_index++; | |
5740 | if (stripe_index == map->num_stripes) { | |
5741 | /* This could only happen for RAID0/10 */ | |
5742 | stripe_index = 0; | |
5743 | stripe_nr++; | |
5744 | } | |
5745 | } | |
5746 | } else { | |
5747 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5748 | bbio->stripes[i].physical = |
212a17ab LT |
5749 | map->stripes[stripe_index].physical + |
5750 | stripe_offset + | |
5751 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5752 | bbio->stripes[i].dev = |
212a17ab | 5753 | map->stripes[stripe_index].dev; |
fce3bb9a | 5754 | stripe_index++; |
f2d8d74d | 5755 | } |
593060d7 | 5756 | } |
de11cc12 | 5757 | |
b3d3fa51 | 5758 | if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) |
d20983b4 | 5759 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5760 | |
8e5cfb55 ZL |
5761 | if (bbio->raid_map) |
5762 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5763 | |
2c8cdd6e | 5764 | tgtdev_indexes = 0; |
b3d3fa51 MC |
5765 | if (dev_replace_is_ongoing && |
5766 | (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) && | |
472262f3 SB |
5767 | dev_replace->tgtdev != NULL) { |
5768 | int index_where_to_add; | |
5769 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5770 | ||
5771 | /* | |
5772 | * duplicate the write operations while the dev replace | |
5773 | * procedure is running. Since the copying of the old disk | |
5774 | * to the new disk takes place at run time while the | |
5775 | * filesystem is mounted writable, the regular write | |
5776 | * operations to the old disk have to be duplicated to go | |
5777 | * to the new disk as well. | |
5778 | * Note that device->missing is handled by the caller, and | |
5779 | * that the write to the old disk is already set up in the | |
5780 | * stripes array. | |
5781 | */ | |
5782 | index_where_to_add = num_stripes; | |
5783 | for (i = 0; i < num_stripes; i++) { | |
5784 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5785 | /* write to new disk, too */ | |
5786 | struct btrfs_bio_stripe *new = | |
5787 | bbio->stripes + index_where_to_add; | |
5788 | struct btrfs_bio_stripe *old = | |
5789 | bbio->stripes + i; | |
5790 | ||
5791 | new->physical = old->physical; | |
5792 | new->length = old->length; | |
5793 | new->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5794 | bbio->tgtdev_map[i] = index_where_to_add; |
472262f3 SB |
5795 | index_where_to_add++; |
5796 | max_errors++; | |
2c8cdd6e | 5797 | tgtdev_indexes++; |
472262f3 SB |
5798 | } |
5799 | } | |
5800 | num_stripes = index_where_to_add; | |
b3d3fa51 | 5801 | } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) && |
ad6d620e SB |
5802 | dev_replace->tgtdev != NULL) { |
5803 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5804 | int index_srcdev = 0; | |
5805 | int found = 0; | |
5806 | u64 physical_of_found = 0; | |
5807 | ||
5808 | /* | |
5809 | * During the dev-replace procedure, the target drive can | |
5810 | * also be used to read data in case it is needed to repair | |
5811 | * a corrupt block elsewhere. This is possible if the | |
5812 | * requested area is left of the left cursor. In this area, | |
5813 | * the target drive is a full copy of the source drive. | |
5814 | */ | |
5815 | for (i = 0; i < num_stripes; i++) { | |
5816 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5817 | /* | |
5818 | * In case of DUP, in order to keep it | |
5819 | * simple, only add the mirror with the | |
5820 | * lowest physical address | |
5821 | */ | |
5822 | if (found && | |
5823 | physical_of_found <= | |
5824 | bbio->stripes[i].physical) | |
5825 | continue; | |
5826 | index_srcdev = i; | |
5827 | found = 1; | |
5828 | physical_of_found = bbio->stripes[i].physical; | |
5829 | } | |
5830 | } | |
5831 | if (found) { | |
22ab04e8 FM |
5832 | struct btrfs_bio_stripe *tgtdev_stripe = |
5833 | bbio->stripes + num_stripes; | |
ad6d620e | 5834 | |
22ab04e8 FM |
5835 | tgtdev_stripe->physical = physical_of_found; |
5836 | tgtdev_stripe->length = | |
5837 | bbio->stripes[index_srcdev].length; | |
5838 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5839 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
ad6d620e | 5840 | |
22ab04e8 FM |
5841 | tgtdev_indexes++; |
5842 | num_stripes++; | |
ad6d620e | 5843 | } |
472262f3 SB |
5844 | } |
5845 | ||
de11cc12 | 5846 | *bbio_ret = bbio; |
10f11900 | 5847 | bbio->map_type = map->type; |
de11cc12 LZ |
5848 | bbio->num_stripes = num_stripes; |
5849 | bbio->max_errors = max_errors; | |
5850 | bbio->mirror_num = mirror_num; | |
2c8cdd6e | 5851 | bbio->num_tgtdevs = tgtdev_indexes; |
ad6d620e SB |
5852 | |
5853 | /* | |
5854 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5855 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5856 | * available as a mirror | |
5857 | */ | |
5858 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5859 | WARN_ON(num_stripes > 1); | |
5860 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5861 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5862 | bbio->mirror_num = map->num_stripes + 1; | |
5863 | } | |
cea9e445 | 5864 | out: |
73beece9 LB |
5865 | if (dev_replace_is_ongoing) { |
5866 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5867 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5868 | } | |
0b86a832 | 5869 | free_extent_map(em); |
de11cc12 | 5870 | return ret; |
0b86a832 CM |
5871 | } |
5872 | ||
b3d3fa51 | 5873 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int op, |
f2d8d74d | 5874 | u64 logical, u64 *length, |
a1d3c478 | 5875 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5876 | { |
b3d3fa51 | 5877 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5878 | mirror_num, 0); |
f2d8d74d CM |
5879 | } |
5880 | ||
af8e2d1d | 5881 | /* For Scrub/replace */ |
b3d3fa51 | 5882 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op, |
af8e2d1d MX |
5883 | u64 logical, u64 *length, |
5884 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5885 | int need_raid_map) |
af8e2d1d | 5886 | { |
b3d3fa51 | 5887 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5888 | mirror_num, need_raid_map); |
af8e2d1d MX |
5889 | } |
5890 | ||
ab8d0fc4 | 5891 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
5892 | u64 chunk_start, u64 physical, u64 devid, |
5893 | u64 **logical, int *naddrs, int *stripe_len) | |
5894 | { | |
ab8d0fc4 | 5895 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
a512bbf8 YZ |
5896 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5897 | struct extent_map *em; | |
5898 | struct map_lookup *map; | |
5899 | u64 *buf; | |
5900 | u64 bytenr; | |
5901 | u64 length; | |
5902 | u64 stripe_nr; | |
53b381b3 | 5903 | u64 rmap_len; |
a512bbf8 YZ |
5904 | int i, j, nr = 0; |
5905 | ||
890871be | 5906 | read_lock(&em_tree->lock); |
a512bbf8 | 5907 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5908 | read_unlock(&em_tree->lock); |
