]>
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, | |
51 | .devs_increment = 2, | |
52 | .ncopies = 2, | |
53 | }, | |
54 | [BTRFS_RAID_RAID1] = { | |
55 | .sub_stripes = 1, | |
56 | .dev_stripes = 1, | |
57 | .devs_max = 2, | |
58 | .devs_min = 2, | |
59 | .devs_increment = 2, | |
60 | .ncopies = 2, | |
61 | }, | |
62 | [BTRFS_RAID_DUP] = { | |
63 | .sub_stripes = 1, | |
64 | .dev_stripes = 2, | |
65 | .devs_max = 1, | |
66 | .devs_min = 1, | |
67 | .devs_increment = 1, | |
68 | .ncopies = 2, | |
69 | }, | |
70 | [BTRFS_RAID_RAID0] = { | |
71 | .sub_stripes = 1, | |
72 | .dev_stripes = 1, | |
73 | .devs_max = 0, | |
74 | .devs_min = 2, | |
75 | .devs_increment = 1, | |
76 | .ncopies = 1, | |
77 | }, | |
78 | [BTRFS_RAID_SINGLE] = { | |
79 | .sub_stripes = 1, | |
80 | .dev_stripes = 1, | |
81 | .devs_max = 1, | |
82 | .devs_min = 1, | |
83 | .devs_increment = 1, | |
84 | .ncopies = 1, | |
85 | }, | |
86 | [BTRFS_RAID_RAID5] = { | |
87 | .sub_stripes = 1, | |
88 | .dev_stripes = 1, | |
89 | .devs_max = 0, | |
90 | .devs_min = 2, | |
91 | .devs_increment = 1, | |
92 | .ncopies = 2, | |
93 | }, | |
94 | [BTRFS_RAID_RAID6] = { | |
95 | .sub_stripes = 1, | |
96 | .dev_stripes = 1, | |
97 | .devs_max = 0, | |
98 | .devs_min = 3, | |
99 | .devs_increment = 1, | |
100 | .ncopies = 3, | |
101 | }, | |
102 | }; | |
103 | ||
104 | const u64 const btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { | |
105 | [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10, | |
106 | [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1, | |
107 | [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP, | |
108 | [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0, | |
109 | [BTRFS_RAID_SINGLE] = 0, | |
110 | [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5, | |
111 | [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6, | |
112 | }; | |
113 | ||
2b82032c YZ |
114 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
115 | struct btrfs_root *root, | |
116 | struct btrfs_device *device); | |
117 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 118 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 119 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 120 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 121 | |
67a2c45e | 122 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 123 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
124 | struct list_head *btrfs_get_fs_uuids(void) |
125 | { | |
126 | return &fs_uuids; | |
127 | } | |
8a4b83cc | 128 | |
2208a378 ID |
129 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
130 | { | |
131 | struct btrfs_fs_devices *fs_devs; | |
132 | ||
133 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS); | |
134 | if (!fs_devs) | |
135 | return ERR_PTR(-ENOMEM); | |
136 | ||
137 | mutex_init(&fs_devs->device_list_mutex); | |
138 | ||
139 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 140 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
141 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
142 | INIT_LIST_HEAD(&fs_devs->list); | |
143 | ||
144 | return fs_devs; | |
145 | } | |
146 | ||
147 | /** | |
148 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
149 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
150 | * generated. | |
151 | * | |
152 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
153 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
154 | * can be destroyed with kfree() right away. | |
155 | */ | |
156 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
157 | { | |
158 | struct btrfs_fs_devices *fs_devs; | |
159 | ||
160 | fs_devs = __alloc_fs_devices(); | |
161 | if (IS_ERR(fs_devs)) | |
162 | return fs_devs; | |
163 | ||
164 | if (fsid) | |
165 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
166 | else | |
167 | generate_random_uuid(fs_devs->fsid); | |
168 | ||
169 | return fs_devs; | |
170 | } | |
171 | ||
e4404d6e YZ |
172 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
173 | { | |
174 | struct btrfs_device *device; | |
175 | WARN_ON(fs_devices->opened); | |
176 | while (!list_empty(&fs_devices->devices)) { | |
177 | device = list_entry(fs_devices->devices.next, | |
178 | struct btrfs_device, dev_list); | |
179 | list_del(&device->dev_list); | |
606686ee | 180 | rcu_string_free(device->name); |
e4404d6e YZ |
181 | kfree(device); |
182 | } | |
183 | kfree(fs_devices); | |
184 | } | |
185 | ||
b8b8ff59 LC |
186 | static void btrfs_kobject_uevent(struct block_device *bdev, |
187 | enum kobject_action action) | |
188 | { | |
189 | int ret; | |
190 | ||
191 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
192 | if (ret) | |
efe120a0 | 193 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
194 | action, |
195 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
196 | &disk_to_dev(bdev->bd_disk)->kobj); | |
197 | } | |
198 | ||
143bede5 | 199 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
200 | { |
201 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 202 | |
2b82032c YZ |
203 | while (!list_empty(&fs_uuids)) { |
204 | fs_devices = list_entry(fs_uuids.next, | |
205 | struct btrfs_fs_devices, list); | |
206 | list_del(&fs_devices->list); | |
e4404d6e | 207 | free_fs_devices(fs_devices); |
8a4b83cc | 208 | } |
8a4b83cc CM |
209 | } |
210 | ||
12bd2fc0 ID |
211 | static struct btrfs_device *__alloc_device(void) |
212 | { | |
213 | struct btrfs_device *dev; | |
214 | ||
215 | dev = kzalloc(sizeof(*dev), GFP_NOFS); | |
216 | if (!dev) | |
217 | return ERR_PTR(-ENOMEM); | |
218 | ||
219 | INIT_LIST_HEAD(&dev->dev_list); | |
220 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 221 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
222 | |
223 | spin_lock_init(&dev->io_lock); | |
224 | ||
225 | spin_lock_init(&dev->reada_lock); | |
226 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 227 | atomic_set(&dev->dev_stats_ccnt, 0); |
12bd2fc0 ID |
228 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT); |
229 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
230 | ||
231 | return dev; | |
232 | } | |
233 | ||
a1b32a59 CM |
234 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
235 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
236 | { |
237 | struct btrfs_device *dev; | |
8a4b83cc | 238 | |
c6e30871 | 239 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 240 | if (dev->devid == devid && |
8f18cf13 | 241 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 242 | return dev; |
a443755f | 243 | } |
8a4b83cc CM |
244 | } |
245 | return NULL; | |
246 | } | |
247 | ||
a1b32a59 | 248 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 249 | { |
8a4b83cc CM |
250 | struct btrfs_fs_devices *fs_devices; |
251 | ||
c6e30871 | 252 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
253 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
254 | return fs_devices; | |
255 | } | |
256 | return NULL; | |
257 | } | |
258 | ||
beaf8ab3 SB |
259 | static int |
260 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
261 | int flush, struct block_device **bdev, | |
262 | struct buffer_head **bh) | |
263 | { | |
264 | int ret; | |
265 | ||
266 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
267 | ||
268 | if (IS_ERR(*bdev)) { | |
269 | ret = PTR_ERR(*bdev); | |
efe120a0 | 270 | printk(KERN_INFO "BTRFS: open %s failed\n", device_path); |
beaf8ab3 SB |
271 | goto error; |
272 | } | |
273 | ||
274 | if (flush) | |
275 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
276 | ret = set_blocksize(*bdev, 4096); | |
277 | if (ret) { | |
278 | blkdev_put(*bdev, flags); | |
279 | goto error; | |
280 | } | |
281 | invalidate_bdev(*bdev); | |
282 | *bh = btrfs_read_dev_super(*bdev); | |
283 | if (!*bh) { | |
284 | ret = -EINVAL; | |
285 | blkdev_put(*bdev, flags); | |
286 | goto error; | |
287 | } | |
288 | ||
289 | return 0; | |
290 | ||
291 | error: | |
292 | *bdev = NULL; | |
293 | *bh = NULL; | |
294 | return ret; | |
295 | } | |
296 | ||
ffbd517d CM |
297 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
298 | struct bio *head, struct bio *tail) | |
299 | { | |
300 | ||
301 | struct bio *old_head; | |
302 | ||
303 | old_head = pending_bios->head; | |
304 | pending_bios->head = head; | |
305 | if (pending_bios->tail) | |
306 | tail->bi_next = old_head; | |
307 | else | |
308 | pending_bios->tail = tail; | |
309 | } | |
310 | ||
8b712842 CM |
311 | /* |
312 | * we try to collect pending bios for a device so we don't get a large | |
313 | * number of procs sending bios down to the same device. This greatly | |
314 | * improves the schedulers ability to collect and merge the bios. | |
315 | * | |
316 | * But, it also turns into a long list of bios to process and that is sure | |
317 | * to eventually make the worker thread block. The solution here is to | |
318 | * make some progress and then put this work struct back at the end of | |
319 | * the list if the block device is congested. This way, multiple devices | |
320 | * can make progress from a single worker thread. | |
321 | */ | |
143bede5 | 322 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
323 | { |
324 | struct bio *pending; | |
325 | struct backing_dev_info *bdi; | |
b64a2851 | 326 | struct btrfs_fs_info *fs_info; |
ffbd517d | 327 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
328 | struct bio *tail; |
329 | struct bio *cur; | |
330 | int again = 0; | |
ffbd517d | 331 | unsigned long num_run; |
d644d8a1 | 332 | unsigned long batch_run = 0; |
b64a2851 | 333 | unsigned long limit; |
b765ead5 | 334 | unsigned long last_waited = 0; |
d84275c9 | 335 | int force_reg = 0; |
0e588859 | 336 | int sync_pending = 0; |
211588ad CM |
337 | struct blk_plug plug; |
338 | ||
339 | /* | |
340 | * this function runs all the bios we've collected for | |
341 | * a particular device. We don't want to wander off to | |
342 | * another device without first sending all of these down. | |
343 | * So, setup a plug here and finish it off before we return | |
344 | */ | |
345 | blk_start_plug(&plug); | |
8b712842 | 346 | |
bedf762b | 347 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
348 | fs_info = device->dev_root->fs_info; |
349 | limit = btrfs_async_submit_limit(fs_info); | |
350 | limit = limit * 2 / 3; | |
351 | ||
8b712842 CM |
352 | loop: |
353 | spin_lock(&device->io_lock); | |
354 | ||
a6837051 | 355 | loop_lock: |
d84275c9 | 356 | num_run = 0; |
ffbd517d | 357 | |
8b712842 CM |
358 | /* take all the bios off the list at once and process them |
359 | * later on (without the lock held). But, remember the | |
360 | * tail and other pointers so the bios can be properly reinserted | |
361 | * into the list if we hit congestion | |
362 | */ | |
d84275c9 | 363 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 364 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
365 | force_reg = 1; |
366 | } else { | |
ffbd517d | 367 | pending_bios = &device->pending_bios; |
d84275c9 CM |
368 | force_reg = 0; |
369 | } | |
ffbd517d CM |
370 | |
371 | pending = pending_bios->head; | |
372 | tail = pending_bios->tail; | |
8b712842 | 373 | WARN_ON(pending && !tail); |
8b712842 CM |
374 | |
375 | /* | |
376 | * if pending was null this time around, no bios need processing | |
