Commit | Line | Data |
---|---|---|
0b86a832 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8a4b83cc | 21 | #include <linux/buffer_head.h> |
f2d8d74d | 22 | #include <linux/blkdev.h> |
788f20eb | 23 | #include <linux/random.h> |
b765ead5 | 24 | #include <linux/iocontext.h> |
6f88a440 | 25 | #include <linux/capability.h> |
442a4f63 | 26 | #include <linux/ratelimit.h> |
59641015 | 27 | #include <linux/kthread.h> |
53b381b3 | 28 | #include <linux/raid/pq.h> |
803b2f54 | 29 | #include <linux/semaphore.h> |
53b381b3 | 30 | #include <asm/div64.h> |
0b86a832 CM |
31 | #include "ctree.h" |
32 | #include "extent_map.h" | |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "print-tree.h" | |
36 | #include "volumes.h" | |
53b381b3 | 37 | #include "raid56.h" |
8b712842 | 38 | #include "async-thread.h" |
21adbd5c | 39 | #include "check-integrity.h" |
606686ee | 40 | #include "rcu-string.h" |
3fed40cc | 41 | #include "math.h" |
8dabb742 | 42 | #include "dev-replace.h" |
99994cde | 43 | #include "sysfs.h" |
0b86a832 | 44 | |
af902047 ZL |
45 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
46 | [BTRFS_RAID_RAID10] = { | |
47 | .sub_stripes = 2, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 0, /* 0 == as many as possible */ | |
50 | .devs_min = 4, | |
8789f4fe | 51 | .tolerated_failures = 1, |
af902047 ZL |
52 | .devs_increment = 2, |
53 | .ncopies = 2, | |
54 | }, | |
55 | [BTRFS_RAID_RAID1] = { | |
56 | .sub_stripes = 1, | |
57 | .dev_stripes = 1, | |
58 | .devs_max = 2, | |
59 | .devs_min = 2, | |
8789f4fe | 60 | .tolerated_failures = 1, |
af902047 ZL |
61 | .devs_increment = 2, |
62 | .ncopies = 2, | |
63 | }, | |
64 | [BTRFS_RAID_DUP] = { | |
65 | .sub_stripes = 1, | |
66 | .dev_stripes = 2, | |
67 | .devs_max = 1, | |
68 | .devs_min = 1, | |
8789f4fe | 69 | .tolerated_failures = 0, |
af902047 ZL |
70 | .devs_increment = 1, |
71 | .ncopies = 2, | |
72 | }, | |
73 | [BTRFS_RAID_RAID0] = { | |
74 | .sub_stripes = 1, | |
75 | .dev_stripes = 1, | |
76 | .devs_max = 0, | |
77 | .devs_min = 2, | |
8789f4fe | 78 | .tolerated_failures = 0, |
af902047 ZL |
79 | .devs_increment = 1, |
80 | .ncopies = 1, | |
81 | }, | |
82 | [BTRFS_RAID_SINGLE] = { | |
83 | .sub_stripes = 1, | |
84 | .dev_stripes = 1, | |
85 | .devs_max = 1, | |
86 | .devs_min = 1, | |
8789f4fe | 87 | .tolerated_failures = 0, |
af902047 ZL |
88 | .devs_increment = 1, |
89 | .ncopies = 1, | |
90 | }, | |
91 | [BTRFS_RAID_RAID5] = { | |
92 | .sub_stripes = 1, | |
93 | .dev_stripes = 1, | |
94 | .devs_max = 0, | |
95 | .devs_min = 2, | |
8789f4fe | 96 | .tolerated_failures = 1, |
af902047 ZL |
97 | .devs_increment = 1, |
98 | .ncopies = 2, | |
99 | }, | |
100 | [BTRFS_RAID_RAID6] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 3, | |
8789f4fe | 105 | .tolerated_failures = 2, |
af902047 ZL |
106 | .devs_increment = 1, |
107 | .ncopies = 3, | |
108 | }, | |
109 | }; | |
110 | ||
111 | const u64 const btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { | |
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 | ||
2b82032c YZ |
121 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
122 | struct btrfs_root *root, | |
123 | struct btrfs_device *device); | |
124 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 125 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 126 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 127 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
87ad58c5 | 128 | static void btrfs_close_one_device(struct btrfs_device *device); |
2b82032c | 129 | |
67a2c45e | 130 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 131 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
132 | struct list_head *btrfs_get_fs_uuids(void) |
133 | { | |
134 | return &fs_uuids; | |
135 | } | |
8a4b83cc | 136 | |
2208a378 ID |
137 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
138 | { | |
139 | struct btrfs_fs_devices *fs_devs; | |
140 | ||
141 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS); | |
142 | if (!fs_devs) | |
143 | return ERR_PTR(-ENOMEM); | |
144 | ||
145 | mutex_init(&fs_devs->device_list_mutex); | |
146 | ||
147 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 148 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
149 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
150 | INIT_LIST_HEAD(&fs_devs->list); | |
151 | ||
152 | return fs_devs; | |
153 | } | |
154 | ||
155 | /** | |
156 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
157 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
158 | * generated. | |
159 | * | |
160 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
161 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
162 | * can be destroyed with kfree() right away. | |
163 | */ | |
164 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
165 | { | |
166 | struct btrfs_fs_devices *fs_devs; | |
167 | ||
168 | fs_devs = __alloc_fs_devices(); | |
169 | if (IS_ERR(fs_devs)) | |
170 | return fs_devs; | |
171 | ||
172 | if (fsid) | |
173 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
174 | else | |
175 | generate_random_uuid(fs_devs->fsid); | |
176 | ||
177 | return fs_devs; | |
178 | } | |
179 | ||
e4404d6e YZ |
180 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
181 | { | |
182 | struct btrfs_device *device; | |
183 | WARN_ON(fs_devices->opened); | |
184 | while (!list_empty(&fs_devices->devices)) { | |
185 | device = list_entry(fs_devices->devices.next, | |
186 | struct btrfs_device, dev_list); | |
187 | list_del(&device->dev_list); | |
606686ee | 188 | rcu_string_free(device->name); |
e4404d6e YZ |
189 | kfree(device); |
190 | } | |
191 | kfree(fs_devices); | |
192 | } | |
193 | ||
b8b8ff59 LC |
194 | static void btrfs_kobject_uevent(struct block_device *bdev, |
195 | enum kobject_action action) | |
196 | { | |
197 | int ret; | |
198 | ||
199 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
200 | if (ret) | |
efe120a0 | 201 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
202 | action, |
203 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
204 | &disk_to_dev(bdev->bd_disk)->kobj); | |
205 | } | |
206 | ||
143bede5 | 207 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
208 | { |
209 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 210 | |
2b82032c YZ |
211 | while (!list_empty(&fs_uuids)) { |
212 | fs_devices = list_entry(fs_uuids.next, | |
213 | struct btrfs_fs_devices, list); | |
214 | list_del(&fs_devices->list); | |
e4404d6e | 215 | free_fs_devices(fs_devices); |
8a4b83cc | 216 | } |
8a4b83cc CM |
217 | } |
218 | ||
12bd2fc0 ID |
219 | static struct btrfs_device *__alloc_device(void) |
220 | { | |
221 | struct btrfs_device *dev; | |
222 | ||
223 | dev = kzalloc(sizeof(*dev), GFP_NOFS); | |
224 | if (!dev) | |
225 | return ERR_PTR(-ENOMEM); | |
226 | ||
227 | INIT_LIST_HEAD(&dev->dev_list); | |
228 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 229 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
230 | |
231 | spin_lock_init(&dev->io_lock); | |
232 | ||
233 | spin_lock_init(&dev->reada_lock); | |
234 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 235 | atomic_set(&dev->dev_stats_ccnt, 0); |
d0164adc MG |
236 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
237 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
12bd2fc0 ID |
238 | |
239 | return dev; | |
240 | } | |
241 | ||
a1b32a59 CM |
242 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
243 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
244 | { |
245 | struct btrfs_device *dev; | |
8a4b83cc | 246 | |
c6e30871 | 247 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 248 | if (dev->devid == devid && |
8f18cf13 | 249 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 250 | return dev; |
a443755f | 251 | } |
8a4b83cc CM |
252 | } |
253 | return NULL; | |
254 | } | |
255 | ||
a1b32a59 | 256 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 257 | { |
8a4b83cc CM |
258 | struct btrfs_fs_devices *fs_devices; |
259 | ||
c6e30871 | 260 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
261 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
262 | return fs_devices; | |
263 | } | |
264 | return NULL; | |
265 | } | |
266 | ||
beaf8ab3 SB |
267 | static int |
268 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
269 | int flush, struct block_device **bdev, | |
270 | struct buffer_head **bh) | |
271 | { | |
272 | int ret; | |
273 | ||
274 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
275 | ||
276 | if (IS_ERR(*bdev)) { | |
277 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
278 | goto error; |
279 | } | |
280 | ||
281 | if (flush) | |
282 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
283 | ret = set_blocksize(*bdev, 4096); | |
284 | if (ret) { | |
285 | blkdev_put(*bdev, flags); | |
286 | goto error; | |
287 | } | |
288 | invalidate_bdev(*bdev); | |
289 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
290 | if (IS_ERR(*bh)) { |
291 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
292 | blkdev_put(*bdev, flags); |
293 | goto error; | |
294 | } | |
295 | ||
296 | return 0; | |
297 | ||
298 | error: | |
299 | *bdev = NULL; | |
300 | *bh = NULL; | |
301 | return ret; | |
302 | } | |
303 | ||
ffbd517d CM |
304 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
305 | struct bio *head, struct bio *tail) | |
306 | { | |
307 | ||
308 | struct bio *old_head; | |
309 | ||
310 | old_head = pending_bios->head; | |
311 | pending_bios->head = head; | |
312 | if (pending_bios->tail) | |
313 | tail->bi_next = old_head; | |
314 | else | |
315 | pending_bios->tail = tail; | |
316 | } | |
317 | ||
8b712842 CM |
318 | /* |
319 | * we try to collect pending bios for a device so we don't get a large | |
320 | * number of procs sending bios down to the same device. This greatly | |
321 | * improves the schedulers ability to collect and merge the bios. | |
322 | * | |
323 | * But, it also turns into a long list of bios to process and that is sure | |
324 | * to eventually make the worker thread block. The solution here is to | |
325 | * make some progress and then put this work struct back at the end of | |
326 | * the list if the block device is congested. This way, multiple devices | |
327 | * can make progress from a single worker thread. | |
328 | */ | |
143bede5 | 329 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
330 | { |
331 | struct bio *pending; | |
332 | struct backing_dev_info *bdi; | |
b64a2851 | 333 | struct btrfs_fs_info *fs_info; |
ffbd517d | 334 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
335 | struct bio *tail; |
336 | struct bio *cur; | |
337 | int again = 0; | |
ffbd517d | 338 | unsigned long num_run; |
d644d8a1 | 339 | unsigned long batch_run = 0; |
b64a2851 | 340 | unsigned long limit; |
b765ead5 | 341 | unsigned long last_waited = 0; |
d84275c9 | 342 | int force_reg = 0; |
0e588859 | 343 | int sync_pending = 0; |
211588ad CM |
344 | struct blk_plug plug; |
345 | ||
346 | /* | |
347 | * this function runs all the bios we've collected for | |
348 | * a particular device. We don't want to wander off to | |
349 | * another device without first sending all of these down. | |
350 | * So, setup a plug here and finish it off before we return | |
351 | */ | |
352 | blk_start_plug(&plug); | |
8b712842 | 353 | |
bedf762b | 354 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
355 | fs_info = device->dev_root->fs_info; |
356 | limit = btrfs_async_submit_limit(fs_info); | |
357 | limit = limit * 2 / 3; | |
358 | ||
8b712842 CM |
359 | loop: |
360 | spin_lock(&device->io_lock); | |
361 | ||
a6837051 | 362 | loop_lock: |
d84275c9 | 363 | num_run = 0; |
ffbd517d | 364 | |
8b712842 CM |
365 | /* take all the bios off the list at once and process them |
366 | * later on (without the lock held). But, remember the | |
367 | * tail and other pointers so the bios can be properly reinserted | |
368 | * into the list if we hit congestion | |
369 | */ | |
d84275c9 | 370 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 371 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
372 | force_reg = 1; |
373 | } else { | |
ffbd517d | 374 | pending_bios = &device->pending_bios; |
d84275c9 CM |
375 | force_reg = 0; |
376 | } | |
ffbd517d CM |
377 | |
378 | pending = pending_bios->head; | |
379 | tail = pending_bios->tail; | |
8b712842 | 380 | WARN_ON(pending && !tail); |
8b712842 CM |
381 | |
382 | /* | |
383 | * if pending was null this time around, no bios need processing | |
384 | * at all and we can stop. Otherwise it'll loop back up again | |
385 | * and do an additional check so no bios are missed. | |
386 | * | |
387 | * device->running_pending is used to synchronize with the | |
388 | * schedule_bio code. | |
389 | */ | |
ffbd517d CM |
390 | if (device->pending_sync_bios.head == NULL && |
391 | device->pending_bios.head == NULL) { | |
8b712842 CM |
392 | again = 0; |
393 | device->running_pending = 0; | |
ffbd517d CM |
394 | } else { |
395 | again = 1; | |
396 | device->running_pending = 1; | |
8b712842 | 397 | } |
ffbd517d CM |
398 | |
399 | pending_bios->head = NULL; | |
400 | pending_bios->tail = NULL; | |
401 | ||
8b712842 CM |
402 | spin_unlock(&device->io_lock); |
403 | ||
d397712b | 404 | while (pending) { |
ffbd517d CM |
405 | |
406 | rmb(); | |
d84275c9 CM |
407 | /* we want to work on both lists, but do more bios on the |
408 | * sync list than the regular list | |
409 | */ | |
410 | if ((num_run > 32 && | |
411 | pending_bios != &device->pending_sync_bios && | |
412 | device->pending_sync_bios.head) || | |
413 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
414 | device->pending_bios.head)) { | |
ffbd517d CM |
415 | spin_lock(&device->io_lock); |
416 | requeue_list(pending_bios, pending, tail); | |
417 | goto loop_lock; | |
418 | } | |
419 | ||
8b712842 CM |
420 | cur = pending; |
421 | pending = pending->bi_next; | |
422 | cur->bi_next = NULL; | |
b64a2851 | 423 | |
ee863954 DS |
424 | /* |
425 | * atomic_dec_return implies a barrier for waitqueue_active | |
426 | */ | |
66657b31 | 427 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
428 | waitqueue_active(&fs_info->async_submit_wait)) |
429 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 430 | |
dac56212 | 431 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 432 | |
2ab1ba68 CM |
433 | /* |
434 | * if we're doing the sync list, record that our | |
435 | * plug has some sync requests on it | |
436 | * | |
437 | * If we're doing the regular list and there are | |
438 | * sync requests sitting around, unplug before | |
439 | * we add more | |
440 | */ | |
441 | if (pending_bios == &device->pending_sync_bios) { | |
442 | sync_pending = 1; | |
443 | } else if (sync_pending) { | |
444 | blk_finish_plug(&plug); | |
445 | blk_start_plug(&plug); | |
446 | sync_pending = 0; | |
447 | } | |
448 | ||
21adbd5c | 449 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
450 | num_run++; |
451 | batch_run++; | |
853d8ec4 DS |
452 | |
453 | cond_resched(); | |
8b712842 CM |
454 | |
455 | /* | |
456 | * we made progress, there is more work to do and the bdi | |
457 | * is now congested. Back off and let other work structs | |
458 | * run instead | |
459 | */ | |
57fd5a5f | 460 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 461 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 462 | struct io_context *ioc; |
8b712842 | 463 | |
b765ead5 CM |
464 | ioc = current->io_context; |
465 | ||
466 | /* | |
467 | * the main goal here is that we don't want to | |
468 | * block if we're going to be able to submit | |
469 | * more requests without blocking. | |
470 | * | |
471 | * This code does two great things, it pokes into | |
472 | * the elevator code from a filesystem _and_ | |
473 | * it makes assumptions about how batching works. | |
474 | */ | |
475 | if (ioc && ioc->nr_batch_requests > 0 && | |
476 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
477 | (last_waited == 0 || | |
478 | ioc->last_waited == last_waited)) { | |
479 | /* | |
480 | * we want to go through our batch of | |
481 | * requests and stop. So, we copy out | |
482 | * the ioc->last_waited time and test | |
483 | * against it before looping | |
484 | */ | |
485 | last_waited = ioc->last_waited; | |
853d8ec4 | 486 | cond_resched(); |
b765ead5 CM |
487 | continue; |
488 | } | |
8b712842 | 489 | spin_lock(&device->io_lock); |
ffbd517d | 490 | requeue_list(pending_bios, pending, tail); |
a6837051 | 491 | device->running_pending = 1; |
8b712842 CM |
492 | |
493 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
494 | btrfs_queue_work(fs_info->submit_workers, |
495 | &device->work); | |
8b712842 CM |
496 | goto done; |
497 | } | |
d85c8a6f CM |
498 | /* unplug every 64 requests just for good measure */ |
499 | if (batch_run % 64 == 0) { | |
500 | blk_finish_plug(&plug); | |
501 | blk_start_plug(&plug); | |
502 | sync_pending = 0; | |
503 | } | |
8b712842 | 504 | } |
ffbd517d | 505 | |
51684082 CM |
506 | cond_resched(); |
507 | if (again) | |
508 | goto loop; | |
509 | ||
510 | spin_lock(&device->io_lock); | |
511 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
512 | goto loop_lock; | |
513 | spin_unlock(&device->io_lock); | |
514 | ||
8b712842 | 515 | done: |
211588ad | 516 | blk_finish_plug(&plug); |
8b712842 CM |
517 | } |
518 | ||
b2950863 | 519 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
520 | { |
521 | struct btrfs_device *device; | |
522 | ||
523 | device = container_of(work, struct btrfs_device, work); | |
524 | run_scheduled_bios(device); | |
525 | } | |
526 | ||
4fde46f0 AJ |
527 | |
528 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
529 | { | |
530 | struct btrfs_fs_devices *fs_devs; | |
531 | struct btrfs_device *dev; | |
532 | ||
533 | if (!cur_dev->name) | |
534 | return; | |
535 | ||
536 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
537 | int del = 1; | |
538 | ||
539 | if (fs_devs->opened) | |
540 | continue; | |
541 | if (fs_devs->seeding) | |
542 | continue; | |
543 | ||
544 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
545 | ||
546 | if (dev == cur_dev) | |
547 | continue; | |
548 | if (!dev->name) | |
549 | continue; | |
550 | ||
551 | /* | |
552 | * Todo: This won't be enough. What if the same device | |
553 | * comes back (with new uuid and) with its mapper path? | |
554 | * But for now, this does help as mostly an admin will | |
555 | * either use mapper or non mapper path throughout. | |
556 | */ | |
557 | rcu_read_lock(); | |
558 | del = strcmp(rcu_str_deref(dev->name), | |
559 | rcu_str_deref(cur_dev->name)); | |
560 | rcu_read_unlock(); | |
561 | if (!del) | |
562 | break; | |
563 | } | |
564 | ||
565 | if (!del) { | |
566 | /* delete the stale device */ | |
567 | if (fs_devs->num_devices == 1) { | |
568 | btrfs_sysfs_remove_fsid(fs_devs); | |
569 | list_del(&fs_devs->list); | |
570 | free_fs_devices(fs_devs); | |
571 | } else { | |
572 | fs_devs->num_devices--; | |
573 | list_del(&dev->dev_list); | |
574 | rcu_string_free(dev->name); | |
575 | kfree(dev); | |
576 | } | |
577 | break; | |
578 | } | |
579 | } | |
580 | } | |
581 | ||
60999ca4 DS |
582 | /* |
583 | * Add new device to list of registered devices | |
584 | * | |
585 | * Returns: | |
586 | * 1 - first time device is seen | |
587 | * 0 - device already known | |
588 | * < 0 - error | |
589 | */ | |
a1b32a59 | 590 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
591 | struct btrfs_super_block *disk_super, |
592 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
593 | { | |
594 | struct btrfs_device *device; | |
595 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 596 | struct rcu_string *name; |
60999ca4 | 597 | int ret = 0; |
8a4b83cc CM |
598 | u64 found_transid = btrfs_super_generation(disk_super); |
599 | ||
600 | fs_devices = find_fsid(disk_super->fsid); | |
601 | if (!fs_devices) { | |
2208a378 ID |
602 | fs_devices = alloc_fs_devices(disk_super->fsid); |
603 | if (IS_ERR(fs_devices)) | |
604 | return PTR_ERR(fs_devices); | |
605 | ||
8a4b83cc | 606 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 607 | |
8a4b83cc CM |
608 | device = NULL; |
609 | } else { | |
a443755f CM |
610 | device = __find_device(&fs_devices->devices, devid, |
611 | disk_super->dev_item.uuid); | |
8a4b83cc | 612 | } |
443f24fe | 613 | |
8a4b83cc | 614 | if (!device) { |
2b82032c YZ |
615 | if (fs_devices->opened) |
616 | return -EBUSY; | |
617 | ||
12bd2fc0 ID |
618 | device = btrfs_alloc_device(NULL, &devid, |
619 | disk_super->dev_item.uuid); | |
620 | if (IS_ERR(device)) { | |
8a4b83cc | 621 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 622 | return PTR_ERR(device); |
8a4b83cc | 623 | } |
606686ee JB |
624 | |
625 | name = rcu_string_strdup(path, GFP_NOFS); | |
626 | if (!name) { | |
8a4b83cc CM |
627 | kfree(device); |
628 | return -ENOMEM; | |
629 | } | |
606686ee | 630 | rcu_assign_pointer(device->name, name); |
90519d66 | 631 | |
e5e9a520 | 632 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 633 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 634 | fs_devices->num_devices++; |
e5e9a520 CM |
635 | mutex_unlock(&fs_devices->device_list_mutex); |
636 | ||
60999ca4 | 637 | ret = 1; |
2b82032c | 638 | device->fs_devices = fs_devices; |
606686ee | 639 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
640 | /* |
641 | * When FS is already mounted. | |
642 | * 1. If you are here and if the device->name is NULL that | |
643 | * means this device was missing at time of FS mount. | |
644 | * 2. If you are here and if the device->name is different | |
645 | * from 'path' that means either | |
646 | * a. The same device disappeared and reappeared with | |
647 | * different name. or | |
648 | * b. The missing-disk-which-was-replaced, has | |
649 | * reappeared now. | |
650 | * | |
651 | * We must allow 1 and 2a above. But 2b would be a spurious | |
652 | * and unintentional. | |
653 | * | |
654 | * Further in case of 1 and 2a above, the disk at 'path' | |
655 | * would have missed some transaction when it was away and | |
656 | * in case of 2a the stale bdev has to be updated as well. | |
657 | * 2b must not be allowed at all time. | |
658 | */ | |
659 | ||
660 | /* | |
0f23ae74 CM |
661 | * For now, we do allow update to btrfs_fs_device through the |
662 | * btrfs dev scan cli after FS has been mounted. We're still | |
663 | * tracking a problem where systems fail mount by subvolume id | |
664 | * when we reject replacement on a mounted FS. | |
b96de000 | 665 | */ |
0f23ae74 | 666 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
667 | /* |
668 | * That is if the FS is _not_ mounted and if you | |
669 | * are here, that means there is more than one | |
670 | * disk with same uuid and devid.We keep the one | |
671 | * with larger generation number or the last-in if | |
672 | * generation are equal. | |
673 | */ | |
0f23ae74 | 674 | return -EEXIST; |
77bdae4d | 675 | } |
b96de000 | 676 | |
606686ee | 677 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
678 | if (!name) |
679 | return -ENOMEM; | |
606686ee JB |
680 | rcu_string_free(device->name); |
681 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
682 | if (device->missing) { |
683 | fs_devices->missing_devices--; | |
684 | device->missing = 0; | |
685 | } | |
8a4b83cc CM |
686 | } |
687 | ||
77bdae4d AJ |
688 | /* |
689 | * Unmount does not free the btrfs_device struct but would zero | |
690 | * generation along with most of the other members. So just update | |
691 | * it back. We need it to pick the disk with largest generation | |
692 | * (as above). | |
693 | */ | |
694 | if (!fs_devices->opened) | |
695 | device->generation = found_transid; | |
696 | ||
4fde46f0 AJ |
697 | /* |
698 | * if there is new btrfs on an already registered device, | |
699 | * then remove the stale device entry. | |
700 | */ | |
701 | btrfs_free_stale_device(device); | |
702 | ||
8a4b83cc | 703 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
704 | |
705 | return ret; | |
8a4b83cc CM |
706 | } |
707 | ||
e4404d6e YZ |
708 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
709 | { | |
710 | struct btrfs_fs_devices *fs_devices; | |
711 | struct btrfs_device *device; | |
712 | struct btrfs_device *orig_dev; | |
713 | ||
2208a378 ID |
714 | fs_devices = alloc_fs_devices(orig->fsid); |
715 | if (IS_ERR(fs_devices)) | |
716 | return fs_devices; | |
e4404d6e | 717 | |
adbbb863 | 718 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 719 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 720 | |
46224705 | 721 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 722 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
723 | struct rcu_string *name; |
724 | ||
12bd2fc0 ID |
725 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
726 | orig_dev->uuid); | |
727 | if (IS_ERR(device)) | |
e4404d6e YZ |
728 | goto error; |
729 | ||
606686ee JB |
730 | /* |
731 | * This is ok to do without rcu read locked because we hold the | |
732 | * uuid mutex so nothing we touch in here is going to disappear. | |
733 | */ | |
e755f780 AJ |
734 | if (orig_dev->name) { |
735 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
736 | if (!name) { | |
737 | kfree(device); | |
738 | goto error; | |
739 | } | |
740 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 741 | } |
e4404d6e | 742 | |
e4404d6e YZ |
743 | list_add(&device->dev_list, &fs_devices->devices); |
744 | device->fs_devices = fs_devices; | |
745 | fs_devices->num_devices++; | |
746 | } | |
adbbb863 | 747 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
748 | return fs_devices; |
749 | error: | |
adbbb863 | 750 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
751 | free_fs_devices(fs_devices); |
752 | return ERR_PTR(-ENOMEM); | |
753 | } | |
754 | ||
9eaed21e | 755 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 756 | { |
c6e30871 | 757 | struct btrfs_device *device, *next; |
443f24fe | 758 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 759 | |
dfe25020 CM |
760 | mutex_lock(&uuid_mutex); |
761 | again: | |
46224705 | 762 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 763 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 764 | if (device->in_fs_metadata) { |
63a212ab | 765 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
766 | (!latest_dev || |
767 | device->generation > latest_dev->generation)) { | |
768 | latest_dev = device; | |
a6b0d5c8 | 769 | } |
2b82032c | 770 | continue; |
a6b0d5c8 | 771 | } |
2b82032c | 772 | |
8dabb742 SB |
773 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
774 | /* | |
775 | * In the first step, keep the device which has | |
776 | * the correct fsid and the devid that is used | |
777 | * for the dev_replace procedure. | |
778 | * In the second step, the dev_replace state is | |
779 | * read from the device tree and it is known | |
780 | * whether the procedure is really active or | |
781 | * not, which means whether this device is | |
782 | * used or whether it should be removed. | |
783 | */ | |
784 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
785 | continue; | |
786 | } | |
787 | } | |
2b82032c | 788 | if (device->bdev) { |
d4d77629 | 789 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
790 | device->bdev = NULL; |
791 | fs_devices->open_devices--; | |
792 | } | |
793 | if (device->writeable) { | |
794 | list_del_init(&device->dev_alloc_list); | |
795 | device->writeable = 0; | |
8dabb742 SB |
796 | if (!device->is_tgtdev_for_dev_replace) |
797 | fs_devices->rw_devices--; | |
2b82032c | 798 | } |
e4404d6e YZ |
799 | list_del_init(&device->dev_list); |
800 | fs_devices->num_devices--; | |
606686ee | 801 | rcu_string_free(device->name); |
e4404d6e | 802 | kfree(device); |
dfe25020 | 803 | } |
2b82032c YZ |
804 | |
805 | if (fs_devices->seed) { | |
806 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
807 | goto again; |
808 | } | |
809 | ||
443f24fe | 810 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 811 | |
dfe25020 | 812 | mutex_unlock(&uuid_mutex); |
dfe25020 | 813 | } |
a0af469b | 814 | |
1f78160c XG |
815 | static void __free_device(struct work_struct *work) |
816 | { | |
817 | struct btrfs_device *device; | |
818 | ||
819 | device = container_of(work, struct btrfs_device, rcu_work); | |
820 | ||
821 | if (device->bdev) | |
822 | blkdev_put(device->bdev, device->mode); | |
823 | ||
606686ee | 824 | rcu_string_free(device->name); |
1f78160c XG |
825 | kfree(device); |
826 | } | |
827 | ||
828 | static void free_device(struct rcu_head *head) | |
829 | { | |
830 | struct btrfs_device *device; | |
831 | ||
832 | device = container_of(head, struct btrfs_device, rcu); | |
833 | ||
834 | INIT_WORK(&device->rcu_work, __free_device); | |
835 | schedule_work(&device->rcu_work); | |
836 | } | |
837 | ||
2b82032c | 838 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 839 | { |
2037a093 | 840 | struct btrfs_device *device, *tmp; |
e4404d6e | 841 | |
2b82032c YZ |
842 | if (--fs_devices->opened > 0) |
843 | return 0; | |
8a4b83cc | 844 | |
c9513edb | 845 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 846 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
f190aa47 | 847 | btrfs_close_one_device(device); |
8a4b83cc | 848 | } |
c9513edb XG |
849 | mutex_unlock(&fs_devices->device_list_mutex); |
850 | ||
e4404d6e YZ |
851 | WARN_ON(fs_devices->open_devices); |
852 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
853 | fs_devices->opened = 0; |
854 | fs_devices->seeding = 0; | |
2b82032c | 855 | |
8a4b83cc CM |
856 | return 0; |
857 | } | |
858 | ||
2b82032c YZ |
859 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
860 | { | |
e4404d6e | 861 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
862 | int ret; |
863 | ||
864 | mutex_lock(&uuid_mutex); | |
865 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
866 | if (!fs_devices->opened) { |
867 | seed_devices = fs_devices->seed; | |
868 | fs_devices->seed = NULL; | |
869 | } | |
2b82032c | 870 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
871 | |
872 | while (seed_devices) { | |
873 | fs_devices = seed_devices; | |
874 | seed_devices = fs_devices->seed; | |
875 | __btrfs_close_devices(fs_devices); | |
876 | free_fs_devices(fs_devices); | |
877 | } | |
bc178622 ES |
878 | /* |
879 | * Wait for rcu kworkers under __btrfs_close_devices | |
880 | * to finish all blkdev_puts so device is really | |
881 | * free when umount is done. | |
882 | */ | |
883 | rcu_barrier(); | |
2b82032c YZ |
884 | return ret; |
885 | } | |
886 | ||
e4404d6e YZ |
887 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
888 | fmode_t flags, void *holder) | |
8a4b83cc | 889 | { |
d5e2003c | 890 | struct request_queue *q; |
8a4b83cc CM |
891 | struct block_device *bdev; |
892 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 893 | struct btrfs_device *device; |
443f24fe | 894 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
895 | struct buffer_head *bh; |
896 | struct btrfs_super_block *disk_super; | |
a0af469b | 897 | u64 devid; |
2b82032c | 898 | int seeding = 1; |
a0af469b | 899 | int ret = 0; |
8a4b83cc | 900 | |
d4d77629 TH |
901 | flags |= FMODE_EXCL; |
902 | ||
c6e30871 | 903 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
904 | if (device->bdev) |
905 | continue; | |
dfe25020 CM |
906 | if (!device->name) |
907 | continue; | |
908 | ||
f63e0cca ES |
909 | /* Just open everything we can; ignore failures here */ |
910 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
911 | &bdev, &bh)) | |
beaf8ab3 | 912 | continue; |
a0af469b CM |
913 | |
914 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 915 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
916 | if (devid != device->devid) |
917 | goto error_brelse; | |
918 | ||
2b82032c YZ |
919 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
920 | BTRFS_UUID_SIZE)) | |
921 | goto error_brelse; | |
922 | ||
923 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
924 | if (!latest_dev || |
925 | device->generation > latest_dev->generation) | |
926 | latest_dev = device; | |
a0af469b | 927 | |
2b82032c YZ |
928 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
929 | device->writeable = 0; | |
930 | } else { | |
931 | device->writeable = !bdev_read_only(bdev); | |
932 | seeding = 0; | |
933 | } | |
934 | ||
d5e2003c | 935 | q = bdev_get_queue(bdev); |
90180da4 | 936 | if (blk_queue_discard(q)) |
d5e2003c | 937 | device->can_discard = 1; |
d5e2003c | 938 | |
8a4b83cc | 939 | device->bdev = bdev; |
dfe25020 | 940 | device->in_fs_metadata = 0; |
15916de8 CM |
941 | device->mode = flags; |
942 | ||
c289811c CM |
943 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
944 | fs_devices->rotating = 1; | |
945 | ||
a0af469b | 946 | fs_devices->open_devices++; |
55e50e45 ID |