a512bbf8 | 5909 | |
835d974f | 5910 | if (!em) { |
ab8d0fc4 JM |
5911 | btrfs_err(fs_info, "couldn't find em for chunk %Lu", |
5912 | chunk_start); | |
835d974f JB |
5913 | return -EIO; |
5914 | } | |
5915 | ||
5916 | if (em->start != chunk_start) { | |
ab8d0fc4 | 5917 | btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu", |
835d974f JB |
5918 | em->start, chunk_start); |
5919 | free_extent_map(em); | |
5920 | return -EIO; | |
5921 | } | |
95617d69 | 5922 | map = em->map_lookup; |
a512bbf8 YZ |
5923 | |
5924 | length = em->len; | |
53b381b3 DW |
5925 | rmap_len = map->stripe_len; |
5926 | ||
a512bbf8 | 5927 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5928 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5929 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5930 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5931 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5932 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5933 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5934 | } | |
a512bbf8 | 5935 | |
31e818fe | 5936 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5937 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5938 | |
5939 | for (i = 0; i < map->num_stripes; i++) { | |
5940 | if (devid && map->stripes[i].dev->devid != devid) | |
5941 | continue; | |
5942 | if (map->stripes[i].physical > physical || | |
5943 | map->stripes[i].physical + length <= physical) | |
5944 | continue; | |
5945 | ||
5946 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 5947 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5948 | |
5949 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5950 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5951 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5952 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5953 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5954 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5955 | * Alternatively, just use rmap_len below instead of | |
5956 | * map->stripe_len */ | |
5957 | ||
5958 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5959 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5960 | for (j = 0; j < nr; j++) { |
5961 | if (buf[j] == bytenr) | |
5962 | break; | |
5963 | } | |
934d375b CM |
5964 | if (j == nr) { |
5965 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5966 | buf[nr++] = bytenr; |
934d375b | 5967 | } |
a512bbf8 YZ |
5968 | } |
5969 | ||
a512bbf8 YZ |
5970 | *logical = buf; |
5971 | *naddrs = nr; | |
53b381b3 | 5972 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5973 | |
5974 | free_extent_map(em); | |
5975 | return 0; | |
f2d8d74d CM |
5976 | } |
5977 | ||
4246a0b6 | 5978 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 5979 | { |
326e1dbb MS |
5980 | bio->bi_private = bbio->private; |
5981 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 5982 | bio_endio(bio); |
326e1dbb | 5983 | |
6e9606d2 | 5984 | btrfs_put_bbio(bbio); |
8408c716 MX |
5985 | } |
5986 | ||
4246a0b6 | 5987 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 5988 | { |
9be3395b | 5989 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 5990 | int is_orig_bio = 0; |
8790d502 | 5991 | |
4246a0b6 | 5992 | if (bio->bi_error) { |
a1d3c478 | 5993 | atomic_inc(&bbio->error); |
4246a0b6 | 5994 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { |
442a4f63 | 5995 | unsigned int stripe_index = |
9be3395b | 5996 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 5997 | struct btrfs_device *dev; |
442a4f63 SB |
5998 | |
5999 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6000 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6001 | if (dev->bdev) { |
37226b21 | 6002 | if (bio_op(bio) == REQ_OP_WRITE) |
597a60fa SB |
6003 | btrfs_dev_stat_inc(dev, |
6004 | BTRFS_DEV_STAT_WRITE_ERRS); | |
6005 | else | |
6006 | btrfs_dev_stat_inc(dev, | |
6007 | BTRFS_DEV_STAT_READ_ERRS); | |
1eff9d32 | 6008 | if ((bio->bi_opf & WRITE_FLUSH) == WRITE_FLUSH) |
597a60fa SB |
6009 | btrfs_dev_stat_inc(dev, |
6010 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
6011 | btrfs_dev_stat_print_on_error(dev); | |
6012 | } | |
442a4f63 SB |
6013 | } |
6014 | } | |
8790d502 | 6015 | |
a1d3c478 | 6016 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6017 | is_orig_bio = 1; |
6018 | ||
c404e0dc MX |
6019 | btrfs_bio_counter_dec(bbio->fs_info); |
6020 | ||
a1d3c478 | 6021 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6022 | if (!is_orig_bio) { |
6023 | bio_put(bio); | |
a1d3c478 | 6024 | bio = bbio->orig_bio; |
7d2b4daa | 6025 | } |
c7b22bb1 | 6026 | |
9be3395b | 6027 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6028 | /* only send an error to the higher layers if it is |
53b381b3 | 6029 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6030 | */ |
a1d3c478 | 6031 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4246a0b6 | 6032 | bio->bi_error = -EIO; |
5dbc8fca | 6033 | } else { |
1259ab75 CM |
6034 | /* |
6035 | * this bio is actually up to date, we didn't | |
6036 | * go over the max number of errors | |
6037 | */ | |
4246a0b6 | 6038 | bio->bi_error = 0; |
1259ab75 | 6039 | } |
c55f1396 | 6040 | |
4246a0b6 | 6041 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6042 | } else if (!is_orig_bio) { |
8790d502 CM |
6043 | bio_put(bio); |
6044 | } | |
8790d502 CM |
6045 | } |
6046 | ||
8b712842 CM |
6047 | /* |
6048 | * see run_scheduled_bios for a description of why bios are collected for | |
6049 | * async submit. | |
6050 | * | |
6051 | * This will add one bio to the pending list for a device and make sure | |
6052 | * the work struct is scheduled. | |
6053 | */ | |
48a3b636 ES |
6054 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
6055 | struct btrfs_device *device, | |
4e49ea4a | 6056 | struct bio *bio) |
8b712842 CM |
6057 | { |
6058 | int should_queue = 1; | |
ffbd517d | 6059 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6060 | |
53b381b3 | 6061 | if (device->missing || !device->bdev) { |
4246a0b6 | 6062 | bio_io_error(bio); |
53b381b3 DW |
6063 | return; |
6064 | } | |
6065 | ||
8b712842 | 6066 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6067 | if (bio_op(bio) == REQ_OP_READ) { |
492bb6de | 6068 | bio_get(bio); |
4e49ea4a | 6069 | btrfsic_submit_bio(bio); |
492bb6de | 6070 | bio_put(bio); |
143bede5 | 6071 | return; |
8b712842 CM |
6072 | } |
6073 | ||
6074 | /* | |
0986fe9e | 6075 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
6076 | * higher layers. Otherwise, the async bio makes it appear we have |
6077 | * made progress against dirty pages when we've really just put it | |
6078 | * on a queue for later | |
6079 | */ | |
0986fe9e | 6080 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 6081 | WARN_ON(bio->bi_next); |
8b712842 | 6082 | bio->bi_next = NULL; |
8b712842 CM |
6083 | |