377 | * at all and we can stop. Otherwise it'll loop back up again | |
378 | * and do an additional check so no bios are missed. | |
379 | * | |
380 | * device->running_pending is used to synchronize with the | |
381 | * schedule_bio code. | |
382 | */ | |
ffbd517d CM |
383 | if (device->pending_sync_bios.head == NULL && |
384 | device->pending_bios.head == NULL) { | |
8b712842 CM |
385 | again = 0; |
386 | device->running_pending = 0; | |
ffbd517d CM |
387 | } else { |
388 | again = 1; | |
389 | device->running_pending = 1; | |
8b712842 | 390 | } |
ffbd517d CM |
391 | |
392 | pending_bios->head = NULL; | |
393 | pending_bios->tail = NULL; | |
394 | ||
8b712842 CM |
395 | spin_unlock(&device->io_lock); |
396 | ||
d397712b | 397 | while (pending) { |
ffbd517d CM |
398 | |
399 | rmb(); | |
d84275c9 CM |
400 | /* we want to work on both lists, but do more bios on the |
401 | * sync list than the regular list | |
402 | */ | |
403 | if ((num_run > 32 && | |
404 | pending_bios != &device->pending_sync_bios && | |
405 | device->pending_sync_bios.head) || | |
406 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
407 | device->pending_bios.head)) { | |
ffbd517d CM |
408 | spin_lock(&device->io_lock); |
409 | requeue_list(pending_bios, pending, tail); | |
410 | goto loop_lock; | |
411 | } | |
412 | ||
8b712842 CM |
413 | cur = pending; |
414 | pending = pending->bi_next; | |
415 | cur->bi_next = NULL; | |
b64a2851 | 416 | |
66657b31 | 417 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
418 | waitqueue_active(&fs_info->async_submit_wait)) |
419 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 420 | |
dac56212 | 421 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 422 | |
2ab1ba68 CM |
423 | /* |
424 | * if we're doing the sync list, record that our | |
425 | * plug has some sync requests on it | |
426 | * | |
427 | * If we're doing the regular list and there are | |
428 | * sync requests sitting around, unplug before | |
429 | * we add more | |
430 | */ | |
431 | if (pending_bios == &device->pending_sync_bios) { | |
432 | sync_pending = 1; | |
433 | } else if (sync_pending) { | |
434 | blk_finish_plug(&plug); | |
435 | blk_start_plug(&plug); | |
436 | sync_pending = 0; | |
437 | } | |
438 | ||
21adbd5c | 439 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
440 | num_run++; |
441 | batch_run++; | |
853d8ec4 DS |
442 | |
443 | cond_resched(); | |
8b712842 CM |
444 | |
445 | /* | |
446 | * we made progress, there is more work to do and the bdi | |
447 | * is now congested. Back off and let other work structs | |
448 | * run instead | |
449 | */ | |
57fd5a5f | 450 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 451 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 452 | struct io_context *ioc; |
8b712842 | 453 | |
b765ead5 CM |
454 | ioc = current->io_context; |
455 | ||
456 | /* | |
457 | * the main goal here is that we don't want to | |
458 | * block if we're going to be able to submit | |
459 | * more requests without blocking. | |
460 | * | |
461 | * This code does two great things, it pokes into | |
462 | * the elevator code from a filesystem _and_ | |
463 | * it makes assumptions about how batching works. | |
464 | */ | |
465 | if (ioc && ioc->nr_batch_requests > 0 && | |
466 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
467 | (last_waited == 0 || | |
468 | ioc->last_waited == last_waited)) { | |
469 | /* | |
470 | * we want to go through our batch of | |
471 | * requests and stop. So, we copy out | |
472 | * the ioc->last_waited time and test | |
473 | * against it before looping | |
474 | */ | |
475 | last_waited = ioc->last_waited; | |
853d8ec4 | 476 | cond_resched(); |
b765ead5 CM |
477 | continue; |
478 | } | |
8b712842 | 479 | spin_lock(&device->io_lock); |
ffbd517d | 480 | requeue_list(pending_bios, pending, tail); |
a6837051 | 481 | device->running_pending = 1; |
8b712842 CM |
482 | |
483 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
484 | btrfs_queue_work(fs_info->submit_workers, |
485 | &device->work); | |
8b712842 CM |
486 | goto done; |
487 | } | |
d85c8a6f CM |
488 | /* unplug every 64 requests just for good measure */ |
489 | if (batch_run % 64 == 0) { | |
490 | blk_finish_plug(&plug); | |
491 | blk_start_plug(&plug); | |
492 | sync_pending = 0; | |
493 | } | |
8b712842 | 494 | } |
ffbd517d | 495 | |
51684082 CM |
496 | cond_resched(); |
497 | if (again) | |
498 | goto loop; | |
499 | ||
500 | spin_lock(&device->io_lock); | |
501 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
502 | goto loop_lock; | |
503 | spin_unlock(&device->io_lock); | |
504 | ||
8b712842 | 505 | done: |
211588ad | 506 | blk_finish_plug(&plug); |
8b712842 CM |
507 | } |
508 | ||
b2950863 | 509 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
510 | { |
511 | struct btrfs_device *device; | |
512 | ||
513 | device = container_of(work, struct btrfs_device, work); | |
514 | run_scheduled_bios(device); | |
515 | } | |
516 | ||
4fde46f0 AJ |
517 | |
518 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
519 | { | |
520 | struct btrfs_fs_devices *fs_devs; | |
521 | struct btrfs_device *dev; | |
522 | ||
523 | if (!cur_dev->name) | |
524 | return; | |
525 | ||
526 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
527 | int del = 1; | |
528 | ||
529 | if (fs_devs->opened) | |
530 | continue; | |
531 | if (fs_devs->seeding) | |
532 | continue; | |
533 | ||
534 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
535 | ||
536 | if (dev == cur_dev) | |
537 | continue; | |
538 | if (!dev->name) | |
539 | continue; | |
540 | ||
541 | /* | |
542 | * Todo: This won't be enough. What if the same device | |
543 | * comes back (with new uuid and) with its mapper path? | |
544 | * But for now, this does help as mostly an admin will | |
545 | * either use mapper or non mapper path throughout. | |
546 | */ | |
547 | rcu_read_lock(); | |
548 | del = strcmp(rcu_str_deref(dev->name), | |
549 | rcu_str_deref(cur_dev->name)); | |
550 | rcu_read_unlock(); | |
551 | if (!del) | |
552 | break; | |
553 | } | |
554 | ||
555 | if (!del) { | |
556 | /* delete the stale device */ | |
557 | if (fs_devs->num_devices == 1) { | |
558 | btrfs_sysfs_remove_fsid(fs_devs); | |
559 | list_del(&fs_devs->list); | |
560 | free_fs_devices(fs_devs); | |
561 | } else { | |
562 | fs_devs->num_devices--; | |
563 | list_del(&dev->dev_list); | |
564 | rcu_string_free(dev->name); | |
565 | kfree(dev); | |
566 | } | |
567 | break; | |
568 | } | |
569 | } | |
570 | } | |
571 | ||
60999ca4 DS |
572 | /* |
573 | * Add new device to list of registered devices | |
574 | * | |
575 | * Returns: | |
576 | * 1 - first time device is seen | |
577 | * 0 - device already known | |
578 | * < 0 - error | |
579 | */ | |
a1b32a59 | 580 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
581 | struct btrfs_super_block *disk_super, |
582 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
583 | { | |
584 | struct btrfs_device *device; | |
585 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 586 | struct rcu_string *name; |
60999ca4 | 587 | int ret = 0; |
8a4b83cc CM |
588 | u64 found_transid = btrfs_super_generation(disk_super); |
589 | ||
590 | fs_devices = find_fsid(disk_super->fsid); | |
591 | if (!fs_devices) { | |
2208a378 ID |
592 | fs_devices = alloc_fs_devices(disk_super->fsid); |
593 | if (IS_ERR(fs_devices)) | |
594 | return PTR_ERR(fs_devices); | |
595 | ||
8a4b83cc | 596 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 597 | |
8a4b83cc CM |
598 | device = NULL; |
599 | } else { | |
a443755f CM |
600 | device = __find_device(&fs_devices->devices, devid, |
601 | disk_super->dev_item.uuid); | |
8a4b83cc | 602 | } |
443f24fe | 603 | |
8a4b83cc | 604 | if (!device) { |
2b82032c YZ |
605 | if (fs_devices->opened) |
606 | return -EBUSY; | |
607 | ||
12bd2fc0 ID |
608 | device = btrfs_alloc_device(NULL, &devid, |
609 | disk_super->dev_item.uuid); | |
610 | if (IS_ERR(device)) { | |
8a4b83cc | 611 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 612 | return PTR_ERR(device); |
8a4b83cc | 613 | } |
606686ee JB |
614 | |
615 | name = rcu_string_strdup(path, GFP_NOFS); | |
616 | if (!name) { | |
8a4b83cc CM |
617 | kfree(device); |
618 | return -ENOMEM; | |
619 | } | |
606686ee | 620 | rcu_assign_pointer(device->name, name); |
90519d66 | 621 | |
e5e9a520 | 622 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 623 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 624 | fs_devices->num_devices++; |
e5e9a520 CM |
625 | mutex_unlock(&fs_devices->device_list_mutex); |
626 | ||
60999ca4 | 627 | ret = 1; |
2b82032c | 628 | device->fs_devices = fs_devices; |
606686ee | 629 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
630 | /* |
631 | * When FS is already mounted. | |
632 | * 1. If you are here and if the device->name is NULL that | |
633 | * means this device was missing at time of FS mount. | |
634 | * 2. If you are here and if the device->name is different | |
635 | * from 'path' that means either | |
636 | * a. The same device disappeared and reappeared with | |
637 | * different name. or | |
638 | * b. The missing-disk-which-was-replaced, has | |
639 | * reappeared now. | |
640 | * | |
641 | * We must allow 1 and 2a above. But 2b would be a spurious | |
642 | * and unintentional. | |
643 | * | |
644 | * Further in case of 1 and 2a above, the disk at 'path' | |
645 | * would have missed some transaction when it was away and | |
646 | * in case of 2a the stale bdev has to be updated as well. | |
647 | * 2b must not be allowed at all time. | |
648 | */ | |
649 | ||
650 | /* | |
0f23ae74 CM |
651 | * For now, we do allow update to btrfs_fs_device through the |
652 | * btrfs dev scan cli after FS has been mounted. We're still | |
653 | * tracking a problem where systems fail mount by subvolume id | |
654 | * when we reject replacement on a mounted FS. | |
b96de000 | 655 | */ |
0f23ae74 | 656 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
657 | /* |
658 | * That is if the FS is _not_ mounted and if you | |
659 | * are here, that means there is more than one | |
660 | * disk with same uuid and devid.We keep the one | |
661 | * with larger generation number or the last-in if | |
662 | * generation are equal. | |
663 | */ | |
0f23ae74 | 664 | return -EEXIST; |
77bdae4d | 665 | } |
b96de000 | 666 | |
606686ee | 667 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
668 | if (!name) |
669 | return -ENOMEM; | |
606686ee JB |
670 | rcu_string_free(device->name); |
671 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
672 | if (device->missing) { |
673 | fs_devices->missing_devices--; | |
674 | device->missing = 0; | |
675 | } | |
8a4b83cc CM |
676 | } |
677 | ||
77bdae4d AJ |
678 | /* |
679 | * Unmount does not free the btrfs_device struct but would zero | |
680 | * generation along with most of the other members. So just update | |
681 | * it back. We need it to pick the disk with largest generation | |
682 | * (as above). | |
683 | */ | |
684 | if (!fs_devices->opened) | |
685 | device->generation = found_transid; | |
686 | ||
4fde46f0 AJ |
687 | /* |
688 | * if there is new btrfs on an already registered device, | |
689 | * then remove the stale device entry. | |