947 | if (device->writeable && |
948 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
949 | fs_devices->rw_devices++; |
950 | list_add(&device->dev_alloc_list, | |
951 | &fs_devices->alloc_list); | |
952 | } | |
4f6c9328 | 953 | brelse(bh); |
a0af469b | 954 | continue; |
a061fc8d | 955 | |
a0af469b CM |
956 | error_brelse: |
957 | brelse(bh); | |
d4d77629 | 958 | blkdev_put(bdev, flags); |
a0af469b | 959 | continue; |
8a4b83cc | 960 | } |
a0af469b | 961 | if (fs_devices->open_devices == 0) { |
20bcd649 | 962 | ret = -EINVAL; |
a0af469b CM |
963 | goto out; |
964 | } | |
2b82032c YZ |
965 | fs_devices->seeding = seeding; |
966 | fs_devices->opened = 1; | |
443f24fe | 967 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 968 | fs_devices->total_rw_bytes = 0; |
a0af469b | 969 | out: |
2b82032c YZ |
970 | return ret; |
971 | } | |
972 | ||
973 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 974 | fmode_t flags, void *holder) |
2b82032c YZ |
975 | { |
976 | int ret; | |
977 | ||
978 | mutex_lock(&uuid_mutex); | |
979 | if (fs_devices->opened) { | |
e4404d6e YZ |
980 | fs_devices->opened++; |
981 | ret = 0; | |
2b82032c | 982 | } else { |
15916de8 | 983 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 984 | } |
8a4b83cc | 985 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
986 | return ret; |
987 | } | |
988 | ||
6f60cbd3 DS |
989 | /* |
990 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
991 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
992 | * is read via pagecache | |
993 | */ | |
97288f2c | 994 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
995 | struct btrfs_fs_devices **fs_devices_ret) |
996 | { | |
997 | struct btrfs_super_block *disk_super; | |
998 | struct block_device *bdev; | |
6f60cbd3 DS |
999 | struct page *page; |
1000 | void *p; | |
1001 | int ret = -EINVAL; | |
8a4b83cc | 1002 | u64 devid; |
f2984462 | 1003 | u64 transid; |
02db0844 | 1004 | u64 total_devices; |
6f60cbd3 DS |
1005 | u64 bytenr; |
1006 | pgoff_t index; | |
8a4b83cc | 1007 | |
6f60cbd3 DS |
1008 | /* |
1009 | * we would like to check all the supers, but that would make | |
1010 | * a btrfs mount succeed after a mkfs from a different FS. | |
1011 | * So, we need to add a special mount option to scan for | |
1012 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1013 | */ | |
1014 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1015 | flags |= FMODE_EXCL; |
10f6327b | 1016 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1017 | |
1018 | bdev = blkdev_get_by_path(path, flags, holder); | |
1019 | ||
1020 | if (IS_ERR(bdev)) { | |
1021 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1022 | goto error; |
6f60cbd3 DS |
1023 | } |
1024 | ||
1025 | /* make sure our super fits in the device */ | |
1026 | if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode)) | |
1027 | goto error_bdev_put; | |
1028 | ||
1029 | /* make sure our super fits in the page */ | |
1030 | if (sizeof(*disk_super) > PAGE_CACHE_SIZE) | |
1031 | goto error_bdev_put; | |
1032 | ||
1033 | /* make sure our super doesn't straddle pages on disk */ | |
1034 | index = bytenr >> PAGE_CACHE_SHIFT; | |
1035 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index) | |
1036 | goto error_bdev_put; | |
1037 | ||
1038 | /* pull in the page with our super */ | |
1039 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1040 | index, GFP_NOFS); | |
1041 | ||
1042 | if (IS_ERR_OR_NULL(page)) | |
1043 | goto error_bdev_put; | |
1044 | ||
1045 | p = kmap(page); | |
1046 | ||
1047 | /* align our pointer to the offset of the super block */ | |
1048 | disk_super = p + (bytenr & ~PAGE_CACHE_MASK); | |
1049 | ||
1050 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
3cae210f | 1051 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) |
6f60cbd3 DS |
1052 | goto error_unmap; |
1053 | ||
a343832f | 1054 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1055 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1056 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1057 | |
8a4b83cc | 1058 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1059 | if (ret > 0) { |
1060 | if (disk_super->label[0]) { | |
1061 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
1062 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1063 | printk(KERN_INFO "BTRFS: device label %s ", disk_super->label); | |
1064 | } else { | |
1065 | printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid); | |
1066 | } | |
1067 | ||
1068 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); | |
1069 | ret = 0; | |
1070 | } | |
02db0844 JB |
1071 | if (!ret && fs_devices_ret) |
1072 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 DS |
1073 | |
1074 | error_unmap: | |
1075 | kunmap(page); | |
1076 | page_cache_release(page); | |
1077 | ||
1078 | error_bdev_put: | |
d4d77629 | 1079 | blkdev_put(bdev, flags); |
8a4b83cc | 1080 | error: |
beaf8ab3 | 1081 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1082 | return ret; |
1083 | } | |
0b86a832 | 1084 | |
6d07bcec MX |
1085 | /* helper to account the used device space in the range */ |
1086 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1087 | u64 end, u64 *length) | |
1088 | { | |
1089 | struct btrfs_key key; | |
1090 | struct btrfs_root *root = device->dev_root; | |
1091 | struct btrfs_dev_extent *dev_extent; | |
1092 | struct btrfs_path *path; | |
1093 | u64 extent_end; | |
1094 | int ret; | |
1095 | int slot; | |
1096 | struct extent_buffer *l; | |
1097 | ||
1098 | *length = 0; | |
1099 | ||
63a212ab | 1100 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1101 | return 0; |
1102 | ||
1103 | path = btrfs_alloc_path(); | |
1104 | if (!path) | |
1105 | return -ENOMEM; | |
1106 | path->reada = 2; | |
1107 | ||
1108 | key.objectid = device->devid; | |
1109 | key.offset = start; | |
1110 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1111 | ||
1112 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1113 | if (ret < 0) | |
1114 | goto out; | |
1115 | if (ret > 0) { | |
1116 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1117 | if (ret < 0) | |
1118 | goto out; | |
1119 | } | |
1120 | ||
1121 | while (1) { | |
1122 | l = path->nodes[0]; | |
1123 | slot = path->slots[0]; | |
1124 | if (slot >= btrfs_header_nritems(l)) { | |
1125 | ret = btrfs_next_leaf(root, path); | |
1126 | if (ret == 0) | |
1127 | continue; | |
1128 | if (ret < 0) | |
1129 | goto out; | |
1130 | ||
1131 | break; | |
1132 | } | |
1133 | btrfs_item_key_to_cpu(l, &key, slot); | |
1134 | ||
1135 | if (key.objectid < device->devid) | |
1136 | goto next; | |
1137 | ||
1138 | if (key.objectid > device->devid) | |
1139 | break; | |
1140 | ||
962a298f | 1141 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1142 | goto next; |
1143 | ||
1144 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1145 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1146 | dev_extent); | |
1147 | if (key.offset <= start && extent_end > end) { | |
1148 | *length = end - start + 1; | |
1149 | break; | |
1150 | } else if (key.offset <= start && extent_end > start) | |
1151 | *length += extent_end - start; | |
1152 | else if (key.offset > start && extent_end <= end) | |
1153 | *length += extent_end - key.offset; | |
1154 | else if (key.offset > start && key.offset <= end) { | |
1155 | *length += end - key.offset + 1; | |
1156 | break; | |
1157 | } else if (key.offset > end) | |
1158 | break; | |
1159 | ||
1160 | next: | |
1161 | path->slots[0]++; | |
1162 | } | |
1163 | ret = 0; | |
1164 | out: | |
1165 | btrfs_free_path(path); | |
1166 | return ret; | |
1167 | } | |
1168 | ||
499f377f | 1169 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1170 | struct btrfs_device *device, |
1171 | u64 *start, u64 len) | |
1172 | { | |
499f377f | 1173 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; |
6df9a95e | 1174 | struct extent_map *em; |
499f377f | 1175 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1176 | int ret = 0; |
1b984508 | 1177 | u64 physical_start = *start; |
6df9a95e | 1178 | |
499f377f JM |
1179 | if (transaction) |
1180 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1181 | again: |
1182 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1183 | struct map_lookup *map; |
1184 | int i; | |
1185 | ||
1186 | map = (struct map_lookup *)em->bdev; | |
1187 | for (i = 0; i < map->num_stripes; i++) { | |
c152b63e FM |
1188 | u64 end; |
1189 | ||
6df9a95e JB |
1190 | if (map->stripes[i].dev != device) |
1191 | continue; | |
1b984508 | 1192 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1193 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1194 | physical_start) |
6df9a95e | 1195 | continue; |
c152b63e FM |
1196 | /* |
1197 | * Make sure that while processing the pinned list we do | |
1198 | * not override our *start with a lower value, because | |
1199 | * we can have pinned chunks that fall within this | |
1200 | * device hole and that have lower physical addresses | |
1201 | * than the pending chunks we processed before. If we | |
1202 | * do not take this special care we can end up getting | |
1203 | * 2 pending chunks that start at the same physical | |
1204 | * device offsets because the end offset of a pinned | |
1205 | * chunk can be equal to the start offset of some | |
1206 | * pending chunk. | |
1207 | */ | |
1208 | end = map->stripes[i].physical + em->orig_block_len; | |
1209 | if (end > *start) { | |
1210 | *start = end; | |
1211 | ret = 1; | |
1212 | } | |
6df9a95e JB |
1213 | } |
1214 | } | |
499f377f JM |
1215 | if (search_list != &fs_info->pinned_chunks) { |
1216 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1217 | goto again; |
1218 | } | |
6df9a95e JB |
1219 | |
1220 | return ret; | |
1221 | } | |
1222 | ||
1223 | ||
0b86a832 | 1224 | /* |
499f377f JM |
1225 | * find_free_dev_extent_start - find free space in the specified device |
1226 | * @device: the device which we search the free space in | |
1227 | * @num_bytes: the size of the free space that we need | |
1228 | * @search_start: the position from which to begin the search | |
1229 | * @start: store the start of the free space. | |
1230 | * @len: the size of the free space. that we find, or the size | |
1231 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1232 | * |
0b86a832 CM |
1233 | * this uses a pretty simple search, the expectation is that it is |
1234 | * called very infrequently and that a given device has a small number | |
1235 | * of extents | |
7bfc837d MX |
1236 | * |
1237 | * @start is used to store the start of the free space if we find. But if we | |
1238 | * don't find suitable free space, it will be used to store the start position | |
1239 | * of the max free space. | |
1240 | * | |
1241 | * @len is used to store the size of the free space that we find. | |
1242 | * But if we don't find suitable free space, it is used to store the size of | |
1243 | * the max free space. | |
0b86a832 | 1244 | */ |
499f377f JM |
1245 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1246 | struct btrfs_device *device, u64 num_bytes, | |
1247 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 CM |
1248 | { |
1249 | struct btrfs_key key; | |
1250 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1251 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1252 | struct btrfs_path *path; |
7bfc837d MX |
1253 | u64 hole_size; |
1254 | u64 max_hole_start; | |
1255 | u64 max_hole_size; | |
1256 | u64 extent_end; | |
0b86a832 CM |
1257 | u64 search_end = device->total_bytes; |
1258 | int ret; | |
7bfc837d | 1259 | int slot; |
0b86a832 CM |
1260 | struct extent_buffer *l; |
1261 | ||
6df9a95e JB |
1262 | path = btrfs_alloc_path(); |
1263 | if (!path) | |
1264 | return -ENOMEM; | |
f2ab7618 | 1265 | |
7bfc837d MX |
1266 | max_hole_start = search_start; |
1267 | max_hole_size = 0; | |
1268 | ||
f2ab7618 | 1269 | again: |
63a212ab | 1270 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1271 | ret = -ENOSPC; |
6df9a95e | 1272 | goto out; |
7bfc837d MX |
1273 | } |
1274 | ||
7bfc837d | 1275 | path->reada = 2; |
6df9a95e JB |
1276 | path->search_commit_root = 1; |
1277 | path->skip_locking = 1; | |
7bfc837d | 1278 | |
0b86a832 CM |
1279 | key.objectid = device->devid; |
1280 | key.offset = search_start; | |
1281 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1282 | |
125ccb0a | 1283 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1284 | if (ret < 0) |
7bfc837d | 1285 | goto out; |
1fcbac58 YZ |
1286 | if (ret > 0) { |
1287 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1288 | if (ret < 0) | |
7bfc837d | 1289 | goto out; |
1fcbac58 | 1290 | } |
7bfc837d | 1291 | |
0b86a832 CM |
1292 | while (1) { |
1293 | l = path->nodes[0]; | |
1294 | slot = path->slots[0]; | |
1295 | if (slot >= btrfs_header_nritems(l)) { | |
1296 | ret = btrfs_next_leaf(root, path); | |
1297 | if (ret == 0) | |
1298 | continue; | |
1299 | if (ret < 0) | |
7bfc837d MX |
1300 | goto out; |
1301 | ||
1302 | break; | |
0b86a832 CM |
1303 | } |
1304 | btrfs_item_key_to_cpu(l, &key, slot); | |
1305 | ||
1306 | if (key.objectid < device->devid) | |
1307 | goto next; | |
1308 | ||
1309 | if (key.objectid > device->devid) | |
7bfc837d | 1310 | break; |
0b86a832 | 1311 | |
962a298f | 1312 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1313 | goto next; |
9779b72f | 1314 | |
7bfc837d MX |
1315 | if (key.offset > search_start) { |
1316 | hole_size = key.offset - search_start; | |
9779b72f | 1317 | |
6df9a95e JB |
1318 | /* |
1319 | * Have to check before we set max_hole_start, otherwise | |
1320 | * we could end up sending back this offset anyway. | |
1321 | */ | |
499f377f | 1322 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1323 | &search_start, |
1b984508 FL |
1324 | hole_size)) { |
1325 | if (key.offset >= search_start) { | |
1326 | hole_size = key.offset - search_start; | |
1327 | } else { | |
1328 | WARN_ON_ONCE(1); | |
1329 | hole_size = 0; | |
1330 | } | |
1331 | } | |
6df9a95e | 1332 | |
7bfc837d MX |
1333 | if (hole_size > max_hole_size) { |
1334 | max_hole_start = search_start; | |
1335 | max_hole_size = hole_size; | |
1336 | } | |
9779b72f | 1337 | |
7bfc837d MX |
1338 | /* |
1339 | * If this free space is greater than which we need, | |
1340 | * it must be the max free space that we have found | |
1341 | * until now, so max_hole_start must point to the start | |
1342 | * of this free space and the length of this free space | |
1343 | * is stored in max_hole_size. Thus, we return | |
1344 | * max_hole_start and max_hole_size and go back to the | |
1345 | * caller. | |
1346 | */ | |
1347 | if (hole_size >= num_bytes) { | |
1348 | ret = 0; | |
1349 | goto out; | |
0b86a832 CM |
1350 | } |
1351 | } | |
0b86a832 | 1352 | |
0b86a832 | 1353 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1354 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1355 | dev_extent); | |
1356 | if (extent_end > search_start) | |
1357 | search_start = extent_end; | |
0b86a832 CM |
1358 | next: |
1359 | path->slots[0]++; | |
1360 | cond_resched(); | |
1361 | } | |
0b86a832 | 1362 | |
38c01b96 | 1363 | /* |
1364 | * At this point, search_start should be the end of | |
1365 | * allocated dev extents, and when shrinking the device, | |
1366 | * search_end may be smaller than search_start. | |
1367 | */ | |
f2ab7618 | 1368 | if (search_end > search_start) { |
38c01b96 | 1369 | hole_size = search_end - search_start; |
1370 | ||
499f377f | 1371 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1372 | hole_size)) { |
1373 | btrfs_release_path(path); | |
1374 | goto again; | |
1375 | } | |
0b86a832 | 1376 | |
f2ab7618 ZL |
1377 | if (hole_size > max_hole_size) { |
1378 | max_hole_start = search_start; | |
1379 | max_hole_size = hole_size; | |
1380 | } | |
6df9a95e JB |
1381 | } |
1382 | ||
7bfc837d | 1383 | /* See above. */ |
f2ab7618 | 1384 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1385 | ret = -ENOSPC; |
1386 | else | |
1387 | ret = 0; | |
1388 | ||
1389 | out: | |
2b82032c | 1390 | btrfs_free_path(path); |
7bfc837d | 1391 | *start = max_hole_start; |
b2117a39 | 1392 | if (len) |
7bfc837d | 1393 | *len = max_hole_size; |
0b86a832 CM |
1394 | return ret; |
1395 | } | |
1396 | ||
499f377f JM |
1397 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1398 | struct btrfs_device *device, u64 num_bytes, | |
1399 | u64 *start, u64 *len) | |
1400 | { | |
1401 | struct btrfs_root *root = device->dev_root; | |
1402 | u64 search_start; | |
1403 | ||
1404 | /* FIXME use last free of some kind */ | |
1405 | ||
1406 | /* | |
1407 | * we don't want to overwrite the superblock on the drive, | |
1408 | * so we make sure to start at an offset of at least 1MB | |
1409 | */ | |
1410 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); | |
1411 | return find_free_dev_extent_start(trans->transaction, device, | |
1412 | num_bytes, search_start, start, len); | |
1413 | } | |
1414 | ||
b2950863 | 1415 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1416 | struct btrfs_device *device, |
2196d6e8 | 1417 | u64 start, u64 *dev_extent_len) |
8f18cf13 CM |
1418 | { |
1419 | int ret; | |
1420 | struct btrfs_path *path; | |
1421 | struct btrfs_root *root = device->dev_root; | |
1422 | struct btrfs_key key; | |
a061fc8d CM |
1423 | struct btrfs_key found_key; |
1424 | struct extent_buffer *leaf = NULL; | |
1425 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1426 | |
1427 | path = btrfs_alloc_path(); | |
1428 | if (!path) | |
1429 | return -ENOMEM; | |
1430 | ||
1431 | key.objectid = device->devid; | |
1432 | key.offset = start; | |
1433 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1434 | again: |
8f18cf13 | 1435 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1436 | if (ret > 0) { |
1437 | ret = btrfs_previous_item(root, path, key.objectid, | |
1438 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1439 | if (ret) |
1440 | goto out; | |
a061fc8d CM |
1441 | leaf = path->nodes[0]; |
1442 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1443 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1444 | struct btrfs_dev_extent); | |
1445 | BUG_ON(found_key.offset > start || found_key.offset + | |
1446 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1447 | key = found_key; |
1448 | btrfs_release_path(path); | |
1449 | goto again; | |
a061fc8d CM |
1450 | } else if (ret == 0) { |
1451 | leaf = path->nodes[0]; | |
1452 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1453 | struct btrfs_dev_extent); | |
79787eaa | 1454 | } else { |
a4553fef | 1455 | btrfs_std_error(root->fs_info, ret, "Slot search failed"); |
79787eaa | 1456 | goto out; |
a061fc8d | 1457 | } |
8f18cf13 | 1458 | |
2196d6e8 MX |
1459 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1460 | ||
8f18cf13 | 1461 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1462 | if (ret) { |
a4553fef | 1463 | btrfs_std_error(root->fs_info, ret, |
79787eaa | 1464 | "Failed to remove dev extent item"); |
13212b54 | 1465 | } else { |
3204d33c | 1466 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1467 | } |
b0b802d7 | 1468 | out: |
8f18cf13 CM |
1469 | btrfs_free_path(path); |
1470 | return ret; | |
1471 | } | |
1472 | ||
48a3b636 ES |
1473 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1474 | struct btrfs_device *device, | |
1475 | u64 chunk_tree, u64 chunk_objectid, | |
1476 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1477 | { |
1478 | int ret; | |
1479 | struct btrfs_path *path; | |
1480 | struct btrfs_root *root = device->dev_root; | |
1481 | struct btrfs_dev_extent *extent; | |
1482 | struct extent_buffer *leaf; | |
1483 | struct btrfs_key key; | |
1484 | ||
dfe25020 | 1485 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1486 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1487 | path = btrfs_alloc_path(); |
1488 | if (!path) | |
1489 | return -ENOMEM; | |
1490 | ||
0b86a832 | 1491 | key.objectid = device->devid; |
2b82032c | 1492 | key.offset = start; |
0b86a832 CM |
1493 | key.type = BTRFS_DEV_EXTENT_KEY; |
1494 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1495 | sizeof(*extent)); | |
2cdcecbc MF |
1496 | if (ret) |
1497 | goto out; | |
0b86a832 CM |
1498 | |
1499 | leaf = path->nodes[0]; | |
1500 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1501 | struct btrfs_dev_extent); | |
e17cade2 CM |
1502 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1503 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1504 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1505 | ||
1506 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1507 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1508 | |
0b86a832 CM |
1509 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1510 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1511 | out: |
0b86a832 CM |
1512 | btrfs_free_path(path); |
1513 | return ret; | |
1514 | } | |
1515 | ||
6df9a95e | 1516 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1517 | { |
6df9a95e JB |
1518 | struct extent_map_tree *em_tree; |
1519 | struct extent_map *em; | |
1520 | struct rb_node *n; | |
1521 | u64 ret = 0; | |
0b86a832 | 1522 | |
6df9a95e JB |
1523 | em_tree = &fs_info->mapping_tree.map_tree; |
1524 | read_lock(&em_tree->lock); | |
1525 | n = rb_last(&em_tree->map); | |
1526 | if (n) { | |
1527 | em = rb_entry(n, struct extent_map, rb_node); | |
1528 | ret = em->start + em->len; | |
0b86a832 | 1529 | } |
6df9a95e JB |
1530 | read_unlock(&em_tree->lock); |
1531 | ||
0b86a832 CM |
1532 | return ret; |
1533 | } | |
1534 | ||
53f10659 ID |
1535 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1536 | u64 *devid_ret) | |
0b86a832 CM |
1537 | { |
1538 | int ret; | |
1539 | struct btrfs_key key; | |
1540 | struct btrfs_key found_key; | |
2b82032c YZ |
1541 | struct btrfs_path *path; |
1542 | ||
2b82032c YZ |
1543 | path = btrfs_alloc_path(); |
1544 | if (!path) | |
1545 | return -ENOMEM; | |
0b86a832 CM |
1546 | |
1547 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1548 | key.type = BTRFS_DEV_ITEM_KEY; | |
1549 | key.offset = (u64)-1; | |
1550 | ||
53f10659 | 1551 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1552 | if (ret < 0) |
1553 | goto error; | |
1554 | ||
79787eaa | 1555 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1556 | |
53f10659 ID |
1557 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1558 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1559 | BTRFS_DEV_ITEM_KEY); |
1560 | if (ret) { | |
53f10659 | 1561 | *devid_ret = 1; |
0b86a832 CM |
1562 | } else { |
1563 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1564 | path->slots[0]); | |
53f10659 | 1565 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1566 | } |
1567 | ret = 0; | |
1568 | error: | |
2b82032c | 1569 | btrfs_free_path(path); |
0b86a832 CM |
1570 | return ret; |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * the device information is stored in the chunk root | |
1575 | * the btrfs_device struct should be fully filled in | |
1576 | */ | |
48a3b636 ES |
1577 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1578 | struct btrfs_root *root, | |
1579 | struct btrfs_device *device) | |
0b86a832 CM |
1580 | { |
1581 | int ret; | |
1582 | struct btrfs_path *path; | |
1583 | struct btrfs_dev_item *dev_item; | |
1584 | struct extent_buffer *leaf; | |
1585 | struct btrfs_key key; | |
1586 | unsigned long ptr; | |
0b86a832 CM |
1587 | |
1588 | root = root->fs_info->chunk_root; | |
1589 | ||
1590 | path = btrfs_alloc_path(); | |
1591 | if (!path) | |
1592 | return -ENOMEM; | |
1593 | ||
0b86a832 CM |
1594 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1595 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1596 | key.offset = device->devid; |
0b86a832 CM |
1597 | |
1598 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1599 | sizeof(*dev_item)); |
0b86a832 CM |
1600 | if (ret) |
1601 | goto out; | |
1602 | ||
1603 | leaf = path->nodes[0]; | |
1604 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1605 | ||
1606 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1607 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1608 | btrfs_set_device_type(leaf, dev_item, device->type); |
1609 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1610 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1611 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1612 | btrfs_set_device_total_bytes(leaf, dev_item, |
1613 | btrfs_device_get_disk_total_bytes(device)); | |
1614 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1615 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1616 | btrfs_set_device_group(leaf, dev_item, 0); |
1617 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1618 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1619 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1620 | |
410ba3a2 | 1621 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1622 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1623 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1624 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1625 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1626 | |
2b82032c | 1627 | ret = 0; |
0b86a832 CM |
1628 | out: |
1629 | btrfs_free_path(path); | |
1630 | return ret; | |
1631 | } | |
8f18cf13 | 1632 | |
5a1972bd QW |
1633 | /* |
1634 | * Function to update ctime/mtime for a given device path. | |
1635 | * Mainly used for ctime/mtime based probe like libblkid. | |
1636 | */ | |
1637 | static void update_dev_time(char *path_name) | |
1638 | { | |
1639 | struct file *filp; | |
1640 | ||
1641 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1642 | if (IS_ERR(filp)) |
5a1972bd QW |
1643 | return; |
1644 | file_update_time(filp); | |
1645 | filp_close(filp, NULL); | |
1646 | return; | |
1647 | } | |
1648 | ||
a061fc8d CM |
1649 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1650 | struct btrfs_device *device) | |
1651 | { | |
1652 | int ret; | |
1653 | struct btrfs_path *path; | |
a061fc8d | 1654 | struct btrfs_key key; |
a061fc8d CM |
1655 | struct btrfs_trans_handle *trans; |
1656 | ||
1657 | root = root->fs_info->chunk_root; | |
1658 | ||
1659 | path = btrfs_alloc_path(); | |
1660 | if (!path) | |
1661 | return -ENOMEM; | |
1662 | ||
a22285a6 | 1663 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1664 | if (IS_ERR(trans)) { |
1665 | btrfs_free_path(path); | |
1666 | return PTR_ERR(trans); | |
1667 | } | |
a061fc8d CM |
1668 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1669 | key.type = BTRFS_DEV_ITEM_KEY; | |
1670 | key.offset = device->devid; | |
1671 | ||
1672 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1673 | if (ret < 0) | |
1674 | goto out; | |
1675 | ||
1676 | if (ret > 0) { | |
1677 | ret = -ENOENT; | |
1678 | goto out; | |
1679 | } | |
1680 | ||
1681 | ret = btrfs_del_item(trans, root, path); | |
1682 | if (ret) | |
1683 | goto out; | |
a061fc8d CM |
1684 | out: |
1685 | btrfs_free_path(path); | |
1686 | btrfs_commit_transaction(trans, root); | |
1687 | return ret; | |
1688 | } | |
1689 | ||
1690 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1691 | { | |
1692 | struct btrfs_device *device; | |
2b82032c | 1693 | struct btrfs_device *next_device; |
a061fc8d | 1694 | struct block_device *bdev; |
dfe25020 | 1695 | struct buffer_head *bh = NULL; |
a061fc8d | 1696 | struct btrfs_super_block *disk_super; |
1f78160c | 1697 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1698 | u64 all_avail; |
1699 | u64 devid; | |
2b82032c YZ |
1700 | u64 num_devices; |
1701 | u8 *dev_uuid; | |
de98ced9 | 1702 | unsigned seq; |
a061fc8d | 1703 | int ret = 0; |
1f78160c | 1704 | bool clear_super = false; |
a061fc8d | 1705 | |
a061fc8d CM |
1706 | mutex_lock(&uuid_mutex); |
1707 | ||
de98ced9 MX |
1708 | do { |
1709 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
1710 | ||
1711 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1712 | root->fs_info->avail_system_alloc_bits | | |
1713 | root->fs_info->avail_metadata_alloc_bits; | |
1714 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
a061fc8d | 1715 | |
8dabb742 SB |
1716 | num_devices = root->fs_info->fs_devices->num_devices; |
1717 | btrfs_dev_replace_lock(&root->fs_info->dev_replace); | |
1718 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1719 | WARN_ON(num_devices < 1); | |
1720 | num_devices--; | |
1721 | } | |
1722 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace); | |
1723 | ||
1724 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) { | |
183860f6 | 1725 | ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET; |
a061fc8d CM |
1726 | goto out; |
1727 | } | |
1728 | ||
8dabb742 | 1729 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) { |
183860f6 | 1730 | ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET; |
a061fc8d CM |
1731 | goto out; |
1732 | } | |
1733 | ||
53b381b3 DW |
1734 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && |
1735 | root->fs_info->fs_devices->rw_devices <= 2) { | |
183860f6 | 1736 | ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET; |
53b381b3 DW |
1737 | goto out; |
1738 | } | |
1739 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1740 | root->fs_info->fs_devices->rw_devices <= 3) { | |
183860f6 | 1741 | ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET; |
53b381b3 DW |
1742 | goto out; |
1743 | } | |
1744 | ||
dfe25020 | 1745 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1746 | struct list_head *devices; |
1747 | struct btrfs_device *tmp; | |
a061fc8d | 1748 | |
dfe25020 CM |
1749 | device = NULL; |
1750 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1751 | /* |
1752 | * It is safe to read the devices since the volume_mutex | |
1753 | * is held. | |
1754 | */ | |
c6e30871 | 1755 | list_for_each_entry(tmp, devices, dev_list) { |
63a212ab SB |
1756 | if (tmp->in_fs_metadata && |
1757 | !tmp->is_tgtdev_for_dev_replace && | |
1758 | !tmp->bdev) { | |
dfe25020 CM |
1759 | device = tmp; |
1760 | break; | |
1761 | } | |
1762 | } | |
1763 | bdev = NULL; | |
1764 | bh = NULL; | |
1765 | disk_super = NULL; | |
1766 | if (!device) { | |
183860f6 | 1767 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
dfe25020 CM |
1768 | goto out; |
1769 | } | |
dfe25020 | 1770 | } else { |
beaf8ab3 | 1771 | ret = btrfs_get_bdev_and_sb(device_path, |
cc975eb4 | 1772 | FMODE_WRITE | FMODE_EXCL, |
beaf8ab3 SB |
1773 | root->fs_info->bdev_holder, 0, |
1774 | &bdev, &bh); | |
1775 | if (ret) | |
dfe25020 | 1776 | goto out; |
dfe25020 | 1777 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1778 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c | 1779 | dev_uuid = disk_super->dev_item.uuid; |
aa1b8cd4 | 1780 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2b82032c | 1781 | disk_super->fsid); |
dfe25020 CM |
1782 | if (!device) { |
1783 | ret = -ENOENT; | |
1784 | goto error_brelse; | |
1785 | } | |
2b82032c | 1786 | } |
dfe25020 | 1787 | |
63a212ab | 1788 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1789 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
63a212ab SB |
1790 | goto error_brelse; |
1791 | } | |
1792 | ||
2b82032c | 1793 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1794 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
2b82032c YZ |
1795 | goto error_brelse; |
1796 | } | |
1797 | ||
1798 | if (device->writeable) { | |
0c1daee0 | 1799 | lock_chunks(root); |
2b82032c | 1800 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1801 | device->fs_devices->rw_devices--; |
0c1daee0 | 1802 | unlock_chunks(root); |
1f78160c | 1803 | clear_super = true; |
dfe25020 | 1804 | } |
a061fc8d | 1805 | |
d7901554 | 1806 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1807 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1808 | mutex_lock(&uuid_mutex); |
a061fc8d | 1809 | if (ret) |
9b3517e9 | 1810 | goto error_undo; |
a061fc8d | 1811 | |
63a212ab SB |
1812 | /* |
1813 | * TODO: the superblock still includes this device in its num_devices | |
1814 | * counter although write_all_supers() is not locked out. This | |
1815 | * could give a filesystem state which requires a degraded mount. | |
1816 | */ | |
a061fc8d CM |
1817 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1818 | if (ret) | |
9b3517e9 | 1819 | goto error_undo; |
a061fc8d | 1820 | |
2b82032c | 1821 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1822 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1823 | |
1824 | /* | |
1825 | * the device list mutex makes sure that we don't change | |
1826 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1827 | * the device supers. Whoever is writing all supers, should |
1828 | * lock the device list mutex before getting the number of | |
1829 | * devices in the super block (super_copy). Conversely, | |