6084 | spin_lock(&device->io_lock); | |
1eff9d32 | 6085 | if (bio->bi_opf & REQ_SYNC) |
ffbd517d CM |
6086 | pending_bios = &device->pending_sync_bios; |
6087 | else | |
6088 | pending_bios = &device->pending_bios; | |
8b712842 | 6089 | |
ffbd517d CM |
6090 | if (pending_bios->tail) |
6091 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6092 | |
ffbd517d CM |
6093 | pending_bios->tail = bio; |
6094 | if (!pending_bios->head) | |
6095 | pending_bios->head = bio; | |
8b712842 CM |
6096 | if (device->running_pending) |
6097 | should_queue = 0; | |
6098 | ||
6099 | spin_unlock(&device->io_lock); | |
6100 | ||
6101 | if (should_queue) | |
a8c93d4e QW |
6102 | btrfs_queue_work(root->fs_info->submit_workers, |
6103 | &device->work); | |
8b712842 CM |
6104 | } |
6105 | ||
de1ee92a JB |
6106 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, |
6107 | struct bio *bio, u64 physical, int dev_nr, | |
81a75f67 | 6108 | int async) |
de1ee92a JB |
6109 | { |
6110 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
6111 | ||
6112 | bio->bi_private = bbio; | |
9be3395b | 6113 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6114 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6115 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6116 | #ifdef DEBUG |
6117 | { | |
6118 | struct rcu_string *name; | |
6119 | ||
6120 | rcu_read_lock(); | |
6121 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6122 | btrfs_debug(fs_info, |
6123 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6124 | bio_op(bio), bio->bi_opf, | |
6125 | (u64)bio->bi_iter.bi_sector, | |
6126 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6127 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6128 | rcu_read_unlock(); |
6129 | } | |
6130 | #endif | |
6131 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
6132 | |
6133 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
6134 | ||
de1ee92a | 6135 | if (async) |
4e49ea4a | 6136 | btrfs_schedule_bio(root, dev, bio); |
de1ee92a | 6137 | else |
4e49ea4a | 6138 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6139 | } |
6140 | ||
de1ee92a JB |
6141 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6142 | { | |
6143 | atomic_inc(&bbio->error); | |
6144 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6145 | /* Should be the original bio. */ |
8408c716 MX |
6146 | WARN_ON(bio != bbio->orig_bio); |
6147 | ||
9be3395b | 6148 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6149 | bio->bi_iter.bi_sector = logical >> 9; |
4246a0b6 CH |
6150 | bio->bi_error = -EIO; |
6151 | btrfs_end_bbio(bbio, bio); | |
de1ee92a JB |
6152 | } |
6153 | } | |
6154 | ||
81a75f67 | 6155 | int btrfs_map_bio(struct btrfs_root *root, struct bio *bio, |
8b712842 | 6156 | int mirror_num, int async_submit) |
0b86a832 | 6157 | { |
0b86a832 | 6158 | struct btrfs_device *dev; |
8790d502 | 6159 | struct bio *first_bio = bio; |
4f024f37 | 6160 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6161 | u64 length = 0; |
6162 | u64 map_length; | |
0b86a832 | 6163 | int ret; |
08da757d ZL |
6164 | int dev_nr; |
6165 | int total_devs; | |
a1d3c478 | 6166 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6167 | |
4f024f37 | 6168 | length = bio->bi_iter.bi_size; |
0b86a832 | 6169 | map_length = length; |
cea9e445 | 6170 | |
c404e0dc | 6171 | btrfs_bio_counter_inc_blocked(root->fs_info); |
37226b21 MC |
6172 | ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical, |
6173 | &map_length, &bbio, mirror_num, 1); | |
c404e0dc MX |
6174 | if (ret) { |
6175 | btrfs_bio_counter_dec(root->fs_info); | |
79787eaa | 6176 | return ret; |
c404e0dc | 6177 | } |
cea9e445 | 6178 | |
a1d3c478 | 6179 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6180 | bbio->orig_bio = first_bio; |
6181 | bbio->private = first_bio->bi_private; | |
6182 | bbio->end_io = first_bio->bi_end_io; | |
c404e0dc | 6183 | bbio->fs_info = root->fs_info; |
53b381b3 DW |
6184 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6185 | ||
ad1ba2a0 | 6186 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6187 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6188 | /* In this case, map_length has been set to the length of |
6189 | a single stripe; not the whole write */ | |
37226b21 | 6190 | if (bio_op(bio) == REQ_OP_WRITE) { |
8e5cfb55 | 6191 | ret = raid56_parity_write(root, bio, bbio, map_length); |
53b381b3 | 6192 | } else { |
8e5cfb55 | 6193 | ret = raid56_parity_recover(root, bio, bbio, map_length, |
4245215d | 6194 | mirror_num, 1); |
53b381b3 | 6195 | } |
4245215d | 6196 | |
c404e0dc MX |
6197 | btrfs_bio_counter_dec(root->fs_info); |
6198 | return ret; | |
53b381b3 DW |
6199 | } |
6200 | ||
cea9e445 | 6201 | if (map_length < length) { |
5d163e0e JM |
6202 | btrfs_crit(root->fs_info, |
6203 | "mapping failed logical %llu bio len %llu len %llu", | |
6204 | logical, length, map_length); | |
cea9e445 CM |
6205 | BUG(); |
6206 | } | |
a1d3c478 | 6207 | |
08da757d | 6208 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6209 | dev = bbio->stripes[dev_nr].dev; |
37226b21 MC |
6210 | if (!dev || !dev->bdev || |
6211 | (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) { | |
de1ee92a | 6212 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6213 | continue; |
6214 | } | |
6215 | ||
a1d3c478 | 6216 | if (dev_nr < total_devs - 1) { |
9be3395b | 6217 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 6218 | BUG_ON(!bio); /* -ENOMEM */ |
326e1dbb | 6219 | } else |
a1d3c478 | 6220 | bio = first_bio; |
de1ee92a JB |
6221 | |
6222 | submit_stripe_bio(root, bbio, bio, | |
81a75f67 | 6223 | bbio->stripes[dev_nr].physical, dev_nr, |
de1ee92a | 6224 | async_submit); |
8790d502 | 6225 | } |
c404e0dc | 6226 | btrfs_bio_counter_dec(root->fs_info); |
0b86a832 CM |
6227 | return 0; |
6228 | } | |
6229 | ||
aa1b8cd4 | 6230 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6231 | u8 *uuid, u8 *fsid) |
0b86a832 | 6232 | { |
2b82032c YZ |
6233 | struct btrfs_device *device; |
6234 | struct btrfs_fs_devices *cur_devices; | |
6235 | ||
aa1b8cd4 | 6236 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6237 | while (cur_devices) { |
6238 | if (!fsid || | |
6239 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6240 | device = __find_device(&cur_devices->devices, | |
6241 | devid, uuid); | |
6242 | if (device) | |
6243 | return device; | |
6244 | } | |
6245 | cur_devices = cur_devices->seed; | |
6246 | } | |
6247 | return NULL; | |
0b86a832 CM |
6248 | } |
6249 | ||
dfe25020 | 6250 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 6251 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6252 | u64 devid, u8 *dev_uuid) |
6253 | { | |
6254 | struct btrfs_device *device; | |
dfe25020 | 6255 | |
12bd2fc0 ID |
6256 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6257 | if (IS_ERR(device)) | |
7cbd8a83 | 6258 | return NULL; |
12bd2fc0 ID |
6259 | |