690 | */ | |
691 | btrfs_free_stale_device(device); | |
692 | ||
8a4b83cc | 693 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
694 | |
695 | return ret; | |
8a4b83cc CM |
696 | } |
697 | ||
e4404d6e YZ |
698 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
699 | { | |
700 | struct btrfs_fs_devices *fs_devices; | |
701 | struct btrfs_device *device; | |
702 | struct btrfs_device *orig_dev; | |
703 | ||
2208a378 ID |
704 | fs_devices = alloc_fs_devices(orig->fsid); |
705 | if (IS_ERR(fs_devices)) | |
706 | return fs_devices; | |
e4404d6e | 707 | |
adbbb863 | 708 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 709 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 710 | |
46224705 | 711 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 712 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
713 | struct rcu_string *name; |
714 | ||
12bd2fc0 ID |
715 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
716 | orig_dev->uuid); | |
717 | if (IS_ERR(device)) | |
e4404d6e YZ |
718 | goto error; |
719 | ||
606686ee JB |
720 | /* |
721 | * This is ok to do without rcu read locked because we hold the | |
722 | * uuid mutex so nothing we touch in here is going to disappear. | |
723 | */ | |
e755f780 AJ |
724 | if (orig_dev->name) { |
725 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
726 | if (!name) { | |
727 | kfree(device); | |
728 | goto error; | |
729 | } | |
730 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 731 | } |
e4404d6e | 732 | |
e4404d6e YZ |
733 | list_add(&device->dev_list, &fs_devices->devices); |
734 | device->fs_devices = fs_devices; | |
735 | fs_devices->num_devices++; | |
736 | } | |
adbbb863 | 737 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
738 | return fs_devices; |
739 | error: | |
adbbb863 | 740 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
741 | free_fs_devices(fs_devices); |
742 | return ERR_PTR(-ENOMEM); | |
743 | } | |
744 | ||
9eaed21e | 745 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 746 | { |
c6e30871 | 747 | struct btrfs_device *device, *next; |
443f24fe | 748 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 749 | |
dfe25020 CM |
750 | mutex_lock(&uuid_mutex); |
751 | again: | |
46224705 | 752 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 753 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 754 | if (device->in_fs_metadata) { |
63a212ab | 755 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
756 | (!latest_dev || |
757 | device->generation > latest_dev->generation)) { | |
758 | latest_dev = device; | |
a6b0d5c8 | 759 | } |
2b82032c | 760 | continue; |
a6b0d5c8 | 761 | } |
2b82032c | 762 | |
8dabb742 SB |
763 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
764 | /* | |
765 | * In the first step, keep the device which has | |
766 | * the correct fsid and the devid that is used | |
767 | * for the dev_replace procedure. | |
768 | * In the second step, the dev_replace state is | |
769 | * read from the device tree and it is known | |
770 | * whether the procedure is really active or | |
771 | * not, which means whether this device is | |
772 | * used or whether it should be removed. | |
773 | */ | |
774 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
775 | continue; | |
776 | } | |
777 | } | |
2b82032c | 778 | if (device->bdev) { |
d4d77629 | 779 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
780 | device->bdev = NULL; |
781 | fs_devices->open_devices--; | |
782 | } | |
783 | if (device->writeable) { | |
784 | list_del_init(&device->dev_alloc_list); | |
785 | device->writeable = 0; | |
8dabb742 SB |
786 | if (!device->is_tgtdev_for_dev_replace) |
787 | fs_devices->rw_devices--; | |
2b82032c | 788 | } |
e4404d6e YZ |
789 | list_del_init(&device->dev_list); |
790 | fs_devices->num_devices--; | |
606686ee | 791 | rcu_string_free(device->name); |
e4404d6e | 792 | kfree(device); |
dfe25020 | 793 | } |
2b82032c YZ |
794 | |
795 | if (fs_devices->seed) { | |
796 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
797 | goto again; |
798 | } | |
799 | ||
443f24fe | 800 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 801 | |
dfe25020 | 802 | mutex_unlock(&uuid_mutex); |
dfe25020 | 803 | } |
a0af469b | 804 | |
1f78160c XG |
805 | static void __free_device(struct work_struct *work) |
806 | { | |
807 | struct btrfs_device *device; | |
808 | ||
809 | device = container_of(work, struct btrfs_device, rcu_work); | |
810 | ||
811 | if (device->bdev) | |
812 | blkdev_put(device->bdev, device->mode); | |
813 | ||
606686ee | 814 | rcu_string_free(device->name); |
1f78160c XG |
815 | kfree(device); |
816 | } | |
817 | ||
818 | static void free_device(struct rcu_head *head) | |
819 | { | |
820 | struct btrfs_device *device; | |
821 | ||
822 | device = container_of(head, struct btrfs_device, rcu); | |
823 | ||
824 | INIT_WORK(&device->rcu_work, __free_device); | |
825 | schedule_work(&device->rcu_work); | |
826 | } | |
827 | ||
2b82032c | 828 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 829 | { |
2037a093 | 830 | struct btrfs_device *device, *tmp; |
e4404d6e | 831 | |
2b82032c YZ |
832 | if (--fs_devices->opened > 0) |
833 | return 0; | |
8a4b83cc | 834 | |
c9513edb | 835 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 836 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
1f78160c | 837 | struct btrfs_device *new_device; |
606686ee | 838 | struct rcu_string *name; |
1f78160c XG |
839 | |
840 | if (device->bdev) | |
a0af469b | 841 | fs_devices->open_devices--; |
1f78160c | 842 | |
f747cab7 ID |
843 | if (device->writeable && |
844 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
845 | list_del_init(&device->dev_alloc_list); |
846 | fs_devices->rw_devices--; | |
847 | } | |
848 | ||
726551eb JB |
849 | if (device->missing) |
850 | fs_devices->missing_devices--; | |
d5e2003c | 851 | |
a1e8780a ID |
852 | new_device = btrfs_alloc_device(NULL, &device->devid, |
853 | device->uuid); | |
854 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
606686ee JB |
855 | |
856 | /* Safe because we are under uuid_mutex */ | |
99f5944b JB |
857 | if (device->name) { |
858 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
a1e8780a | 859 | BUG_ON(!name); /* -ENOMEM */ |
99f5944b JB |
860 | rcu_assign_pointer(new_device->name, name); |
861 | } | |
a1e8780a | 862 | |
1f78160c | 863 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
a1e8780a | 864 | new_device->fs_devices = device->fs_devices; |
1f78160c XG |
865 | |
866 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 867 | } |
c9513edb XG |
868 | mutex_unlock(&fs_devices->device_list_mutex); |
869 | ||
e4404d6e YZ |
870 | WARN_ON(fs_devices->open_devices); |
871 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
872 | fs_devices->opened = 0; |
873 | fs_devices->seeding = 0; | |
2b82032c | 874 | |
8a4b83cc CM |
875 | return 0; |
876 | } | |
877 | ||
2b82032c YZ |
878 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
879 | { | |
e4404d6e | 880 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
881 | int ret; |
882 | ||
883 | mutex_lock(&uuid_mutex); | |
884 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
885 | if (!fs_devices->opened) { |
886 | seed_devices = fs_devices->seed; | |
887 | fs_devices->seed = NULL; | |
888 | } | |
2b82032c | 889 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
890 | |
891 | while (seed_devices) { | |
892 | fs_devices = seed_devices; | |
893 | seed_devices = fs_devices->seed; | |
894 | __btrfs_close_devices(fs_devices); | |
895 | free_fs_devices(fs_devices); | |
896 | } | |
bc178622 ES |
897 | /* |
898 | * Wait for rcu kworkers under __btrfs_close_devices | |
899 | * to finish all blkdev_puts so device is really | |
900 | * free when umount is done. | |
901 | */ | |
902 | rcu_barrier(); | |
2b82032c YZ |
903 | return ret; |
904 | } | |
905 | ||
e4404d6e YZ |
906 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
907 | fmode_t flags, void *holder) | |
8a4b83cc | 908 | { |
d5e2003c | 909 | struct request_queue *q; |
8a4b83cc CM |
910 | struct block_device *bdev; |
911 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 912 | struct btrfs_device *device; |
443f24fe | 913 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
914 | struct buffer_head *bh; |
915 | struct btrfs_super_block *disk_super; | |
a0af469b | 916 | u64 devid; |
2b82032c | 917 | int seeding = 1; |
a0af469b | 918 | int ret = 0; |
8a4b83cc | 919 | |
d4d77629 TH |
920 | flags |= FMODE_EXCL; |
921 | ||
c6e30871 | 922 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
923 | if (device->bdev) |
924 | continue; | |
dfe25020 CM |
925 | if (!device->name) |
926 | continue; | |
927 | ||
f63e0cca ES |
928 | /* Just open everything we can; ignore failures here */ |
929 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
930 | &bdev, &bh)) | |
beaf8ab3 | 931 | continue; |
a0af469b CM |
932 | |
933 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 934 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
935 | if (devid != device->devid) |
936 | goto error_brelse; | |
937 | ||
2b82032c YZ |
938 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
939 | BTRFS_UUID_SIZE)) | |
940 | goto error_brelse; | |
941 | ||
942 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
943 | if (!latest_dev || |
944 | device->generation > latest_dev->generation) | |
945 | latest_dev = device; | |
a0af469b | 946 | |
2b82032c YZ |
947 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
948 | device->writeable = 0; | |
949 | } else { | |
950 | device->writeable = !bdev_read_only(bdev); | |
951 | seeding = 0; | |
952 | } | |
953 | ||
d5e2003c | 954 | q = bdev_get_queue(bdev); |
90180da4 | 955 | if (blk_queue_discard(q)) |
d5e2003c | 956 | device->can_discard = 1; |
d5e2003c | 957 | |
8a4b83cc | 958 | device->bdev = bdev; |
dfe25020 | 959 | device->in_fs_metadata = 0; |
15916de8 CM |
960 | device->mode = flags; |
961 | ||
c289811c CM |
962 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
963 | fs_devices->rotating = 1; | |
964 | ||
a0af469b | 965 | fs_devices->open_devices++; |
55e50e45 ID |
966 | if (device->writeable && |
967 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
968 | fs_devices->rw_devices++; |
969 | list_add(&device->dev_alloc_list, | |
970 | &fs_devices->alloc_list); | |
971 | } | |
4f6c9328 | 972 | brelse(bh); |
a0af469b | 973 | continue; |
a061fc8d | 974 | |
a0af469b CM |
975 | error_brelse: |
976 | brelse(bh); | |
d4d77629 | 977 | blkdev_put(bdev, flags); |
a0af469b | 978 | continue; |
8a4b83cc | 979 | } |
a0af469b | 980 | if (fs_devices->open_devices == 0) { |
20bcd649 | 981 | ret = -EINVAL; |
a0af469b CM |
982 | goto out; |
983 | } | |
2b82032c YZ |
984 | fs_devices->seeding = seeding; |
985 | fs_devices->opened = 1; | |
443f24fe | 986 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 987 | fs_devices->total_rw_bytes = 0; |
a0af469b | 988 | out: |
2b82032c YZ |
989 | return ret; |
990 | } | |
991 | ||
992 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 993 | fmode_t flags, void *holder) |
2b82032c YZ |
994 | { |
995 | int ret; | |
996 | ||
997 | mutex_lock(&uuid_mutex); | |
998 | if (fs_devices->opened) { | |
e4404d6e YZ |
999 | fs_devices->opened++; |