1830 | * whoever updates the number of devices in the super block | |
1831 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1832 | */ |
1f78160c XG |
1833 | |
1834 | cur_devices = device->fs_devices; | |
e5e9a520 | 1835 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1836 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1837 | |
e4404d6e | 1838 | device->fs_devices->num_devices--; |
02db0844 | 1839 | device->fs_devices->total_devices--; |
2b82032c | 1840 | |
cd02dca5 | 1841 | if (device->missing) |
3a7d55c8 | 1842 | device->fs_devices->missing_devices--; |
cd02dca5 | 1843 | |
2b82032c YZ |
1844 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1845 | struct btrfs_device, dev_list); | |
1846 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1847 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1848 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1849 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1850 | ||
0bfaa9c5 | 1851 | if (device->bdev) { |
e4404d6e | 1852 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1853 | /* remove sysfs entry */ |
32576040 | 1854 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); |
0bfaa9c5 | 1855 | } |
99994cde | 1856 | |
1f78160c | 1857 | call_rcu(&device->rcu, free_device); |
e4404d6e | 1858 | |
6c41761f DS |
1859 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1860 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
d7306801 | 1861 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2b82032c | 1862 | |
1f78160c | 1863 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1864 | struct btrfs_fs_devices *fs_devices; |
1865 | fs_devices = root->fs_info->fs_devices; | |
1866 | while (fs_devices) { | |
8321cf25 RS |
1867 | if (fs_devices->seed == cur_devices) { |
1868 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1869 | break; |
8321cf25 | 1870 | } |
e4404d6e | 1871 | fs_devices = fs_devices->seed; |
2b82032c | 1872 | } |
1f78160c | 1873 | cur_devices->seed = NULL; |
1f78160c | 1874 | __btrfs_close_devices(cur_devices); |
1f78160c | 1875 | free_fs_devices(cur_devices); |
2b82032c YZ |
1876 | } |
1877 | ||
5af3e8cc SB |
1878 | root->fs_info->num_tolerated_disk_barrier_failures = |
1879 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1880 | ||
2b82032c YZ |
1881 | /* |
1882 | * at this point, the device is zero sized. We want to | |
1883 | * remove it from the devices list and zero out the old super | |
1884 | */ | |
aa1b8cd4 | 1885 | if (clear_super && disk_super) { |
4d90d28b AJ |
1886 | u64 bytenr; |
1887 | int i; | |
1888 | ||
dfe25020 CM |
1889 | /* make sure this device isn't detected as part of |
1890 | * the FS anymore | |
1891 | */ | |
1892 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1893 | set_buffer_dirty(bh); | |
1894 | sync_dirty_buffer(bh); | |
4d90d28b AJ |
1895 | |
1896 | /* clear the mirror copies of super block on the disk | |
1897 | * being removed, 0th copy is been taken care above and | |
1898 | * the below would take of the rest | |
1899 | */ | |
1900 | for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1901 | bytenr = btrfs_sb_offset(i); | |
1902 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
1903 | i_size_read(bdev->bd_inode)) | |
1904 | break; | |
1905 | ||
1906 | brelse(bh); | |
1907 | bh = __bread(bdev, bytenr / 4096, | |
1908 | BTRFS_SUPER_INFO_SIZE); | |
1909 | if (!bh) | |
1910 | continue; | |
1911 | ||
1912 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1913 | ||
1914 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
1915 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { | |
1916 | continue; | |
1917 | } | |
1918 | memset(&disk_super->magic, 0, | |
1919 | sizeof(disk_super->magic)); | |
1920 | set_buffer_dirty(bh); | |
1921 | sync_dirty_buffer(bh); | |
1922 | } | |
dfe25020 | 1923 | } |
a061fc8d | 1924 | |
a061fc8d | 1925 | ret = 0; |
a061fc8d | 1926 | |
5a1972bd QW |
1927 | if (bdev) { |
1928 | /* Notify udev that device has changed */ | |
3c911608 | 1929 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); |
b8b8ff59 | 1930 | |
5a1972bd QW |
1931 | /* Update ctime/mtime for device path for libblkid */ |
1932 | update_dev_time(device_path); | |
1933 | } | |
1934 | ||
a061fc8d CM |
1935 | error_brelse: |
1936 | brelse(bh); | |
dfe25020 | 1937 | if (bdev) |
e525fd89 | 1938 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1939 | out: |
1940 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1941 | return ret; |
9b3517e9 ID |
1942 | error_undo: |
1943 | if (device->writeable) { | |
0c1daee0 | 1944 | lock_chunks(root); |
9b3517e9 ID |
1945 | list_add(&device->dev_alloc_list, |
1946 | &root->fs_info->fs_devices->alloc_list); | |
c3929c36 | 1947 | device->fs_devices->rw_devices++; |
0c1daee0 | 1948 | unlock_chunks(root); |
9b3517e9 ID |
1949 | } |
1950 | goto error_brelse; | |
a061fc8d CM |
1951 | } |
1952 | ||
084b6e7c QW |
1953 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1954 | struct btrfs_device *srcdev) | |
e93c89c1 | 1955 | { |
d51908ce AJ |
1956 | struct btrfs_fs_devices *fs_devices; |
1957 | ||
e93c89c1 | 1958 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 1959 | |
25e8e911 AJ |
1960 | /* |
1961 | * in case of fs with no seed, srcdev->fs_devices will point | |
1962 | * to fs_devices of fs_info. However when the dev being replaced is | |
1963 | * a seed dev it will point to the seed's local fs_devices. In short | |
1964 | * srcdev will have its correct fs_devices in both the cases. | |
1965 | */ | |
1966 | fs_devices = srcdev->fs_devices; | |
d51908ce | 1967 | |
e93c89c1 SB |
1968 | list_del_rcu(&srcdev->dev_list); |
1969 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 1970 | fs_devices->num_devices--; |
82372bc8 | 1971 | if (srcdev->missing) |
d51908ce | 1972 | fs_devices->missing_devices--; |
e93c89c1 | 1973 | |
82372bc8 MX |
1974 | if (srcdev->writeable) { |
1975 | fs_devices->rw_devices--; | |
1976 | /* zero out the old super if it is writable */ | |
31388ab2 | 1977 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); |
1357272f ID |
1978 | } |
1979 | ||
82372bc8 | 1980 | if (srcdev->bdev) |
d51908ce | 1981 | fs_devices->open_devices--; |
084b6e7c QW |
1982 | } |
1983 | ||
1984 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
1985 | struct btrfs_device *srcdev) | |
1986 | { | |
1987 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 SB |
1988 | |
1989 | call_rcu(&srcdev->rcu, free_device); | |
94d5f0c2 AJ |
1990 | |
1991 | /* | |
1992 | * unless fs_devices is seed fs, num_devices shouldn't go | |
1993 | * zero | |
1994 | */ | |
1995 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
1996 | ||
1997 | /* if this is no devs we rather delete the fs_devices */ | |
1998 | if (!fs_devices->num_devices) { | |
1999 | struct btrfs_fs_devices *tmp_fs_devices; | |
2000 | ||
2001 | tmp_fs_devices = fs_info->fs_devices; | |
2002 | while (tmp_fs_devices) { | |
2003 | if (tmp_fs_devices->seed == fs_devices) { | |
2004 | tmp_fs_devices->seed = fs_devices->seed; | |
2005 | break; | |
2006 | } | |
2007 | tmp_fs_devices = tmp_fs_devices->seed; | |
2008 | } | |
2009 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2010 | __btrfs_close_devices(fs_devices); |
2011 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2012 | } |
e93c89c1 SB |
2013 | } |
2014 | ||
2015 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2016 | struct btrfs_device *tgtdev) | |
2017 | { | |
2018 | struct btrfs_device *next_device; | |
2019 | ||
67a2c45e | 2020 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2021 | WARN_ON(!tgtdev); |
2022 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2023 | |
32576040 | 2024 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2025 | |
e93c89c1 | 2026 | if (tgtdev->bdev) { |
31388ab2 | 2027 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); |
e93c89c1 SB |
2028 | fs_info->fs_devices->open_devices--; |
2029 | } | |
2030 | fs_info->fs_devices->num_devices--; | |
e93c89c1 SB |
2031 | |
2032 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
2033 | struct btrfs_device, dev_list); | |
2034 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
2035 | fs_info->sb->s_bdev = next_device->bdev; | |
2036 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
2037 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2038 | list_del_rcu(&tgtdev->dev_list); | |
2039 | ||
2040 | call_rcu(&tgtdev->rcu, free_device); | |
2041 | ||
2042 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2043 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2044 | } |
2045 | ||
48a3b636 ES |
2046 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
2047 | struct btrfs_device **device) | |
7ba15b7d SB |
2048 | { |
2049 | int ret = 0; | |
2050 | struct btrfs_super_block *disk_super; | |
2051 | u64 devid; | |
2052 | u8 *dev_uuid; | |
2053 | struct block_device *bdev; | |
2054 | struct buffer_head *bh; | |
2055 | ||
2056 | *device = NULL; | |
2057 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
2058 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
2059 | if (ret) | |
2060 | return ret; | |
2061 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2062 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2063 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 2064 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
2065 | disk_super->fsid); |
2066 | brelse(bh); | |
2067 | if (!*device) | |
2068 | ret = -ENOENT; | |
2069 | blkdev_put(bdev, FMODE_READ); | |
2070 | return ret; | |
2071 | } | |
2072 | ||
2073 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
2074 | char *device_path, | |
2075 | struct btrfs_device **device) | |
2076 | { | |
2077 | *device = NULL; | |
2078 | if (strcmp(device_path, "missing") == 0) { | |
2079 | struct list_head *devices; | |
2080 | struct btrfs_device *tmp; | |
2081 | ||
2082 | devices = &root->fs_info->fs_devices->devices; | |
2083 | /* | |
2084 | * It is safe to read the devices since the volume_mutex | |
2085 | * is held by the caller. | |
2086 | */ | |
2087 | list_for_each_entry(tmp, devices, dev_list) { | |
2088 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2089 | *device = tmp; | |
2090 | break; | |
2091 | } | |
2092 | } | |
2093 | ||
d74a6259 AJ |
2094 | if (!*device) |
2095 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2096 | |
2097 | return 0; | |
2098 | } else { | |
2099 | return btrfs_find_device_by_path(root, device_path, device); | |
2100 | } | |
2101 | } | |
2102 | ||
2b82032c YZ |
2103 | /* |
2104 | * does all the dirty work required for changing file system's UUID. | |
2105 | */ | |
125ccb0a | 2106 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
2107 | { |
2108 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
2109 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 2110 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 2111 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
2112 | struct btrfs_device *device; |
2113 | u64 super_flags; | |
2114 | ||
2115 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2116 | if (!fs_devices->seeding) |
2b82032c YZ |
2117 | return -EINVAL; |
2118 | ||
2208a378 ID |
2119 | seed_devices = __alloc_fs_devices(); |
2120 | if (IS_ERR(seed_devices)) | |
2121 | return PTR_ERR(seed_devices); | |
2b82032c | 2122 | |
e4404d6e YZ |
2123 | old_devices = clone_fs_devices(fs_devices); |
2124 | if (IS_ERR(old_devices)) { | |
2125 | kfree(seed_devices); | |
2126 | return PTR_ERR(old_devices); | |
2b82032c | 2127 | } |
e4404d6e | 2128 | |
2b82032c YZ |
2129 | list_add(&old_devices->list, &fs_uuids); |
2130 | ||
e4404d6e YZ |
2131 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2132 | seed_devices->opened = 1; | |
2133 | INIT_LIST_HEAD(&seed_devices->devices); | |
2134 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2135 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
2136 | |
2137 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
2138 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2139 | synchronize_rcu); | |
2196d6e8 MX |
2140 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2141 | device->fs_devices = seed_devices; | |
c9513edb | 2142 | |
2196d6e8 | 2143 | lock_chunks(root); |
e4404d6e | 2144 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
2196d6e8 | 2145 | unlock_chunks(root); |
e4404d6e | 2146 | |
2b82032c YZ |
2147 | fs_devices->seeding = 0; |
2148 | fs_devices->num_devices = 0; | |
2149 | fs_devices->open_devices = 0; | |
69611ac8 | 2150 | fs_devices->missing_devices = 0; |
69611ac8 | 2151 | fs_devices->rotating = 0; |
e4404d6e | 2152 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2153 | |
2154 | generate_random_uuid(fs_devices->fsid); | |
2155 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2156 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
f7171750 FDBM |
2157 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2158 | ||
2b82032c YZ |
2159 | super_flags = btrfs_super_flags(disk_super) & |
2160 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2161 | btrfs_set_super_flags(disk_super, super_flags); | |
2162 | ||
2163 | return 0; | |
2164 | } | |
2165 | ||
2166 | /* | |
2167 | * strore the expected generation for seed devices in device items. | |
2168 | */ | |
2169 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2170 | struct btrfs_root *root) | |
2171 | { | |
2172 | struct btrfs_path *path; | |
2173 | struct extent_buffer *leaf; | |
2174 | struct btrfs_dev_item *dev_item; | |
2175 | struct btrfs_device *device; | |
2176 | struct btrfs_key key; | |
2177 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2178 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2179 | u64 devid; | |
2180 | int ret; | |
2181 | ||
2182 | path = btrfs_alloc_path(); | |
2183 | if (!path) | |
2184 | return -ENOMEM; | |
2185 | ||
2186 | root = root->fs_info->chunk_root; | |
2187 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2188 | key.offset = 0; | |
2189 | key.type = BTRFS_DEV_ITEM_KEY; | |
2190 | ||
2191 | while (1) { | |
2192 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2193 | if (ret < 0) | |
2194 | goto error; | |
2195 | ||
2196 | leaf = path->nodes[0]; | |
2197 | next_slot: | |
2198 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2199 | ret = btrfs_next_leaf(root, path); | |
2200 | if (ret > 0) | |
2201 | break; | |
2202 | if (ret < 0) | |
2203 | goto error; | |
2204 | leaf = path->nodes[0]; | |
2205 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2206 | btrfs_release_path(path); |
2b82032c YZ |
2207 | continue; |
2208 | } | |
2209 | ||
2210 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2211 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2212 | key.type != BTRFS_DEV_ITEM_KEY) | |
2213 | break; | |
2214 | ||
2215 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2216 | struct btrfs_dev_item); | |
2217 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2218 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2219 | BTRFS_UUID_SIZE); |
1473b24e | 2220 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2221 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
2222 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2223 | fs_uuid); | |
79787eaa | 2224 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2225 | |
2226 | if (device->fs_devices->seeding) { | |
2227 | btrfs_set_device_generation(leaf, dev_item, | |
2228 | device->generation); | |
2229 | btrfs_mark_buffer_dirty(leaf); | |
2230 | } | |
2231 | ||
2232 | path->slots[0]++; | |
2233 | goto next_slot; | |
2234 | } | |
2235 | ret = 0; | |
2236 | error: | |
2237 | btrfs_free_path(path); | |
2238 | return ret; | |
2239 | } | |
2240 | ||
788f20eb CM |
2241 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
2242 | { | |
d5e2003c | 2243 | struct request_queue *q; |
788f20eb CM |
2244 | struct btrfs_trans_handle *trans; |
2245 | struct btrfs_device *device; | |
2246 | struct block_device *bdev; | |
788f20eb | 2247 | struct list_head *devices; |
2b82032c | 2248 | struct super_block *sb = root->fs_info->sb; |
606686ee | 2249 | struct rcu_string *name; |
3c1dbdf5 | 2250 | u64 tmp; |
2b82032c | 2251 | int seeding_dev = 0; |
788f20eb CM |
2252 | int ret = 0; |
2253 | ||
2b82032c | 2254 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 2255 | return -EROFS; |
788f20eb | 2256 | |
a5d16333 | 2257 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 2258 | root->fs_info->bdev_holder); |
7f59203a JB |
2259 | if (IS_ERR(bdev)) |
2260 | return PTR_ERR(bdev); | |
a2135011 | 2261 | |
2b82032c YZ |
2262 | if (root->fs_info->fs_devices->seeding) { |
2263 | seeding_dev = 1; | |
2264 | down_write(&sb->s_umount); | |
2265 | mutex_lock(&uuid_mutex); | |
2266 | } | |
2267 | ||
8c8bee1d | 2268 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2269 | |
788f20eb | 2270 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
2271 | |
2272 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 2273 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2274 | if (device->bdev == bdev) { |
2275 | ret = -EEXIST; | |
d25628bd LB |
2276 | mutex_unlock( |
2277 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2278 | goto error; |
788f20eb CM |
2279 | } |
2280 | } | |
d25628bd | 2281 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2282 | |
12bd2fc0 ID |
2283 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
2284 | if (IS_ERR(device)) { | |
788f20eb | 2285 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2286 | ret = PTR_ERR(device); |
2b82032c | 2287 | goto error; |
788f20eb CM |
2288 | } |
2289 | ||
606686ee JB |
2290 | name = rcu_string_strdup(device_path, GFP_NOFS); |
2291 | if (!name) { | |
788f20eb | 2292 | kfree(device); |
2b82032c YZ |
2293 | ret = -ENOMEM; |
2294 | goto error; | |
788f20eb | 2295 | } |
606686ee | 2296 | rcu_assign_pointer(device->name, name); |
2b82032c | 2297 | |
a22285a6 | 2298 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2299 | if (IS_ERR(trans)) { |
606686ee | 2300 | rcu_string_free(device->name); |
98d5dc13 TI |
2301 | kfree(device); |
2302 | ret = PTR_ERR(trans); | |
2303 | goto error; | |
2304 | } | |
2305 | ||
d5e2003c JB |
2306 | q = bdev_get_queue(bdev); |
2307 | if (blk_queue_discard(q)) | |
2308 | device->can_discard = 1; | |
2b82032c | 2309 | device->writeable = 1; |
2b82032c | 2310 | device->generation = trans->transid; |
788f20eb CM |
2311 | device->io_width = root->sectorsize; |
2312 | device->io_align = root->sectorsize; | |
2313 | device->sector_size = root->sectorsize; | |
2314 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2315 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2316 | device->commit_total_bytes = device->total_bytes; |
788f20eb CM |
2317 | device->dev_root = root->fs_info->dev_root; |
2318 | device->bdev = bdev; | |
dfe25020 | 2319 | device->in_fs_metadata = 1; |
63a212ab | 2320 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2321 | device->mode = FMODE_EXCL; |
27087f37 | 2322 | device->dev_stats_valid = 1; |
2b82032c | 2323 | set_blocksize(device->bdev, 4096); |
788f20eb | 2324 | |
2b82032c YZ |
2325 | if (seeding_dev) { |
2326 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2327 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2328 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2329 | } |
788f20eb | 2330 | |
2b82032c | 2331 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2332 | |
e5e9a520 | 2333 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
2196d6e8 | 2334 | lock_chunks(root); |
1f78160c | 2335 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2336 | list_add(&device->dev_alloc_list, |
2337 | &root->fs_info->fs_devices->alloc_list); | |
2338 | root->fs_info->fs_devices->num_devices++; | |
2339 | root->fs_info->fs_devices->open_devices++; | |
2340 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2341 | root->fs_info->fs_devices->total_devices++; |
2b82032c | 2342 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2343 | |
2bf64758 JB |
2344 | spin_lock(&root->fs_info->free_chunk_lock); |
2345 | root->fs_info->free_chunk_space += device->total_bytes; | |
2346 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2347 | ||
c289811c CM |
2348 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2349 | root->fs_info->fs_devices->rotating = 1; | |
2350 | ||
3c1dbdf5 | 2351 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); |
6c41761f | 2352 | btrfs_set_super_total_bytes(root->fs_info->super_copy, |
3c1dbdf5 | 2353 | tmp + device->total_bytes); |
788f20eb | 2354 | |
3c1dbdf5 | 2355 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); |
6c41761f | 2356 | btrfs_set_super_num_devices(root->fs_info->super_copy, |
3c1dbdf5 | 2357 | tmp + 1); |
0d39376a AJ |
2358 | |
2359 | /* add sysfs device entry */ | |
e3bd6973 | 2360 | btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device); |
0d39376a | 2361 | |
2196d6e8 MX |
2362 | /* |
2363 | * we've got more storage, clear any full flags on the space | |
2364 | * infos | |
2365 | */ | |
2366 | btrfs_clear_space_info_full(root->fs_info); | |
2367 | ||
2368 | unlock_chunks(root); | |
e5e9a520 | 2369 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2370 | |
2b82032c | 2371 | if (seeding_dev) { |
2196d6e8 | 2372 | lock_chunks(root); |
2b82032c | 2373 | ret = init_first_rw_device(trans, root, device); |
2196d6e8 | 2374 | unlock_chunks(root); |
005d6427 DS |
2375 | if (ret) { |
2376 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2377 | goto error_trans; |
005d6427 | 2378 | } |
2196d6e8 MX |
2379 | } |
2380 | ||
2381 | ret = btrfs_add_device(trans, root, device); | |
2382 | if (ret) { | |
2383 | btrfs_abort_transaction(trans, root, ret); | |
2384 | goto error_trans; | |
2385 | } | |
2386 | ||
2387 | if (seeding_dev) { | |
2388 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2389 | ||
2b82032c | 2390 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2391 | if (ret) { |
2392 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2393 | goto error_trans; |
005d6427 | 2394 | } |
b2373f25 AJ |
2395 | |
2396 | /* Sprouting would change fsid of the mounted root, | |
2397 | * so rename the fsid on the sysfs | |
2398 | */ | |
2399 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2400 | root->fs_info->fsid); | |
c1b7e474 | 2401 | if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj, |
2e7910d6 | 2402 | fsid_buf)) |
f14d104d DS |
2403 | btrfs_warn(root->fs_info, |
2404 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2405 | } |
2406 | ||
5af3e8cc SB |
2407 | root->fs_info->num_tolerated_disk_barrier_failures = |
2408 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2409 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2410 | |
2b82032c YZ |
2411 | if (seeding_dev) { |
2412 | mutex_unlock(&uuid_mutex); | |
2413 | up_write(&sb->s_umount); | |
788f20eb | 2414 | |
79787eaa JM |
2415 | if (ret) /* transaction commit */ |
2416 | return ret; | |
2417 | ||
2b82032c | 2418 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa | 2419 | if (ret < 0) |
a4553fef | 2420 | btrfs_std_error(root->fs_info, ret, |
79787eaa JM |
2421 | "Failed to relocate sys chunks after " |
2422 | "device initialization. This can be fixed " | |
2423 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2424 | trans = btrfs_attach_transaction(root); |
2425 | if (IS_ERR(trans)) { | |
2426 | if (PTR_ERR(trans) == -ENOENT) | |
2427 | return 0; | |
2428 | return PTR_ERR(trans); | |
2429 | } | |
2430 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2431 | } |
c9e9f97b | 2432 | |
5a1972bd QW |
2433 | /* Update ctime/mtime for libblkid */ |
2434 | update_dev_time(device_path); | |
2b82032c | 2435 | return ret; |
79787eaa JM |
2436 | |
2437 | error_trans: | |
79787eaa | 2438 | btrfs_end_transaction(trans, root); |
606686ee | 2439 | rcu_string_free(device->name); |
32576040 | 2440 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); |
79787eaa | 2441 | kfree(device); |
2b82032c | 2442 | error: |
e525fd89 | 2443 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2444 | if (seeding_dev) { |
2445 | mutex_unlock(&uuid_mutex); | |
2446 | up_write(&sb->s_umount); | |
2447 | } | |
c9e9f97b | 2448 | return ret; |
788f20eb CM |
2449 | } |
2450 | ||
e93c89c1 | 2451 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2452 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2453 | struct btrfs_device **device_out) |
2454 | { | |
2455 | struct request_queue *q; | |
2456 | struct btrfs_device *device; | |
2457 | struct block_device *bdev; | |
2458 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2459 | struct list_head *devices; | |
2460 | struct rcu_string *name; | |
12bd2fc0 | 2461 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2462 | int ret = 0; |
2463 | ||
2464 | *device_out = NULL; | |
1c43366d MX |
2465 | if (fs_info->fs_devices->seeding) { |
2466 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2467 | return -EINVAL; |
1c43366d | 2468 | } |
e93c89c1 SB |
2469 | |
2470 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2471 | fs_info->bdev_holder); | |
1c43366d MX |
2472 | if (IS_ERR(bdev)) { |
2473 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2474 | return PTR_ERR(bdev); |
1c43366d | 2475 | } |
e93c89c1 SB |
2476 | |
2477 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2478 | ||
2479 | devices = &fs_info->fs_devices->devices; | |
2480 | list_for_each_entry(device, devices, dev_list) { | |
2481 | if (device->bdev == bdev) { | |
1c43366d | 2482 | btrfs_err(fs_info, "target device is in the filesystem!"); |
e93c89c1 SB |
2483 | ret = -EEXIST; |
2484 | goto error; | |
2485 | } | |
2486 | } | |
2487 | ||
1c43366d | 2488 | |
7cc8e58d MX |
2489 | if (i_size_read(bdev->bd_inode) < |
2490 | btrfs_device_get_total_bytes(srcdev)) { | |
1c43366d MX |
2491 | btrfs_err(fs_info, "target device is smaller than source device!"); |
2492 | ret = -EINVAL; | |
2493 | goto error; | |
2494 | } | |
2495 | ||
2496 | ||
12bd2fc0 ID |
2497 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2498 | if (IS_ERR(device)) { | |
2499 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2500 | goto error; |
2501 | } | |
2502 | ||
2503 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2504 | if (!name) { | |
2505 | kfree(device); | |
2506 | ret = -ENOMEM; | |
2507 | goto error; | |
2508 | } | |
2509 | rcu_assign_pointer(device->name, name); | |
2510 | ||
2511 | q = bdev_get_queue(bdev); | |
2512 | if (blk_queue_discard(q)) | |
2513 | device->can_discard = 1; | |
2514 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2515 | device->writeable = 1; | |
e93c89c1 SB |
2516 | device->generation = 0; |
2517 | device->io_width = root->sectorsize; | |
2518 | device->io_align = root->sectorsize; | |
2519 | device->sector_size = root->sectorsize; | |
7cc8e58d MX |
2520 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2521 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2522 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2523 | ASSERT(list_empty(&srcdev->resized_list)); |
2524 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2525 | device->commit_bytes_used = device->bytes_used; |
e93c89c1 SB |
2526 | device->dev_root = fs_info->dev_root; |
2527 | device->bdev = bdev; | |
2528 | device->in_fs_metadata = 1; | |
2529 | device->is_tgtdev_for_dev_replace = 1; | |
2530 | device->mode = FMODE_EXCL; | |
27087f37 | 2531 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2532 | set_blocksize(device->bdev, 4096); |
2533 | device->fs_devices = fs_info->fs_devices; | |
2534 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2535 | fs_info->fs_devices->num_devices++; | |
2536 | fs_info->fs_devices->open_devices++; | |
e93c89c1 SB |
2537 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2538 | ||
2539 | *device_out = device; | |
2540 | return ret; | |
2541 | ||
2542 | error: | |
2543 | blkdev_put(bdev, FMODE_EXCL); | |
2544 | return ret; | |
2545 | } | |
2546 | ||
2547 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2548 | struct btrfs_device *tgtdev) | |
2549 | { | |
2550 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2551 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2552 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2553 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2554 | tgtdev->dev_root = fs_info->dev_root; | |
2555 | tgtdev->in_fs_metadata = 1; | |
2556 | } | |
2557 | ||
d397712b CM |
2558 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2559 | struct btrfs_device *device) | |
0b86a832 CM |
2560 | { |
2561 | int ret; | |
2562 | struct btrfs_path *path; | |
2563 | struct btrfs_root *root; | |
2564 | struct btrfs_dev_item *dev_item; | |
2565 | struct extent_buffer *leaf; | |
2566 | struct btrfs_key key; | |
2567 | ||
2568 | root = device->dev_root->fs_info->chunk_root; | |
2569 | ||
2570 | path = btrfs_alloc_path(); | |
2571 | if (!path) | |
2572 | return -ENOMEM; | |
2573 | ||
2574 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2575 | key.type = BTRFS_DEV_ITEM_KEY; | |
2576 | key.offset = device->devid; | |
2577 | ||
2578 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2579 | if (ret < 0) | |
2580 | goto out; | |
2581 | ||
2582 | if (ret > 0) { | |
2583 | ret = -ENOENT; | |
2584 | goto out; | |
2585 | } | |
2586 | ||
2587 | leaf = path->nodes[0]; | |
2588 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2589 | ||
2590 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2591 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2592 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2593 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2594 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2595 | btrfs_set_device_total_bytes(leaf, dev_item, |
2596 | btrfs_device_get_disk_total_bytes(device)); | |
2597 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2598 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2599 | btrfs_mark_buffer_dirty(leaf); |
2600 | ||
2601 | out: | |
2602 | btrfs_free_path(path); | |
2603 | return ret; | |
2604 | } | |
2605 | ||
2196d6e8 | 2606 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2607 | struct btrfs_device *device, u64 new_size) |
2608 | { | |
2609 | struct btrfs_super_block *super_copy = | |
6c41761f | 2610 | device->dev_root->fs_info->super_copy; |
935e5cc9 | 2611 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2612 | u64 old_total; |
2613 | u64 diff; | |
8f18cf13 | 2614 | |
2b82032c YZ |
2615 | if (!device->writeable) |
2616 | return -EACCES; | |
2196d6e8 MX |
2617 | |
2618 | lock_chunks(device->dev_root); | |
2619 | old_total = btrfs_super_total_bytes(super_copy); | |
2620 | diff = new_size - device->total_bytes; | |
2621 | ||
63a212ab | 2622 | if (new_size <= device->total_bytes || |
2196d6e8 MX |
2623 | device->is_tgtdev_for_dev_replace) { |
2624 | unlock_chunks(device->dev_root); | |
2b82032c | 2625 | return -EINVAL; |
2196d6e8 | 2626 | } |
2b82032c | 2627 | |
935e5cc9 | 2628 | fs_devices = device->dev_root->fs_info->fs_devices; |
2b82032c | 2629 | |
8f18cf13 | 2630 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2631 | device->fs_devices->total_rw_bytes += diff; |
2632 | ||
7cc8e58d MX |
2633 | btrfs_device_set_total_bytes(device, new_size); |
2634 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4184ea7f | 2635 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
935e5cc9 MX |
2636 | if (list_empty(&device->resized_list)) |
2637 | list_add_tail(&device->resized_list, | |
2638 | &fs_devices->resized_devices); | |
2196d6e8 | 2639 | unlock_chunks(device->dev_root); |
4184ea7f | 2640 | |
8f18cf13 CM |
2641 | return btrfs_update_device(trans, device); |
2642 | } | |
2643 | ||
2644 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
a688a04a | 2645 | struct btrfs_root *root, u64 chunk_objectid, |
8f18cf13 CM |
2646 | u64 chunk_offset) |
2647 | { | |
2648 | int ret; | |
2649 | struct btrfs_path *path; | |
2650 | struct btrfs_key key; | |
2651 | ||
2652 | root = root->fs_info->chunk_root; | |
2653 | path = btrfs_alloc_path(); | |
2654 | if (!path) | |
2655 | return -ENOMEM; | |
2656 | ||
2657 | key.objectid = chunk_objectid; | |
2658 | key.offset = chunk_offset; | |
2659 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2660 | ||
2661 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2662 | if (ret < 0) |
2663 | goto out; | |
2664 | else if (ret > 0) { /* Logic error or corruption */ | |
a4553fef | 2665 | btrfs_std_error(root->fs_info, -ENOENT, |
79787eaa JM |
2666 | "Failed lookup while freeing chunk."); |
2667 | ret = -ENOENT; | |
2668 | goto out; | |
2669 | } | |
8f18cf13 CM |
2670 | |
2671 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2672 | if (ret < 0) |
a4553fef | 2673 | btrfs_std_error(root->fs_info, ret, |
79787eaa JM |
2674 | "Failed to delete chunk item."); |
2675 | out: | |