6260 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6261 | device->fs_devices = fs_devices; |
dfe25020 | 6262 | fs_devices->num_devices++; |
12bd2fc0 ID |
6263 | |
6264 | device->missing = 1; | |
cd02dca5 | 6265 | fs_devices->missing_devices++; |
12bd2fc0 | 6266 | |
dfe25020 CM |
6267 | return device; |
6268 | } | |
6269 | ||
12bd2fc0 ID |
6270 | /** |
6271 | * btrfs_alloc_device - allocate struct btrfs_device | |
6272 | * @fs_info: used only for generating a new devid, can be NULL if | |
6273 | * devid is provided (i.e. @devid != NULL). | |
6274 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6275 | * is generated. | |
6276 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6277 | * is generated. | |
6278 | * | |
6279 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6280 | * on error. Returned struct is not linked onto any lists and can be | |
6281 | * destroyed with kfree() right away. | |
6282 | */ | |
6283 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6284 | const u64 *devid, | |
6285 | const u8 *uuid) | |
6286 | { | |
6287 | struct btrfs_device *dev; | |
6288 | u64 tmp; | |
6289 | ||
fae7f21c | 6290 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6291 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6292 | |
6293 | dev = __alloc_device(); | |
6294 | if (IS_ERR(dev)) | |
6295 | return dev; | |
6296 | ||
6297 | if (devid) | |
6298 | tmp = *devid; | |
6299 | else { | |
6300 | int ret; | |
6301 | ||
6302 | ret = find_next_devid(fs_info, &tmp); | |
6303 | if (ret) { | |
6304 | kfree(dev); | |
6305 | return ERR_PTR(ret); | |
6306 | } | |
6307 | } | |
6308 | dev->devid = tmp; | |
6309 | ||
6310 | if (uuid) | |
6311 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6312 | else | |
6313 | generate_random_uuid(dev->uuid); | |
6314 | ||
9e0af237 LB |
6315 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6316 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6317 | |
6318 | return dev; | |
6319 | } | |
6320 | ||
e06cd3dd LB |
6321 | /* Return -EIO if any error, otherwise return 0. */ |
6322 | static int btrfs_check_chunk_valid(struct btrfs_root *root, | |
6323 | struct extent_buffer *leaf, | |
6324 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6325 | { |
0b86a832 | 6326 | u64 length; |
f04b772b | 6327 | u64 stripe_len; |
e06cd3dd LB |
6328 | u16 num_stripes; |
6329 | u16 sub_stripes; | |
6330 | u64 type; | |
0b86a832 | 6331 | |
e17cade2 | 6332 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6333 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6334 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6335 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6336 | type = btrfs_chunk_type(leaf, chunk); | |
6337 | ||
f04b772b QW |
6338 | if (!num_stripes) { |
6339 | btrfs_err(root->fs_info, "invalid chunk num_stripes: %u", | |
6340 | num_stripes); | |
6341 | return -EIO; | |
6342 | } | |
6343 | if (!IS_ALIGNED(logical, root->sectorsize)) { | |
6344 | btrfs_err(root->fs_info, | |
6345 | "invalid chunk logical %llu", logical); | |
6346 | return -EIO; | |
6347 | } | |
e06cd3dd LB |
6348 | if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) { |
6349 | btrfs_err(root->fs_info, "invalid chunk sectorsize %u", | |
6350 | btrfs_chunk_sector_size(leaf, chunk)); | |
6351 | return -EIO; | |
6352 | } | |
f04b772b QW |
6353 | if (!length || !IS_ALIGNED(length, root->sectorsize)) { |
6354 | btrfs_err(root->fs_info, | |
6355 | "invalid chunk length %llu", length); | |
6356 | return -EIO; | |
6357 | } | |
3d8da678 | 6358 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
f04b772b QW |
6359 | btrfs_err(root->fs_info, "invalid chunk stripe length: %llu", |
6360 | stripe_len); | |
6361 | return -EIO; | |
6362 | } | |
6363 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6364 | type) { |
f04b772b QW |
6365 | btrfs_err(root->fs_info, "unrecognized chunk type: %llu", |
6366 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | | |
6367 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6368 | btrfs_chunk_type(leaf, chunk)); | |
6369 | return -EIO; | |
6370 | } | |
e06cd3dd LB |
6371 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6372 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6373 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6374 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6375 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6376 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6377 | num_stripes != 1)) { | |
6378 | btrfs_err(root->fs_info, | |
6379 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", | |
6380 | num_stripes, sub_stripes, | |
6381 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6382 | return -EIO; | |
6383 | } | |
6384 | ||
6385 | return 0; | |
6386 | } | |
6387 | ||
6388 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, | |
6389 | struct extent_buffer *leaf, | |
6390 | struct btrfs_chunk *chunk) | |
6391 | { | |
6392 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6393 | struct map_lookup *map; | |
6394 | struct extent_map *em; | |
6395 | u64 logical; | |
6396 | u64 length; | |
6397 | u64 stripe_len; | |
6398 | u64 devid; | |
6399 | u8 uuid[BTRFS_UUID_SIZE]; | |
6400 | int num_stripes; | |
6401 | int ret; | |
6402 | int i; | |
6403 | ||
6404 | logical = key->offset; | |
6405 | length = btrfs_chunk_length(leaf, chunk); | |
6406 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6407 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
6408 | ||
6409 | ret = btrfs_check_chunk_valid(root, leaf, chunk, logical); | |
6410 | if (ret) | |
6411 | return ret; | |
a061fc8d | 6412 | |
890871be | 6413 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6414 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6415 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6416 | |
6417 | /* already mapped? */ | |
6418 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6419 | free_extent_map(em); | |
0b86a832 CM |
6420 | return 0; |
6421 | } else if (em) { | |
6422 | free_extent_map(em); | |
6423 | } | |
0b86a832 | 6424 | |
172ddd60 | 6425 | em = alloc_extent_map(); |
0b86a832 CM |
6426 | if (!em) |
6427 | return -ENOMEM; | |
593060d7 | 6428 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6429 | if (!map) { |
6430 | free_extent_map(em); | |
6431 | return -ENOMEM; | |
6432 | } | |
6433 | ||
298a8f9c | 6434 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6435 | em->map_lookup = map; |
0b86a832 CM |
6436 | em->start = logical; |
6437 | em->len = length; | |
70c8a91c | 6438 | em->orig_start = 0; |
0b86a832 | 6439 | em->block_start = 0; |
c8b97818 | 6440 | em->block_len = em->len; |
0b86a832 | 6441 | |
593060d7 CM |
6442 | map->num_stripes = num_stripes; |