1000 | ret = 0; | |
2b82032c | 1001 | } else { |
15916de8 | 1002 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1003 | } |
8a4b83cc | 1004 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1005 | return ret; |
1006 | } | |
1007 | ||
6f60cbd3 DS |
1008 | /* |
1009 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1010 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1011 | * is read via pagecache | |
1012 | */ | |
97288f2c | 1013 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1014 | struct btrfs_fs_devices **fs_devices_ret) |
1015 | { | |
1016 | struct btrfs_super_block *disk_super; | |
1017 | struct block_device *bdev; | |
6f60cbd3 DS |
1018 | struct page *page; |
1019 | void *p; | |
1020 | int ret = -EINVAL; | |
8a4b83cc | 1021 | u64 devid; |
f2984462 | 1022 | u64 transid; |
02db0844 | 1023 | u64 total_devices; |
6f60cbd3 DS |
1024 | u64 bytenr; |
1025 | pgoff_t index; | |
8a4b83cc | 1026 | |
6f60cbd3 DS |
1027 | /* |
1028 | * we would like to check all the supers, but that would make | |
1029 | * a btrfs mount succeed after a mkfs from a different FS. | |
1030 | * So, we need to add a special mount option to scan for | |
1031 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1032 | */ | |
1033 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1034 | flags |= FMODE_EXCL; |
10f6327b | 1035 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1036 | |
1037 | bdev = blkdev_get_by_path(path, flags, holder); | |
1038 | ||
1039 | if (IS_ERR(bdev)) { | |
1040 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1041 | goto error; |
6f60cbd3 DS |
1042 | } |
1043 | ||
1044 | /* make sure our super fits in the device */ | |
1045 | if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode)) | |
1046 | goto error_bdev_put; | |
1047 | ||
1048 | /* make sure our super fits in the page */ | |
1049 | if (sizeof(*disk_super) > PAGE_CACHE_SIZE) | |
1050 | goto error_bdev_put; | |
1051 | ||
1052 | /* make sure our super doesn't straddle pages on disk */ | |
1053 | index = bytenr >> PAGE_CACHE_SHIFT; | |
1054 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index) | |
1055 | goto error_bdev_put; | |
1056 | ||
1057 | /* pull in the page with our super */ | |
1058 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1059 | index, GFP_NOFS); | |
1060 | ||
1061 | if (IS_ERR_OR_NULL(page)) | |
1062 | goto error_bdev_put; | |
1063 | ||
1064 | p = kmap(page); | |
1065 | ||
1066 | /* align our pointer to the offset of the super block */ | |
1067 | disk_super = p + (bytenr & ~PAGE_CACHE_MASK); | |
1068 | ||
1069 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
3cae210f | 1070 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) |
6f60cbd3 DS |
1071 | goto error_unmap; |
1072 | ||
a343832f | 1073 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1074 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1075 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1076 | |
8a4b83cc | 1077 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1078 | if (ret > 0) { |
1079 | if (disk_super->label[0]) { | |
1080 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
1081 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1082 | printk(KERN_INFO "BTRFS: device label %s ", disk_super->label); | |
1083 | } else { | |
1084 | printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid); | |
1085 | } | |
1086 | ||
1087 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); | |
1088 | ret = 0; | |
1089 | } | |
02db0844 JB |
1090 | if (!ret && fs_devices_ret) |
1091 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 DS |
1092 | |
1093 | error_unmap: | |
1094 | kunmap(page); | |
1095 | page_cache_release(page); | |
1096 | ||
1097 | error_bdev_put: | |
d4d77629 | 1098 | blkdev_put(bdev, flags); |
8a4b83cc | 1099 | error: |
beaf8ab3 | 1100 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1101 | return ret; |
1102 | } | |
0b86a832 | 1103 | |
6d07bcec MX |
1104 | /* helper to account the used device space in the range */ |
1105 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1106 | u64 end, u64 *length) | |
1107 | { | |
1108 | struct btrfs_key key; | |
1109 | struct btrfs_root *root = device->dev_root; | |
1110 | struct btrfs_dev_extent *dev_extent; | |
1111 | struct btrfs_path *path; | |
1112 | u64 extent_end; | |
1113 | int ret; | |
1114 | int slot; | |
1115 | struct extent_buffer *l; | |
1116 | ||
1117 | *length = 0; | |
1118 | ||
63a212ab | 1119 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1120 | return 0; |
1121 | ||
1122 | path = btrfs_alloc_path(); | |
1123 | if (!path) | |
1124 | return -ENOMEM; | |
1125 | path->reada = 2; | |
1126 | ||
1127 | key.objectid = device->devid; | |
1128 | key.offset = start; | |
1129 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1130 | ||
1131 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1132 | if (ret < 0) | |
1133 | goto out; | |
1134 | if (ret > 0) { | |
1135 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1136 | if (ret < 0) | |
1137 | goto out; | |
1138 | } | |
1139 | ||
1140 | while (1) { | |
1141 | l = path->nodes[0]; | |
1142 | slot = path->slots[0]; | |
1143 | if (slot >= btrfs_header_nritems(l)) { | |
1144 | ret = btrfs_next_leaf(root, path); | |
1145 | if (ret == 0) | |
1146 | continue; | |
1147 | if (ret < 0) | |
1148 | goto out; | |
1149 | ||
1150 | break; | |
1151 | } | |
1152 | btrfs_item_key_to_cpu(l, &key, slot); | |
1153 | ||
1154 | if (key.objectid < device->devid) | |
1155 | goto next; | |
1156 | ||
1157 | if (key.objectid > device->devid) | |
1158 | break; | |
1159 | ||
962a298f | 1160 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1161 | goto next; |
1162 | ||
1163 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1164 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1165 | dev_extent); | |
1166 | if (key.offset <= start && extent_end > end) { | |
1167 | *length = end - start + 1; | |
1168 | break; | |
1169 | } else if (key.offset <= start && extent_end > start) | |
1170 | *length += extent_end - start; | |
1171 | else if (key.offset > start && extent_end <= end) | |
1172 | *length += extent_end - key.offset; | |
1173 | else if (key.offset > start && key.offset <= end) { | |
1174 | *length += end - key.offset + 1; | |
1175 | break; | |
1176 | } else if (key.offset > end) | |
1177 | break; | |
1178 | ||
1179 | next: | |
1180 | path->slots[0]++; | |
1181 | } | |
1182 | ret = 0; | |
1183 | out: | |
1184 | btrfs_free_path(path); | |
1185 | return ret; | |
1186 | } | |
1187 | ||
499f377f | 1188 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1189 | struct btrfs_device *device, |
1190 | u64 *start, u64 len) | |
1191 | { | |
499f377f | 1192 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; |
6df9a95e | 1193 | struct extent_map *em; |
499f377f | 1194 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1195 | int ret = 0; |
1b984508 | 1196 | u64 physical_start = *start; |
6df9a95e | 1197 | |
499f377f JM |
1198 | if (transaction) |
1199 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1200 | again: |
1201 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1202 | struct map_lookup *map; |
1203 | int i; | |
1204 | ||
1205 | map = (struct map_lookup *)em->bdev; | |
1206 | for (i = 0; i < map->num_stripes; i++) { | |
c152b63e FM |
1207 | u64 end; |
1208 | ||
6df9a95e JB |
1209 | if (map->stripes[i].dev != device) |
1210 | continue; | |
1b984508 | 1211 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1212 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1213 | physical_start) |
6df9a95e | 1214 | continue; |
c152b63e FM |
1215 | /* |
1216 | * Make sure that while processing the pinned list we do | |
1217 | * not override our *start with a lower value, because | |
1218 | * we can have pinned chunks that fall within this | |
1219 | * device hole and that have lower physical addresses | |
1220 | * than the pending chunks we processed before. If we | |
1221 | * do not take this special care we can end up getting | |
1222 | * 2 pending chunks that start at the same physical | |
1223 | * device offsets because the end offset of a pinned | |
1224 | * chunk can be equal to the start offset of some | |
1225 | * pending chunk. | |
1226 | */ | |
1227 | end = map->stripes[i].physical + em->orig_block_len; | |
1228 | if (end > *start) { | |
1229 | *start = end; | |
1230 | ret = 1; | |
1231 | } | |
6df9a95e JB |
1232 | } |
1233 | } | |
499f377f JM |
1234 | if (search_list != &fs_info->pinned_chunks) { |
1235 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1236 | goto again; |
1237 | } | |
6df9a95e JB |
1238 | |
1239 | return ret; | |
1240 | } | |
1241 | ||
1242 | ||
0b86a832 | 1243 | /* |
499f377f JM |
1244 | * find_free_dev_extent_start - find free space in the specified device |
1245 | * @device: the device which we search the free space in | |
1246 | * @num_bytes: the size of the free space that we need | |
1247 | * @search_start: the position from which to begin the search | |
1248 | * @start: store the start of the free space. | |
1249 | * @len: the size of the free space. that we find, or the size | |
1250 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1251 | * |
0b86a832 CM |
1252 | * this uses a pretty simple search, the expectation is that it is |
1253 | * called very infrequently and that a given device has a small number | |
1254 | * of extents | |
7bfc837d MX |
1255 | * |
1256 | * @start is used to store the start of the free space if we find. But if we | |
1257 | * don't find suitable free space, it will be used to store the start position | |
1258 | * of the max free space. | |
1259 | * | |
1260 | * @len is used to store the size of the free space that we find. | |
1261 | * But if we don't find suitable free space, it is used to store the size of | |
1262 | * the max free space. | |
0b86a832 | 1263 | */ |
499f377f JM |
1264 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1265 | struct btrfs_device *device, u64 num_bytes, | |
1266 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 CM |
1267 | { |
1268 | struct btrfs_key key; | |
1269 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1270 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1271 | struct btrfs_path *path; |
7bfc837d MX |
1272 | u64 hole_size; |
1273 | u64 max_hole_start; | |
1274 | u64 max_hole_size; | |
1275 | u64 extent_end; | |
0b86a832 CM |
1276 | u64 search_end = device->total_bytes; |
1277 | int ret; | |
7bfc837d | 1278 | int slot; |
0b86a832 CM |
1279 | struct extent_buffer *l; |
1280 | ||
6df9a95e JB |
1281 | path = btrfs_alloc_path(); |
1282 | if (!path) | |
1283 | return -ENOMEM; | |
f2ab7618 | 1284 | |
7bfc837d MX |
1285 | max_hole_start = search_start; |
1286 | max_hole_size = 0; | |
1287 | ||
f2ab7618 | 1288 | again: |
63a212ab | 1289 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1290 | ret = -ENOSPC; |
6df9a95e | 1291 | goto out; |
7bfc837d MX |
1292 | } |
1293 | ||
7bfc837d | 1294 | path->reada = 2; |
6df9a95e JB |
1295 | path->search_commit_root = 1; |
1296 | path->skip_locking = 1; | |
7bfc837d | 1297 | |
0b86a832 CM |
1298 | key.objectid = device->devid; |
1299 | key.offset = search_start; | |
1300 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1301 | |