8f18cf13 | 2676 | btrfs_free_path(path); |
65a246c5 | 2677 | return ret; |
8f18cf13 CM |
2678 | } |
2679 | ||
b2950863 | 2680 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2681 | chunk_offset) |
2682 | { | |
6c41761f | 2683 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2684 | struct btrfs_disk_key *disk_key; |
2685 | struct btrfs_chunk *chunk; | |
2686 | u8 *ptr; | |
2687 | int ret = 0; | |
2688 | u32 num_stripes; | |
2689 | u32 array_size; | |
2690 | u32 len = 0; | |
2691 | u32 cur; | |
2692 | struct btrfs_key key; | |
2693 | ||
2196d6e8 | 2694 | lock_chunks(root); |
8f18cf13 CM |
2695 | array_size = btrfs_super_sys_array_size(super_copy); |
2696 | ||
2697 | ptr = super_copy->sys_chunk_array; | |
2698 | cur = 0; | |
2699 | ||
2700 | while (cur < array_size) { | |
2701 | disk_key = (struct btrfs_disk_key *)ptr; | |
2702 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2703 | ||
2704 | len = sizeof(*disk_key); | |
2705 | ||
2706 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2707 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2708 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2709 | len += btrfs_chunk_item_size(num_stripes); | |
2710 | } else { | |
2711 | ret = -EIO; | |
2712 | break; | |
2713 | } | |
2714 | if (key.objectid == chunk_objectid && | |
2715 | key.offset == chunk_offset) { | |
2716 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2717 | array_size -= len; | |
2718 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2719 | } else { | |
2720 | ptr += len; | |
2721 | cur += len; | |
2722 | } | |
2723 | } | |
2196d6e8 | 2724 | unlock_chunks(root); |
8f18cf13 CM |
2725 | return ret; |
2726 | } | |
2727 | ||
47ab2a6c JB |
2728 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
2729 | struct btrfs_root *root, u64 chunk_offset) | |
8f18cf13 CM |
2730 | { |
2731 | struct extent_map_tree *em_tree; | |
8f18cf13 | 2732 | struct extent_map *em; |
47ab2a6c | 2733 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
8f18cf13 | 2734 | struct map_lookup *map; |
2196d6e8 | 2735 | u64 dev_extent_len = 0; |
47ab2a6c | 2736 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2737 | int i, ret = 0; |
8f18cf13 | 2738 | |
47ab2a6c | 2739 | /* Just in case */ |
8f18cf13 | 2740 | root = root->fs_info->chunk_root; |
8f18cf13 CM |
2741 | em_tree = &root->fs_info->mapping_tree.map_tree; |
2742 | ||
890871be | 2743 | read_lock(&em_tree->lock); |
8f18cf13 | 2744 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2745 | read_unlock(&em_tree->lock); |
8f18cf13 | 2746 | |
47ab2a6c JB |
2747 | if (!em || em->start > chunk_offset || |
2748 | em->start + em->len < chunk_offset) { | |
2749 | /* | |
2750 | * This is a logic error, but we don't want to just rely on the | |
2751 | * user having built with ASSERT enabled, so if ASSERT doens't | |
2752 | * do anything we still error out. | |
2753 | */ | |
2754 | ASSERT(0); | |
2755 | if (em) | |
2756 | free_extent_map(em); | |
2757 | return -EINVAL; | |
2758 | } | |
8f18cf13 | 2759 | map = (struct map_lookup *)em->bdev; |
39c2d7fa | 2760 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 2761 | check_system_chunk(trans, extent_root, map->type); |
39c2d7fa | 2762 | unlock_chunks(root->fs_info->chunk_root); |
8f18cf13 CM |
2763 | |
2764 | for (i = 0; i < map->num_stripes; i++) { | |
47ab2a6c | 2765 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2766 | ret = btrfs_free_dev_extent(trans, device, |
2767 | map->stripes[i].physical, | |
2768 | &dev_extent_len); | |
47ab2a6c JB |
2769 | if (ret) { |
2770 | btrfs_abort_transaction(trans, root, ret); | |
2771 | goto out; | |
2772 | } | |
a061fc8d | 2773 | |
2196d6e8 MX |
2774 | if (device->bytes_used > 0) { |
2775 | lock_chunks(root); | |
2776 | btrfs_device_set_bytes_used(device, | |
2777 | device->bytes_used - dev_extent_len); | |
2778 | spin_lock(&root->fs_info->free_chunk_lock); | |
2779 | root->fs_info->free_chunk_space += dev_extent_len; | |
2780 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2781 | btrfs_clear_space_info_full(root->fs_info); | |
2782 | unlock_chunks(root); | |
2783 | } | |
a061fc8d | 2784 | |
dfe25020 CM |
2785 | if (map->stripes[i].dev) { |
2786 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c JB |
2787 | if (ret) { |
2788 | btrfs_abort_transaction(trans, root, ret); | |
2789 | goto out; | |
2790 | } | |
dfe25020 | 2791 | } |
8f18cf13 | 2792 | } |
a688a04a | 2793 | ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset); |
47ab2a6c JB |
2794 | if (ret) { |
2795 | btrfs_abort_transaction(trans, root, ret); | |
2796 | goto out; | |
2797 | } | |
8f18cf13 | 2798 | |
1abe9b8a | 2799 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2800 | ||
8f18cf13 CM |
2801 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2802 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
47ab2a6c JB |
2803 | if (ret) { |
2804 | btrfs_abort_transaction(trans, root, ret); | |
2805 | goto out; | |
2806 | } | |
8f18cf13 CM |
2807 | } |
2808 | ||
04216820 | 2809 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em); |
47ab2a6c JB |
2810 | if (ret) { |
2811 | btrfs_abort_transaction(trans, extent_root, ret); | |
2812 | goto out; | |
2813 | } | |
2b82032c | 2814 | |
47ab2a6c | 2815 | out: |
2b82032c YZ |
2816 | /* once for us */ |
2817 | free_extent_map(em); | |
47ab2a6c JB |
2818 | return ret; |
2819 | } | |
2b82032c | 2820 | |
dc2ee4e2 | 2821 | static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset) |
47ab2a6c JB |
2822 | { |
2823 | struct btrfs_root *extent_root; | |
2824 | struct btrfs_trans_handle *trans; | |
2825 | int ret; | |
2b82032c | 2826 | |
47ab2a6c JB |
2827 | root = root->fs_info->chunk_root; |
2828 | extent_root = root->fs_info->extent_root; | |
2829 | ||
67c5e7d4 FM |
2830 | /* |
2831 | * Prevent races with automatic removal of unused block groups. | |
2832 | * After we relocate and before we remove the chunk with offset | |
2833 | * chunk_offset, automatic removal of the block group can kick in, | |
2834 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2835 | * | |
2836 | * Make sure to acquire this mutex before doing a tree search (dev | |
2837 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2838 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2839 | * we release the path used to search the chunk/dev tree and before | |
2840 | * the current task acquires this mutex and calls us. | |
2841 | */ | |
2842 | ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex)); | |
2843 | ||
47ab2a6c JB |
2844 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2845 | if (ret) | |
2846 | return -ENOSPC; | |
2847 | ||
2848 | /* step one, relocate all the extents inside this chunk */ | |
55e3a601 | 2849 | btrfs_scrub_pause(root); |
47ab2a6c | 2850 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
55e3a601 | 2851 | btrfs_scrub_continue(root); |
47ab2a6c JB |
2852 | if (ret) |
2853 | return ret; | |
2854 | ||
7fd01182 FM |
2855 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2856 | chunk_offset); | |
47ab2a6c JB |
2857 | if (IS_ERR(trans)) { |
2858 | ret = PTR_ERR(trans); | |
a4553fef | 2859 | btrfs_std_error(root->fs_info, ret, NULL); |
47ab2a6c JB |
2860 | return ret; |
2861 | } | |
2862 | ||
2863 | /* | |
2864 | * step two, delete the device extents and the | |
2865 | * chunk tree entries | |
2866 | */ | |
2867 | ret = btrfs_remove_chunk(trans, root, chunk_offset); | |
2b82032c | 2868 | btrfs_end_transaction(trans, root); |
47ab2a6c | 2869 | return ret; |
2b82032c YZ |
2870 | } |
2871 | ||
2872 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2873 | { | |
2874 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2875 | struct btrfs_path *path; | |
2876 | struct extent_buffer *leaf; | |
2877 | struct btrfs_chunk *chunk; | |
2878 | struct btrfs_key key; | |
2879 | struct btrfs_key found_key; | |
2b82032c | 2880 | u64 chunk_type; |
ba1bf481 JB |
2881 | bool retried = false; |
2882 | int failed = 0; | |
2b82032c YZ |
2883 | int ret; |
2884 | ||
2885 | path = btrfs_alloc_path(); | |
2886 | if (!path) | |
2887 | return -ENOMEM; | |
2888 | ||
ba1bf481 | 2889 | again: |
2b82032c YZ |
2890 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2891 | key.offset = (u64)-1; | |
2892 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2893 | ||
2894 | while (1) { | |
67c5e7d4 | 2895 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
2b82032c | 2896 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
2897 | if (ret < 0) { |
2898 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c | 2899 | goto error; |
67c5e7d4 | 2900 | } |
79787eaa | 2901 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2902 | |
2903 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2904 | key.type); | |
67c5e7d4 FM |
2905 | if (ret) |
2906 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c YZ |
2907 | if (ret < 0) |
2908 | goto error; | |
2909 | if (ret > 0) | |
2910 | break; | |
1a40e23b | 2911 | |
2b82032c YZ |
2912 | leaf = path->nodes[0]; |
2913 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2914 | |
2b82032c YZ |
2915 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2916 | struct btrfs_chunk); | |
2917 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2918 | btrfs_release_path(path); |
8f18cf13 | 2919 | |
2b82032c | 2920 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
a688a04a | 2921 | ret = btrfs_relocate_chunk(chunk_root, |
2b82032c | 2922 | found_key.offset); |
ba1bf481 JB |
2923 | if (ret == -ENOSPC) |
2924 | failed++; | |
14586651 HS |
2925 | else |
2926 | BUG_ON(ret); | |
2b82032c | 2927 | } |
67c5e7d4 | 2928 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 2929 | |
2b82032c YZ |
2930 | if (found_key.offset == 0) |
2931 | break; | |
2932 | key.offset = found_key.offset - 1; | |
2933 | } | |
2934 | ret = 0; | |
ba1bf481 JB |
2935 | if (failed && !retried) { |
2936 | failed = 0; | |
2937 | retried = true; | |
2938 | goto again; | |
fae7f21c | 2939 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2940 | ret = -ENOSPC; |
2941 | } | |
2b82032c YZ |
2942 | error: |
2943 | btrfs_free_path(path); | |
2944 | return ret; | |
8f18cf13 CM |
2945 | } |
2946 | ||
0940ebf6 ID |
2947 | static int insert_balance_item(struct btrfs_root *root, |
2948 | struct btrfs_balance_control *bctl) | |
2949 | { | |
2950 | struct btrfs_trans_handle *trans; | |
2951 | struct btrfs_balance_item *item; | |
2952 | struct btrfs_disk_balance_args disk_bargs; | |
2953 | struct btrfs_path *path; | |
2954 | struct extent_buffer *leaf; | |
2955 | struct btrfs_key key; | |
2956 | int ret, err; | |
2957 | ||
2958 | path = btrfs_alloc_path(); | |
2959 | if (!path) | |
2960 | return -ENOMEM; | |
2961 | ||
2962 | trans = btrfs_start_transaction(root, 0); | |
2963 | if (IS_ERR(trans)) { | |
2964 | btrfs_free_path(path); | |
2965 | return PTR_ERR(trans); | |
2966 | } | |
2967 | ||
2968 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2969 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2970 | key.offset = 0; | |
2971 | ||
2972 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2973 | sizeof(*item)); | |
2974 | if (ret) | |
2975 | goto out; | |
2976 | ||
2977 | leaf = path->nodes[0]; | |
2978 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2979 | ||
2980 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2981 | ||
2982 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2983 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2984 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2985 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2986 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2987 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2988 | ||
2989 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2990 | ||
2991 | btrfs_mark_buffer_dirty(leaf); | |
2992 | out: | |
2993 | btrfs_free_path(path); | |
2994 | err = btrfs_commit_transaction(trans, root); | |
2995 | if (err && !ret) | |
2996 | ret = err; | |
2997 | return ret; | |
2998 | } | |
2999 | ||
3000 | static int del_balance_item(struct btrfs_root *root) | |
3001 | { | |
3002 | struct btrfs_trans_handle *trans; | |
3003 | struct btrfs_path *path; | |
3004 | struct btrfs_key key; | |
3005 | int ret, err; | |
3006 | ||
3007 | path = btrfs_alloc_path(); | |
3008 | if (!path) | |
3009 | return -ENOMEM; | |
3010 | ||
3011 | trans = btrfs_start_transaction(root, 0); | |
3012 | if (IS_ERR(trans)) { | |
3013 | btrfs_free_path(path); | |
3014 | return PTR_ERR(trans); | |
3015 | } | |
3016 | ||
3017 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
3018 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3019 | key.offset = 0; | |
3020 | ||
3021 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3022 | if (ret < 0) | |
3023 | goto out; | |
3024 | if (ret > 0) { | |
3025 | ret = -ENOENT; | |
3026 | goto out; | |
3027 | } | |
3028 | ||
3029 | ret = btrfs_del_item(trans, root, path); | |
3030 | out: | |
3031 | btrfs_free_path(path); | |
3032 | err = btrfs_commit_transaction(trans, root); | |
3033 | if (err && !ret) | |
3034 | ret = err; | |
3035 | return ret; | |
3036 | } | |
3037 | ||
59641015 ID |
3038 | /* |
3039 | * This is a heuristic used to reduce the number of chunks balanced on | |
3040 | * resume after balance was interrupted. | |
3041 | */ | |
3042 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3043 | { | |
3044 | /* | |
3045 | * Turn on soft mode for chunk types that were being converted. | |
3046 | */ | |
3047 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3048 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3049 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3050 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3051 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3052 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3053 | ||
3054 | /* | |
3055 | * Turn on usage filter if is not already used. The idea is | |
3056 | * that chunks that we have already balanced should be | |
3057 | * reasonably full. Don't do it for chunks that are being | |
3058 | * converted - that will keep us from relocating unconverted | |
3059 | * (albeit full) chunks. | |
3060 | */ | |
3061 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3062 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3063 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3064 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3065 | bctl->data.usage = 90; | |
3066 | } | |
3067 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3068 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3069 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3070 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3071 | bctl->sys.usage = 90; | |
3072 | } | |
3073 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3074 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3075 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3076 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3077 | bctl->meta.usage = 90; | |
3078 | } | |
3079 | } | |
3080 | ||
c9e9f97b ID |
3081 | /* |
3082 | * Should be called with both balance and volume mutexes held to | |
3083 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3084 | * restriper. Same goes for unset_balance_control. | |
3085 | */ | |
3086 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3087 | { | |
3088 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3089 | ||
3090 | BUG_ON(fs_info->balance_ctl); | |
3091 | ||
3092 | spin_lock(&fs_info->balance_lock); | |
3093 | fs_info->balance_ctl = bctl; | |
3094 | spin_unlock(&fs_info->balance_lock); | |
3095 | } | |
3096 | ||
3097 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3098 | { | |
3099 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3100 | ||
3101 | BUG_ON(!fs_info->balance_ctl); | |
3102 | ||
3103 | spin_lock(&fs_info->balance_lock); | |
3104 | fs_info->balance_ctl = NULL; | |
3105 | spin_unlock(&fs_info->balance_lock); | |
3106 | ||
3107 | kfree(bctl); | |
3108 | } | |
3109 | ||
ed25e9b2 ID |
3110 | /* |
3111 | * Balance filters. Return 1 if chunk should be filtered out | |
3112 | * (should not be balanced). | |
3113 | */ | |
899c81ea | 3114 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3115 | struct btrfs_balance_args *bargs) |
3116 | { | |
899c81ea ID |
3117 | chunk_type = chunk_to_extended(chunk_type) & |
3118 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3119 | |
899c81ea | 3120 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3121 | return 0; |
3122 | ||
3123 | return 1; | |
3124 | } | |
3125 | ||
dba72cb3 | 3126 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3127 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3128 | { |
3129 | struct btrfs_block_group_cache *cache; | |
3130 | u64 chunk_used; | |
3131 | u64 user_thresh_min; | |
3132 | u64 user_thresh_max; | |
3133 | int ret = 1; | |
3134 | ||
3135 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3136 | chunk_used = btrfs_block_group_used(&cache->item); | |
3137 | ||
3138 | if (bargs->usage_min == 0) | |
3139 | user_thresh_min = 0; | |
3140 | else | |
3141 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3142 | bargs->usage_min); | |
3143 | ||
3144 | if (bargs->usage_max == 0) | |
3145 | user_thresh_max = 1; | |
3146 | else if (bargs->usage_max > 100) | |
3147 | user_thresh_max = cache->key.offset; | |
3148 | else | |
3149 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3150 | bargs->usage_max); | |
3151 | ||
3152 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3153 | ret = 0; | |
3154 | ||
3155 | btrfs_put_block_group(cache); | |
3156 | return ret; | |
3157 | } | |
3158 | ||
dba72cb3 | 3159 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3160 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3161 | { |
3162 | struct btrfs_block_group_cache *cache; | |
3163 | u64 chunk_used, user_thresh; | |
3164 | int ret = 1; | |
3165 | ||
3166 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3167 | chunk_used = btrfs_block_group_used(&cache->item); | |
3168 | ||
bc309467 | 3169 | if (bargs->usage_min == 0) |
3e39cea6 | 3170 | user_thresh = 1; |
a105bb88 ID |
3171 | else if (bargs->usage > 100) |
3172 | user_thresh = cache->key.offset; | |
3173 | else | |
3174 | user_thresh = div_factor_fine(cache->key.offset, | |
3175 | bargs->usage); | |
3176 | ||
5ce5b3c0 ID |
3177 | if (chunk_used < user_thresh) |
3178 | ret = 0; | |
3179 | ||
3180 | btrfs_put_block_group(cache); | |
3181 | return ret; | |
3182 | } | |
3183 | ||
409d404b ID |
3184 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3185 | struct btrfs_chunk *chunk, | |
3186 | struct btrfs_balance_args *bargs) | |
3187 | { | |
3188 | struct btrfs_stripe *stripe; | |
3189 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3190 | int i; | |
3191 | ||
3192 | for (i = 0; i < num_stripes; i++) { | |
3193 | stripe = btrfs_stripe_nr(chunk, i); | |
3194 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3195 | return 0; | |
3196 | } | |
3197 | ||
3198 | return 1; | |
3199 | } | |
3200 | ||
94e60d5a ID |
3201 | /* [pstart, pend) */ |
3202 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3203 | struct btrfs_chunk *chunk, | |
3204 | u64 chunk_offset, | |
3205 | struct btrfs_balance_args *bargs) | |
3206 | { | |
3207 | struct btrfs_stripe *stripe; | |
3208 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3209 | u64 stripe_offset; | |
3210 | u64 stripe_length; | |
3211 | int factor; | |
3212 | int i; | |
3213 | ||
3214 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3215 | return 0; | |
3216 | ||
3217 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3218 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3219 | factor = num_stripes / 2; | |
3220 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3221 | factor = num_stripes - 1; | |
3222 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3223 | factor = num_stripes - 2; | |
3224 | } else { | |
3225 | factor = num_stripes; | |
3226 | } | |
94e60d5a ID |
3227 | |
3228 | for (i = 0; i < num_stripes; i++) { | |
3229 | stripe = btrfs_stripe_nr(chunk, i); | |
3230 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3231 | continue; | |
3232 | ||
3233 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3234 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3235 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3236 | |
3237 | if (stripe_offset < bargs->pend && | |
3238 | stripe_offset + stripe_length > bargs->pstart) | |
3239 | return 0; | |
3240 | } | |
3241 | ||
3242 | return 1; | |
3243 | } | |
3244 | ||
ea67176a ID |
3245 | /* [vstart, vend) */ |
3246 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3247 | struct btrfs_chunk *chunk, | |
3248 | u64 chunk_offset, | |
3249 | struct btrfs_balance_args *bargs) | |
3250 | { | |
3251 | if (chunk_offset < bargs->vend && | |
3252 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3253 | /* at least part of the chunk is inside this vrange */ | |
3254 | return 0; | |
3255 | ||
3256 | return 1; | |
3257 | } | |
3258 | ||
dee32d0a GAP |
3259 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3260 | struct btrfs_chunk *chunk, | |
3261 | struct btrfs_balance_args *bargs) | |
3262 | { | |
3263 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3264 | ||
3265 | if (bargs->stripes_min <= num_stripes | |
3266 | && num_stripes <= bargs->stripes_max) | |
3267 | return 0; | |
3268 | ||
3269 | return 1; | |
3270 | } | |
3271 | ||
899c81ea | 3272 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3273 | struct btrfs_balance_args *bargs) |
3274 | { | |
3275 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3276 | return 0; | |
3277 | ||
899c81ea ID |
3278 | chunk_type = chunk_to_extended(chunk_type) & |
3279 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3280 | |
899c81ea | 3281 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3282 | return 1; |
3283 | ||
3284 | return 0; | |
3285 | } | |
3286 | ||
f43ffb60 ID |
3287 | static int should_balance_chunk(struct btrfs_root *root, |
3288 | struct extent_buffer *leaf, | |
3289 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3290 | { | |
3291 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3292 | struct btrfs_balance_args *bargs = NULL; | |
3293 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3294 | ||
3295 | /* type filter */ | |
3296 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3297 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3298 | return 0; | |
3299 | } | |
3300 | ||
3301 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3302 | bargs = &bctl->data; | |
3303 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3304 | bargs = &bctl->sys; | |
3305 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3306 | bargs = &bctl->meta; | |
3307 | ||
ed25e9b2 ID |
3308 | /* profiles filter */ |
3309 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3310 | chunk_profiles_filter(chunk_type, bargs)) { | |
3311 | return 0; | |
5ce5b3c0 ID |
3312 | } |
3313 | ||
3314 | /* usage filter */ | |
3315 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3316 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3317 | return 0; | |
bc309467 DS |
3318 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
3319 | chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3320 | return 0; | |
409d404b ID |
3321 | } |
3322 | ||
3323 | /* devid filter */ | |
3324 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3325 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3326 | return 0; | |
94e60d5a ID |
3327 | } |
3328 | ||
3329 | /* drange filter, makes sense only with devid filter */ | |
3330 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3331 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3332 | return 0; | |
ea67176a ID |
3333 | } |
3334 | ||
3335 | /* vrange filter */ | |
3336 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3337 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3338 | return 0; | |
ed25e9b2 ID |
3339 | } |
3340 | ||
dee32d0a GAP |
3341 | /* stripes filter */ |
3342 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3343 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3344 | return 0; | |
3345 | } | |
3346 | ||
cfa4c961 ID |
3347 | /* soft profile changing mode */ |
3348 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3349 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3350 | return 0; | |
3351 | } | |
3352 | ||
7d824b6f DS |
3353 | /* |
3354 | * limited by count, must be the last filter | |
3355 | */ | |
3356 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3357 | if (bargs->limit == 0) | |
3358 | return 0; | |
3359 | else | |
3360 | bargs->limit--; | |
12907fc7 DS |
3361 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3362 | /* | |
3363 | * Same logic as the 'limit' filter; the minimum cannot be | |
3364 | * determined here because we do not have the global informatoin | |
3365 | * about the count of all chunks that satisfy the filters. | |
3366 | */ | |
3367 | if (bargs->limit_max == 0) | |
3368 | return 0; | |
3369 | else | |
3370 | bargs->limit_max--; | |
7d824b6f DS |
3371 | } |
3372 | ||
f43ffb60 ID |
3373 | return 1; |
3374 | } | |
3375 | ||
c9e9f97b | 3376 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3377 | { |
19a39dce | 3378 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3379 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3380 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3381 | struct list_head *devices; | |
ec44a35c CM |
3382 | struct btrfs_device *device; |
3383 | u64 old_size; | |
3384 | u64 size_to_free; | |
12907fc7 | 3385 | u64 chunk_type; |
f43ffb60 | 3386 | struct btrfs_chunk *chunk; |
ec44a35c CM |
3387 | struct btrfs_path *path; |
3388 | struct btrfs_key key; | |
ec44a35c | 3389 | struct btrfs_key found_key; |
c9e9f97b | 3390 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3391 | struct extent_buffer *leaf; |
3392 | int slot; | |
c9e9f97b ID |
3393 | int ret; |
3394 | int enospc_errors = 0; | |
19a39dce | 3395 | bool counting = true; |
12907fc7 | 3396 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3397 | u64 limit_data = bctl->data.limit; |
3398 | u64 limit_meta = bctl->meta.limit; | |
3399 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3400 | u32 count_data = 0; |
3401 | u32 count_meta = 0; | |
3402 | u32 count_sys = 0; | |
2c9fe835 | 3403 | int chunk_reserved = 0; |
ec44a35c | 3404 | |
ec44a35c | 3405 | /* step one make some room on all the devices */ |
c9e9f97b | 3406 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3407 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3408 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c CM |
3409 | size_to_free = div_factor(old_size, 1); |
3410 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c | 3411 | if (!device->writeable || |
7cc8e58d MX |
3412 | btrfs_device_get_total_bytes(device) - |
3413 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3414 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3415 | continue; |
3416 | ||
3417 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3418 | if (ret == -ENOSPC) |
3419 | break; | |
ec44a35c CM |
3420 | BUG_ON(ret); |
3421 | ||
a22285a6 | 3422 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 3423 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
3424 | |
3425 | ret = btrfs_grow_device(trans, device, old_size); | |
3426 | BUG_ON(ret); | |
3427 | ||
3428 | btrfs_end_transaction(trans, dev_root); | |
3429 | } | |
3430 | ||
3431 | /* step two, relocate all the chunks */ | |
3432 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3433 | if (!path) { |
3434 | ret = -ENOMEM; | |
3435 | goto error; | |
3436 | } | |
19a39dce ID |
3437 | |
3438 | /* zero out stat counters */ | |
3439 | spin_lock(&fs_info->balance_lock); | |
3440 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3441 | spin_unlock(&fs_info->balance_lock); | |
3442 | again: | |
7d824b6f | 3443 | if (!counting) { |
12907fc7 DS |
3444 | /* |
3445 | * The single value limit and min/max limits use the same bytes | |
3446 | * in the | |
3447 | */ | |
7d824b6f DS |
3448 | bctl->data.limit = limit_data; |
3449 | bctl->meta.limit = limit_meta; | |
3450 | bctl->sys.limit = limit_sys; | |
3451 | } | |
ec44a35c CM |
3452 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3453 | key.offset = (u64)-1; | |
3454 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3455 | ||
d397712b | 3456 | while (1) { |
19a39dce | 3457 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3458 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3459 | ret = -ECANCELED; |
3460 | goto error; | |
3461 | } | |
3462 | ||
67c5e7d4 | 3463 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3464 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3465 | if (ret < 0) { |
3466 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3467 | goto error; |
67c5e7d4 | 3468 | } |
ec44a35c CM |
3469 | |
3470 | /* | |
3471 | * this shouldn't happen, it means the last relocate | |
3472 | * failed | |
3473 | */ | |
3474 | if (ret == 0) | |
c9e9f97b | 3475 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3476 | |
3477 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3478 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3479 | if (ret) { |
67c5e7d4 | 3480 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3481 | ret = 0; |
ec44a35c | 3482 | break; |
c9e9f97b | 3483 | } |
7d9eb12c | 3484 | |
f43ffb60 ID |
3485 | leaf = path->nodes[0]; |
3486 | slot = path->slots[0]; | |
3487 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3488 | |
67c5e7d4 FM |
3489 | if (found_key.objectid != key.objectid) { |
3490 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3491 | break; |
67c5e7d4 | 3492 | } |
7d9eb12c | 3493 | |
f43ffb60 | 3494 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3495 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3496 | |
19a39dce ID |
3497 | if (!counting) { |
3498 | spin_lock(&fs_info->balance_lock); | |
3499 | bctl->stat.considered++; | |
3500 | spin_unlock(&fs_info->balance_lock); | |
3501 | } | |
3502 | ||
f43ffb60 ID |
3503 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3504 | found_key.offset); | |
2c9fe835 | 3505 | |
b3b4aa74 | 3506 | btrfs_release_path(path); |
67c5e7d4 FM |
3507 | if (!ret) { |
3508 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3509 | goto loop; |
67c5e7d4 | 3510 | } |
f43ffb60 | 3511 | |
19a39dce | 3512 | if (counting) { |
67c5e7d4 | 3513 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3514 | spin_lock(&fs_info->balance_lock); |
3515 | bctl->stat.expected++; | |
3516 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3517 | |
3518 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3519 | count_data++; | |
3520 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3521 | count_sys++; | |
3522 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3523 | count_meta++; | |
3524 | ||
3525 | goto loop; | |
3526 | } | |
3527 | ||
3528 | /* | |
3529 | * Apply limit_min filter, no need to check if the LIMITS | |
3530 | * filter is used, limit_min is 0 by default | |
3531 | */ | |
3532 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3533 | count_data < bctl->data.limit_min) | |
3534 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3535 | count_meta < bctl->meta.limit_min) | |
3536 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3537 | count_sys < bctl->sys.limit_min)) { | |
3538 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3539 | goto loop; |
3540 | } | |
3541 | ||
2c9fe835 ZL |
3542 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) { |
3543 | trans = btrfs_start_transaction(chunk_root, 0); | |
3544 | if (IS_ERR(trans)) { | |
3545 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3546 | ret = PTR_ERR(trans); | |
3547 | goto error; | |
3548 | } | |
3549 | ||
3550 | ret = btrfs_force_chunk_alloc(trans, chunk_root, | |
3551 | BTRFS_BLOCK_GROUP_DATA); | |
8a7d656f | 3552 | btrfs_end_transaction(trans, chunk_root); |
2c9fe835 ZL |
3553 | if (ret < 0) { |
3554 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3555 | goto error; | |
3556 | } | |
2c9fe835 ZL |
3557 | chunk_reserved = 1; |
3558 | } | |
3559 | ||
ec44a35c | 3560 | ret = btrfs_relocate_chunk(chunk_root, |
ec44a35c | 3561 | found_key.offset); |
67c5e7d4 | 3562 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3563 | if (ret && ret != -ENOSPC) |
3564 | goto error; | |
19a39dce | 3565 | if (ret == -ENOSPC) { |
c9e9f97b | 3566 | enospc_errors++; |
19a39dce ID |
3567 | } else { |
3568 | spin_lock(&fs_info->balance_lock); | |
3569 | bctl->stat.completed++; | |
3570 | spin_unlock(&fs_info->balance_lock); | |
3571 | } | |
f43ffb60 | 3572 | loop: |
795a3321 ID |
3573 | if (found_key.offset == 0) |
3574 | break; | |