6443 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6444 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6445 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6446 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6447 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6448 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6449 | for (i = 0; i < num_stripes; i++) { |
6450 | map->stripes[i].physical = | |
6451 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6452 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6453 | read_extent_buffer(leaf, uuid, (unsigned long) |
6454 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6455 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
6456 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
6457 | uuid, NULL); | |
3cdde224 JM |
6458 | if (!map->stripes[i].dev && |
6459 | !btrfs_test_opt(root->fs_info, DEGRADED)) { | |
593060d7 CM |
6460 | free_extent_map(em); |
6461 | return -EIO; | |
6462 | } | |
dfe25020 CM |
6463 | if (!map->stripes[i].dev) { |
6464 | map->stripes[i].dev = | |
5f375835 MX |
6465 | add_missing_dev(root, root->fs_info->fs_devices, |
6466 | devid, uuid); | |
dfe25020 | 6467 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6468 | free_extent_map(em); |
6469 | return -EIO; | |
6470 | } | |
5d163e0e JM |
6471 | btrfs_warn(root->fs_info, |
6472 | "devid %llu uuid %pU is missing", | |
6473 | 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 | ||
5f375835 MX |
6510 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
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)); |
2b82032c YZ |
6517 | |
6518 | fs_devices = root->fs_info->fs_devices->seed; | |
6519 | while (fs_devices) { | |
5f375835 MX |
6520 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6521 | return fs_devices; | |
6522 | ||
2b82032c YZ |
6523 | fs_devices = fs_devices->seed; |
6524 | } | |
6525 | ||
6526 | fs_devices = find_fsid(fsid); | |
6527 | if (!fs_devices) { | |
3cdde224 | 6528 | if (!btrfs_test_opt(root->fs_info, DEGRADED)) |
5f375835 MX |
6529 | return ERR_PTR(-ENOENT); |
6530 | ||
6531 | fs_devices = alloc_fs_devices(fsid); | |
6532 | if (IS_ERR(fs_devices)) | |
6533 | return fs_devices; | |
6534 | ||
6535 | fs_devices->seeding = 1; | |
6536 | fs_devices->opened = 1; | |
6537 | return fs_devices; | |
2b82032c | 6538 | } |
e4404d6e YZ |
6539 | |
6540 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6541 | if (IS_ERR(fs_devices)) |
6542 | return fs_devices; | |
2b82032c | 6543 | |
97288f2c | 6544 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 6545 | root->fs_info->bdev_holder); |
48d28232 JL |
6546 | if (ret) { |
6547 | free_fs_devices(fs_devices); | |
5f375835 | 6548 | fs_devices = ERR_PTR(ret); |
2b82032c | 6549 | goto out; |
48d28232 | 6550 | } |
2b82032c YZ |
6551 | |
6552 | if (!fs_devices->seeding) { | |
6553 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6554 | free_fs_devices(fs_devices); |
5f375835 | 6555 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6556 | goto out; |
6557 | } | |
6558 | ||
6559 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6560 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6561 | out: |
5f375835 | 6562 | return fs_devices; |
2b82032c YZ |
6563 | } |
6564 | ||
0d81ba5d | 6565 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6566 | struct extent_buffer *leaf, |
6567 | struct btrfs_dev_item *dev_item) | |
6568 | { | |
5f375835 | 6569 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
0b86a832 CM |
6570 | struct btrfs_device *device; |
6571 | u64 devid; | |
6572 | int ret; | |
2b82032c | 6573 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6574 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6575 | ||
0b86a832 | 6576 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6577 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6578 | BTRFS_UUID_SIZE); |
1473b24e | 6579 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6580 | BTRFS_UUID_SIZE); |
6581 | ||
6582 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5f375835 MX |
6583 | fs_devices = open_seed_devices(root, fs_uuid); |
6584 | if (IS_ERR(fs_devices)) | |
6585 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6586 | } |
6587 | ||
aa1b8cd4 | 6588 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6589 | if (!device) { |
3cdde224 | 6590 | if (!btrfs_test_opt(root->fs_info, DEGRADED)) |
2b82032c YZ |
6591 | return -EIO; |
6592 | ||
5f375835 MX |
6593 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); |
6594 | if (!device) | |
6595 | return -ENOMEM; | |
33b97e43 AJ |
6596 | btrfs_warn(root->fs_info, "devid %llu uuid %pU missing", |
6597 | devid, dev_uuid); | |
5f375835 | 6598 | } else { |
3cdde224 | 6599 | if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED)) |
5f375835 MX |
6600 | return -EIO; |
6601 | ||
6602 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6603 | /* |
6604 | * this happens when a device that was properly setup | |
6605 | * in the device info lists suddenly goes bad. | |
6606 | * device->bdev is NULL, and so we have to set | |
6607 | * device->missing to one here | |
6608 | */ | |
5f375835 | 6609 | device->fs_devices->missing_devices++; |
cd02dca5 | 6610 | device->missing = 1; |
2b82032c | 6611 | } |
5f375835 MX |
6612 | |
6613 | /* Move the device to its own fs_devices */ | |
6614 | if (device->fs_devices != fs_devices) { | |
6615 | ASSERT(device->missing); | |
6616 | ||
6617 | list_move(&device->dev_list, &fs_devices->devices); | |
6618 | device->fs_devices->num_devices--; | |
6619 | fs_devices->num_devices++; | |
6620 | ||
6621 | device->fs_devices->missing_devices--; | |
6622 | fs_devices->missing_devices++; | |
6623 | ||
6624 | device->fs_devices = fs_devices; | |
6625 | } | |
2b82032c YZ |
6626 | } |
6627 | ||
6628 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6629 | BUG_ON(device->writeable); | |
6630 | if (device->generation != | |
6631 | btrfs_device_generation(leaf, dev_item)) | |
6632 | return -EINVAL; | |
6324fbf3 | 6633 | } |
0b86a832 CM |
6634 | |
6635 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6636 | device->in_fs_metadata = 1; |
63a212ab | 6637 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6638 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
6639 | spin_lock(&root->fs_info->free_chunk_lock); |
6640 | root->fs_info->free_chunk_space += device->total_bytes - | |
6641 | device->bytes_used; | |
6642 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6643 | } | |
0b86a832 | 6644 | ret = 0; |
0b86a832 CM |
6645 | return ret; |
6646 | } | |
6647 | ||
e4404d6e | 6648 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 6649 | { |
ab8d0fc4 JM |
6650 | struct btrfs_fs_info *fs_info = root->fs_info; |
6651 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
a061fc8d | 6652 | struct extent_buffer *sb; |
0b86a832 | 6653 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6654 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6655 | u8 *array_ptr; |