125ccb0a | 1302 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1303 | if (ret < 0) |
7bfc837d | 1304 | goto out; |
1fcbac58 YZ |
1305 | if (ret > 0) { |
1306 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1307 | if (ret < 0) | |
7bfc837d | 1308 | goto out; |
1fcbac58 | 1309 | } |
7bfc837d | 1310 | |
0b86a832 CM |
1311 | while (1) { |
1312 | l = path->nodes[0]; | |
1313 | slot = path->slots[0]; | |
1314 | if (slot >= btrfs_header_nritems(l)) { | |
1315 | ret = btrfs_next_leaf(root, path); | |
1316 | if (ret == 0) | |
1317 | continue; | |
1318 | if (ret < 0) | |
7bfc837d MX |
1319 | goto out; |
1320 | ||
1321 | break; | |
0b86a832 CM |
1322 | } |
1323 | btrfs_item_key_to_cpu(l, &key, slot); | |
1324 | ||
1325 | if (key.objectid < device->devid) | |
1326 | goto next; | |
1327 | ||
1328 | if (key.objectid > device->devid) | |
7bfc837d | 1329 | break; |
0b86a832 | 1330 | |
962a298f | 1331 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1332 | goto next; |
9779b72f | 1333 | |
7bfc837d MX |
1334 | if (key.offset > search_start) { |
1335 | hole_size = key.offset - search_start; | |
9779b72f | 1336 | |
6df9a95e JB |
1337 | /* |
1338 | * Have to check before we set max_hole_start, otherwise | |
1339 | * we could end up sending back this offset anyway. | |
1340 | */ | |
499f377f | 1341 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1342 | &search_start, |
1b984508 FL |
1343 | hole_size)) { |
1344 | if (key.offset >= search_start) { | |
1345 | hole_size = key.offset - search_start; | |
1346 | } else { | |
1347 | WARN_ON_ONCE(1); | |
1348 | hole_size = 0; | |
1349 | } | |
1350 | } | |
6df9a95e | 1351 | |
7bfc837d MX |
1352 | if (hole_size > max_hole_size) { |
1353 | max_hole_start = search_start; | |
1354 | max_hole_size = hole_size; | |
1355 | } | |
9779b72f | 1356 | |
7bfc837d MX |
1357 | /* |
1358 | * If this free space is greater than which we need, | |
1359 | * it must be the max free space that we have found | |
1360 | * until now, so max_hole_start must point to the start | |
1361 | * of this free space and the length of this free space | |
1362 | * is stored in max_hole_size. Thus, we return | |
1363 | * max_hole_start and max_hole_size and go back to the | |
1364 | * caller. | |
1365 | */ | |
1366 | if (hole_size >= num_bytes) { | |
1367 | ret = 0; | |
1368 | goto out; | |
0b86a832 CM |
1369 | } |
1370 | } | |
0b86a832 | 1371 | |
0b86a832 | 1372 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1373 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1374 | dev_extent); | |
1375 | if (extent_end > search_start) | |
1376 | search_start = extent_end; | |
0b86a832 CM |
1377 | next: |
1378 | path->slots[0]++; | |
1379 | cond_resched(); | |
1380 | } | |
0b86a832 | 1381 | |
38c01b96 | 1382 | /* |
1383 | * At this point, search_start should be the end of | |
1384 | * allocated dev extents, and when shrinking the device, | |
1385 | * search_end may be smaller than search_start. | |
1386 | */ | |
f2ab7618 | 1387 | if (search_end > search_start) { |
38c01b96 | 1388 | hole_size = search_end - search_start; |
1389 | ||
499f377f | 1390 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1391 | hole_size)) { |
1392 | btrfs_release_path(path); | |
1393 | goto again; | |
1394 | } | |
0b86a832 | 1395 | |
f2ab7618 ZL |
1396 | if (hole_size > max_hole_size) { |
1397 | max_hole_start = search_start; | |
1398 | max_hole_size = hole_size; | |
1399 | } | |
6df9a95e JB |
1400 | } |
1401 | ||
7bfc837d | 1402 | /* See above. */ |
f2ab7618 | 1403 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1404 | ret = -ENOSPC; |
1405 | else | |
1406 | ret = 0; | |
1407 | ||
1408 | out: | |
2b82032c | 1409 | btrfs_free_path(path); |
7bfc837d | 1410 | *start = max_hole_start; |
b2117a39 | 1411 | if (len) |
7bfc837d | 1412 | *len = max_hole_size; |
0b86a832 CM |
1413 | return ret; |
1414 | } | |
1415 | ||
499f377f JM |
1416 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1417 | struct btrfs_device *device, u64 num_bytes, | |
1418 | u64 *start, u64 *len) | |
1419 | { | |
1420 | struct btrfs_root *root = device->dev_root; | |
1421 | u64 search_start; | |
1422 | ||
1423 | /* FIXME use last free of some kind */ | |
1424 | ||
1425 | /* | |
1426 | * we don't want to overwrite the superblock on the drive, | |
1427 | * so we make sure to start at an offset of at least 1MB | |
1428 | */ | |
1429 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); | |
1430 | return find_free_dev_extent_start(trans->transaction, device, | |
1431 | num_bytes, search_start, start, len); | |
1432 | } | |
1433 | ||
b2950863 | 1434 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1435 | struct btrfs_device *device, |
2196d6e8 | 1436 | u64 start, u64 *dev_extent_len) |
8f18cf13 CM |
1437 | { |
1438 | int ret; | |
1439 | struct btrfs_path *path; | |
1440 | struct btrfs_root *root = device->dev_root; | |
1441 | struct btrfs_key key; | |
a061fc8d CM |
1442 | struct btrfs_key found_key; |
1443 | struct extent_buffer *leaf = NULL; | |
1444 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1445 | |
1446 | path = btrfs_alloc_path(); | |
1447 | if (!path) | |
1448 | return -ENOMEM; | |
1449 | ||
1450 | key.objectid = device->devid; | |
1451 | key.offset = start; | |
1452 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1453 | again: |
8f18cf13 | 1454 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1455 | if (ret > 0) { |
1456 | ret = btrfs_previous_item(root, path, key.objectid, | |
1457 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1458 | if (ret) |
1459 | goto out; | |
a061fc8d CM |
1460 | leaf = path->nodes[0]; |
1461 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1462 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1463 | struct btrfs_dev_extent); | |
1464 | BUG_ON(found_key.offset > start || found_key.offset + | |
1465 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1466 | key = found_key; |
1467 | btrfs_release_path(path); | |
1468 | goto again; | |
a061fc8d CM |
1469 | } else if (ret == 0) { |
1470 | leaf = path->nodes[0]; | |
1471 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1472 | struct btrfs_dev_extent); | |
79787eaa JM |
1473 | } else { |
1474 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1475 | goto out; | |
a061fc8d | 1476 | } |
8f18cf13 | 1477 | |
2196d6e8 MX |
1478 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1479 | ||
8f18cf13 | 1480 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1481 | if (ret) { |
1482 | btrfs_error(root->fs_info, ret, | |
1483 | "Failed to remove dev extent item"); | |
13212b54 ZL |
1484 | } else { |
1485 | trans->transaction->have_free_bgs = 1; | |
79787eaa | 1486 | } |
b0b802d7 | 1487 | out: |
8f18cf13 CM |
1488 | btrfs_free_path(path); |
1489 | return ret; | |
1490 | } | |
1491 | ||
48a3b636 ES |
1492 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1493 | struct btrfs_device *device, | |
1494 | u64 chunk_tree, u64 chunk_objectid, | |
1495 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1496 | { |
1497 | int ret; | |
1498 | struct btrfs_path *path; | |
1499 | struct btrfs_root *root = device->dev_root; | |
1500 | struct btrfs_dev_extent *extent; | |
1501 | struct extent_buffer *leaf; | |
1502 | struct btrfs_key key; | |
1503 | ||
dfe25020 | 1504 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1505 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1506 | path = btrfs_alloc_path(); |
1507 | if (!path) | |
1508 | return -ENOMEM; | |
1509 | ||
0b86a832 | 1510 | key.objectid = device->devid; |
2b82032c | 1511 | key.offset = start; |
0b86a832 CM |
1512 | key.type = BTRFS_DEV_EXTENT_KEY; |
1513 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1514 | sizeof(*extent)); | |
2cdcecbc MF |
1515 | if (ret) |
1516 | goto out; | |
0b86a832 CM |
1517 | |
1518 | leaf = path->nodes[0]; | |
1519 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1520 | struct btrfs_dev_extent); | |
e17cade2 CM |
1521 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1522 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1523 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1524 | ||
1525 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1526 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1527 | |
0b86a832 CM |
1528 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1529 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1530 | out: |
0b86a832 CM |
1531 | btrfs_free_path(path); |
1532 | return ret; | |
1533 | } | |
1534 | ||
6df9a95e | 1535 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1536 | { |
6df9a95e JB |
1537 | struct extent_map_tree *em_tree; |
1538 | struct extent_map *em; | |
1539 | struct rb_node *n; | |
1540 | u64 ret = 0; | |
0b86a832 | 1541 | |
6df9a95e JB |
1542 | em_tree = &fs_info->mapping_tree.map_tree; |
1543 | read_lock(&em_tree->lock); | |
1544 | n = rb_last(&em_tree->map); | |
1545 | if (n) { | |
1546 | em = rb_entry(n, struct extent_map, rb_node); | |
1547 | ret = em->start + em->len; | |
0b86a832 | 1548 | } |
6df9a95e JB |
1549 | read_unlock(&em_tree->lock); |
1550 | ||
0b86a832 CM |
1551 | return ret; |
1552 | } | |
1553 | ||
53f10659 ID |
1554 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1555 | u64 *devid_ret) | |
0b86a832 CM |
1556 | { |
1557 | int ret; | |
1558 | struct btrfs_key key; | |
1559 | struct btrfs_key found_key; | |
2b82032c YZ |
1560 | struct btrfs_path *path; |
1561 | ||
2b82032c YZ |
1562 | path = btrfs_alloc_path(); |
1563 | if (!path) | |
1564 | return -ENOMEM; | |
0b86a832 CM |
1565 | |
1566 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1567 | key.type = BTRFS_DEV_ITEM_KEY; | |
1568 | key.offset = (u64)-1; | |
1569 | ||
53f10659 | 1570 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1571 | if (ret < 0) |
1572 | goto error; | |
1573 | ||
79787eaa | 1574 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1575 | |
53f10659 ID |
1576 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1577 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1578 | BTRFS_DEV_ITEM_KEY); |
1579 | if (ret) { | |
53f10659 | 1580 | *devid_ret = 1; |
0b86a832 CM |
1581 | } else { |
1582 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1583 | path->slots[0]); | |
53f10659 | 1584 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1585 | } |
1586 | ret = 0; | |
1587 | error: | |
2b82032c | 1588 | btrfs_free_path(path); |
0b86a832 CM |
1589 | return ret; |
1590 | } | |
1591 | ||
1592 | /* | |
1593 | * the device information is stored in the chunk root | |
1594 | * the btrfs_device struct should be fully filled in | |
1595 | */ | |
48a3b636 ES |
1596 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1597 | struct btrfs_root *root, | |
1598 | struct btrfs_device *device) | |
0b86a832 CM |
1599 | { |
1600 | int ret; | |
1601 | struct btrfs_path *path; | |
1602 | struct btrfs_dev_item *dev_item; | |
1603 | struct extent_buffer *leaf; | |
1604 | struct btrfs_key key; | |
1605 | unsigned long ptr; | |
0b86a832 CM |
1606 | |
1607 | root = root->fs_info->chunk_root; | |
1608 | ||
1609 | path = btrfs_alloc_path(); | |
1610 | if (!path) | |
1611 | return -ENOMEM; | |
1612 | ||
0b86a832 CM |