ba1bf481 | 3575 | key.offset = found_key.offset - 1; |
ec44a35c | 3576 | } |
c9e9f97b | 3577 | |
19a39dce ID |
3578 | if (counting) { |
3579 | btrfs_release_path(path); | |
3580 | counting = false; | |
3581 | goto again; | |
3582 | } | |
ec44a35c CM |
3583 | error: |
3584 | btrfs_free_path(path); | |
c9e9f97b | 3585 | if (enospc_errors) { |
efe120a0 | 3586 | btrfs_info(fs_info, "%d enospc errors during balance", |
c9e9f97b ID |
3587 | enospc_errors); |
3588 | if (!ret) | |
3589 | ret = -ENOSPC; | |
3590 | } | |
3591 | ||
ec44a35c CM |
3592 | return ret; |
3593 | } | |
3594 | ||
0c460c0d ID |
3595 | /** |
3596 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3597 | * @flags: profile to validate | |
3598 | * @extended: if true @flags is treated as an extended profile | |
3599 | */ | |
3600 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3601 | { | |
3602 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3603 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3604 | ||
3605 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3606 | ||
3607 | /* 1) check that all other bits are zeroed */ | |
3608 | if (flags & ~mask) | |
3609 | return 0; | |
3610 | ||
3611 | /* 2) see if profile is reduced */ | |
3612 | if (flags == 0) | |
3613 | return !extended; /* "0" is valid for usual profiles */ | |
3614 | ||
3615 | /* true if exactly one bit set */ | |
3616 | return (flags & (flags - 1)) == 0; | |
3617 | } | |
3618 | ||
837d5b6e ID |
3619 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3620 | { | |
a7e99c69 ID |
3621 | /* cancel requested || normal exit path */ |
3622 | return atomic_read(&fs_info->balance_cancel_req) || | |
3623 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3624 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3625 | } |
3626 | ||
c9e9f97b ID |
3627 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3628 | { | |
0940ebf6 ID |
3629 | int ret; |
3630 | ||
c9e9f97b | 3631 | unset_balance_control(fs_info); |
0940ebf6 | 3632 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 | 3633 | if (ret) |
a4553fef | 3634 | btrfs_std_error(fs_info, ret, NULL); |
ed0fb78f ID |
3635 | |
3636 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3637 | } |
3638 | ||
bdcd3c97 AM |
3639 | /* Non-zero return value signifies invalidity */ |
3640 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3641 | u64 allowed) | |
3642 | { | |
3643 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3644 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3645 | (bctl_arg->target & ~allowed))); | |
3646 | } | |
3647 | ||
c9e9f97b ID |
3648 | /* |
3649 | * Should be called with both balance and volume mutexes held | |
3650 | */ | |
3651 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3652 | struct btrfs_ioctl_balance_args *bargs) | |
3653 | { | |
3654 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3655 | u64 allowed; |
e4837f8f | 3656 | int mixed = 0; |
c9e9f97b | 3657 | int ret; |
8dabb742 | 3658 | u64 num_devices; |
de98ced9 | 3659 | unsigned seq; |
c9e9f97b | 3660 | |
837d5b6e | 3661 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3662 | atomic_read(&fs_info->balance_pause_req) || |
3663 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3664 | ret = -EINVAL; |
3665 | goto out; | |
3666 | } | |
3667 | ||
e4837f8f ID |
3668 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3669 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3670 | mixed = 1; | |
3671 | ||
f43ffb60 ID |
3672 | /* |
3673 | * In case of mixed groups both data and meta should be picked, | |
3674 | * and identical options should be given for both of them. | |
3675 | */ | |
e4837f8f ID |
3676 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3677 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3678 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3679 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3680 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
efe120a0 FH |
3681 | btrfs_err(fs_info, "with mixed groups data and " |
3682 | "metadata balance options must be the same"); | |
f43ffb60 ID |
3683 | ret = -EINVAL; |
3684 | goto out; | |
3685 | } | |
3686 | } | |
3687 | ||
8dabb742 SB |
3688 | num_devices = fs_info->fs_devices->num_devices; |
3689 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
3690 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3691 | BUG_ON(num_devices < 1); | |
3692 | num_devices--; | |
3693 | } | |
3694 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
e4d8ec0f | 3695 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3696 | if (num_devices == 1) |
e4d8ec0f | 3697 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8250dabe | 3698 | else if (num_devices > 1) |
e4d8ec0f | 3699 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3700 | if (num_devices > 2) |
3701 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3702 | if (num_devices > 3) | |
3703 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3704 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3705 | if (validate_convert_profile(&bctl->data, allowed)) { |
efe120a0 FH |
3706 | btrfs_err(fs_info, "unable to start balance with target " |
3707 | "data profile %llu", | |
c1c9ff7c | 3708 | bctl->data.target); |
e4d8ec0f ID |
3709 | ret = -EINVAL; |
3710 | goto out; | |
3711 | } | |
bdcd3c97 | 3712 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 FH |
3713 | btrfs_err(fs_info, |
3714 | "unable to start balance with target metadata profile %llu", | |
c1c9ff7c | 3715 | bctl->meta.target); |
e4d8ec0f ID |
3716 | ret = -EINVAL; |
3717 | goto out; | |
3718 | } | |
bdcd3c97 | 3719 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 FH |
3720 | btrfs_err(fs_info, |
3721 | "unable to start balance with target system profile %llu", | |
c1c9ff7c | 3722 | bctl->sys.target); |
e4d8ec0f ID |
3723 | ret = -EINVAL; |
3724 | goto out; | |
3725 | } | |
3726 | ||
e4837f8f ID |
3727 | /* allow dup'ed data chunks only in mixed mode */ |
3728 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 3729 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
efe120a0 | 3730 | btrfs_err(fs_info, "dup for data is not allowed"); |
e4d8ec0f ID |
3731 | ret = -EINVAL; |
3732 | goto out; | |
3733 | } | |
3734 | ||
3735 | /* allow to reduce meta or sys integrity only if force set */ | |
3736 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3737 | BTRFS_BLOCK_GROUP_RAID10 | |
3738 | BTRFS_BLOCK_GROUP_RAID5 | | |
3739 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3740 | do { |
3741 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3742 | ||
3743 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3744 | (fs_info->avail_system_alloc_bits & allowed) && | |
3745 | !(bctl->sys.target & allowed)) || | |
3746 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3747 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3748 | !(bctl->meta.target & allowed))) { | |
3749 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
efe120a0 | 3750 | btrfs_info(fs_info, "force reducing metadata integrity"); |
de98ced9 | 3751 | } else { |
efe120a0 FH |
3752 | btrfs_err(fs_info, "balance will reduce metadata " |
3753 | "integrity, use force if you want this"); | |
de98ced9 MX |
3754 | ret = -EINVAL; |
3755 | goto out; | |
3756 | } | |
e4d8ec0f | 3757 | } |
de98ced9 | 3758 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3759 | |
5af3e8cc | 3760 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
943c6e99 ZL |
3761 | fs_info->num_tolerated_disk_barrier_failures = min( |
3762 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3763 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3764 | bctl->sys.target)); | |
5af3e8cc SB |
3765 | } |
3766 | ||
0940ebf6 | 3767 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3768 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3769 | goto out; |
3770 | ||
59641015 ID |
3771 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3772 | BUG_ON(ret == -EEXIST); | |
3773 | set_balance_control(bctl); | |
3774 | } else { | |
3775 | BUG_ON(ret != -EEXIST); | |
3776 | spin_lock(&fs_info->balance_lock); | |
3777 | update_balance_args(bctl); | |
3778 | spin_unlock(&fs_info->balance_lock); | |
3779 | } | |
c9e9f97b | 3780 | |
837d5b6e | 3781 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3782 | mutex_unlock(&fs_info->balance_mutex); |
3783 | ||
3784 | ret = __btrfs_balance(fs_info); | |
3785 | ||
3786 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3787 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3788 | |
bf023ecf ID |
3789 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3790 | fs_info->num_tolerated_disk_barrier_failures = | |
3791 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3792 | } | |
3793 | ||
c9e9f97b ID |
3794 | if (bargs) { |
3795 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3796 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3797 | } |
3798 | ||
3a01aa7a ID |
3799 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3800 | balance_need_close(fs_info)) { | |
3801 | __cancel_balance(fs_info); | |
3802 | } | |
3803 | ||
837d5b6e | 3804 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3805 | |
3806 | return ret; | |
3807 | out: | |
59641015 ID |
3808 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3809 | __cancel_balance(fs_info); | |
ed0fb78f | 3810 | else { |
59641015 | 3811 | kfree(bctl); |
ed0fb78f ID |
3812 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3813 | } | |
59641015 ID |
3814 | return ret; |
3815 | } | |
3816 | ||
3817 | static int balance_kthread(void *data) | |
3818 | { | |
2b6ba629 | 3819 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3820 | int ret = 0; |
59641015 ID |
3821 | |
3822 | mutex_lock(&fs_info->volume_mutex); | |
3823 | mutex_lock(&fs_info->balance_mutex); | |
3824 | ||
2b6ba629 | 3825 | if (fs_info->balance_ctl) { |
efe120a0 | 3826 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3827 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3828 | } |
59641015 ID |
3829 | |
3830 | mutex_unlock(&fs_info->balance_mutex); | |
3831 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3832 | |
59641015 ID |
3833 | return ret; |
3834 | } | |
3835 | ||
2b6ba629 ID |
3836 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3837 | { | |
3838 | struct task_struct *tsk; | |
3839 | ||
3840 | spin_lock(&fs_info->balance_lock); | |
3841 | if (!fs_info->balance_ctl) { | |
3842 | spin_unlock(&fs_info->balance_lock); | |
3843 | return 0; | |
3844 | } | |
3845 | spin_unlock(&fs_info->balance_lock); | |
3846 | ||
3847 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
efe120a0 | 3848 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3849 | return 0; |
3850 | } | |
3851 | ||
3852 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3853 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3854 | } |
3855 | ||
68310a5e | 3856 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3857 | { |
59641015 ID |
3858 | struct btrfs_balance_control *bctl; |
3859 | struct btrfs_balance_item *item; | |
3860 | struct btrfs_disk_balance_args disk_bargs; | |
3861 | struct btrfs_path *path; | |
3862 | struct extent_buffer *leaf; | |
3863 | struct btrfs_key key; | |
3864 | int ret; | |
3865 | ||
3866 | path = btrfs_alloc_path(); | |
3867 | if (!path) | |
3868 | return -ENOMEM; | |
3869 | ||
59641015 ID |
3870 | key.objectid = BTRFS_BALANCE_OBJECTID; |
3871 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3872 | key.offset = 0; | |
3873 | ||
68310a5e | 3874 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3875 | if (ret < 0) |
68310a5e | 3876 | goto out; |
59641015 ID |
3877 | if (ret > 0) { /* ret = -ENOENT; */ |
3878 | ret = 0; | |
68310a5e ID |
3879 | goto out; |
3880 | } | |
3881 | ||
3882 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3883 | if (!bctl) { | |
3884 | ret = -ENOMEM; | |
3885 | goto out; | |
59641015 ID |
3886 | } |
3887 | ||
3888 | leaf = path->nodes[0]; | |
3889 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3890 | ||
68310a5e ID |
3891 | bctl->fs_info = fs_info; |
3892 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3893 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3894 | |
3895 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3896 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3897 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3898 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3899 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3900 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3901 | ||
ed0fb78f ID |
3902 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3903 | ||
68310a5e ID |
3904 | mutex_lock(&fs_info->volume_mutex); |
3905 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3906 | |
68310a5e ID |
3907 | set_balance_control(bctl); |
3908 | ||
3909 | mutex_unlock(&fs_info->balance_mutex); | |
3910 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3911 | out: |
3912 | btrfs_free_path(path); | |
ec44a35c CM |
3913 | return ret; |
3914 | } | |
3915 | ||
837d5b6e ID |
3916 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3917 | { | |
3918 | int ret = 0; | |
3919 | ||
3920 | mutex_lock(&fs_info->balance_mutex); | |
3921 | if (!fs_info->balance_ctl) { | |
3922 | mutex_unlock(&fs_info->balance_mutex); | |
3923 | return -ENOTCONN; | |
3924 | } | |
3925 | ||
3926 | if (atomic_read(&fs_info->balance_running)) { | |
3927 | atomic_inc(&fs_info->balance_pause_req); | |
3928 | mutex_unlock(&fs_info->balance_mutex); | |
3929 | ||
3930 | wait_event(fs_info->balance_wait_q, | |
3931 | atomic_read(&fs_info->balance_running) == 0); | |
3932 | ||
3933 | mutex_lock(&fs_info->balance_mutex); | |
3934 | /* we are good with balance_ctl ripped off from under us */ | |
3935 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3936 | atomic_dec(&fs_info->balance_pause_req); | |
3937 | } else { | |
3938 | ret = -ENOTCONN; | |
3939 | } | |
3940 | ||
3941 | mutex_unlock(&fs_info->balance_mutex); | |
3942 | return ret; | |
3943 | } | |
3944 | ||
a7e99c69 ID |
3945 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3946 | { | |
e649e587 ID |
3947 | if (fs_info->sb->s_flags & MS_RDONLY) |
3948 | return -EROFS; | |
3949 | ||
a7e99c69 ID |
3950 | mutex_lock(&fs_info->balance_mutex); |
3951 | if (!fs_info->balance_ctl) { | |
3952 | mutex_unlock(&fs_info->balance_mutex); | |
3953 | return -ENOTCONN; | |
3954 | } | |
3955 | ||
3956 | atomic_inc(&fs_info->balance_cancel_req); | |
3957 | /* | |
3958 | * if we are running just wait and return, balance item is | |
3959 | * deleted in btrfs_balance in this case | |
3960 | */ | |
3961 | if (atomic_read(&fs_info->balance_running)) { | |
3962 | mutex_unlock(&fs_info->balance_mutex); | |
3963 | wait_event(fs_info->balance_wait_q, | |
3964 | atomic_read(&fs_info->balance_running) == 0); | |
3965 | mutex_lock(&fs_info->balance_mutex); | |
3966 | } else { | |
3967 | /* __cancel_balance needs volume_mutex */ | |
3968 | mutex_unlock(&fs_info->balance_mutex); | |
3969 | mutex_lock(&fs_info->volume_mutex); | |
3970 | mutex_lock(&fs_info->balance_mutex); | |
3971 | ||
3972 | if (fs_info->balance_ctl) | |
3973 | __cancel_balance(fs_info); | |
3974 | ||
3975 | mutex_unlock(&fs_info->volume_mutex); | |
3976 | } | |
3977 | ||
3978 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3979 | atomic_dec(&fs_info->balance_cancel_req); | |
3980 | mutex_unlock(&fs_info->balance_mutex); | |
3981 | return 0; | |
3982 | } | |
3983 | ||
803b2f54 SB |
3984 | static int btrfs_uuid_scan_kthread(void *data) |
3985 | { | |
3986 | struct btrfs_fs_info *fs_info = data; | |
3987 | struct btrfs_root *root = fs_info->tree_root; | |
3988 | struct btrfs_key key; | |
3989 | struct btrfs_key max_key; | |
3990 | struct btrfs_path *path = NULL; | |
3991 | int ret = 0; | |
3992 | struct extent_buffer *eb; | |
3993 | int slot; | |
3994 | struct btrfs_root_item root_item; | |
3995 | u32 item_size; | |
f45388f3 | 3996 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
3997 | |
3998 | path = btrfs_alloc_path(); | |
3999 | if (!path) { | |
4000 | ret = -ENOMEM; | |
4001 | goto out; | |
4002 | } | |
4003 | ||
4004 | key.objectid = 0; | |
4005 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4006 | key.offset = 0; | |
4007 | ||
4008 | max_key.objectid = (u64)-1; | |
4009 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
4010 | max_key.offset = (u64)-1; | |
4011 | ||
803b2f54 | 4012 | while (1) { |
6174d3cb | 4013 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4014 | if (ret) { |
4015 | if (ret > 0) | |
4016 | ret = 0; | |
4017 | break; | |
4018 | } | |
4019 | ||
4020 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4021 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4022 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4023 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4024 | goto skip; | |
4025 | ||
4026 | eb = path->nodes[0]; | |
4027 | slot = path->slots[0]; | |
4028 | item_size = btrfs_item_size_nr(eb, slot); | |
4029 | if (item_size < sizeof(root_item)) | |
4030 | goto skip; | |
4031 | ||
803b2f54 SB |
4032 | read_extent_buffer(eb, &root_item, |
4033 | btrfs_item_ptr_offset(eb, slot), | |
4034 | (int)sizeof(root_item)); | |
4035 | if (btrfs_root_refs(&root_item) == 0) | |
4036 | goto skip; | |
f45388f3 FDBM |
4037 | |
4038 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4039 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4040 | if (trans) | |
4041 | goto update_tree; | |
4042 | ||
4043 | btrfs_release_path(path); | |
803b2f54 SB |
4044 | /* |
4045 | * 1 - subvol uuid item | |
4046 | * 1 - received_subvol uuid item | |
4047 | */ | |
4048 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4049 | if (IS_ERR(trans)) { | |
4050 | ret = PTR_ERR(trans); | |
4051 | break; | |
4052 | } | |
f45388f3 FDBM |
4053 | continue; |
4054 | } else { | |
4055 | goto skip; | |
4056 | } | |
4057 | update_tree: | |
4058 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
803b2f54 SB |
4059 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
4060 | root_item.uuid, | |
4061 | BTRFS_UUID_KEY_SUBVOL, | |
4062 | key.objectid); | |
4063 | if (ret < 0) { | |
efe120a0 | 4064 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4065 | ret); |
803b2f54 SB |
4066 | break; |
4067 | } | |
4068 | } | |
4069 | ||
4070 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
803b2f54 SB |
4071 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
4072 | root_item.received_uuid, | |
4073 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4074 | key.objectid); | |
4075 | if (ret < 0) { | |
efe120a0 | 4076 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4077 | ret); |
803b2f54 SB |
4078 | break; |
4079 | } | |
4080 | } | |
4081 | ||
f45388f3 | 4082 | skip: |
803b2f54 SB |
4083 | if (trans) { |
4084 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 4085 | trans = NULL; |
803b2f54 SB |
4086 | if (ret) |
4087 | break; | |
4088 | } | |
4089 | ||
803b2f54 SB |
4090 | btrfs_release_path(path); |
4091 | if (key.offset < (u64)-1) { | |
4092 | key.offset++; | |
4093 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4094 | key.offset = 0; | |
4095 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4096 | } else if (key.objectid < (u64)-1) { | |
4097 | key.offset = 0; | |
4098 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4099 | key.objectid++; | |
4100 | } else { | |
4101 | break; | |
4102 | } | |
4103 | cond_resched(); | |
4104 | } | |
4105 | ||
4106 | out: | |
4107 | btrfs_free_path(path); | |
f45388f3 FDBM |
4108 | if (trans && !IS_ERR(trans)) |
4109 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 4110 | if (ret) |
efe120a0 | 4111 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 SB |
4112 | else |
4113 | fs_info->update_uuid_tree_gen = 1; | |
803b2f54 SB |
4114 | up(&fs_info->uuid_tree_rescan_sem); |
4115 | return 0; | |
4116 | } | |
4117 | ||
70f80175 SB |
4118 | /* |
4119 | * Callback for btrfs_uuid_tree_iterate(). | |
4120 | * returns: | |
4121 | * 0 check succeeded, the entry is not outdated. | |
4122 | * < 0 if an error occured. | |
4123 | * > 0 if the check failed, which means the caller shall remove the entry. | |
4124 | */ | |
4125 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4126 | u8 *uuid, u8 type, u64 subid) | |
4127 | { | |
4128 | struct btrfs_key key; | |
4129 | int ret = 0; | |
4130 | struct btrfs_root *subvol_root; | |
4131 | ||
4132 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4133 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4134 | goto out; | |
4135 | ||
4136 | key.objectid = subid; | |
4137 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4138 | key.offset = (u64)-1; | |
4139 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4140 | if (IS_ERR(subvol_root)) { | |
4141 | ret = PTR_ERR(subvol_root); | |
4142 | if (ret == -ENOENT) | |
4143 | ret = 1; | |
4144 | goto out; | |
4145 | } | |
4146 | ||
4147 | switch (type) { | |
4148 | case BTRFS_UUID_KEY_SUBVOL: | |
4149 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4150 | ret = 1; | |
4151 | break; | |
4152 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4153 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4154 | BTRFS_UUID_SIZE)) | |
4155 | ret = 1; | |
4156 | break; | |
4157 | } | |
4158 | ||
4159 | out: | |
4160 | return ret; | |
4161 | } | |
4162 | ||
4163 | static int btrfs_uuid_rescan_kthread(void *data) | |
4164 | { | |
4165 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4166 | int ret; | |
4167 | ||
4168 | /* | |
4169 | * 1st step is to iterate through the existing UUID tree and | |
4170 | * to delete all entries that contain outdated data. | |
4171 | * 2nd step is to add all missing entries to the UUID tree. | |
4172 | */ | |
4173 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4174 | if (ret < 0) { | |
efe120a0 | 4175 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4176 | up(&fs_info->uuid_tree_rescan_sem); |
4177 | return ret; | |
4178 | } | |
4179 | return btrfs_uuid_scan_kthread(data); | |
4180 | } | |
4181 | ||
f7a81ea4 SB |
4182 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4183 | { | |
4184 | struct btrfs_trans_handle *trans; | |
4185 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4186 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4187 | struct task_struct *task; |
4188 | int ret; | |
f7a81ea4 SB |
4189 | |
4190 | /* | |
4191 | * 1 - root node | |
4192 | * 1 - root item | |
4193 | */ | |
4194 | trans = btrfs_start_transaction(tree_root, 2); | |
4195 | if (IS_ERR(trans)) | |
4196 | return PTR_ERR(trans); | |
4197 | ||
4198 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4199 | BTRFS_UUID_TREE_OBJECTID); | |
4200 | if (IS_ERR(uuid_root)) { | |
6d13f549 DS |
4201 | ret = PTR_ERR(uuid_root); |
4202 | btrfs_abort_transaction(trans, tree_root, ret); | |
4203 | return ret; | |
f7a81ea4 SB |
4204 | } |
4205 | ||
4206 | fs_info->uuid_root = uuid_root; | |
4207 | ||
803b2f54 SB |
4208 | ret = btrfs_commit_transaction(trans, tree_root); |
4209 | if (ret) | |
4210 | return ret; | |
4211 | ||
4212 | down(&fs_info->uuid_tree_rescan_sem); | |
4213 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4214 | if (IS_ERR(task)) { | |
70f80175 | 4215 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4216 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4217 | up(&fs_info->uuid_tree_rescan_sem); |
4218 | return PTR_ERR(task); | |
4219 | } | |
4220 | ||
4221 | return 0; | |
f7a81ea4 | 4222 | } |
803b2f54 | 4223 | |
70f80175 SB |
4224 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4225 | { | |
4226 | struct task_struct *task; | |
4227 | ||
4228 | down(&fs_info->uuid_tree_rescan_sem); | |
4229 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4230 | if (IS_ERR(task)) { | |
4231 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4232 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4233 | up(&fs_info->uuid_tree_rescan_sem); |
4234 | return PTR_ERR(task); | |
4235 | } | |
4236 | ||
4237 | return 0; | |
4238 | } | |
4239 | ||
8f18cf13 CM |
4240 | /* |
4241 | * shrinking a device means finding all of the device extents past | |
4242 | * the new size, and then following the back refs to the chunks. | |
4243 | * The chunk relocation code actually frees the device extent | |
4244 | */ | |
4245 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4246 | { | |
4247 | struct btrfs_trans_handle *trans; | |
4248 | struct btrfs_root *root = device->dev_root; | |
4249 | struct btrfs_dev_extent *dev_extent = NULL; | |
4250 | struct btrfs_path *path; | |
4251 | u64 length; | |
8f18cf13 CM |
4252 | u64 chunk_offset; |
4253 | int ret; | |
4254 | int slot; | |
ba1bf481 JB |
4255 | int failed = 0; |
4256 | bool retried = false; | |
53e489bc | 4257 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4258 | struct extent_buffer *l; |
4259 | struct btrfs_key key; | |
6c41761f | 4260 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 4261 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4262 | u64 old_size = btrfs_device_get_total_bytes(device); |
4263 | u64 diff = old_size - new_size; | |
8f18cf13 | 4264 | |
63a212ab SB |
4265 | if (device->is_tgtdev_for_dev_replace) |
4266 | return -EINVAL; | |
4267 | ||
8f18cf13 CM |
4268 | path = btrfs_alloc_path(); |
4269 | if (!path) | |
4270 | return -ENOMEM; | |
4271 | ||
8f18cf13 CM |
4272 | path->reada = 2; |
4273 | ||
7d9eb12c CM |
4274 | lock_chunks(root); |
4275 | ||
7cc8e58d | 4276 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4277 | if (device->writeable) { |
2b82032c | 4278 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
4279 | spin_lock(&root->fs_info->free_chunk_lock); |
4280 | root->fs_info->free_chunk_space -= diff; | |
4281 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4282 | } | |
7d9eb12c | 4283 | unlock_chunks(root); |
8f18cf13 | 4284 | |
ba1bf481 | 4285 | again: |
8f18cf13 CM |
4286 | key.objectid = device->devid; |
4287 | key.offset = (u64)-1; | |
4288 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4289 | ||
213e64da | 4290 | do { |
67c5e7d4 | 4291 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4292 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 FM |
4293 | if (ret < 0) { |
4294 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 | 4295 | goto done; |
67c5e7d4 | 4296 | } |
8f18cf13 CM |
4297 | |
4298 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 FM |
4299 | if (ret) |
4300 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 CM |
4301 | if (ret < 0) |
4302 | goto done; | |
4303 | if (ret) { | |
4304 | ret = 0; | |
b3b4aa74 | 4305 | btrfs_release_path(path); |
bf1fb512 | 4306 | break; |
8f18cf13 CM |
4307 | } |
4308 | ||
4309 | l = path->nodes[0]; | |
4310 | slot = path->slots[0]; | |
4311 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4312 | ||
ba1bf481 | 4313 | if (key.objectid != device->devid) { |
67c5e7d4 | 4314 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4315 | btrfs_release_path(path); |
bf1fb512 | 4316 | break; |
ba1bf481 | 4317 | } |
8f18cf13 CM |
4318 | |
4319 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4320 | length = btrfs_dev_extent_length(l, dev_extent); | |
4321 | ||
ba1bf481 | 4322 | if (key.offset + length <= new_size) { |
67c5e7d4 | 4323 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4324 | btrfs_release_path(path); |
d6397bae | 4325 | break; |
ba1bf481 | 4326 | } |
8f18cf13 | 4327 | |
8f18cf13 | 4328 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4329 | btrfs_release_path(path); |
8f18cf13 | 4330 | |
dc2ee4e2 | 4331 | ret = btrfs_relocate_chunk(root, chunk_offset); |
67c5e7d4 | 4332 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
ba1bf481 | 4333 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4334 | goto done; |
ba1bf481 JB |
4335 | if (ret == -ENOSPC) |
4336 | failed++; | |
213e64da | 4337 | } while (key.offset-- > 0); |
ba1bf481 JB |
4338 | |
4339 | if (failed && !retried) { | |
4340 | failed = 0; | |
4341 | retried = true; | |
4342 | goto again; | |
4343 | } else if (failed && retried) { | |
4344 | ret = -ENOSPC; | |
ba1bf481 | 4345 | goto done; |
8f18cf13 CM |
4346 | } |
4347 | ||
d6397bae | 4348 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4349 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4350 | if (IS_ERR(trans)) { |
4351 | ret = PTR_ERR(trans); | |
4352 | goto done; | |
4353 | } | |
4354 | ||
d6397bae | 4355 | lock_chunks(root); |
53e489bc FM |
4356 | |
4357 | /* | |
4358 | * We checked in the above loop all device extents that were already in | |
4359 | * the device tree. However before we have updated the device's | |
4360 | * total_bytes to the new size, we might have had chunk allocations that | |
4361 | * have not complete yet (new block groups attached to transaction | |
4362 | * handles), and therefore their device extents were not yet in the | |
4363 | * device tree and we missed them in the loop above. So if we have any | |
4364 | * pending chunk using a device extent that overlaps the device range | |
4365 | * that we can not use anymore, commit the current transaction and | |
4366 | * repeat the search on the device tree - this way we guarantee we will | |
4367 | * not have chunks using device extents that end beyond 'new_size'. | |
4368 | */ | |
4369 | if (!checked_pending_chunks) { | |
4370 | u64 start = new_size; | |
4371 | u64 len = old_size - new_size; | |
4372 | ||
499f377f JM |
4373 | if (contains_pending_extent(trans->transaction, device, |
4374 | &start, len)) { | |
53e489bc FM |
4375 | unlock_chunks(root); |
4376 | checked_pending_chunks = true; | |
4377 | failed = 0; | |
4378 | retried = false; | |
4379 | ret = btrfs_commit_transaction(trans, root); | |
4380 | if (ret) | |
4381 | goto done; | |
4382 | goto again; | |
4383 | } | |
4384 | } | |
4385 | ||
7cc8e58d | 4386 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4387 | if (list_empty(&device->resized_list)) |
4388 | list_add_tail(&device->resized_list, | |
4389 | &root->fs_info->fs_devices->resized_devices); | |
d6397bae | 4390 | |
d6397bae CB |
4391 | WARN_ON(diff > old_total); |
4392 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4393 | unlock_chunks(root); | |
2196d6e8 MX |
4394 | |
4395 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4396 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4397 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4398 | done: |
4399 | btrfs_free_path(path); | |
53e489bc FM |
4400 | if (ret) { |
4401 | lock_chunks(root); | |
4402 | btrfs_device_set_total_bytes(device, old_size); | |
4403 | if (device->writeable) | |
4404 | device->fs_devices->total_rw_bytes += diff; | |
4405 | spin_lock(&root->fs_info->free_chunk_lock); | |
4406 | root->fs_info->free_chunk_space += diff; | |
4407 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4408 | unlock_chunks(root); | |
4409 | } | |
8f18cf13 CM |
4410 | return ret; |
4411 | } | |
4412 | ||
125ccb0a | 4413 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4414 | struct btrfs_key *key, |
4415 | struct btrfs_chunk *chunk, int item_size) | |
4416 | { | |
6c41761f | 4417 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
4418 | struct btrfs_disk_key disk_key; |
4419 | u32 array_size; | |
4420 | u8 *ptr; | |
4421 | ||
fe48a5c0 | 4422 | lock_chunks(root); |
0b86a832 | 4423 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4424 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 MX |
4425 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
4426 | unlock_chunks(root); | |
0b86a832 | 4427 | return -EFBIG; |
fe48a5c0 | 4428 | } |
0b86a832 CM |
4429 | |
4430 | ptr = super_copy->sys_chunk_array + array_size; | |
4431 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4432 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4433 | ptr += sizeof(disk_key); | |
4434 | memcpy(ptr, chunk, item_size); | |
4435 | item_size += sizeof(disk_key); | |
4436 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 MX |
4437 | unlock_chunks(root); |
4438 | ||
0b86a832 CM |
4439 | return 0; |
4440 | } | |
4441 | ||
73c5de00 AJ |
4442 | /* |
4443 | * sort the devices in descending order by max_avail, total_avail | |
4444 | */ | |
4445 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4446 | { |
73c5de00 AJ |
4447 | const struct btrfs_device_info *di_a = a; |
4448 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4449 | |
73c5de00 | 4450 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4451 | return -1; |
73c5de00 | 4452 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4453 | return 1; |
73c5de00 AJ |
4454 | if (di_a->total_avail > di_b->total_avail) |
4455 | return -1; | |
4456 | if (di_a->total_avail < di_b->total_avail) | |
4457 | return 1; | |
4458 | return 0; | |
b2117a39 | 4459 | } |
0b86a832 | 4460 | |
53b381b3 DW |