6656 | unsigned long sb_array_offset; | |
84eed90f | 6657 | int ret = 0; |
0b86a832 CM |
6658 | u32 num_stripes; |
6659 | u32 array_size; | |
6660 | u32 len = 0; | |
1ffb22cf | 6661 | u32 cur_offset; |
e06cd3dd | 6662 | u64 type; |
84eed90f | 6663 | struct btrfs_key key; |
0b86a832 | 6664 | |
a83fffb7 DS |
6665 | ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize); |
6666 | /* | |
6667 | * This will create extent buffer of nodesize, superblock size is | |
6668 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6669 | * overallocate but we can keep it as-is, only the first page is used. | |
6670 | */ | |
6671 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
c871b0f2 LB |
6672 | if (IS_ERR(sb)) |
6673 | return PTR_ERR(sb); | |
4db8c528 | 6674 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6675 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6676 | /* |
01327610 | 6677 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6678 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6679 | * pages up-to-date when the page is larger: extent does not cover the |
6680 | * whole page and consequently check_page_uptodate does not find all | |
6681 | * the page's extents up-to-date (the hole beyond sb), | |
6682 | * write_extent_buffer then triggers a WARN_ON. | |
6683 | * | |
6684 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6685 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6686 | * to silence the warning eg. on PowerPC 64. | |
6687 | */ | |
09cbfeaf | 6688 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6689 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6690 | |
a061fc8d | 6691 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6692 | array_size = btrfs_super_sys_array_size(super_copy); |
6693 | ||
1ffb22cf DS |
6694 | array_ptr = super_copy->sys_chunk_array; |
6695 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6696 | cur_offset = 0; | |
0b86a832 | 6697 | |
1ffb22cf DS |
6698 | while (cur_offset < array_size) { |
6699 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6700 | len = sizeof(*disk_key); |
6701 | if (cur_offset + len > array_size) | |
6702 | goto out_short_read; | |
6703 | ||
0b86a832 CM |
6704 | btrfs_disk_key_to_cpu(&key, disk_key); |
6705 | ||
1ffb22cf DS |
6706 | array_ptr += len; |
6707 | sb_array_offset += len; | |
6708 | cur_offset += len; | |
0b86a832 | 6709 | |
0d81ba5d | 6710 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6711 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6712 | /* |
6713 | * At least one btrfs_chunk with one stripe must be | |
6714 | * present, exact stripe count check comes afterwards | |
6715 | */ | |
6716 | len = btrfs_chunk_item_size(1); | |
6717 | if (cur_offset + len > array_size) | |
6718 | goto out_short_read; | |
6719 | ||
6720 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6721 | if (!num_stripes) { |
ab8d0fc4 JM |
6722 | btrfs_err(fs_info, |
6723 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6724 | num_stripes, cur_offset); |
6725 | ret = -EIO; | |
6726 | break; | |
6727 | } | |
6728 | ||
e06cd3dd LB |
6729 | type = btrfs_chunk_type(sb, chunk); |
6730 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6731 | btrfs_err(fs_info, |
e06cd3dd LB |
6732 | "invalid chunk type %llu in sys_array at offset %u", |
6733 | type, cur_offset); | |
6734 | ret = -EIO; | |
6735 | break; | |
6736 | } | |
6737 | ||
e3540eab DS |
6738 | len = btrfs_chunk_item_size(num_stripes); |
6739 | if (cur_offset + len > array_size) | |
6740 | goto out_short_read; | |
6741 | ||
0d81ba5d | 6742 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6743 | if (ret) |
6744 | break; | |
0b86a832 | 6745 | } else { |
ab8d0fc4 JM |
6746 | btrfs_err(fs_info, |
6747 | "unexpected item type %u in sys_array at offset %u", | |
6748 | (u32)key.type, cur_offset); | |
84eed90f CM |
6749 | ret = -EIO; |
6750 | break; | |
0b86a832 | 6751 | } |
1ffb22cf DS |
6752 | array_ptr += len; |
6753 | sb_array_offset += len; | |
6754 | cur_offset += len; | |
0b86a832 | 6755 | } |
d865177a | 6756 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6757 | free_extent_buffer_stale(sb); |
84eed90f | 6758 | return ret; |
e3540eab DS |
6759 | |
6760 | out_short_read: | |
ab8d0fc4 | 6761 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6762 | len, cur_offset); |
d865177a | 6763 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6764 | free_extent_buffer_stale(sb); |
e3540eab | 6765 | return -EIO; |
0b86a832 CM |
6766 | } |
6767 | ||
6768 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6769 | { | |
6770 | struct btrfs_path *path; | |
6771 | struct extent_buffer *leaf; | |
6772 | struct btrfs_key key; | |
6773 | struct btrfs_key found_key; | |
6774 | int ret; | |
6775 | int slot; | |
99e3ecfc | 6776 | u64 total_dev = 0; |
0b86a832 CM |
6777 | |
6778 | root = root->fs_info->chunk_root; | |
6779 | ||
6780 | path = btrfs_alloc_path(); | |
6781 | if (!path) | |
6782 | return -ENOMEM; | |
6783 | ||
b367e47f LZ |
6784 | mutex_lock(&uuid_mutex); |
6785 | lock_chunks(root); | |
6786 | ||
395927a9 FDBM |
6787 | /* |
6788 | * Read all device items, and then all the chunk items. All | |
6789 | * device items are found before any chunk item (their object id | |
6790 | * is smaller than the lowest possible object id for a chunk | |
6791 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6792 | */ |
6793 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6794 | key.offset = 0; | |
6795 | key.type = 0; | |
0b86a832 | 6796 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6797 | if (ret < 0) |
6798 | goto error; | |
d397712b | 6799 | while (1) { |
0b86a832 CM |
6800 | leaf = path->nodes[0]; |
6801 | slot = path->slots[0]; | |
6802 | if (slot >= btrfs_header_nritems(leaf)) { | |
6803 | ret = btrfs_next_leaf(root, path); | |
6804 | if (ret == 0) | |
6805 | continue; | |
6806 | if (ret < 0) | |
6807 | goto error; | |
6808 | break; | |
6809 | } | |
6810 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6811 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6812 | struct btrfs_dev_item *dev_item; | |
6813 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6814 | struct btrfs_dev_item); |
395927a9 FDBM |
6815 | ret = read_one_dev(root, leaf, dev_item); |
6816 | if (ret) | |
6817 | goto error; | |
99e3ecfc | 6818 | total_dev++; |
0b86a832 CM |
6819 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6820 | struct btrfs_chunk *chunk; | |
6821 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6822 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6823 | if (ret) |
6824 | goto error; | |
0b86a832 CM |
6825 | } |
6826 | path->slots[0]++; | |
6827 | } | |
99e3ecfc LB |
6828 | |
6829 | /* | |
6830 | * After loading chunk tree, we've got all device information, | |
6831 | * do another round of validation checks. | |
6832 | */ | |