1613 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1614 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1615 | key.offset = device->devid; |
0b86a832 CM |
1616 | |
1617 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1618 | sizeof(*dev_item)); |
0b86a832 CM |
1619 | if (ret) |
1620 | goto out; | |
1621 | ||
1622 | leaf = path->nodes[0]; | |
1623 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1624 | ||
1625 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1626 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1627 | btrfs_set_device_type(leaf, dev_item, device->type); |
1628 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1629 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1630 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1631 | btrfs_set_device_total_bytes(leaf, dev_item, |
1632 | btrfs_device_get_disk_total_bytes(device)); | |
1633 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1634 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1635 | btrfs_set_device_group(leaf, dev_item, 0); |
1636 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1637 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1638 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1639 | |
410ba3a2 | 1640 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1641 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1642 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1643 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1644 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1645 | |
2b82032c | 1646 | ret = 0; |
0b86a832 CM |
1647 | out: |
1648 | btrfs_free_path(path); | |
1649 | return ret; | |
1650 | } | |
8f18cf13 | 1651 | |
5a1972bd QW |
1652 | /* |
1653 | * Function to update ctime/mtime for a given device path. | |
1654 | * Mainly used for ctime/mtime based probe like libblkid. | |
1655 | */ | |
1656 | static void update_dev_time(char *path_name) | |
1657 | { | |
1658 | struct file *filp; | |
1659 | ||
1660 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1661 | if (IS_ERR(filp)) |
5a1972bd QW |
1662 | return; |
1663 | file_update_time(filp); | |
1664 | filp_close(filp, NULL); | |
1665 | return; | |
1666 | } | |
1667 | ||
a061fc8d CM |
1668 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1669 | struct btrfs_device *device) | |
1670 | { | |
1671 | int ret; | |
1672 | struct btrfs_path *path; | |
a061fc8d | 1673 | struct btrfs_key key; |
a061fc8d CM |
1674 | struct btrfs_trans_handle *trans; |
1675 | ||
1676 | root = root->fs_info->chunk_root; | |
1677 | ||
1678 | path = btrfs_alloc_path(); | |
1679 | if (!path) | |
1680 | return -ENOMEM; | |
1681 | ||
a22285a6 | 1682 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1683 | if (IS_ERR(trans)) { |
1684 | btrfs_free_path(path); | |
1685 | return PTR_ERR(trans); | |
1686 | } | |
a061fc8d CM |
1687 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1688 | key.type = BTRFS_DEV_ITEM_KEY; | |
1689 | key.offset = device->devid; | |
1690 | ||
1691 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1692 | if (ret < 0) | |
1693 | goto out; | |
1694 | ||
1695 | if (ret > 0) { | |
1696 | ret = -ENOENT; | |
1697 | goto out; | |
1698 | } | |
1699 | ||
1700 | ret = btrfs_del_item(trans, root, path); | |
1701 | if (ret) | |
1702 | goto out; | |
a061fc8d CM |
1703 | out: |
1704 | btrfs_free_path(path); | |
1705 | btrfs_commit_transaction(trans, root); | |
1706 | return ret; | |
1707 | } | |
1708 | ||
1709 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1710 | { | |
1711 | struct btrfs_device *device; | |
2b82032c | 1712 | struct btrfs_device *next_device; |
a061fc8d | 1713 | struct block_device *bdev; |
dfe25020 | 1714 | struct buffer_head *bh = NULL; |
a061fc8d | 1715 | struct btrfs_super_block *disk_super; |
1f78160c | 1716 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1717 | u64 all_avail; |
1718 | u64 devid; | |
2b82032c YZ |
1719 | u64 num_devices; |
1720 | u8 *dev_uuid; | |
de98ced9 | 1721 | unsigned seq; |
a061fc8d | 1722 | int ret = 0; |
1f78160c | 1723 | bool clear_super = false; |
a061fc8d | 1724 | |
a061fc8d CM |
1725 | mutex_lock(&uuid_mutex); |
1726 | ||
de98ced9 MX |
1727 | do { |
1728 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
1729 | ||
1730 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1731 | root->fs_info->avail_system_alloc_bits | | |
1732 | root->fs_info->avail_metadata_alloc_bits; | |
1733 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
a061fc8d | 1734 | |
8dabb742 SB |
1735 | num_devices = root->fs_info->fs_devices->num_devices; |
1736 | btrfs_dev_replace_lock(&root->fs_info->dev_replace); | |
1737 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1738 | WARN_ON(num_devices < 1); | |
1739 | num_devices--; | |
1740 | } | |
1741 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace); | |
1742 | ||
1743 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) { | |
183860f6 | 1744 | ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET; |
a061fc8d CM |
1745 | goto out; |
1746 | } | |
1747 | ||
8dabb742 | 1748 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) { |
183860f6 | 1749 | ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET; |
a061fc8d CM |
1750 | goto out; |
1751 | } | |
1752 | ||
53b381b3 DW |
1753 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && |
1754 | root->fs_info->fs_devices->rw_devices <= 2) { | |
183860f6 | 1755 | ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET; |
53b381b3 DW |
1756 | goto out; |
1757 | } | |
1758 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1759 | root->fs_info->fs_devices->rw_devices <= 3) { | |
183860f6 | 1760 | ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET; |
53b381b3 DW |
1761 | goto out; |
1762 | } | |
1763 | ||
dfe25020 | 1764 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1765 | struct list_head *devices; |
1766 | struct btrfs_device *tmp; | |
a061fc8d | 1767 | |
dfe25020 CM |
1768 | device = NULL; |
1769 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1770 | /* |
1771 | * It is safe to read the devices since the volume_mutex | |
1772 | * is held. | |
1773 | */ | |
c6e30871 | 1774 | list_for_each_entry(tmp, devices, dev_list) { |
63a212ab SB |
1775 | if (tmp->in_fs_metadata && |
1776 | !tmp->is_tgtdev_for_dev_replace && | |
1777 | !tmp->bdev) { | |
dfe25020 CM |
1778 | device = tmp; |
1779 | break; | |
1780 | } | |
1781 | } | |
1782 | bdev = NULL; | |
1783 | bh = NULL; | |
1784 | disk_super = NULL; | |
1785 | if (!device) { | |
183860f6 | 1786 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
dfe25020 CM |
1787 | goto out; |
1788 | } | |
dfe25020 | 1789 | } else { |
beaf8ab3 | 1790 | ret = btrfs_get_bdev_and_sb(device_path, |
cc975eb4 | 1791 | FMODE_WRITE | FMODE_EXCL, |
beaf8ab3 SB |
1792 | root->fs_info->bdev_holder, 0, |
1793 | &bdev, &bh); | |
1794 | if (ret) | |
dfe25020 | 1795 | goto out; |
dfe25020 | 1796 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1797 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c | 1798 | dev_uuid = disk_super->dev_item.uuid; |
aa1b8cd4 | 1799 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2b82032c | 1800 | disk_super->fsid); |
dfe25020 CM |
1801 | if (!device) { |
1802 | ret = -ENOENT; | |
1803 | goto error_brelse; | |
1804 | } | |
2b82032c | 1805 | } |
dfe25020 | 1806 | |
63a212ab | 1807 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1808 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
63a212ab SB |
1809 | goto error_brelse; |
1810 | } | |
1811 | ||
2b82032c | 1812 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1813 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
2b82032c YZ |
1814 | goto error_brelse; |
1815 | } | |
1816 | ||
1817 | if (device->writeable) { | |
0c1daee0 | 1818 | lock_chunks(root); |
2b82032c | 1819 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1820 | device->fs_devices->rw_devices--; |
0c1daee0 | 1821 | unlock_chunks(root); |
1f78160c | 1822 | clear_super = true; |
dfe25020 | 1823 | } |
a061fc8d | 1824 | |
d7901554 | 1825 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1826 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1827 | mutex_lock(&uuid_mutex); |
a061fc8d | 1828 | if (ret) |
9b3517e9 | 1829 | goto error_undo; |
a061fc8d | 1830 | |
63a212ab SB |
1831 | /* |
1832 | * TODO: the superblock still includes this device in its num_devices | |
1833 | * counter although write_all_supers() is not locked out. This | |
1834 | * could give a filesystem state which requires a degraded mount. | |
1835 | */ | |
a061fc8d CM |
1836 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1837 | if (ret) | |
9b3517e9 | 1838 | goto error_undo; |
a061fc8d | 1839 | |
2b82032c | 1840 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1841 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1842 | |
1843 | /* | |
1844 | * the device list mutex makes sure that we don't change | |
1845 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1846 | * the device supers. Whoever is writing all supers, should |
1847 | * lock the device list mutex before getting the number of | |
1848 | * devices in the super block (super_copy). Conversely, | |
1849 | * whoever updates the number of devices in the super block | |
1850 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1851 | */ |
1f78160c XG |
1852 | |
1853 | cur_devices = device->fs_devices; | |
e5e9a520 | 1854 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1855 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1856 | |
e4404d6e | 1857 | device->fs_devices->num_devices--; |
02db0844 | 1858 | device->fs_devices->total_devices--; |
2b82032c | 1859 | |
cd02dca5 | 1860 | if (device->missing) |
3a7d55c8 | 1861 | device->fs_devices->missing_devices--; |
cd02dca5 | 1862 | |
2b82032c YZ |
1863 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1864 | struct btrfs_device, dev_list); | |
1865 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1866 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1867 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1868 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1869 | ||
0bfaa9c5 | 1870 | if (device->bdev) { |
e4404d6e | 1871 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1872 | /* remove sysfs entry */ |
6c14a164 | 1873 | btrfs_kobj_rm_device(root->fs_info->fs_devices, device); |
0bfaa9c5 | 1874 | } |
99994cde | 1875 | |
1f78160c | 1876 | call_rcu(&device->rcu, free_device); |
e4404d6e | 1877 | |
6c41761f DS |
1878 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1879 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
d7306801 | 1880 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2b82032c | 1881 | |
1f78160c | 1882 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1883 | struct btrfs_fs_devices *fs_devices; |
1884 | fs_devices = root->fs_info->fs_devices; | |
1885 | while (fs_devices) { | |
8321cf25 RS |
1886 | if (fs_devices->seed == cur_devices) { |
1887 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1888 | break; |
8321cf25 | 1889 | } |
e4404d6e | 1890 | fs_devices = fs_devices->seed; |
2b82032c | 1891 | } |
1f78160c | 1892 | cur_devices->seed = NULL; |
1f78160c | 1893 | __btrfs_close_devices(cur_devices); |