4461 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4462 | { | |
4463 | /* TODO allow them to set a preferred stripe size */ | |
4464 | return 64 * 1024; | |
4465 | } | |
4466 | ||
4467 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4468 | { | |
ffe2d203 | 4469 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4470 | return; |
4471 | ||
ceda0864 | 4472 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4473 | } |
4474 | ||
23f8f9b7 GH |
4475 | #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \ |
4476 | - sizeof(struct btrfs_item) \ | |
4477 | - sizeof(struct btrfs_chunk)) \ | |
4478 | / sizeof(struct btrfs_stripe) + 1) | |
4479 | ||
4480 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4481 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4482 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4483 | / sizeof(struct btrfs_stripe) + 1) | |
4484 | ||
73c5de00 | 4485 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4486 | struct btrfs_root *extent_root, u64 start, |
4487 | u64 type) | |
b2117a39 | 4488 | { |
73c5de00 AJ |
4489 | struct btrfs_fs_info *info = extent_root->fs_info; |
4490 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4491 | struct list_head *cur; | |
4492 | struct map_lookup *map = NULL; | |
4493 | struct extent_map_tree *em_tree; | |
4494 | struct extent_map *em; | |
4495 | struct btrfs_device_info *devices_info = NULL; | |
4496 | u64 total_avail; | |
4497 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4498 | int data_stripes; /* number of stripes that count for |
4499 | block group size */ | |
73c5de00 AJ |
4500 | int sub_stripes; /* sub_stripes info for map */ |
4501 | int dev_stripes; /* stripes per dev */ | |
4502 | int devs_max; /* max devs to use */ | |
4503 | int devs_min; /* min devs needed */ | |
4504 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4505 | int ncopies; /* how many copies to data has */ | |
4506 | int ret; | |
4507 | u64 max_stripe_size; | |
4508 | u64 max_chunk_size; | |
4509 | u64 stripe_size; | |
4510 | u64 num_bytes; | |
53b381b3 | 4511 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4512 | int ndevs; |
4513 | int i; | |
4514 | int j; | |
31e50229 | 4515 | int index; |
593060d7 | 4516 | |
0c460c0d | 4517 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4518 | |
73c5de00 AJ |
4519 | if (list_empty(&fs_devices->alloc_list)) |
4520 | return -ENOSPC; | |
b2117a39 | 4521 | |
31e50229 | 4522 | index = __get_raid_index(type); |
73c5de00 | 4523 | |
31e50229 LB |
4524 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4525 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4526 | devs_max = btrfs_raid_array[index].devs_max; | |
4527 | devs_min = btrfs_raid_array[index].devs_min; | |
4528 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4529 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4530 | |
9b3f68b9 | 4531 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
4532 | max_stripe_size = 1024 * 1024 * 1024; |
4533 | max_chunk_size = 10 * max_stripe_size; | |
23f8f9b7 GH |
4534 | if (!devs_max) |
4535 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4536 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
4537 | /* for larger filesystems, use larger metadata chunks */ |
4538 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
4539 | max_stripe_size = 1024 * 1024 * 1024; | |
4540 | else | |
4541 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 4542 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4543 | if (!devs_max) |
4544 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4545 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 4546 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 | 4547 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4548 | if (!devs_max) |
4549 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4550 | } else { |
351fd353 | 4551 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4552 | type); |
4553 | BUG_ON(1); | |
9b3f68b9 CM |
4554 | } |
4555 | ||
2b82032c YZ |
4556 | /* we don't want a chunk larger than 10% of writeable space */ |
4557 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4558 | max_chunk_size); | |
9b3f68b9 | 4559 | |
31e818fe | 4560 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4561 | GFP_NOFS); |
4562 | if (!devices_info) | |
4563 | return -ENOMEM; | |
0cad8a11 | 4564 | |
73c5de00 | 4565 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4566 | |
9f680ce0 | 4567 | /* |
73c5de00 AJ |
4568 | * in the first pass through the devices list, we gather information |
4569 | * about the available holes on each device. | |
9f680ce0 | 4570 | */ |
73c5de00 AJ |
4571 | ndevs = 0; |
4572 | while (cur != &fs_devices->alloc_list) { | |
4573 | struct btrfs_device *device; | |
4574 | u64 max_avail; | |
4575 | u64 dev_offset; | |
b2117a39 | 4576 | |
73c5de00 | 4577 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4578 | |
73c5de00 | 4579 | cur = cur->next; |
b2117a39 | 4580 | |
73c5de00 | 4581 | if (!device->writeable) { |
31b1a2bd | 4582 | WARN(1, KERN_ERR |
efe120a0 | 4583 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4584 | continue; |
4585 | } | |
b2117a39 | 4586 | |
63a212ab SB |
4587 | if (!device->in_fs_metadata || |
4588 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4589 | continue; |
b2117a39 | 4590 | |
73c5de00 AJ |
4591 | if (device->total_bytes > device->bytes_used) |
4592 | total_avail = device->total_bytes - device->bytes_used; | |
4593 | else | |
4594 | total_avail = 0; | |
38c01b96 | 4595 | |
4596 | /* If there is no space on this device, skip it. */ | |
4597 | if (total_avail == 0) | |
4598 | continue; | |
b2117a39 | 4599 | |
6df9a95e | 4600 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4601 | max_stripe_size * dev_stripes, |
4602 | &dev_offset, &max_avail); | |
4603 | if (ret && ret != -ENOSPC) | |
4604 | goto error; | |
b2117a39 | 4605 | |
73c5de00 AJ |
4606 | if (ret == 0) |
4607 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4608 | |
73c5de00 AJ |
4609 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4610 | continue; | |
b2117a39 | 4611 | |
063d006f ES |
4612 | if (ndevs == fs_devices->rw_devices) { |
4613 | WARN(1, "%s: found more than %llu devices\n", | |
4614 | __func__, fs_devices->rw_devices); | |
4615 | break; | |
4616 | } | |
73c5de00 AJ |
4617 | devices_info[ndevs].dev_offset = dev_offset; |
4618 | devices_info[ndevs].max_avail = max_avail; | |
4619 | devices_info[ndevs].total_avail = total_avail; | |
4620 | devices_info[ndevs].dev = device; | |
4621 | ++ndevs; | |
4622 | } | |
b2117a39 | 4623 | |
73c5de00 AJ |
4624 | /* |
4625 | * now sort the devices by hole size / available space | |
4626 | */ | |
4627 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4628 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4629 | |
73c5de00 AJ |
4630 | /* round down to number of usable stripes */ |
4631 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4632 | |
73c5de00 AJ |
4633 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4634 | ret = -ENOSPC; | |
4635 | goto error; | |
b2117a39 | 4636 | } |
9f680ce0 | 4637 | |
73c5de00 AJ |
4638 | if (devs_max && ndevs > devs_max) |
4639 | ndevs = devs_max; | |
4640 | /* | |
4641 | * the primary goal is to maximize the number of stripes, so use as many | |
4642 | * devices as possible, even if the stripes are not maximum sized. | |
4643 | */ | |
4644 | stripe_size = devices_info[ndevs-1].max_avail; | |
4645 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4646 | |
53b381b3 DW |
4647 | /* |
4648 | * this will have to be fixed for RAID1 and RAID10 over | |
4649 | * more drives | |
4650 | */ | |
4651 | data_stripes = num_stripes / ncopies; | |
4652 | ||
53b381b3 DW |
4653 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4654 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4655 | btrfs_super_stripesize(info->super_copy)); | |
4656 | data_stripes = num_stripes - 1; | |
4657 | } | |
4658 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4659 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4660 | btrfs_super_stripesize(info->super_copy)); | |
4661 | data_stripes = num_stripes - 2; | |
4662 | } | |
86db2578 CM |
4663 | |
4664 | /* | |
4665 | * Use the number of data stripes to figure out how big this chunk | |
4666 | * is really going to be in terms of logical address space, | |
4667 | * and compare that answer with the max chunk size | |
4668 | */ | |
4669 | if (stripe_size * data_stripes > max_chunk_size) { | |
4670 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4671 | |
4672 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4673 | |
4674 | /* bump the answer up to a 16MB boundary */ | |
4675 | stripe_size = (stripe_size + mask) & ~mask; | |
4676 | ||
4677 | /* but don't go higher than the limits we found | |
4678 | * while searching for free extents | |
4679 | */ | |
4680 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4681 | stripe_size = devices_info[ndevs-1].max_avail; | |
4682 | } | |
4683 | ||
b8b93add | 4684 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4685 | |
4686 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4687 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4688 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4689 | |
4690 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4691 | if (!map) { | |
4692 | ret = -ENOMEM; | |
4693 | goto error; | |
4694 | } | |
4695 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4696 | |
73c5de00 AJ |
4697 | for (i = 0; i < ndevs; ++i) { |
4698 | for (j = 0; j < dev_stripes; ++j) { | |
4699 | int s = i * dev_stripes + j; | |
4700 | map->stripes[s].dev = devices_info[i].dev; | |
4701 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4702 | j * stripe_size; | |
6324fbf3 | 4703 | } |
6324fbf3 | 4704 | } |
2b82032c | 4705 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4706 | map->stripe_len = raid_stripe_len; |
4707 | map->io_align = raid_stripe_len; | |
4708 | map->io_width = raid_stripe_len; | |
2b82032c | 4709 | map->type = type; |
2b82032c | 4710 | map->sub_stripes = sub_stripes; |
0b86a832 | 4711 | |
53b381b3 | 4712 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4713 | |
73c5de00 | 4714 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4715 | |
172ddd60 | 4716 | em = alloc_extent_map(); |
2b82032c | 4717 | if (!em) { |
298a8f9c | 4718 | kfree(map); |
b2117a39 MX |
4719 | ret = -ENOMEM; |
4720 | goto error; | |
593060d7 | 4721 | } |
298a8f9c | 4722 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
2b82032c YZ |
4723 | em->bdev = (struct block_device *)map; |
4724 | em->start = start; | |
73c5de00 | 4725 | em->len = num_bytes; |
2b82032c YZ |
4726 | em->block_start = 0; |
4727 | em->block_len = em->len; | |
6df9a95e | 4728 | em->orig_block_len = stripe_size; |
593060d7 | 4729 | |
2b82032c | 4730 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4731 | write_lock(&em_tree->lock); |
09a2a8f9 | 4732 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4733 | if (!ret) { |
4734 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4735 | atomic_inc(&em->refs); | |
4736 | } | |
890871be | 4737 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4738 | if (ret) { |
4739 | free_extent_map(em); | |
1dd4602f | 4740 | goto error; |
0f5d42b2 | 4741 | } |
0b86a832 | 4742 | |
04487488 JB |
4743 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4744 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4745 | start, num_bytes); | |
6df9a95e JB |
4746 | if (ret) |
4747 | goto error_del_extent; | |
2b82032c | 4748 | |
7cc8e58d MX |
4749 | for (i = 0; i < map->num_stripes; i++) { |
4750 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4751 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4752 | } | |
43530c46 | 4753 | |
1c116187 MX |
4754 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4755 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4756 | map->num_stripes); | |
4757 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4758 | ||
0f5d42b2 | 4759 | free_extent_map(em); |
53b381b3 DW |
4760 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4761 | ||
b2117a39 | 4762 | kfree(devices_info); |
2b82032c | 4763 | return 0; |
b2117a39 | 4764 | |
6df9a95e | 4765 | error_del_extent: |
0f5d42b2 JB |
4766 | write_lock(&em_tree->lock); |
4767 | remove_extent_mapping(em_tree, em); | |
4768 | write_unlock(&em_tree->lock); | |
4769 | ||
4770 | /* One for our allocation */ | |
4771 | free_extent_map(em); | |
4772 | /* One for the tree reference */ | |
4773 | free_extent_map(em); | |
495e64f4 FM |
4774 | /* One for the pending_chunks list reference */ |
4775 | free_extent_map(em); | |
b2117a39 | 4776 | error: |
b2117a39 MX |
4777 | kfree(devices_info); |
4778 | return ret; | |
2b82032c YZ |
4779 | } |
4780 | ||
6df9a95e | 4781 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4782 | struct btrfs_root *extent_root, |
6df9a95e | 4783 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4784 | { |
2b82032c YZ |
4785 | struct btrfs_key key; |
4786 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4787 | struct btrfs_device *device; | |
4788 | struct btrfs_chunk *chunk; | |
4789 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4790 | struct extent_map_tree *em_tree; |
4791 | struct extent_map *em; | |
4792 | struct map_lookup *map; | |
4793 | size_t item_size; | |
4794 | u64 dev_offset; | |
4795 | u64 stripe_size; | |
4796 | int i = 0; | |
2b82032c YZ |
4797 | int ret; |
4798 | ||
6df9a95e JB |
4799 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4800 | read_lock(&em_tree->lock); | |
4801 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4802 | read_unlock(&em_tree->lock); | |
4803 | ||
4804 | if (!em) { | |
4805 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4806 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4807 | return -EINVAL; | |
4808 | } | |
4809 | ||
4810 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4811 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
351fd353 | 4812 | " %Lu-%Lu, found %Lu-%Lu", chunk_offset, |
6df9a95e JB |
4813 | chunk_size, em->start, em->len); |
4814 | free_extent_map(em); | |
4815 | return -EINVAL; | |
4816 | } | |
4817 | ||
4818 | map = (struct map_lookup *)em->bdev; | |
4819 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
4820 | stripe_size = em->orig_block_len; | |
4821 | ||
2b82032c | 4822 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4823 | if (!chunk) { |
4824 | ret = -ENOMEM; | |
4825 | goto out; | |
4826 | } | |
4827 | ||
50460e37 FM |
4828 | /* |
4829 | * Take the device list mutex to prevent races with the final phase of | |
4830 | * a device replace operation that replaces the device object associated | |
4831 | * with the map's stripes, because the device object's id can change | |
4832 | * at any time during that final phase of the device replace operation | |
4833 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4834 | */ | |
4835 | mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
6df9a95e JB |
4836 | for (i = 0; i < map->num_stripes; i++) { |
4837 | device = map->stripes[i].dev; | |
4838 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4839 | |
0b86a832 | 4840 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4841 | if (ret) |
50460e37 | 4842 | break; |
6df9a95e JB |
4843 | ret = btrfs_alloc_dev_extent(trans, device, |
4844 | chunk_root->root_key.objectid, | |
4845 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4846 | chunk_offset, dev_offset, | |
4847 | stripe_size); | |
4848 | if (ret) | |
50460e37 FM |
4849 | break; |
4850 | } | |
4851 | if (ret) { | |
4852 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
4853 | goto out; | |
2b82032c YZ |
4854 | } |
4855 | ||
2b82032c | 4856 | stripe = &chunk->stripe; |
6df9a95e JB |
4857 | for (i = 0; i < map->num_stripes; i++) { |
4858 | device = map->stripes[i].dev; | |
4859 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4860 | |
e17cade2 CM |
4861 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4862 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4863 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4864 | stripe++; |
0b86a832 | 4865 | } |
50460e37 | 4866 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4867 | |
2b82032c | 4868 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4869 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4870 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4871 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4872 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4873 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4874 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4875 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4876 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4877 | |
2b82032c YZ |
4878 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4879 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4880 | key.offset = chunk_offset; | |
0b86a832 | 4881 | |
2b82032c | 4882 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4883 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4884 | /* | |
4885 | * TODO: Cleanup of inserted chunk root in case of | |
4886 | * failure. | |
4887 | */ | |
125ccb0a | 4888 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4889 | item_size); |
8f18cf13 | 4890 | } |
1abe9b8a | 4891 | |
6df9a95e | 4892 | out: |
0b86a832 | 4893 | kfree(chunk); |
6df9a95e | 4894 | free_extent_map(em); |
4ed1d16e | 4895 | return ret; |
2b82032c | 4896 | } |
0b86a832 | 4897 | |
2b82032c YZ |
4898 | /* |
4899 | * Chunk allocation falls into two parts. The first part does works | |
4900 | * that make the new allocated chunk useable, but not do any operation | |
4901 | * that modifies the chunk tree. The second part does the works that | |
4902 | * require modifying the chunk tree. This division is important for the | |
4903 | * bootstrap process of adding storage to a seed btrfs. | |
4904 | */ | |
4905 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4906 | struct btrfs_root *extent_root, u64 type) | |
4907 | { | |
4908 | u64 chunk_offset; | |
2b82032c | 4909 | |
a9629596 | 4910 | ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex)); |
6df9a95e JB |
4911 | chunk_offset = find_next_chunk(extent_root->fs_info); |
4912 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
4913 | } |
4914 | ||
d397712b | 4915 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4916 | struct btrfs_root *root, |
4917 | struct btrfs_device *device) | |
4918 | { | |
4919 | u64 chunk_offset; | |
4920 | u64 sys_chunk_offset; | |
2b82032c | 4921 | u64 alloc_profile; |
2b82032c YZ |
4922 | struct btrfs_fs_info *fs_info = root->fs_info; |
4923 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4924 | int ret; | |
4925 | ||
6df9a95e | 4926 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 4927 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
4928 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
4929 | alloc_profile); | |
79787eaa JM |
4930 | if (ret) |
4931 | return ret; | |
2b82032c | 4932 | |
6df9a95e | 4933 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 4934 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
4935 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
4936 | alloc_profile); | |
79787eaa | 4937 | return ret; |
2b82032c YZ |
4938 | } |
4939 | ||
d20983b4 MX |
4940 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
4941 | { | |
4942 | int max_errors; | |
4943 | ||
4944 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4945 | BTRFS_BLOCK_GROUP_RAID10 | | |
4946 | BTRFS_BLOCK_GROUP_RAID5 | | |
4947 | BTRFS_BLOCK_GROUP_DUP)) { | |
4948 | max_errors = 1; | |
4949 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4950 | max_errors = 2; | |
4951 | } else { | |
4952 | max_errors = 0; | |
005d6427 | 4953 | } |
2b82032c | 4954 | |
d20983b4 | 4955 | return max_errors; |
2b82032c YZ |
4956 | } |
4957 | ||
4958 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4959 | { | |
4960 | struct extent_map *em; | |
4961 | struct map_lookup *map; | |
4962 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4963 | int readonly = 0; | |
d20983b4 | 4964 | int miss_ndevs = 0; |
2b82032c YZ |
4965 | int i; |
4966 | ||
890871be | 4967 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4968 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4969 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4970 | if (!em) |
4971 | return 1; | |
4972 | ||
4973 | map = (struct map_lookup *)em->bdev; | |
4974 | for (i = 0; i < map->num_stripes; i++) { | |
d20983b4 MX |
4975 | if (map->stripes[i].dev->missing) { |
4976 | miss_ndevs++; | |
4977 | continue; | |
4978 | } | |
4979 | ||
2b82032c YZ |
4980 | if (!map->stripes[i].dev->writeable) { |
4981 | readonly = 1; | |
d20983b4 | 4982 | goto end; |
2b82032c YZ |
4983 | } |
4984 | } | |
d20983b4 MX |
4985 | |
4986 | /* | |
4987 | * If the number of missing devices is larger than max errors, | |
4988 | * we can not write the data into that chunk successfully, so | |
4989 | * set it readonly. | |
4990 | */ | |
4991 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
4992 | readonly = 1; | |
4993 | end: | |
0b86a832 | 4994 | free_extent_map(em); |
2b82032c | 4995 | return readonly; |
0b86a832 CM |
4996 | } |
4997 | ||
4998 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4999 | { | |
a8067e02 | 5000 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5001 | } |
5002 | ||
5003 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5004 | { | |
5005 | struct extent_map *em; | |
5006 | ||
d397712b | 5007 | while (1) { |
890871be | 5008 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5009 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5010 | if (em) | |
5011 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5012 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5013 | if (!em) |
5014 | break; | |
0b86a832 CM |
5015 | /* once for us */ |
5016 | free_extent_map(em); | |
5017 | /* once for the tree */ | |
5018 | free_extent_map(em); | |
5019 | } | |
5020 | } | |
5021 | ||
5d964051 | 5022 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 5023 | { |
5d964051 | 5024 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
5025 | struct extent_map *em; |
5026 | struct map_lookup *map; | |
5027 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5028 | int ret; | |
5029 | ||
890871be | 5030 | read_lock(&em_tree->lock); |
f188591e | 5031 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 5032 | read_unlock(&em_tree->lock); |
f188591e | 5033 | |
fb7669b5 JB |
5034 | /* |
5035 | * We could return errors for these cases, but that could get ugly and | |
5036 | * we'd probably do the same thing which is just not do anything else | |
5037 | * and exit, so return 1 so the callers don't try to use other copies. | |
5038 | */ | |
5039 | if (!em) { | |
351fd353 | 5040 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
5041 | logical+len); |
5042 | return 1; | |
5043 | } | |
5044 | ||
5045 | if (em->start > logical || em->start + em->len < logical) { | |
ccf39f92 | 5046 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " |
351fd353 | 5047 | "%Lu-%Lu", logical, logical+len, em->start, |
fb7669b5 | 5048 | em->start + em->len); |
7d3d1744 | 5049 | free_extent_map(em); |
fb7669b5 JB |
5050 | return 1; |
5051 | } | |
5052 | ||
f188591e CM |
5053 | map = (struct map_lookup *)em->bdev; |
5054 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
5055 | ret = map->num_stripes; | |
321aecc6 CM |
5056 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5057 | ret = map->sub_stripes; | |
53b381b3 DW |
5058 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5059 | ret = 2; | |
5060 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5061 | ret = 3; | |
f188591e CM |
5062 | else |
5063 | ret = 1; | |
5064 | free_extent_map(em); | |
ad6d620e SB |
5065 | |
5066 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
5067 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
5068 | ret++; | |
5069 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
5070 | ||
f188591e CM |
5071 | return ret; |
5072 | } | |
5073 | ||
53b381b3 DW |
5074 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
5075 | struct btrfs_mapping_tree *map_tree, | |
5076 | u64 logical) | |
5077 | { | |
5078 | struct extent_map *em; | |
5079 | struct map_lookup *map; | |
5080 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5081 | unsigned long len = root->sectorsize; | |
5082 | ||
5083 | read_lock(&em_tree->lock); | |
5084 | em = lookup_extent_mapping(em_tree, logical, len); | |
5085 | read_unlock(&em_tree->lock); | |
5086 | BUG_ON(!em); | |
5087 | ||
5088 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
5089 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 5090 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 5091 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
5092 | free_extent_map(em); |
5093 | return len; | |
5094 | } | |
5095 | ||
5096 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
5097 | u64 logical, u64 len, int mirror_num) | |
5098 | { | |
5099 | struct extent_map *em; | |
5100 | struct map_lookup *map; | |
5101 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5102 | int ret = 0; | |
5103 | ||
5104 | read_lock(&em_tree->lock); | |
5105 | em = lookup_extent_mapping(em_tree, logical, len); | |
5106 | read_unlock(&em_tree->lock); | |
5107 | BUG_ON(!em); | |
5108 | ||
5109 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
5110 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 5111 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
5112 | ret = 1; |
5113 | free_extent_map(em); | |
5114 | return ret; | |
5115 | } | |
5116 | ||
30d9861f SB |
5117 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5118 | struct map_lookup *map, int first, int num, | |
5119 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5120 | { |
5121 | int i; | |
30d9861f SB |
5122 | int tolerance; |
5123 | struct btrfs_device *srcdev; | |
5124 | ||
5125 | if (dev_replace_is_ongoing && | |
5126 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5127 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5128 | srcdev = fs_info->dev_replace.srcdev; | |
5129 | else | |
5130 | srcdev = NULL; | |
5131 | ||
5132 | /* | |
5133 | * try to avoid the drive that is the source drive for a | |
5134 | * dev-replace procedure, only choose it if no other non-missing | |
5135 | * mirror is available | |
5136 | */ | |
5137 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5138 | if (map->stripes[optimal].dev->bdev && | |
5139 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5140 | return optimal; | |
5141 | for (i = first; i < first + num; i++) { | |
5142 | if (map->stripes[i].dev->bdev && | |
5143 | (tolerance || map->stripes[i].dev != srcdev)) | |
5144 | return i; | |
5145 | } | |
dfe25020 | 5146 | } |
30d9861f | 5147 | |
dfe25020 CM |
5148 | /* we couldn't find one that doesn't fail. Just return something |
5149 | * and the io error handling code will clean up eventually | |
5150 | */ | |
5151 | return optimal; | |
5152 | } | |
5153 | ||
53b381b3 DW |
5154 | static inline int parity_smaller(u64 a, u64 b) |
5155 | { | |
5156 | return a > b; | |
5157 | } | |
5158 | ||
5159 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5160 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5161 | { |
5162 | struct btrfs_bio_stripe s; | |
5163 | int i; | |
5164 | u64 l; | |
5165 | int again = 1; | |
5166 | ||
5167 | while (again) { | |
5168 | again = 0; | |
cc7539ed | 5169 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5170 | if (parity_smaller(bbio->raid_map[i], |
5171 | bbio->raid_map[i+1])) { | |
53b381b3 | 5172 | s = bbio->stripes[i]; |
8e5cfb55 | 5173 | l = bbio->raid_map[i]; |
53b381b3 | 5174 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5175 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5176 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5177 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5178 | |
53b381b3 DW |
5179 | again = 1; |
5180 | } | |
5181 | } | |
5182 | } | |
5183 | } | |
5184 | ||
6e9606d2 ZL |
5185 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5186 | { | |
5187 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5188 | /* the size of the btrfs_bio */ |
6e9606d2 | 5189 | sizeof(struct btrfs_bio) + |
e57cf21e | 5190 | /* plus the variable array for the stripes */ |
6e9606d2 | 5191 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5192 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5193 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5194 | /* |
5195 | * plus the raid_map, which includes both the tgt dev | |
5196 | * and the stripes | |
5197 | */ | |
5198 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5199 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5200 | |
5201 | atomic_set(&bbio->error, 0); | |
5202 | atomic_set(&bbio->refs, 1); | |
5203 | ||
5204 | return bbio; | |
5205 | } | |
5206 | ||
5207 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5208 | { | |
5209 | WARN_ON(!atomic_read(&bbio->refs)); | |
5210 | atomic_inc(&bbio->refs); | |
5211 | } | |
5212 | ||
5213 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5214 | { | |
5215 | if (!bbio) | |
5216 | return; | |
5217 | if (atomic_dec_and_test(&bbio->refs)) | |
5218 | kfree(bbio); | |
5219 | } | |
5220 | ||
3ec706c8 | 5221 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5222 | u64 logical, u64 *length, |
a1d3c478 | 5223 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5224 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5225 | { |
5226 | struct extent_map *em; | |
5227 | struct map_lookup *map; | |
3ec706c8 | 5228 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
5229 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5230 | u64 offset; | |
593060d7 | 5231 | u64 stripe_offset; |
fce3bb9a | 5232 | u64 stripe_end_offset; |
593060d7 | 5233 | u64 stripe_nr; |
fce3bb9a LD |
5234 | u64 stripe_nr_orig; |
5235 | u64 stripe_nr_end; | |
53b381b3 | 5236 | u64 stripe_len; |
9d644a62 | 5237 | u32 stripe_index; |
cea9e445 | 5238 | int i; |
de11cc12 | 5239 | int ret = 0; |
f2d8d74d | 5240 | int num_stripes; |
a236aed1 | 5241 | int max_errors = 0; |
2c8cdd6e | 5242 | int tgtdev_indexes = 0; |
a1d3c478 | 5243 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5244 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5245 | int dev_replace_is_ongoing = 0; | |
5246 | int num_alloc_stripes; | |
ad6d620e SB |
5247 | int patch_the_first_stripe_for_dev_replace = 0; |
5248 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5249 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5250 | |
890871be | 5251 | read_lock(&em_tree->lock); |
0b86a832 | 5252 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 5253 | read_unlock(&em_tree->lock); |
f2d8d74d | 5254 | |
3b951516 | 5255 | if (!em) { |
c2cf52eb | 5256 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 5257 | logical, *length); |
9bb91873 JB |
5258 | return -EINVAL; |
5259 | } | |
5260 | ||
5261 | if (em->start > logical || em->start + em->len < logical) { | |
5262 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
351fd353 | 5263 | "found %Lu-%Lu", logical, em->start, |
9bb91873 | 5264 | em->start + em->len); |
7d3d1744 | 5265 | free_extent_map(em); |
9bb91873 | 5266 | return -EINVAL; |
3b951516 | 5267 | } |
0b86a832 | 5268 | |
0b86a832 CM |
5269 | map = (struct map_lookup *)em->bdev; |
5270 | offset = logical - em->start; | |
593060d7 | 5271 | |
53b381b3 | 5272 | stripe_len = map->stripe_len; |
593060d7 CM |
5273 | stripe_nr = offset; |
5274 | /* | |
5275 | * stripe_nr counts the total number of stripes we have to stride | |
5276 | * to get to this block | |
5277 | */ | |
47c5713f | 5278 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5279 | |
53b381b3 | 5280 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
5281 | BUG_ON(offset < stripe_offset); |
5282 | ||
5283 | /* stripe_offset is the offset of this block in its stripe*/ | |
5284 | stripe_offset = offset - stripe_offset; | |
5285 | ||
53b381b3 | 5286 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5287 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5288 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5289 | raid56_full_stripe_start = offset; | |
5290 | ||
5291 | /* allow a write of a full stripe, but make sure we don't | |
5292 | * allow straddling of stripes | |
5293 | */ | |
47c5713f DS |
5294 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5295 | full_stripe_len); | |
53b381b3 DW |
5296 | raid56_full_stripe_start *= full_stripe_len; |
5297 | } | |
5298 | ||
5299 | if (rw & REQ_DISCARD) { | |
5300 | /* we don't discard raid56 yet */ | |