6833 | if (total_dev != root->fs_info->fs_devices->total_devices) { | |
6834 | btrfs_err(root->fs_info, | |
6835 | "super_num_devices %llu mismatch with num_devices %llu found here", | |
6836 | btrfs_super_num_devices(root->fs_info->super_copy), | |
6837 | total_dev); | |
6838 | ret = -EINVAL; | |
6839 | goto error; | |
6840 | } | |
6841 | if (btrfs_super_total_bytes(root->fs_info->super_copy) < | |
6842 | root->fs_info->fs_devices->total_rw_bytes) { | |
6843 | btrfs_err(root->fs_info, | |
6844 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", | |
6845 | btrfs_super_total_bytes(root->fs_info->super_copy), | |
6846 | root->fs_info->fs_devices->total_rw_bytes); | |
6847 | ret = -EINVAL; | |
6848 | goto error; | |
6849 | } | |
0b86a832 CM |
6850 | ret = 0; |
6851 | error: | |
b367e47f LZ |
6852 | unlock_chunks(root); |
6853 | mutex_unlock(&uuid_mutex); | |
6854 | ||
2b82032c | 6855 | btrfs_free_path(path); |
0b86a832 CM |
6856 | return ret; |
6857 | } | |
442a4f63 | 6858 | |
cb517eab MX |
6859 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6860 | { | |
6861 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6862 | struct btrfs_device *device; | |
6863 | ||
29cc83f6 LB |
6864 | while (fs_devices) { |
6865 | mutex_lock(&fs_devices->device_list_mutex); | |
6866 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6867 | device->dev_root = fs_info->dev_root; | |
6868 | mutex_unlock(&fs_devices->device_list_mutex); | |
6869 | ||
6870 | fs_devices = fs_devices->seed; | |
6871 | } | |
cb517eab MX |
6872 | } |
6873 | ||
733f4fbb SB |
6874 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6875 | { | |
6876 | int i; | |
6877 | ||
6878 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6879 | btrfs_dev_stat_reset(dev, i); | |
6880 | } | |
6881 | ||
6882 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6883 | { | |
6884 | struct btrfs_key key; | |
6885 | struct btrfs_key found_key; | |
6886 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6887 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6888 | struct extent_buffer *eb; | |
6889 | int slot; | |
6890 | int ret = 0; | |
6891 | struct btrfs_device *device; | |
6892 | struct btrfs_path *path = NULL; | |
6893 | int i; | |
6894 | ||
6895 | path = btrfs_alloc_path(); | |
6896 | if (!path) { | |
6897 | ret = -ENOMEM; | |
6898 | goto out; | |
6899 | } | |
6900 | ||
6901 | mutex_lock(&fs_devices->device_list_mutex); | |
6902 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6903 | int item_size; | |
6904 | struct btrfs_dev_stats_item *ptr; | |
6905 | ||
242e2956 DS |
6906 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6907 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6908 | key.offset = device->devid; |
6909 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6910 | if (ret) { | |
733f4fbb SB |
6911 | __btrfs_reset_dev_stats(device); |
6912 | device->dev_stats_valid = 1; | |
6913 | btrfs_release_path(path); | |
6914 | continue; | |
6915 | } | |
6916 | slot = path->slots[0]; | |
6917 | eb = path->nodes[0]; | |
6918 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6919 | item_size = btrfs_item_size_nr(eb, slot); | |
6920 | ||
6921 | ptr = btrfs_item_ptr(eb, slot, | |
6922 | struct btrfs_dev_stats_item); | |
6923 | ||
6924 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6925 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6926 | btrfs_dev_stat_set(device, i, | |
6927 | btrfs_dev_stats_value(eb, ptr, i)); | |
6928 | else | |
6929 | btrfs_dev_stat_reset(device, i); | |
6930 | } | |
6931 | ||
6932 | device->dev_stats_valid = 1; | |
6933 | btrfs_dev_stat_print_on_load(device); | |
6934 | btrfs_release_path(path); | |
6935 | } | |
6936 | mutex_unlock(&fs_devices->device_list_mutex); | |
6937 | ||
6938 | out: | |
6939 | btrfs_free_path(path); | |
6940 | return ret < 0 ? ret : 0; | |
6941 | } | |
6942 | ||
6943 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6944 | struct btrfs_root *dev_root, | |
6945 | struct btrfs_device *device) | |
6946 | { | |
6947 | struct btrfs_path *path; | |
6948 | struct btrfs_key key; | |
6949 | struct extent_buffer *eb; | |
6950 | struct btrfs_dev_stats_item *ptr; | |
6951 | int ret; | |
6952 | int i; | |
6953 | ||
242e2956 DS |
6954 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6955 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6956 | key.offset = device->devid; |
6957 | ||
6958 | path = btrfs_alloc_path(); | |
6959 | BUG_ON(!path); | |
6960 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6961 | if (ret < 0) { | |
ecaeb14b DS |
6962 | btrfs_warn_in_rcu(dev_root->fs_info, |
6963 | "error %d while searching for dev_stats item for device %s", | |
606686ee | 6964 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6965 | goto out; |
6966 | } | |
6967 | ||
6968 | if (ret == 0 && | |
6969 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6970 | /* need to delete old one and insert a new one */ | |
6971 | ret = btrfs_del_item(trans, dev_root, path); | |
6972 | if (ret != 0) { | |
ecaeb14b DS |
6973 | btrfs_warn_in_rcu(dev_root->fs_info, |
6974 | "delete too small dev_stats item for device %s failed %d", | |
606686ee | 6975 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6976 | goto out; |
6977 | } | |
6978 | ret = 1; | |
6979 | } | |
6980 | ||
6981 | if (ret == 1) { | |
6982 | /* need to insert a new item */ | |
6983 | btrfs_release_path(path); | |
6984 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6985 | &key, sizeof(*ptr)); | |
6986 | if (ret < 0) { | |
ecaeb14b DS |
6987 | btrfs_warn_in_rcu(dev_root->fs_info, |
6988 | "insert dev_stats item for device %s failed %d", | |
6989 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
6990 | goto out; |
6991 | } | |
6992 | } | |
6993 | ||
6994 | eb = path->nodes[0]; | |
6995 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6996 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6997 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6998 | btrfs_dev_stat_read(device, i)); | |
6999 | btrfs_mark_buffer_dirty(eb); | |
7000 | ||
7001 | out: | |
7002 | btrfs_free_path(path); | |
7003 | return ret; | |
7004 | } | |
7005 | ||
7006 | /* | |
7007 | * called from commit_transaction. Writes all changed device stats to disk. | |
7008 | */ | |
7009 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7010 | struct btrfs_fs_info *fs_info) | |
7011 | { | |
7012 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7013 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7014 | struct btrfs_device *device; | |
addc3fa7 | 7015 | int stats_cnt; |
733f4fbb SB |
7016 | int ret = 0; |
7017 | ||
7018 | mutex_lock(&fs_devices->device_list_mutex); | |
7019 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 7020 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
7021 | continue; |
7022 | ||
addc3fa7 | 7023 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