1f78160c | 1894 | free_fs_devices(cur_devices); |
2b82032c YZ |
1895 | } |
1896 | ||
5af3e8cc SB |
1897 | root->fs_info->num_tolerated_disk_barrier_failures = |
1898 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1899 | ||
2b82032c YZ |
1900 | /* |
1901 | * at this point, the device is zero sized. We want to | |
1902 | * remove it from the devices list and zero out the old super | |
1903 | */ | |
aa1b8cd4 | 1904 | if (clear_super && disk_super) { |
4d90d28b AJ |
1905 | u64 bytenr; |
1906 | int i; | |
1907 | ||
dfe25020 CM |
1908 | /* make sure this device isn't detected as part of |
1909 | * the FS anymore | |
1910 | */ | |
1911 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1912 | set_buffer_dirty(bh); | |
1913 | sync_dirty_buffer(bh); | |
4d90d28b AJ |
1914 | |
1915 | /* clear the mirror copies of super block on the disk | |
1916 | * being removed, 0th copy is been taken care above and | |
1917 | * the below would take of the rest | |
1918 | */ | |
1919 | for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1920 | bytenr = btrfs_sb_offset(i); | |
1921 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
1922 | i_size_read(bdev->bd_inode)) | |
1923 | break; | |
1924 | ||
1925 | brelse(bh); | |
1926 | bh = __bread(bdev, bytenr / 4096, | |
1927 | BTRFS_SUPER_INFO_SIZE); | |
1928 | if (!bh) | |
1929 | continue; | |
1930 | ||
1931 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1932 | ||
1933 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
1934 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { | |
1935 | continue; | |
1936 | } | |
1937 | memset(&disk_super->magic, 0, | |
1938 | sizeof(disk_super->magic)); | |
1939 | set_buffer_dirty(bh); | |
1940 | sync_dirty_buffer(bh); | |
1941 | } | |
dfe25020 | 1942 | } |
a061fc8d | 1943 | |
a061fc8d | 1944 | ret = 0; |
a061fc8d | 1945 | |
5a1972bd QW |
1946 | if (bdev) { |
1947 | /* Notify udev that device has changed */ | |
3c911608 | 1948 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); |
b8b8ff59 | 1949 | |
5a1972bd QW |
1950 | /* Update ctime/mtime for device path for libblkid */ |
1951 | update_dev_time(device_path); | |
1952 | } | |
1953 | ||
a061fc8d CM |
1954 | error_brelse: |
1955 | brelse(bh); | |
dfe25020 | 1956 | if (bdev) |
e525fd89 | 1957 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1958 | out: |
1959 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1960 | return ret; |
9b3517e9 ID |
1961 | error_undo: |
1962 | if (device->writeable) { | |
0c1daee0 | 1963 | lock_chunks(root); |
9b3517e9 ID |
1964 | list_add(&device->dev_alloc_list, |
1965 | &root->fs_info->fs_devices->alloc_list); | |
c3929c36 | 1966 | device->fs_devices->rw_devices++; |
0c1daee0 | 1967 | unlock_chunks(root); |
9b3517e9 ID |
1968 | } |
1969 | goto error_brelse; | |
a061fc8d CM |
1970 | } |
1971 | ||
084b6e7c QW |
1972 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1973 | struct btrfs_device *srcdev) | |
e93c89c1 | 1974 | { |
d51908ce AJ |
1975 | struct btrfs_fs_devices *fs_devices; |
1976 | ||
e93c89c1 | 1977 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 1978 | |
25e8e911 AJ |
1979 | /* |
1980 | * in case of fs with no seed, srcdev->fs_devices will point | |
1981 | * to fs_devices of fs_info. However when the dev being replaced is | |
1982 | * a seed dev it will point to the seed's local fs_devices. In short | |
1983 | * srcdev will have its correct fs_devices in both the cases. | |
1984 | */ | |
1985 | fs_devices = srcdev->fs_devices; | |
d51908ce | 1986 | |
e93c89c1 SB |
1987 | list_del_rcu(&srcdev->dev_list); |
1988 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 1989 | fs_devices->num_devices--; |
82372bc8 | 1990 | if (srcdev->missing) |
d51908ce | 1991 | fs_devices->missing_devices--; |
e93c89c1 | 1992 | |
82372bc8 MX |
1993 | if (srcdev->writeable) { |
1994 | fs_devices->rw_devices--; | |
1995 | /* zero out the old super if it is writable */ | |
1996 | btrfs_scratch_superblock(srcdev); | |
1357272f ID |
1997 | } |
1998 | ||
82372bc8 | 1999 | if (srcdev->bdev) |
d51908ce | 2000 | fs_devices->open_devices--; |
084b6e7c QW |
2001 | } |
2002 | ||
2003 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2004 | struct btrfs_device *srcdev) | |
2005 | { | |
2006 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 SB |
2007 | |
2008 | call_rcu(&srcdev->rcu, free_device); | |
94d5f0c2 AJ |
2009 | |
2010 | /* | |
2011 | * unless fs_devices is seed fs, num_devices shouldn't go | |
2012 | * zero | |
2013 | */ | |
2014 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
2015 | ||
2016 | /* if this is no devs we rather delete the fs_devices */ | |
2017 | if (!fs_devices->num_devices) { | |
2018 | struct btrfs_fs_devices *tmp_fs_devices; | |
2019 | ||
2020 | tmp_fs_devices = fs_info->fs_devices; | |
2021 | while (tmp_fs_devices) { | |
2022 | if (tmp_fs_devices->seed == fs_devices) { | |
2023 | tmp_fs_devices->seed = fs_devices->seed; | |
2024 | break; | |
2025 | } | |
2026 | tmp_fs_devices = tmp_fs_devices->seed; | |
2027 | } | |
2028 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2029 | __btrfs_close_devices(fs_devices); |
2030 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2031 | } |
e93c89c1 SB |
2032 | } |
2033 | ||
2034 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2035 | struct btrfs_device *tgtdev) | |
2036 | { | |
2037 | struct btrfs_device *next_device; | |
2038 | ||
67a2c45e | 2039 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2040 | WARN_ON(!tgtdev); |
2041 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 AJ |
2042 | |
2043 | btrfs_kobj_rm_device(fs_info->fs_devices, tgtdev); | |
2044 | ||
e93c89c1 SB |
2045 | if (tgtdev->bdev) { |
2046 | btrfs_scratch_superblock(tgtdev); | |
2047 | fs_info->fs_devices->open_devices--; | |
2048 | } | |
2049 | fs_info->fs_devices->num_devices--; | |
e93c89c1 SB |
2050 | |
2051 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
2052 | struct btrfs_device, dev_list); | |
2053 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
2054 | fs_info->sb->s_bdev = next_device->bdev; | |
2055 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
2056 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2057 | list_del_rcu(&tgtdev->dev_list); | |
2058 | ||
2059 | call_rcu(&tgtdev->rcu, free_device); | |
2060 | ||
2061 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2062 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2063 | } |
2064 | ||
48a3b636 ES |
2065 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
2066 | struct btrfs_device **device) | |
7ba15b7d SB |
2067 | { |
2068 | int ret = 0; | |
2069 | struct btrfs_super_block *disk_super; | |
2070 | u64 devid; | |
2071 | u8 *dev_uuid; | |
2072 | struct block_device *bdev; | |
2073 | struct buffer_head *bh; | |
2074 | ||
2075 | *device = NULL; | |
2076 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
2077 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
2078 | if (ret) | |
2079 | return ret; | |
2080 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2081 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2082 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 2083 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
2084 | disk_super->fsid); |
2085 | brelse(bh); | |
2086 | if (!*device) | |
2087 | ret = -ENOENT; | |
2088 | blkdev_put(bdev, FMODE_READ); | |
2089 | return ret; | |
2090 | } | |
2091 | ||
2092 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
2093 | char *device_path, | |
2094 | struct btrfs_device **device) | |
2095 | { | |
2096 | *device = NULL; | |
2097 | if (strcmp(device_path, "missing") == 0) { | |
2098 | struct list_head *devices; | |
2099 | struct btrfs_device *tmp; | |
2100 | ||
2101 | devices = &root->fs_info->fs_devices->devices; | |
2102 | /* | |
2103 | * It is safe to read the devices since the volume_mutex | |
2104 | * is held by the caller. | |
2105 | */ | |
2106 | list_for_each_entry(tmp, devices, dev_list) { | |
2107 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2108 | *device = tmp; | |
2109 | break; | |
2110 | } | |
2111 | } | |
2112 | ||
2113 | if (!*device) { | |
efe120a0 | 2114 | btrfs_err(root->fs_info, "no missing device found"); |
7ba15b7d SB |
2115 | return -ENOENT; |
2116 | } | |
2117 | ||
2118 | return 0; | |
2119 | } else { | |
2120 | return btrfs_find_device_by_path(root, device_path, device); | |
2121 | } | |
2122 | } | |
2123 | ||
2b82032c YZ |
2124 | /* |
2125 | * does all the dirty work required for changing file system's UUID. | |
2126 | */ | |
125ccb0a | 2127 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
2128 | { |
2129 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
2130 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 2131 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 2132 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
2133 | struct btrfs_device *device; |
2134 | u64 super_flags; | |
2135 | ||
2136 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2137 | if (!fs_devices->seeding) |
2b82032c YZ |
2138 | return -EINVAL; |
2139 | ||
2208a378 ID |
2140 | seed_devices = __alloc_fs_devices(); |
2141 | if (IS_ERR(seed_devices)) | |
2142 | return PTR_ERR(seed_devices); | |
2b82032c | 2143 | |
e4404d6e YZ |
2144 | old_devices = clone_fs_devices(fs_devices); |
2145 | if (IS_ERR(old_devices)) { | |
2146 | kfree(seed_devices); | |
2147 | return PTR_ERR(old_devices); | |
2b82032c | 2148 | } |
e4404d6e | 2149 | |
2b82032c YZ |
2150 | list_add(&old_devices->list, &fs_uuids); |
2151 | ||
e4404d6e YZ |
2152 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2153 | seed_devices->opened = 1; | |
2154 | INIT_LIST_HEAD(&seed_devices->devices); | |
2155 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2156 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
2157 | |
2158 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
2159 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2160 | synchronize_rcu); | |
2196d6e8 MX |
2161 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2162 | device->fs_devices = seed_devices; | |
c9513edb | 2163 | |
2196d6e8 | 2164 | lock_chunks(root); |
e4404d6e | 2165 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
2196d6e8 | 2166 | unlock_chunks(root); |
e4404d6e | 2167 | |
2b82032c YZ |
2168 | fs_devices->seeding = 0; |
2169 | fs_devices->num_devices = 0; | |
2170 | fs_devices->open_devices = 0; | |
69611ac8 | 2171 | fs_devices->missing_devices = 0; |
69611ac8 | 2172 | fs_devices->rotating = 0; |
e4404d6e | 2173 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2174 | |
2175 | generate_random_uuid(fs_devices->fsid); | |
2176 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2177 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
f7171750 FDBM |
2178 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2179 | ||
2b82032c YZ |
2180 | super_flags = btrfs_super_flags(disk_super) & |
2181 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2182 | btrfs_set_super_flags(disk_super, super_flags); | |
2183 | ||
2184 | return 0; | |
2185 | } | |
2186 | ||
2187 | /* | |
2188 | * strore the expected generation for seed devices in device items. | |
2189 | */ | |
2190 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2191 | struct btrfs_root *root) | |