ffe2d203 | 5301 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5302 | ret = -EOPNOTSUPP; |
5303 | goto out; | |
5304 | } | |
fce3bb9a | 5305 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
5306 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
5307 | u64 max_len; | |
5308 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5309 | For other RAID types and for RAID[56] reads, just allow a single | |
5310 | stripe (on a single disk). */ | |
ffe2d203 | 5311 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
53b381b3 DW |
5312 | (rw & REQ_WRITE)) { |
5313 | max_len = stripe_len * nr_data_stripes(map) - | |
5314 | (offset - raid56_full_stripe_start); | |
5315 | } else { | |
5316 | /* we limit the length of each bio to what fits in a stripe */ | |
5317 | max_len = stripe_len - stripe_offset; | |
5318 | } | |
5319 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5320 | } else { |
5321 | *length = em->len - offset; | |
5322 | } | |
f2d8d74d | 5323 | |
53b381b3 DW |
5324 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5325 | it cares about is the length */ | |
a1d3c478 | 5326 | if (!bbio_ret) |
cea9e445 CM |
5327 | goto out; |
5328 | ||
472262f3 SB |
5329 | btrfs_dev_replace_lock(dev_replace); |
5330 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
5331 | if (!dev_replace_is_ongoing) | |
5332 | btrfs_dev_replace_unlock(dev_replace); | |
5333 | ||
ad6d620e SB |
5334 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
5335 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
5336 | dev_replace->tgtdev != NULL) { | |
5337 | /* | |
5338 | * in dev-replace case, for repair case (that's the only | |
5339 | * case where the mirror is selected explicitly when | |
5340 | * calling btrfs_map_block), blocks left of the left cursor | |
5341 | * can also be read from the target drive. | |
5342 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5343 | * the last one to the array of stripes. For READ, it also | |
5344 | * needs to be supported using the same mirror number. | |
5345 | * If the requested block is not left of the left cursor, | |
5346 | * EIO is returned. This can happen because btrfs_num_copies() | |
5347 | * returns one more in the dev-replace case. | |
5348 | */ | |
5349 | u64 tmp_length = *length; | |
5350 | struct btrfs_bio *tmp_bbio = NULL; | |
5351 | int tmp_num_stripes; | |
5352 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5353 | int index_srcdev = 0; | |
5354 | int found = 0; | |
5355 | u64 physical_of_found = 0; | |
5356 | ||
5357 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
8e5cfb55 | 5358 | logical, &tmp_length, &tmp_bbio, 0, 0); |
ad6d620e SB |
5359 | if (ret) { |
5360 | WARN_ON(tmp_bbio != NULL); | |
5361 | goto out; | |
5362 | } | |
5363 | ||
5364 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5365 | if (mirror_num > tmp_num_stripes) { | |
5366 | /* | |
5367 | * REQ_GET_READ_MIRRORS does not contain this | |
5368 | * mirror, that means that the requested area | |
5369 | * is not left of the left cursor | |
5370 | */ | |
5371 | ret = -EIO; | |
6e9606d2 | 5372 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5373 | goto out; |
5374 | } | |
5375 | ||
5376 | /* | |
5377 | * process the rest of the function using the mirror_num | |
5378 | * of the source drive. Therefore look it up first. | |
5379 | * At the end, patch the device pointer to the one of the | |
5380 | * target drive. | |
5381 | */ | |
5382 | for (i = 0; i < tmp_num_stripes; i++) { | |
5383 | if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) { | |
5384 | /* | |
5385 | * In case of DUP, in order to keep it | |
5386 | * simple, only add the mirror with the | |
5387 | * lowest physical address | |
5388 | */ | |
5389 | if (found && | |
5390 | physical_of_found <= | |
5391 | tmp_bbio->stripes[i].physical) | |
5392 | continue; | |
5393 | index_srcdev = i; | |
5394 | found = 1; | |
5395 | physical_of_found = | |
5396 | tmp_bbio->stripes[i].physical; | |
5397 | } | |
5398 | } | |
5399 | ||
5400 | if (found) { | |
5401 | mirror_num = index_srcdev + 1; | |
5402 | patch_the_first_stripe_for_dev_replace = 1; | |
5403 | physical_to_patch_in_first_stripe = physical_of_found; | |
5404 | } else { | |
5405 | WARN_ON(1); | |
5406 | ret = -EIO; | |
6e9606d2 | 5407 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5408 | goto out; |
5409 | } | |
5410 | ||
6e9606d2 | 5411 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5412 | } else if (mirror_num > map->num_stripes) { |
5413 | mirror_num = 0; | |
5414 | } | |
5415 | ||
f2d8d74d | 5416 | num_stripes = 1; |
cea9e445 | 5417 | stripe_index = 0; |
fce3bb9a | 5418 | stripe_nr_orig = stripe_nr; |
fda2832f | 5419 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
b8b93add | 5420 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); |
fce3bb9a LD |
5421 | stripe_end_offset = stripe_nr_end * map->stripe_len - |
5422 | (offset + *length); | |
53b381b3 | 5423 | |
fce3bb9a LD |
5424 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5425 | if (rw & REQ_DISCARD) | |
5426 | num_stripes = min_t(u64, map->num_stripes, | |
5427 | stripe_nr_end - stripe_nr_orig); | |
47c5713f DS |
5428 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5429 | &stripe_index); | |
28e1cc7d MX |
5430 | if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))) |
5431 | mirror_num = 1; | |
fce3bb9a | 5432 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
29a8d9a0 | 5433 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5434 | num_stripes = map->num_stripes; |
2fff734f | 5435 | else if (mirror_num) |
f188591e | 5436 | stripe_index = mirror_num - 1; |
dfe25020 | 5437 | else { |
30d9861f | 5438 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5439 | map->num_stripes, |
30d9861f SB |
5440 | current->pid % map->num_stripes, |
5441 | dev_replace_is_ongoing); | |
a1d3c478 | 5442 | mirror_num = stripe_index + 1; |
dfe25020 | 5443 | } |
2fff734f | 5444 | |
611f0e00 | 5445 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 5446 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 5447 | num_stripes = map->num_stripes; |
a1d3c478 | 5448 | } else if (mirror_num) { |
f188591e | 5449 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5450 | } else { |
5451 | mirror_num = 1; | |
5452 | } | |
2fff734f | 5453 | |
321aecc6 | 5454 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5455 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5456 | |
47c5713f | 5457 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5458 | stripe_index *= map->sub_stripes; |
5459 | ||
29a8d9a0 | 5460 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5461 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
5462 | else if (rw & REQ_DISCARD) |
5463 | num_stripes = min_t(u64, map->sub_stripes * | |
5464 | (stripe_nr_end - stripe_nr_orig), | |
5465 | map->num_stripes); | |
321aecc6 CM |
5466 | else if (mirror_num) |
5467 | stripe_index += mirror_num - 1; | |
dfe25020 | 5468 | else { |
3e74317a | 5469 | int old_stripe_index = stripe_index; |
30d9861f SB |
5470 | stripe_index = find_live_mirror(fs_info, map, |
5471 | stripe_index, | |
dfe25020 | 5472 | map->sub_stripes, stripe_index + |
30d9861f SB |
5473 | current->pid % map->sub_stripes, |
5474 | dev_replace_is_ongoing); | |
3e74317a | 5475 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5476 | } |
53b381b3 | 5477 | |
ffe2d203 | 5478 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5479 | if (need_raid_map && |
af8e2d1d MX |
5480 | ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5481 | mirror_num > 1)) { | |
53b381b3 | 5482 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5483 | stripe_nr = div_u64(raid56_full_stripe_start, |
5484 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5485 | |
5486 | /* RAID[56] write or recovery. Return all stripes */ | |
5487 | num_stripes = map->num_stripes; | |
5488 | max_errors = nr_parity_stripes(map); | |
5489 | ||
53b381b3 DW |
5490 | *length = map->stripe_len; |
5491 | stripe_index = 0; | |
5492 | stripe_offset = 0; | |
5493 | } else { | |
5494 | /* | |
5495 | * Mirror #0 or #1 means the original data block. | |
5496 | * Mirror #2 is RAID5 parity block. | |
5497 | * Mirror #3 is RAID6 Q block. | |
5498 | */ | |
47c5713f DS |
5499 | stripe_nr = div_u64_rem(stripe_nr, |
5500 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5501 | if (mirror_num > 1) |
5502 | stripe_index = nr_data_stripes(map) + | |
5503 | mirror_num - 2; | |
5504 | ||
5505 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5506 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5507 | &stripe_index); | |
28e1cc7d MX |
5508 | if (!(rw & (REQ_WRITE | REQ_DISCARD | |
5509 | REQ_GET_READ_MIRRORS)) && mirror_num <= 1) | |
5510 | mirror_num = 1; | |
53b381b3 | 5511 | } |
8790d502 CM |
5512 | } else { |
5513 | /* | |
47c5713f DS |
5514 | * after this, stripe_nr is the number of stripes on this |
5515 | * device we have to walk to find the data, and stripe_index is | |
5516 | * the number of our device in the stripe array | |
8790d502 | 5517 | */ |
47c5713f DS |
5518 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5519 | &stripe_index); | |
a1d3c478 | 5520 | mirror_num = stripe_index + 1; |
8790d502 | 5521 | } |
593060d7 | 5522 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 5523 | |
472262f3 | 5524 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
5525 | if (dev_replace_is_ongoing) { |
5526 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
5527 | num_alloc_stripes <<= 1; | |
5528 | if (rw & REQ_GET_READ_MIRRORS) | |
5529 | num_alloc_stripes++; | |
2c8cdd6e | 5530 | tgtdev_indexes = num_stripes; |
ad6d620e | 5531 | } |
2c8cdd6e | 5532 | |
6e9606d2 | 5533 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5534 | if (!bbio) { |
5535 | ret = -ENOMEM; | |
5536 | goto out; | |
5537 | } | |
2c8cdd6e MX |
5538 | if (dev_replace_is_ongoing) |
5539 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
de11cc12 | 5540 | |
8e5cfb55 | 5541 | /* build raid_map */ |
ffe2d203 | 5542 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
8e5cfb55 ZL |
5543 | need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5544 | mirror_num > 1)) { | |
5545 | u64 tmp; | |
9d644a62 | 5546 | unsigned rot; |
8e5cfb55 ZL |
5547 | |
5548 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5549 | sizeof(struct btrfs_bio_stripe) * | |
5550 | num_alloc_stripes + | |
5551 | sizeof(int) * tgtdev_indexes); | |
5552 | ||
5553 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5554 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5555 | |
5556 | /* Fill in the logical address of each stripe */ | |
5557 | tmp = stripe_nr * nr_data_stripes(map); | |
5558 | for (i = 0; i < nr_data_stripes(map); i++) | |
5559 | bbio->raid_map[(i+rot) % num_stripes] = | |
5560 | em->start + (tmp + i) * map->stripe_len; | |
5561 | ||
5562 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5563 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5564 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5565 | RAID6_Q_STRIPE; | |
5566 | } | |
5567 | ||
fce3bb9a | 5568 | if (rw & REQ_DISCARD) { |
9d644a62 DS |
5569 | u32 factor = 0; |
5570 | u32 sub_stripes = 0; | |
ec9ef7a1 LZ |
5571 | u64 stripes_per_dev = 0; |
5572 | u32 remaining_stripes = 0; | |
b89203f7 | 5573 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5574 | |
5575 | if (map->type & | |
5576 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5577 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5578 | sub_stripes = 1; | |
5579 | else | |
5580 | sub_stripes = map->sub_stripes; | |
5581 | ||
5582 | factor = map->num_stripes / sub_stripes; | |
5583 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5584 | stripe_nr_orig, | |
5585 | factor, | |
5586 | &remaining_stripes); | |
b89203f7 LB |
5587 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5588 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5589 | } |
5590 | ||
fce3bb9a | 5591 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5592 | bbio->stripes[i].physical = |
f2d8d74d CM |
5593 | map->stripes[stripe_index].physical + |
5594 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5595 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5596 | |
ec9ef7a1 LZ |
5597 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5598 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5599 | bbio->stripes[i].length = stripes_per_dev * | |
5600 | map->stripe_len; | |
b89203f7 | 5601 | |
ec9ef7a1 LZ |
5602 | if (i / sub_stripes < remaining_stripes) |
5603 | bbio->stripes[i].length += | |
5604 | map->stripe_len; | |
b89203f7 LB |
5605 | |
5606 | /* | |
5607 | * Special for the first stripe and | |
5608 | * the last stripe: | |
5609 | * | |
5610 | * |-------|...|-------| | |
5611 | * |----------| | |
5612 | * off end_off | |
5613 | */ | |
ec9ef7a1 | 5614 | if (i < sub_stripes) |
a1d3c478 | 5615 | bbio->stripes[i].length -= |
fce3bb9a | 5616 | stripe_offset; |
b89203f7 LB |
5617 | |
5618 | if (stripe_index >= last_stripe && | |
5619 | stripe_index <= (last_stripe + | |
5620 | sub_stripes - 1)) | |
a1d3c478 | 5621 | bbio->stripes[i].length -= |
fce3bb9a | 5622 | stripe_end_offset; |
b89203f7 | 5623 | |
ec9ef7a1 LZ |
5624 | if (i == sub_stripes - 1) |
5625 | stripe_offset = 0; | |
fce3bb9a | 5626 | } else |
a1d3c478 | 5627 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5628 | |
5629 | stripe_index++; | |
5630 | if (stripe_index == map->num_stripes) { | |
5631 | /* This could only happen for RAID0/10 */ | |
5632 | stripe_index = 0; | |
5633 | stripe_nr++; | |
5634 | } | |
5635 | } | |
5636 | } else { | |
5637 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5638 | bbio->stripes[i].physical = |
212a17ab LT |
5639 | map->stripes[stripe_index].physical + |
5640 | stripe_offset + | |
5641 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5642 | bbio->stripes[i].dev = |
212a17ab | 5643 | map->stripes[stripe_index].dev; |
fce3bb9a | 5644 | stripe_index++; |
f2d8d74d | 5645 | } |
593060d7 | 5646 | } |
de11cc12 | 5647 | |
d20983b4 MX |
5648 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
5649 | max_errors = btrfs_chunk_max_errors(map); | |
de11cc12 | 5650 | |
8e5cfb55 ZL |
5651 | if (bbio->raid_map) |
5652 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5653 | |
2c8cdd6e | 5654 | tgtdev_indexes = 0; |
472262f3 SB |
5655 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
5656 | dev_replace->tgtdev != NULL) { | |
5657 | int index_where_to_add; | |
5658 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5659 | ||
5660 | /* | |
5661 | * duplicate the write operations while the dev replace | |
5662 | * procedure is running. Since the copying of the old disk | |
5663 | * to the new disk takes place at run time while the | |
5664 | * filesystem is mounted writable, the regular write | |
5665 | * operations to the old disk have to be duplicated to go | |
5666 | * to the new disk as well. | |
5667 | * Note that device->missing is handled by the caller, and | |
5668 | * that the write to the old disk is already set up in the | |
5669 | * stripes array. | |
5670 | */ | |
5671 | index_where_to_add = num_stripes; | |
5672 | for (i = 0; i < num_stripes; i++) { | |
5673 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5674 | /* write to new disk, too */ | |
5675 | struct btrfs_bio_stripe *new = | |
5676 | bbio->stripes + index_where_to_add; | |
5677 | struct btrfs_bio_stripe *old = | |
5678 | bbio->stripes + i; | |
5679 | ||
5680 | new->physical = old->physical; | |
5681 | new->length = old->length; | |
5682 | new->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5683 | bbio->tgtdev_map[i] = index_where_to_add; |
472262f3 SB |
5684 | index_where_to_add++; |
5685 | max_errors++; | |
2c8cdd6e | 5686 | tgtdev_indexes++; |
472262f3 SB |
5687 | } |
5688 | } | |
5689 | num_stripes = index_where_to_add; | |
ad6d620e SB |
5690 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
5691 | dev_replace->tgtdev != NULL) { | |
5692 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5693 | int index_srcdev = 0; | |
5694 | int found = 0; | |
5695 | u64 physical_of_found = 0; | |
5696 | ||
5697 | /* | |
5698 | * During the dev-replace procedure, the target drive can | |
5699 | * also be used to read data in case it is needed to repair | |
5700 | * a corrupt block elsewhere. This is possible if the | |
5701 | * requested area is left of the left cursor. In this area, | |
5702 | * the target drive is a full copy of the source drive. | |
5703 | */ | |
5704 | for (i = 0; i < num_stripes; i++) { | |
5705 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5706 | /* | |
5707 | * In case of DUP, in order to keep it | |
5708 | * simple, only add the mirror with the | |
5709 | * lowest physical address | |
5710 | */ | |
5711 | if (found && | |
5712 | physical_of_found <= | |
5713 | bbio->stripes[i].physical) | |
5714 | continue; | |
5715 | index_srcdev = i; | |
5716 | found = 1; | |
5717 | physical_of_found = bbio->stripes[i].physical; | |
5718 | } | |
5719 | } | |
5720 | if (found) { | |
258ece02 | 5721 | if (physical_of_found + map->stripe_len <= |
ad6d620e SB |
5722 | dev_replace->cursor_left) { |
5723 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5724 | bbio->stripes + num_stripes; | |
5725 | ||
5726 | tgtdev_stripe->physical = physical_of_found; | |
5727 | tgtdev_stripe->length = | |
5728 | bbio->stripes[index_srcdev].length; | |
5729 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5730 | bbio->tgtdev_map[index_srcdev] = num_stripes; |
ad6d620e | 5731 | |
2c8cdd6e | 5732 | tgtdev_indexes++; |
ad6d620e SB |
5733 | num_stripes++; |
5734 | } | |
5735 | } | |
472262f3 SB |
5736 | } |
5737 | ||
de11cc12 | 5738 | *bbio_ret = bbio; |
10f11900 | 5739 | bbio->map_type = map->type; |
de11cc12 LZ |
5740 | bbio->num_stripes = num_stripes; |
5741 | bbio->max_errors = max_errors; | |
5742 | bbio->mirror_num = mirror_num; | |
2c8cdd6e | 5743 | bbio->num_tgtdevs = tgtdev_indexes; |
ad6d620e SB |
5744 | |
5745 | /* | |
5746 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5747 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5748 | * available as a mirror | |
5749 | */ | |
5750 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5751 | WARN_ON(num_stripes > 1); | |
5752 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5753 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5754 | bbio->mirror_num = map->num_stripes + 1; | |
5755 | } | |
cea9e445 | 5756 | out: |
472262f3 SB |
5757 | if (dev_replace_is_ongoing) |
5758 | btrfs_dev_replace_unlock(dev_replace); | |
0b86a832 | 5759 | free_extent_map(em); |
de11cc12 | 5760 | return ret; |
0b86a832 CM |
5761 | } |
5762 | ||
3ec706c8 | 5763 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5764 | u64 logical, u64 *length, |
a1d3c478 | 5765 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5766 | { |
3ec706c8 | 5767 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
8e5cfb55 | 5768 | mirror_num, 0); |
f2d8d74d CM |
5769 | } |
5770 | ||
af8e2d1d MX |
5771 | /* For Scrub/replace */ |
5772 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw, | |
5773 | u64 logical, u64 *length, | |
5774 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5775 | int need_raid_map) |
af8e2d1d MX |
5776 | { |
5777 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, | |
8e5cfb55 | 5778 | mirror_num, need_raid_map); |
af8e2d1d MX |
5779 | } |
5780 | ||
a512bbf8 YZ |
5781 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
5782 | u64 chunk_start, u64 physical, u64 devid, | |
5783 | u64 **logical, int *naddrs, int *stripe_len) | |
5784 | { | |
5785 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5786 | struct extent_map *em; | |
5787 | struct map_lookup *map; | |
5788 | u64 *buf; | |
5789 | u64 bytenr; | |
5790 | u64 length; | |
5791 | u64 stripe_nr; | |
53b381b3 | 5792 | u64 rmap_len; |
a512bbf8 YZ |
5793 | int i, j, nr = 0; |
5794 | ||
890871be | 5795 | read_lock(&em_tree->lock); |
a512bbf8 | 5796 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5797 | read_unlock(&em_tree->lock); |
a512bbf8 | 5798 | |
835d974f | 5799 | if (!em) { |
efe120a0 | 5800 | printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n", |
835d974f JB |
5801 | chunk_start); |
5802 | return -EIO; | |
5803 | } | |
5804 | ||
5805 | if (em->start != chunk_start) { | |
efe120a0 | 5806 | printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n", |
835d974f JB |
5807 | em->start, chunk_start); |
5808 | free_extent_map(em); | |
5809 | return -EIO; | |
5810 | } | |
a512bbf8 YZ |
5811 | map = (struct map_lookup *)em->bdev; |
5812 | ||
5813 | length = em->len; | |
53b381b3 DW |
5814 | rmap_len = map->stripe_len; |
5815 | ||
a512bbf8 | 5816 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5817 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5818 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5819 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5820 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5821 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5822 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5823 | } | |
a512bbf8 | 5824 | |
31e818fe | 5825 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5826 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5827 | |
5828 | for (i = 0; i < map->num_stripes; i++) { | |
5829 | if (devid && map->stripes[i].dev->devid != devid) | |
5830 | continue; | |
5831 | if (map->stripes[i].physical > physical || | |
5832 | map->stripes[i].physical + length <= physical) | |
5833 | continue; | |
5834 | ||
5835 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 5836 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5837 | |
5838 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5839 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5840 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5841 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5842 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5843 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5844 | * Alternatively, just use rmap_len below instead of | |
5845 | * map->stripe_len */ | |
5846 | ||
5847 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5848 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5849 | for (j = 0; j < nr; j++) { |
5850 | if (buf[j] == bytenr) | |
5851 | break; | |
5852 | } | |
934d375b CM |
5853 | if (j == nr) { |
5854 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5855 | buf[nr++] = bytenr; |
934d375b | 5856 | } |
a512bbf8 YZ |
5857 | } |
5858 | ||
a512bbf8 YZ |
5859 | *logical = buf; |
5860 | *naddrs = nr; | |
53b381b3 | 5861 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5862 | |
5863 | free_extent_map(em); | |
5864 | return 0; | |
f2d8d74d CM |
5865 | } |
5866 | ||
4246a0b6 | 5867 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 5868 | { |
326e1dbb MS |
5869 | bio->bi_private = bbio->private; |
5870 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 5871 | bio_endio(bio); |
326e1dbb | 5872 | |
6e9606d2 | 5873 | btrfs_put_bbio(bbio); |
8408c716 MX |
5874 | } |
5875 | ||
4246a0b6 | 5876 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 5877 | { |
9be3395b | 5878 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 5879 | int is_orig_bio = 0; |
8790d502 | 5880 | |
4246a0b6 | 5881 | if (bio->bi_error) { |
a1d3c478 | 5882 | atomic_inc(&bbio->error); |
4246a0b6 | 5883 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { |
442a4f63 | 5884 | unsigned int stripe_index = |
9be3395b | 5885 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 5886 | struct btrfs_device *dev; |
442a4f63 SB |
5887 | |
5888 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5889 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5890 | if (dev->bdev) { |
5891 | if (bio->bi_rw & WRITE) | |
5892 | btrfs_dev_stat_inc(dev, | |
5893 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5894 | else | |
5895 | btrfs_dev_stat_inc(dev, | |
5896 | BTRFS_DEV_STAT_READ_ERRS); | |
5897 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5898 | btrfs_dev_stat_inc(dev, | |
5899 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5900 | btrfs_dev_stat_print_on_error(dev); | |
5901 | } | |
442a4f63 SB |
5902 | } |
5903 | } | |
8790d502 | 5904 | |
a1d3c478 | 5905 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5906 | is_orig_bio = 1; |
5907 | ||
c404e0dc MX |
5908 | btrfs_bio_counter_dec(bbio->fs_info); |
5909 | ||
a1d3c478 | 5910 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5911 | if (!is_orig_bio) { |
5912 | bio_put(bio); | |
a1d3c478 | 5913 | bio = bbio->orig_bio; |
7d2b4daa | 5914 | } |
c7b22bb1 | 5915 | |
9be3395b | 5916 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 5917 | /* only send an error to the higher layers if it is |
53b381b3 | 5918 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5919 | */ |
a1d3c478 | 5920 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4246a0b6 | 5921 | bio->bi_error = -EIO; |
5dbc8fca | 5922 | } else { |
1259ab75 CM |
5923 | /* |
5924 | * this bio is actually up to date, we didn't | |
5925 | * go over the max number of errors | |
5926 | */ | |
4246a0b6 | 5927 | bio->bi_error = 0; |
1259ab75 | 5928 | } |
c55f1396 | 5929 | |
4246a0b6 | 5930 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 5931 | } else if (!is_orig_bio) { |
8790d502 CM |
5932 | bio_put(bio); |
5933 | } | |
8790d502 CM |
5934 | } |
5935 | ||
8b712842 CM |
5936 | /* |
5937 | * see run_scheduled_bios for a description of why bios are collected for | |
5938 | * async submit. | |
5939 | * | |
5940 | * This will add one bio to the pending list for a device and make sure | |
5941 | * the work struct is scheduled. | |
5942 | */ | |
48a3b636 ES |
5943 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
5944 | struct btrfs_device *device, | |
5945 | int rw, struct bio *bio) | |
8b712842 CM |
5946 | { |
5947 | int should_queue = 1; | |
ffbd517d | 5948 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5949 | |
53b381b3 | 5950 | if (device->missing || !device->bdev) { |
4246a0b6 | 5951 | bio_io_error(bio); |
53b381b3 DW |
5952 | return; |
5953 | } | |
5954 | ||
8b712842 | 5955 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5956 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5957 | bio_get(bio); |
21adbd5c | 5958 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5959 | bio_put(bio); |
143bede5 | 5960 | return; |
8b712842 CM |
5961 | } |
5962 | ||
5963 | /* | |
0986fe9e | 5964 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5965 | * higher layers. Otherwise, the async bio makes it appear we have |
5966 | * made progress against dirty pages when we've really just put it | |
5967 | * on a queue for later | |
5968 | */ | |
0986fe9e | 5969 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5970 | WARN_ON(bio->bi_next); |
8b712842 CM |
5971 | bio->bi_next = NULL; |
5972 | bio->bi_rw |= rw; | |
5973 | ||
5974 | spin_lock(&device->io_lock); | |
7b6d91da | 5975 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5976 | pending_bios = &device->pending_sync_bios; |
5977 | else | |
5978 | pending_bios = &device->pending_bios; | |
8b712842 | 5979 | |
ffbd517d CM |
5980 | if (pending_bios->tail) |
5981 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5982 | |
ffbd517d CM |
5983 | pending_bios->tail = bio; |
5984 | if (!pending_bios->head) | |
5985 | pending_bios->head = bio; | |
8b712842 CM |
5986 | if (device->running_pending) |
5987 | should_queue = 0; | |
5988 | ||
5989 | spin_unlock(&device->io_lock); | |
5990 | ||
5991 | if (should_queue) | |
a8c93d4e QW |
5992 | btrfs_queue_work(root->fs_info->submit_workers, |
5993 | &device->work); | |
8b712842 CM |
5994 | } |
5995 | ||
de1ee92a JB |
5996 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, |
5997 | struct bio *bio, u64 physical, int dev_nr, | |
5998 | int rw, int async) | |
5999 | { | |
6000 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
6001 | ||
6002 | bio->bi_private = bbio; | |
9be3395b | 6003 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6004 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6005 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6006 | #ifdef DEBUG |
6007 | { | |
6008 | struct rcu_string *name; | |
6009 | ||
6010 | rcu_read_lock(); | |
6011 | name = rcu_dereference(dev->name); | |
d1423248 | 6012 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a | 6013 | "(%s id %llu), size=%u\n", rw, |
1b6e4469 FF |
6014 | (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev, |
6015 | name->str, dev->devid, bio->bi_iter.bi_size); | |
de1ee92a JB |
6016 | rcu_read_unlock(); |
6017 | } | |
6018 | #endif | |
6019 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
6020 | |
6021 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
6022 | ||
de1ee92a | 6023 | if (async) |
53b381b3 | 6024 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
6025 | else |
6026 | btrfsic_submit_bio(rw, bio); | |
6027 | } | |
6028 | ||
de1ee92a JB |
6029 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6030 | { | |
6031 | atomic_inc(&bbio->error); | |
6032 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
8408c716 MX |
6033 | /* Shoud be the original bio. */ |
6034 | WARN_ON(bio != bbio->orig_bio); | |
6035 | ||
9be3395b | 6036 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6037 | bio->bi_iter.bi_sector = logical >> 9; |
4246a0b6 CH |
6038 | bio->bi_error = -EIO; |
6039 | btrfs_end_bbio(bbio, bio); | |
de1ee92a JB |
6040 | } |
6041 | } | |
6042 | ||
f188591e | 6043 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 6044 | int mirror_num, int async_submit) |
0b86a832 | 6045 | { |
0b86a832 | 6046 | struct btrfs_device *dev; |
8790d502 | 6047 | struct bio *first_bio = bio; |
4f024f37 | 6048 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6049 | u64 length = 0; |
6050 | u64 map_length; | |
0b86a832 | 6051 | int ret; |
08da757d ZL |
6052 | int dev_nr; |
6053 | int total_devs; | |
a1d3c478 | 6054 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6055 | |
4f024f37 | 6056 | length = bio->bi_iter.bi_size; |
0b86a832 | 6057 | map_length = length; |
cea9e445 | 6058 | |
c404e0dc | 6059 | btrfs_bio_counter_inc_blocked(root->fs_info); |
53b381b3 | 6060 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
8e5cfb55 | 6061 | mirror_num, 1); |
c404e0dc MX |
6062 | if (ret) { |
6063 | btrfs_bio_counter_dec(root->fs_info); | |
79787eaa | 6064 | return ret; |
c404e0dc | 6065 | } |
cea9e445 | 6066 | |
a1d3c478 | 6067 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6068 | bbio->orig_bio = first_bio; |
6069 | bbio->private = first_bio->bi_private; | |
6070 | bbio->end_io = first_bio->bi_end_io; | |
c404e0dc | 6071 | bbio->fs_info = root->fs_info; |
53b381b3 DW |
6072 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6073 | ||
8e5cfb55 | 6074 | if (bbio->raid_map) { |
53b381b3 DW |
6075 | /* In this case, map_length has been set to the length of |
6076 | a single stripe; not the whole write */ | |
6077 | if (rw & WRITE) { | |
8e5cfb55 | 6078 | ret = raid56_parity_write(root, bio, bbio, map_length); |
53b381b3 | 6079 | } else { |
8e5cfb55 | 6080 | ret = raid56_parity_recover(root, bio, bbio, map_length, |
4245215d | 6081 | mirror_num, 1); |
53b381b3 | 6082 | } |
4245215d | 6083 | |
c404e0dc MX |
6084 | btrfs_bio_counter_dec(root->fs_info); |
6085 | return ret; | |
53b381b3 DW |
6086 | } |
6087 | ||
cea9e445 | 6088 | if (map_length < length) { |
c2cf52eb | 6089 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", |
c1c9ff7c | 6090 | logical, length, map_length); |
cea9e445 CM |
6091 | BUG(); |
6092 | } | |
a1d3c478 | 6093 | |
08da757d | 6094 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a JB |
6095 | dev = bbio->stripes[dev_nr].dev; |
6096 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
6097 | bbio_error(bbio, first_bio, logical); | |
de1ee92a JB |
6098 | continue; |
6099 | } | |
6100 | ||
a1d3c478 | 6101 | if (dev_nr < total_devs - 1) { |
9be3395b | 6102 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 6103 | BUG_ON(!bio); /* -ENOMEM */ |
326e1dbb | 6104 | } else |
a1d3c478 | 6105 | bio = first_bio; |
de1ee92a JB |
6106 | |
6107 | submit_stripe_bio(root, bbio, bio, | |
6108 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
6109 | async_submit); | |
8790d502 | 6110 | } |
c404e0dc | 6111 | btrfs_bio_counter_dec(root->fs_info); |
0b86a832 CM |
6112 | return 0; |
6113 | } | |
6114 | ||
aa1b8cd4 | 6115 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6116 | u8 *uuid, u8 *fsid) |
0b86a832 | 6117 | { |
2b82032c YZ |
6118 | struct btrfs_device *device; |
6119 | struct btrfs_fs_devices *cur_devices; | |
6120 | ||
aa1b8cd4 | 6121 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6122 | while (cur_devices) { |