733f4fbb SB |
7024 | ret = update_dev_stat_item(trans, dev_root, device); |
7025 | if (!ret) | |
addc3fa7 | 7026 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7027 | } |
7028 | mutex_unlock(&fs_devices->device_list_mutex); | |
7029 | ||
7030 | return ret; | |
7031 | } | |
7032 | ||
442a4f63 SB |
7033 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7034 | { | |
7035 | btrfs_dev_stat_inc(dev, index); | |
7036 | btrfs_dev_stat_print_on_error(dev); | |
7037 | } | |
7038 | ||
48a3b636 | 7039 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7040 | { |
733f4fbb SB |
7041 | if (!dev->dev_stats_valid) |
7042 | return; | |
b14af3b4 DS |
7043 | btrfs_err_rl_in_rcu(dev->dev_root->fs_info, |
7044 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
606686ee | 7045 | rcu_str_deref(dev->name), |
442a4f63 SB |
7046 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7047 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7048 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7049 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7050 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7051 | } |
c11d2c23 | 7052 | |
733f4fbb SB |
7053 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7054 | { | |
a98cdb85 SB |
7055 | int i; |
7056 | ||
7057 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7058 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7059 | break; | |
7060 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7061 | return; /* all values == 0, suppress message */ | |
7062 | ||
ecaeb14b DS |
7063 | btrfs_info_in_rcu(dev->dev_root->fs_info, |
7064 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
606686ee | 7065 | rcu_str_deref(dev->name), |
733f4fbb SB |
7066 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7067 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7068 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7069 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7070 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7071 | } | |
7072 | ||
c11d2c23 | 7073 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 7074 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7075 | { |
7076 | struct btrfs_device *dev; | |
7077 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7078 | int i; | |
7079 | ||
7080 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 7081 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7082 | mutex_unlock(&fs_devices->device_list_mutex); |
7083 | ||
7084 | if (!dev) { | |
5d163e0e JM |
7085 | btrfs_warn(root->fs_info, |
7086 | "get dev_stats failed, device not found"); | |
c11d2c23 | 7087 | return -ENODEV; |
733f4fbb | 7088 | } else if (!dev->dev_stats_valid) { |
5d163e0e JM |
7089 | btrfs_warn(root->fs_info, |
7090 | "get dev_stats failed, not yet valid"); | |
733f4fbb | 7091 | return -ENODEV; |
b27f7c0c | 7092 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7093 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7094 | if (stats->nr_items > i) | |
7095 | stats->values[i] = | |
7096 | btrfs_dev_stat_read_and_reset(dev, i); | |
7097 | else | |
7098 | btrfs_dev_stat_reset(dev, i); | |
7099 | } | |
7100 | } else { | |
7101 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7102 | if (stats->nr_items > i) | |
7103 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7104 | } | |
7105 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7106 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7107 | return 0; | |
7108 | } | |
a8a6dab7 | 7109 | |
12b1c263 | 7110 | void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path) |
a8a6dab7 SB |
7111 | { |
7112 | struct buffer_head *bh; | |
7113 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7114 | int copy_num; |
a8a6dab7 | 7115 | |
12b1c263 AJ |
7116 | if (!bdev) |
7117 | return; | |
a8a6dab7 | 7118 | |
12b1c263 AJ |
7119 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7120 | copy_num++) { | |
a8a6dab7 | 7121 | |
12b1c263 AJ |
7122 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7123 | continue; | |
7124 | ||
7125 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7126 | ||
7127 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7128 | set_buffer_dirty(bh); | |
7129 | sync_dirty_buffer(bh); | |
7130 | brelse(bh); | |
7131 | } | |
7132 | ||
7133 | /* Notify udev that device has changed */ | |
7134 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7135 | ||
7136 | /* Update ctime/mtime for device path for libblkid */ | |
7137 | update_dev_time(device_path); | |
a8a6dab7 | 7138 | } |
935e5cc9 MX |
7139 | |
7140 | /* | |
7141 | * Update the size of all devices, which is used for writing out the | |
7142 | * super blocks. | |
7143 | */ | |
7144 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7145 | { | |
7146 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7147 | struct btrfs_device *curr, *next; | |
7148 | ||
7149 | if (list_empty(&fs_devices->resized_devices)) | |
7150 | return; | |
7151 | ||
7152 | mutex_lock(&fs_devices->device_list_mutex); | |
7153 | lock_chunks(fs_info->dev_root); | |
7154 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
7155 | resized_list) { | |
7156 | list_del_init(&curr->resized_list); | |
7157 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7158 | } | |
7159 | unlock_chunks(fs_info->dev_root); | |
7160 | mutex_unlock(&fs_devices->device_list_mutex); | |
7161 | } | |
ce7213c7 MX |
7162 | |
7163 | /* Must be invoked during the transaction commit */ | |
7164 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
7165 | struct btrfs_transaction *transaction) | |
7166 | { | |
7167 | struct extent_map *em; | |
7168 | struct map_lookup *map; | |
7169 | struct btrfs_device *dev; | |
7170 | int i; | |
7171 | ||
7172 | if (list_empty(&transaction->pending_chunks)) | |
7173 | return; | |
7174 | ||
7175 | /* In order to kick the device replace finish process */ | |
7176 | lock_chunks(root); | |
7177 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
95617d69 | 7178 | map = em->map_lookup; |
ce7213c7 MX |
7179 | |
7180 | for (i = 0; i < map->num_stripes; i++) { | |
7181 | dev = map->stripes[i].dev; | |
7182 | dev->commit_bytes_used = dev->bytes_used; | |
7183 | } | |
7184 | } | |
7185 | unlock_chunks(root); | |
7186 | } | |
5a13f430 AJ |
7187 | |
7188 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7189 | { | |
7190 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7191 | while (fs_devices) { | |
7192 | fs_devices->fs_info = fs_info; | |
7193 | fs_devices = fs_devices->seed; | |
7194 | } | |
7195 | } | |
7196 | ||
7197 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7198 | { | |
7199 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7200 | while (fs_devices) { | |
7201 | fs_devices->fs_info = NULL; | |
7202 | fs_devices = fs_devices->seed; | |
7203 | } | |
7204 | } |