6123 | if (!fsid || | |
6124 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6125 | device = __find_device(&cur_devices->devices, | |
6126 | devid, uuid); | |
6127 | if (device) | |
6128 | return device; | |
6129 | } | |
6130 | cur_devices = cur_devices->seed; | |
6131 | } | |
6132 | return NULL; | |
0b86a832 CM |
6133 | } |
6134 | ||
dfe25020 | 6135 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 6136 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6137 | u64 devid, u8 *dev_uuid) |
6138 | { | |
6139 | struct btrfs_device *device; | |
dfe25020 | 6140 | |
12bd2fc0 ID |
6141 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6142 | if (IS_ERR(device)) | |
7cbd8a83 | 6143 | return NULL; |
12bd2fc0 ID |
6144 | |
6145 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6146 | device->fs_devices = fs_devices; |
dfe25020 | 6147 | fs_devices->num_devices++; |
12bd2fc0 ID |
6148 | |
6149 | device->missing = 1; | |
cd02dca5 | 6150 | fs_devices->missing_devices++; |
12bd2fc0 | 6151 | |
dfe25020 CM |
6152 | return device; |
6153 | } | |
6154 | ||
12bd2fc0 ID |
6155 | /** |
6156 | * btrfs_alloc_device - allocate struct btrfs_device | |
6157 | * @fs_info: used only for generating a new devid, can be NULL if | |
6158 | * devid is provided (i.e. @devid != NULL). | |
6159 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6160 | * is generated. | |
6161 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6162 | * is generated. | |
6163 | * | |
6164 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6165 | * on error. Returned struct is not linked onto any lists and can be | |
6166 | * destroyed with kfree() right away. | |
6167 | */ | |
6168 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6169 | const u64 *devid, | |
6170 | const u8 *uuid) | |
6171 | { | |
6172 | struct btrfs_device *dev; | |
6173 | u64 tmp; | |
6174 | ||
fae7f21c | 6175 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6176 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6177 | |
6178 | dev = __alloc_device(); | |
6179 | if (IS_ERR(dev)) | |
6180 | return dev; | |
6181 | ||
6182 | if (devid) | |
6183 | tmp = *devid; | |
6184 | else { | |
6185 | int ret; | |
6186 | ||
6187 | ret = find_next_devid(fs_info, &tmp); | |
6188 | if (ret) { | |
6189 | kfree(dev); | |
6190 | return ERR_PTR(ret); | |
6191 | } | |
6192 | } | |
6193 | dev->devid = tmp; | |
6194 | ||
6195 | if (uuid) | |
6196 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6197 | else | |
6198 | generate_random_uuid(dev->uuid); | |
6199 | ||
9e0af237 LB |
6200 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6201 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6202 | |
6203 | return dev; | |
6204 | } | |
6205 | ||
0b86a832 CM |
6206 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
6207 | struct extent_buffer *leaf, | |
6208 | struct btrfs_chunk *chunk) | |
6209 | { | |
6210 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6211 | struct map_lookup *map; | |
6212 | struct extent_map *em; | |
6213 | u64 logical; | |
6214 | u64 length; | |
6215 | u64 devid; | |
a443755f | 6216 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 6217 | int num_stripes; |
0b86a832 | 6218 | int ret; |
593060d7 | 6219 | int i; |
0b86a832 | 6220 | |
e17cade2 CM |
6221 | logical = key->offset; |
6222 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 6223 | |
890871be | 6224 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6225 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6226 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6227 | |
6228 | /* already mapped? */ | |
6229 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6230 | free_extent_map(em); | |
0b86a832 CM |
6231 | return 0; |
6232 | } else if (em) { | |
6233 | free_extent_map(em); | |
6234 | } | |
0b86a832 | 6235 | |
172ddd60 | 6236 | em = alloc_extent_map(); |
0b86a832 CM |
6237 | if (!em) |
6238 | return -ENOMEM; | |
593060d7 CM |
6239 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6240 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
6241 | if (!map) { |
6242 | free_extent_map(em); | |
6243 | return -ENOMEM; | |
6244 | } | |
6245 | ||
298a8f9c | 6246 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
0b86a832 CM |
6247 | em->bdev = (struct block_device *)map; |
6248 | em->start = logical; | |
6249 | em->len = length; | |
70c8a91c | 6250 | em->orig_start = 0; |
0b86a832 | 6251 | em->block_start = 0; |
c8b97818 | 6252 | em->block_len = em->len; |
0b86a832 | 6253 | |
593060d7 CM |
6254 | map->num_stripes = num_stripes; |
6255 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6256 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6257 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6258 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6259 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6260 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6261 | for (i = 0; i < num_stripes; i++) { |
6262 | map->stripes[i].physical = | |
6263 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6264 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6265 | read_extent_buffer(leaf, uuid, (unsigned long) |
6266 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6267 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
6268 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
6269 | uuid, NULL); | |
dfe25020 | 6270 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
6271 | free_extent_map(em); |
6272 | return -EIO; | |
6273 | } | |
dfe25020 CM |
6274 | if (!map->stripes[i].dev) { |
6275 | map->stripes[i].dev = | |
5f375835 MX |
6276 | add_missing_dev(root, root->fs_info->fs_devices, |
6277 | devid, uuid); | |
dfe25020 | 6278 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6279 | free_extent_map(em); |
6280 | return -EIO; | |
6281 | } | |
816fcebe AJ |
6282 | btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing", |
6283 | devid, uuid); | |
dfe25020 CM |
6284 | } |
6285 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6286 | } |
6287 | ||
890871be | 6288 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6289 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6290 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6291 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6292 | free_extent_map(em); |
6293 | ||
6294 | return 0; | |
6295 | } | |
6296 | ||
143bede5 | 6297 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6298 | struct btrfs_dev_item *dev_item, |
6299 | struct btrfs_device *device) | |
6300 | { | |
6301 | unsigned long ptr; | |
0b86a832 CM |
6302 | |
6303 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6304 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6305 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6306 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6307 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6308 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6309 | device->type = btrfs_device_type(leaf, dev_item); |
6310 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6311 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6312 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6313 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6314 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6315 | |
410ba3a2 | 6316 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6317 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6318 | } |
6319 | ||
5f375835 MX |
6320 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
6321 | u8 *fsid) | |
2b82032c YZ |
6322 | { |
6323 | struct btrfs_fs_devices *fs_devices; | |
6324 | int ret; | |
6325 | ||
b367e47f | 6326 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
6327 | |
6328 | fs_devices = root->fs_info->fs_devices->seed; | |
6329 | while (fs_devices) { | |
5f375835 MX |
6330 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6331 | return fs_devices; | |
6332 | ||
2b82032c YZ |
6333 | fs_devices = fs_devices->seed; |
6334 | } | |
6335 | ||
6336 | fs_devices = find_fsid(fsid); | |
6337 | if (!fs_devices) { | |
5f375835 MX |
6338 | if (!btrfs_test_opt(root, DEGRADED)) |
6339 | return ERR_PTR(-ENOENT); | |
6340 | ||
6341 | fs_devices = alloc_fs_devices(fsid); | |
6342 | if (IS_ERR(fs_devices)) | |
6343 | return fs_devices; | |
6344 | ||
6345 | fs_devices->seeding = 1; | |
6346 | fs_devices->opened = 1; | |
6347 | return fs_devices; | |
2b82032c | 6348 | } |
e4404d6e YZ |
6349 | |
6350 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6351 | if (IS_ERR(fs_devices)) |
6352 | return fs_devices; | |
2b82032c | 6353 | |
97288f2c | 6354 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 6355 | root->fs_info->bdev_holder); |
48d28232 JL |
6356 | if (ret) { |
6357 | free_fs_devices(fs_devices); | |
5f375835 | 6358 | fs_devices = ERR_PTR(ret); |
2b82032c | 6359 | goto out; |
48d28232 | 6360 | } |
2b82032c YZ |
6361 | |
6362 | if (!fs_devices->seeding) { | |
6363 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6364 | free_fs_devices(fs_devices); |
5f375835 | 6365 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6366 | goto out; |
6367 | } | |
6368 | ||
6369 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6370 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6371 | out: |
5f375835 | 6372 | return fs_devices; |
2b82032c YZ |
6373 | } |
6374 | ||
0d81ba5d | 6375 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6376 | struct extent_buffer *leaf, |
6377 | struct btrfs_dev_item *dev_item) | |
6378 | { | |
5f375835 | 6379 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
0b86a832 CM |
6380 | struct btrfs_device *device; |
6381 | u64 devid; | |
6382 | int ret; | |
2b82032c | 6383 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6384 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6385 | ||
0b86a832 | 6386 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6387 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6388 | BTRFS_UUID_SIZE); |
1473b24e | 6389 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6390 | BTRFS_UUID_SIZE); |
6391 | ||
6392 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5f375835 MX |
6393 | fs_devices = open_seed_devices(root, fs_uuid); |
6394 | if (IS_ERR(fs_devices)) | |
6395 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6396 | } |
6397 | ||
aa1b8cd4 | 6398 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6399 | if (!device) { |
e4404d6e | 6400 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
6401 | return -EIO; |
6402 | ||
5f375835 MX |
6403 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); |
6404 | if (!device) | |
6405 | return -ENOMEM; | |
33b97e43 AJ |
6406 | btrfs_warn(root->fs_info, "devid %llu uuid %pU missing", |
6407 | devid, dev_uuid); | |
5f375835 MX |
6408 | } else { |
6409 | if (!device->bdev && !btrfs_test_opt(root, DEGRADED)) | |
6410 | return -EIO; | |
6411 | ||
6412 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6413 | /* |
6414 | * this happens when a device that was properly setup | |
6415 | * in the device info lists suddenly goes bad. | |
6416 | * device->bdev is NULL, and so we have to set | |
6417 | * device->missing to one here | |
6418 | */ | |
5f375835 | 6419 | device->fs_devices->missing_devices++; |
cd02dca5 | 6420 | device->missing = 1; |
2b82032c | 6421 | } |
5f375835 MX |
6422 | |
6423 | /* Move the device to its own fs_devices */ | |
6424 | if (device->fs_devices != fs_devices) { | |
6425 | ASSERT(device->missing); | |
6426 | ||
6427 | list_move(&device->dev_list, &fs_devices->devices); | |
6428 | device->fs_devices->num_devices--; | |
6429 | fs_devices->num_devices++; | |
6430 | ||
6431 | device->fs_devices->missing_devices--; | |
6432 | fs_devices->missing_devices++; | |
6433 | ||
6434 | device->fs_devices = fs_devices; | |
6435 | } | |
2b82032c YZ |
6436 | } |
6437 | ||
6438 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6439 | BUG_ON(device->writeable); | |
6440 | if (device->generation != | |
6441 | btrfs_device_generation(leaf, dev_item)) | |
6442 | return -EINVAL; | |
6324fbf3 | 6443 | } |
0b86a832 CM |
6444 | |
6445 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6446 | device->in_fs_metadata = 1; |
63a212ab | 6447 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6448 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
6449 | spin_lock(&root->fs_info->free_chunk_lock); |
6450 | root->fs_info->free_chunk_space += device->total_bytes - | |
6451 | device->bytes_used; | |
6452 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6453 | } | |
0b86a832 | 6454 | ret = 0; |
0b86a832 CM |
6455 | return ret; |
6456 | } | |
6457 | ||
e4404d6e | 6458 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 6459 | { |
6c41761f | 6460 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 6461 | struct extent_buffer *sb; |
0b86a832 | 6462 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6463 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6464 | u8 *array_ptr; |
6465 | unsigned long sb_array_offset; | |
84eed90f | 6466 | int ret = 0; |
0b86a832 CM |
6467 | u32 num_stripes; |
6468 | u32 array_size; | |
6469 | u32 len = 0; | |
1ffb22cf | 6470 | u32 cur_offset; |
84eed90f | 6471 | struct btrfs_key key; |
0b86a832 | 6472 | |
a83fffb7 DS |
6473 | ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize); |
6474 | /* | |
6475 | * This will create extent buffer of nodesize, superblock size is | |
6476 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6477 | * overallocate but we can keep it as-is, only the first page is used. | |
6478 | */ | |
6479 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
a061fc8d CM |
6480 | if (!sb) |
6481 | return -ENOMEM; | |
4db8c528 | 6482 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6483 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
6484 | /* |
6485 | * The sb extent buffer is artifical and just used to read the system array. | |
4db8c528 | 6486 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6487 | * pages up-to-date when the page is larger: extent does not cover the |
6488 | * whole page and consequently check_page_uptodate does not find all | |
6489 | * the page's extents up-to-date (the hole beyond sb), | |
6490 | * write_extent_buffer then triggers a WARN_ON. | |
6491 | * | |
6492 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6493 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6494 | * to silence the warning eg. on PowerPC 64. | |
6495 | */ | |
6496 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 6497 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6498 | |
a061fc8d | 6499 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6500 | array_size = btrfs_super_sys_array_size(super_copy); |
6501 | ||
1ffb22cf DS |
6502 | array_ptr = super_copy->sys_chunk_array; |
6503 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6504 | cur_offset = 0; | |
0b86a832 | 6505 | |
1ffb22cf DS |
6506 | while (cur_offset < array_size) { |
6507 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6508 | len = sizeof(*disk_key); |
6509 | if (cur_offset + len > array_size) | |
6510 | goto out_short_read; | |
6511 | ||
0b86a832 CM |
6512 | btrfs_disk_key_to_cpu(&key, disk_key); |
6513 | ||
1ffb22cf DS |
6514 | array_ptr += len; |
6515 | sb_array_offset += len; | |
6516 | cur_offset += len; | |
0b86a832 | 6517 | |
0d81ba5d | 6518 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6519 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6520 | /* |
6521 | * At least one btrfs_chunk with one stripe must be | |
6522 | * present, exact stripe count check comes afterwards | |
6523 | */ | |
6524 | len = btrfs_chunk_item_size(1); | |
6525 | if (cur_offset + len > array_size) | |
6526 | goto out_short_read; | |
6527 | ||
6528 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
6529 | len = btrfs_chunk_item_size(num_stripes); | |
6530 | if (cur_offset + len > array_size) | |
6531 | goto out_short_read; | |
6532 | ||
0d81ba5d | 6533 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6534 | if (ret) |
6535 | break; | |
0b86a832 | 6536 | } else { |
84eed90f CM |
6537 | ret = -EIO; |
6538 | break; | |
0b86a832 | 6539 | } |
1ffb22cf DS |
6540 | array_ptr += len; |
6541 | sb_array_offset += len; | |
6542 | cur_offset += len; | |
0b86a832 | 6543 | } |
a061fc8d | 6544 | free_extent_buffer(sb); |
84eed90f | 6545 | return ret; |
e3540eab DS |
6546 | |
6547 | out_short_read: | |
6548 | printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n", | |
6549 | len, cur_offset); | |
6550 | free_extent_buffer(sb); | |
6551 | return -EIO; | |
0b86a832 CM |
6552 | } |
6553 | ||
6554 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6555 | { | |
6556 | struct btrfs_path *path; | |
6557 | struct extent_buffer *leaf; | |
6558 | struct btrfs_key key; | |
6559 | struct btrfs_key found_key; | |
6560 | int ret; | |
6561 | int slot; | |
6562 | ||
6563 | root = root->fs_info->chunk_root; | |
6564 | ||
6565 | path = btrfs_alloc_path(); | |
6566 | if (!path) | |
6567 | return -ENOMEM; | |
6568 | ||
b367e47f LZ |
6569 | mutex_lock(&uuid_mutex); |
6570 | lock_chunks(root); | |
6571 | ||
395927a9 FDBM |
6572 | /* |
6573 | * Read all device items, and then all the chunk items. All | |
6574 | * device items are found before any chunk item (their object id | |
6575 | * is smaller than the lowest possible object id for a chunk | |
6576 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6577 | */ |
6578 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6579 | key.offset = 0; | |
6580 | key.type = 0; | |
0b86a832 | 6581 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6582 | if (ret < 0) |
6583 | goto error; | |
d397712b | 6584 | while (1) { |
0b86a832 CM |
6585 | leaf = path->nodes[0]; |
6586 | slot = path->slots[0]; | |
6587 | if (slot >= btrfs_header_nritems(leaf)) { | |
6588 | ret = btrfs_next_leaf(root, path); | |
6589 | if (ret == 0) | |
6590 | continue; | |
6591 | if (ret < 0) | |
6592 | goto error; | |
6593 | break; | |
6594 | } | |
6595 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6596 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6597 | struct btrfs_dev_item *dev_item; | |
6598 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6599 | struct btrfs_dev_item); |
395927a9 FDBM |
6600 | ret = read_one_dev(root, leaf, dev_item); |
6601 | if (ret) | |
6602 | goto error; | |
0b86a832 CM |
6603 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6604 | struct btrfs_chunk *chunk; | |
6605 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6606 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6607 | if (ret) |
6608 | goto error; | |
0b86a832 CM |
6609 | } |
6610 | path->slots[0]++; | |
6611 | } | |
0b86a832 CM |
6612 | ret = 0; |
6613 | error: | |
b367e47f LZ |
6614 | unlock_chunks(root); |
6615 | mutex_unlock(&uuid_mutex); | |
6616 | ||
2b82032c | 6617 | btrfs_free_path(path); |
0b86a832 CM |
6618 | return ret; |
6619 | } | |
442a4f63 | 6620 | |
cb517eab MX |
6621 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6622 | { | |
6623 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6624 | struct btrfs_device *device; | |
6625 | ||
29cc83f6 LB |
6626 | while (fs_devices) { |
6627 | mutex_lock(&fs_devices->device_list_mutex); | |
6628 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6629 | device->dev_root = fs_info->dev_root; | |
6630 | mutex_unlock(&fs_devices->device_list_mutex); | |
6631 | ||
6632 | fs_devices = fs_devices->seed; | |
6633 | } | |
cb517eab MX |
6634 | } |
6635 | ||
733f4fbb SB |
6636 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6637 | { | |
6638 | int i; | |
6639 | ||
6640 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6641 | btrfs_dev_stat_reset(dev, i); | |
6642 | } | |
6643 | ||
6644 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6645 | { | |
6646 | struct btrfs_key key; | |
6647 | struct btrfs_key found_key; | |
6648 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6649 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6650 | struct extent_buffer *eb; | |
6651 | int slot; | |
6652 | int ret = 0; | |
6653 | struct btrfs_device *device; | |
6654 | struct btrfs_path *path = NULL; | |
6655 | int i; | |
6656 | ||
6657 | path = btrfs_alloc_path(); | |
6658 | if (!path) { | |
6659 | ret = -ENOMEM; | |
6660 | goto out; | |
6661 | } | |
6662 | ||
6663 | mutex_lock(&fs_devices->device_list_mutex); | |
6664 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6665 | int item_size; | |
6666 | struct btrfs_dev_stats_item *ptr; | |
6667 | ||
6668 | key.objectid = 0; | |
6669 | key.type = BTRFS_DEV_STATS_KEY; | |
6670 | key.offset = device->devid; | |
6671 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6672 | if (ret) { | |
733f4fbb SB |
6673 | __btrfs_reset_dev_stats(device); |
6674 | device->dev_stats_valid = 1; | |
6675 | btrfs_release_path(path); | |
6676 | continue; | |
6677 | } | |
6678 | slot = path->slots[0]; | |
6679 | eb = path->nodes[0]; | |
6680 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6681 | item_size = btrfs_item_size_nr(eb, slot); | |
6682 | ||
6683 | ptr = btrfs_item_ptr(eb, slot, | |
6684 | struct btrfs_dev_stats_item); | |
6685 | ||
6686 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6687 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6688 | btrfs_dev_stat_set(device, i, | |
6689 | btrfs_dev_stats_value(eb, ptr, i)); | |
6690 | else | |
6691 | btrfs_dev_stat_reset(device, i); | |
6692 | } | |
6693 | ||
6694 | device->dev_stats_valid = 1; | |
6695 | btrfs_dev_stat_print_on_load(device); | |
6696 | btrfs_release_path(path); | |
6697 | } | |
6698 | mutex_unlock(&fs_devices->device_list_mutex); | |
6699 | ||
6700 | out: | |
6701 | btrfs_free_path(path); | |
6702 | return ret < 0 ? ret : 0; | |
6703 | } | |
6704 | ||
6705 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6706 | struct btrfs_root *dev_root, | |
6707 | struct btrfs_device *device) | |
6708 | { | |
6709 | struct btrfs_path *path; | |
6710 | struct btrfs_key key; | |
6711 | struct extent_buffer *eb; | |
6712 | struct btrfs_dev_stats_item *ptr; | |
6713 | int ret; | |
6714 | int i; | |
6715 | ||
6716 | key.objectid = 0; | |
6717 | key.type = BTRFS_DEV_STATS_KEY; | |
6718 | key.offset = device->devid; | |
6719 | ||
6720 | path = btrfs_alloc_path(); | |
6721 | BUG_ON(!path); | |
6722 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6723 | if (ret < 0) { | |
ecaeb14b DS |
6724 | btrfs_warn_in_rcu(dev_root->fs_info, |
6725 | "error %d while searching for dev_stats item for device %s", | |
606686ee | 6726 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6727 | goto out; |
6728 | } | |
6729 | ||
6730 | if (ret == 0 && | |
6731 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6732 | /* need to delete old one and insert a new one */ | |
6733 | ret = btrfs_del_item(trans, dev_root, path); | |
6734 | if (ret != 0) { | |
ecaeb14b DS |
6735 | btrfs_warn_in_rcu(dev_root->fs_info, |
6736 | "delete too small dev_stats item for device %s failed %d", | |
606686ee | 6737 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6738 | goto out; |
6739 | } | |
6740 | ret = 1; | |
6741 | } | |
6742 | ||
6743 | if (ret == 1) { | |
6744 | /* need to insert a new item */ | |
6745 | btrfs_release_path(path); | |
6746 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6747 | &key, sizeof(*ptr)); | |
6748 | if (ret < 0) { | |
ecaeb14b DS |
6749 | btrfs_warn_in_rcu(dev_root->fs_info, |
6750 | "insert dev_stats item for device %s failed %d", | |
6751 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
6752 | goto out; |
6753 | } | |
6754 | } | |
6755 | ||
6756 | eb = path->nodes[0]; | |
6757 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6758 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6759 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6760 | btrfs_dev_stat_read(device, i)); | |
6761 | btrfs_mark_buffer_dirty(eb); | |
6762 | ||
6763 | out: | |
6764 | btrfs_free_path(path); | |
6765 | return ret; | |
6766 | } | |
6767 | ||
6768 | /* | |
6769 | * called from commit_transaction. Writes all changed device stats to disk. | |
6770 | */ | |
6771 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6772 | struct btrfs_fs_info *fs_info) | |
6773 | { | |
6774 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6775 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6776 | struct btrfs_device *device; | |
addc3fa7 | 6777 | int stats_cnt; |
733f4fbb SB |
6778 | int ret = 0; |
6779 | ||
6780 | mutex_lock(&fs_devices->device_list_mutex); | |
6781 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 6782 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
6783 | continue; |
6784 | ||
addc3fa7 | 6785 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
733f4fbb SB |
6786 | ret = update_dev_stat_item(trans, dev_root, device); |
6787 | if (!ret) | |
addc3fa7 | 6788 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
6789 | } |
6790 | mutex_unlock(&fs_devices->device_list_mutex); | |
6791 | ||
6792 | return ret; | |
6793 | } | |
6794 | ||
442a4f63 SB |
6795 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
6796 | { | |
6797 | btrfs_dev_stat_inc(dev, index); | |
6798 | btrfs_dev_stat_print_on_error(dev); | |
6799 | } | |
6800 | ||
48a3b636 | 6801 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 6802 | { |
733f4fbb SB |
6803 | if (!dev->dev_stats_valid) |
6804 | return; | |
b14af3b4 DS |
6805 | btrfs_err_rl_in_rcu(dev->dev_root->fs_info, |
6806 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
606686ee | 6807 | rcu_str_deref(dev->name), |
442a4f63 SB |
6808 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6809 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6810 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
6811 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
6812 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 6813 | } |
c11d2c23 | 6814 | |
733f4fbb SB |
6815 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
6816 | { | |
a98cdb85 SB |
6817 | int i; |
6818 | ||
6819 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6820 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6821 | break; | |
6822 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6823 | return; /* all values == 0, suppress message */ | |
6824 | ||
ecaeb14b DS |
6825 | btrfs_info_in_rcu(dev->dev_root->fs_info, |
6826 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
606686ee | 6827 | rcu_str_deref(dev->name), |
733f4fbb SB |
6828 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6829 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6830 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6831 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6832 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6833 | } | |
6834 | ||
c11d2c23 | 6835 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 6836 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
6837 | { |
6838 | struct btrfs_device *dev; | |
6839 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6840 | int i; | |
6841 | ||
6842 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 6843 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
6844 | mutex_unlock(&fs_devices->device_list_mutex); |
6845 | ||
6846 | if (!dev) { | |
efe120a0 | 6847 | btrfs_warn(root->fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 6848 | return -ENODEV; |
733f4fbb | 6849 | } else if (!dev->dev_stats_valid) { |
efe120a0 | 6850 | btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 6851 | return -ENODEV; |
b27f7c0c | 6852 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
6853 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
6854 | if (stats->nr_items > i) | |
6855 | stats->values[i] = | |
6856 | btrfs_dev_stat_read_and_reset(dev, i); | |
6857 | else | |
6858 | btrfs_dev_stat_reset(dev, i); | |
6859 | } | |
6860 | } else { | |
6861 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6862 | if (stats->nr_items > i) | |
6863 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6864 | } | |
6865 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6866 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6867 | return 0; | |
6868 | } | |
a8a6dab7 | 6869 | |
12b1c263 | 6870 | void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path) |
a8a6dab7 SB |
6871 | { |
6872 | struct buffer_head *bh; | |
6873 | struct btrfs_super_block *disk_super; | |
12b1c263 | 6874 | int copy_num; |
a8a6dab7 | 6875 | |
12b1c263 AJ |
6876 | if (!bdev) |
6877 | return; | |
a8a6dab7 | 6878 | |
12b1c263 AJ |
6879 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
6880 | copy_num++) { | |
a8a6dab7 | 6881 | |
12b1c263 AJ |
6882 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
6883 | continue; | |
6884 | ||
6885 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6886 | ||
6887 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6888 | set_buffer_dirty(bh); | |
6889 | sync_dirty_buffer(bh); | |
6890 | brelse(bh); | |
6891 | } | |
6892 | ||
6893 | /* Notify udev that device has changed */ | |
6894 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
6895 | ||
6896 | /* Update ctime/mtime for device path for libblkid */ | |
6897 | update_dev_time(device_path); | |
a8a6dab7 | 6898 | } |
935e5cc9 MX |
6899 | |
6900 | /* | |
6901 | * Update the size of all devices, which is used for writing out the | |
6902 | * super blocks. | |
6903 | */ | |
6904 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
6905 | { | |
6906 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6907 | struct btrfs_device *curr, *next; | |
6908 | ||
6909 | if (list_empty(&fs_devices->resized_devices)) | |
6910 | return; | |
6911 | ||
6912 | mutex_lock(&fs_devices->device_list_mutex); | |
6913 | lock_chunks(fs_info->dev_root); | |
6914 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
6915 | resized_list) { | |
6916 | list_del_init(&curr->resized_list); | |
6917 | curr->commit_total_bytes = curr->disk_total_bytes; | |
6918 | } | |
6919 | unlock_chunks(fs_info->dev_root); | |
6920 | mutex_unlock(&fs_devices->device_list_mutex); | |
6921 | } | |
ce7213c7 MX |
6922 | |
6923 | /* Must be invoked during the transaction commit */ | |
6924 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
6925 | struct btrfs_transaction *transaction) | |
6926 | { | |
6927 | struct extent_map *em; | |
6928 | struct map_lookup *map; | |
6929 | struct btrfs_device *dev; | |
6930 | int i; | |
6931 | ||
6932 | if (list_empty(&transaction->pending_chunks)) | |
6933 | return; | |
6934 | ||
6935 | /* In order to kick the device replace finish process */ | |
6936 | lock_chunks(root); | |
6937 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
6938 | map = (struct map_lookup *)em->bdev; | |
6939 | ||
6940 | for (i = 0; i < map->num_stripes; i++) { | |
6941 | dev = map->stripes[i].dev; | |
6942 | dev->commit_bytes_used = dev->bytes_used; | |
6943 | } | |
6944 | } | |
6945 | unlock_chunks(root); | |
6946 | } | |
5a13f430 AJ |
6947 | |
6948 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6949 | { | |
6950 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6951 | while (fs_devices) { | |
6952 | fs_devices->fs_info = fs_info; | |
6953 | fs_devices = fs_devices->seed; | |
6954 | } | |
6955 | } | |
6956 | ||
6957 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6958 | { | |
6959 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6960 | while (fs_devices) { | |
6961 | fs_devices->fs_info = NULL; | |
6962 | fs_devices = fs_devices->seed; | |
6963 | } | |
6964 | } | |
f190aa47 | 6965 | |
87ad58c5 | 6966 | static void btrfs_close_one_device(struct btrfs_device *device) |
f190aa47 AJ |
6967 | { |
6968 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
6969 | struct btrfs_device *new_device; | |
6970 | struct rcu_string *name; | |
6971 | ||
6972 | if (device->bdev) | |
6973 | fs_devices->open_devices--; | |
6974 | ||
6975 | if (device->writeable && | |
6976 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
6977 | list_del_init(&device->dev_alloc_list); | |
6978 | fs_devices->rw_devices--; | |
6979 | } | |
6980 | ||
6981 | if (device->missing) | |
6982 | fs_devices->missing_devices--; | |
6983 | ||
6984 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
6985 | device->uuid); | |
6986 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
6987 | ||
6988 | /* Safe because we are under uuid_mutex */ | |
6989 | if (device->name) { | |
6990 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
6991 | BUG_ON(!name); /* -ENOMEM */ | |
6992 | rcu_assign_pointer(new_device->name, name); | |
6993 | } | |
6994 | ||
6995 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
6996 | new_device->fs_devices = device->fs_devices; | |
6997 | ||
6998 | call_rcu(&device->rcu, free_device); | |
6999 | } |