2192 | { | |
2193 | struct btrfs_path *path; | |
2194 | struct extent_buffer *leaf; | |
2195 | struct btrfs_dev_item *dev_item; | |
2196 | struct btrfs_device *device; | |
2197 | struct btrfs_key key; | |
2198 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2199 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2200 | u64 devid; | |
2201 | int ret; | |
2202 | ||
2203 | path = btrfs_alloc_path(); | |
2204 | if (!path) | |
2205 | return -ENOMEM; | |
2206 | ||
2207 | root = root->fs_info->chunk_root; | |
2208 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2209 | key.offset = 0; | |
2210 | key.type = BTRFS_DEV_ITEM_KEY; | |
2211 | ||
2212 | while (1) { | |
2213 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2214 | if (ret < 0) | |
2215 | goto error; | |
2216 | ||
2217 | leaf = path->nodes[0]; | |
2218 | next_slot: | |
2219 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2220 | ret = btrfs_next_leaf(root, path); | |
2221 | if (ret > 0) | |
2222 | break; | |
2223 | if (ret < 0) | |
2224 | goto error; | |
2225 | leaf = path->nodes[0]; | |
2226 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2227 | btrfs_release_path(path); |
2b82032c YZ |
2228 | continue; |
2229 | } | |
2230 | ||
2231 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2232 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2233 | key.type != BTRFS_DEV_ITEM_KEY) | |
2234 | break; | |
2235 | ||
2236 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2237 | struct btrfs_dev_item); | |
2238 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2239 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2240 | BTRFS_UUID_SIZE); |
1473b24e | 2241 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2242 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
2243 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2244 | fs_uuid); | |
79787eaa | 2245 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2246 | |
2247 | if (device->fs_devices->seeding) { | |
2248 | btrfs_set_device_generation(leaf, dev_item, | |
2249 | device->generation); | |
2250 | btrfs_mark_buffer_dirty(leaf); | |
2251 | } | |
2252 | ||
2253 | path->slots[0]++; | |
2254 | goto next_slot; | |
2255 | } | |
2256 | ret = 0; | |
2257 | error: | |
2258 | btrfs_free_path(path); | |
2259 | return ret; | |
2260 | } | |
2261 | ||
788f20eb CM |
2262 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
2263 | { | |
d5e2003c | 2264 | struct request_queue *q; |
788f20eb CM |
2265 | struct btrfs_trans_handle *trans; |
2266 | struct btrfs_device *device; | |
2267 | struct block_device *bdev; | |
788f20eb | 2268 | struct list_head *devices; |
2b82032c | 2269 | struct super_block *sb = root->fs_info->sb; |
606686ee | 2270 | struct rcu_string *name; |
3c1dbdf5 | 2271 | u64 tmp; |
2b82032c | 2272 | int seeding_dev = 0; |
788f20eb CM |
2273 | int ret = 0; |
2274 | ||
2b82032c | 2275 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 2276 | return -EROFS; |
788f20eb | 2277 | |
a5d16333 | 2278 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 2279 | root->fs_info->bdev_holder); |
7f59203a JB |
2280 | if (IS_ERR(bdev)) |
2281 | return PTR_ERR(bdev); | |
a2135011 | 2282 | |
2b82032c YZ |
2283 | if (root->fs_info->fs_devices->seeding) { |
2284 | seeding_dev = 1; | |
2285 | down_write(&sb->s_umount); | |
2286 | mutex_lock(&uuid_mutex); | |
2287 | } | |
2288 | ||
8c8bee1d | 2289 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2290 | |
788f20eb | 2291 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
2292 | |
2293 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 2294 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2295 | if (device->bdev == bdev) { |
2296 | ret = -EEXIST; | |
d25628bd LB |
2297 | mutex_unlock( |
2298 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2299 | goto error; |
788f20eb CM |
2300 | } |
2301 | } | |
d25628bd | 2302 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2303 | |
12bd2fc0 ID |
2304 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
2305 | if (IS_ERR(device)) { | |
788f20eb | 2306 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2307 | ret = PTR_ERR(device); |
2b82032c | 2308 | goto error; |
788f20eb CM |
2309 | } |
2310 | ||
606686ee JB |
2311 | name = rcu_string_strdup(device_path, GFP_NOFS); |
2312 | if (!name) { | |
788f20eb | 2313 | kfree(device); |
2b82032c YZ |
2314 | ret = -ENOMEM; |
2315 | goto error; | |
788f20eb | 2316 | } |
606686ee | 2317 | rcu_assign_pointer(device->name, name); |
2b82032c | 2318 | |
a22285a6 | 2319 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2320 | if (IS_ERR(trans)) { |
606686ee | 2321 | rcu_string_free(device->name); |
98d5dc13 TI |
2322 | kfree(device); |
2323 | ret = PTR_ERR(trans); | |
2324 | goto error; | |
2325 | } | |
2326 | ||
d5e2003c JB |
2327 | q = bdev_get_queue(bdev); |
2328 | if (blk_queue_discard(q)) | |
2329 | device->can_discard = 1; | |
2b82032c | 2330 | device->writeable = 1; |
2b82032c | 2331 | device->generation = trans->transid; |
788f20eb CM |
2332 | device->io_width = root->sectorsize; |
2333 | device->io_align = root->sectorsize; | |
2334 | device->sector_size = root->sectorsize; | |
2335 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2336 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2337 | device->commit_total_bytes = device->total_bytes; |
788f20eb CM |
2338 | device->dev_root = root->fs_info->dev_root; |
2339 | device->bdev = bdev; | |
dfe25020 | 2340 | device->in_fs_metadata = 1; |
63a212ab | 2341 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2342 | device->mode = FMODE_EXCL; |
27087f37 | 2343 | device->dev_stats_valid = 1; |
2b82032c | 2344 | set_blocksize(device->bdev, 4096); |
788f20eb | 2345 | |
2b82032c YZ |
2346 | if (seeding_dev) { |
2347 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2348 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2349 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2350 | } |
788f20eb | 2351 | |
2b82032c | 2352 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2353 | |
e5e9a520 | 2354 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
2196d6e8 | 2355 | lock_chunks(root); |
1f78160c | 2356 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2357 | list_add(&device->dev_alloc_list, |
2358 | &root->fs_info->fs_devices->alloc_list); | |
2359 | root->fs_info->fs_devices->num_devices++; | |
2360 | root->fs_info->fs_devices->open_devices++; | |
2361 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2362 | root->fs_info->fs_devices->total_devices++; |
2b82032c | 2363 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2364 | |
2bf64758 JB |
2365 | spin_lock(&root->fs_info->free_chunk_lock); |
2366 | root->fs_info->free_chunk_space += device->total_bytes; | |
2367 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2368 | ||
c289811c CM |
2369 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2370 | root->fs_info->fs_devices->rotating = 1; | |
2371 | ||
3c1dbdf5 | 2372 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); |
6c41761f | 2373 | btrfs_set_super_total_bytes(root->fs_info->super_copy, |
3c1dbdf5 | 2374 | tmp + device->total_bytes); |
788f20eb | 2375 | |
3c1dbdf5 | 2376 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); |
6c41761f | 2377 | btrfs_set_super_num_devices(root->fs_info->super_copy, |
3c1dbdf5 | 2378 | tmp + 1); |
0d39376a AJ |
2379 | |
2380 | /* add sysfs device entry */ | |
1ba43816 | 2381 | btrfs_kobj_add_device(root->fs_info->fs_devices, device); |
0d39376a | 2382 | |
2196d6e8 MX |
2383 | /* |
2384 | * we've got more storage, clear any full flags on the space | |
2385 | * infos | |
2386 | */ | |
2387 | btrfs_clear_space_info_full(root->fs_info); | |
2388 | ||
2389 | unlock_chunks(root); | |
e5e9a520 | 2390 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2391 | |
2b82032c | 2392 | if (seeding_dev) { |
2196d6e8 | 2393 | lock_chunks(root); |
2b82032c | 2394 | ret = init_first_rw_device(trans, root, device); |
2196d6e8 | 2395 | unlock_chunks(root); |
005d6427 DS |
2396 | if (ret) { |
2397 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2398 | goto error_trans; |
005d6427 | 2399 | } |
2196d6e8 MX |
2400 | } |
2401 | ||
2402 | ret = btrfs_add_device(trans, root, device); | |
2403 | if (ret) { | |
2404 | btrfs_abort_transaction(trans, root, ret); | |
2405 | goto error_trans; | |
2406 | } | |
2407 | ||
2408 | if (seeding_dev) { | |
2409 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2410 | ||
2b82032c | 2411 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2412 | if (ret) { |
2413 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2414 | goto error_trans; |
005d6427 | 2415 | } |
b2373f25 AJ |
2416 | |
2417 | /* Sprouting would change fsid of the mounted root, | |
2418 | * so rename the fsid on the sysfs | |
2419 | */ | |
2420 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2421 | root->fs_info->fsid); | |
2e7910d6 AJ |
2422 | if (kobject_rename(&root->fs_info->fs_devices->super_kobj, |
2423 | fsid_buf)) | |
2421a8cd | 2424 | pr_warn("BTRFS: sysfs: failed to create fsid for sprout\n"); |
2b82032c YZ |
2425 | } |
2426 | ||
5af3e8cc SB |
2427 | root->fs_info->num_tolerated_disk_barrier_failures = |
2428 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2429 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2430 | |
2b82032c YZ |
2431 | if (seeding_dev) { |
2432 | mutex_unlock(&uuid_mutex); | |
2433 | up_write(&sb->s_umount); | |
788f20eb | 2434 | |
79787eaa JM |
2435 | if (ret) /* transaction commit */ |
2436 | return ret; | |
2437 | ||
2b82032c | 2438 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
2439 | if (ret < 0) |
2440 | btrfs_error(root->fs_info, ret, | |
2441 | "Failed to relocate sys chunks after " | |
2442 | "device initialization. This can be fixed " | |
2443 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2444 | trans = btrfs_attach_transaction(root); |
2445 | if (IS_ERR(trans)) { | |
2446 | if (PTR_ERR(trans) == -ENOENT) | |
2447 | return 0; | |
2448 | return PTR_ERR(trans); | |
2449 | } | |
2450 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2451 | } |
c9e9f97b | 2452 | |
5a1972bd QW |
2453 | /* Update ctime/mtime for libblkid */ |
2454 | update_dev_time(device_path); | |
2b82032c | 2455 | return ret; |
79787eaa JM |
2456 | |
2457 | error_trans: | |
79787eaa | 2458 | btrfs_end_transaction(trans, root); |
606686ee | 2459 | rcu_string_free(device->name); |
6c14a164 | 2460 | btrfs_kobj_rm_device(root->fs_info->fs_devices, device); |
79787eaa | 2461 | kfree(device); |
2b82032c | 2462 | error: |
e525fd89 | 2463 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2464 | if (seeding_dev) { |
2465 | mutex_unlock(&uuid_mutex); | |
2466 | up_write(&sb->s_umount); | |
2467 | } | |
c9e9f97b | 2468 | return ret; |
788f20eb CM |
2469 | } |
2470 | ||
e93c89c1 | 2471 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2472 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2473 | struct btrfs_device **device_out) |
2474 | { | |
2475 | struct request_queue *q; | |
2476 | struct btrfs_device *device; | |
2477 | struct block_device *bdev; | |
2478 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2479 | struct list_head *devices; | |
2480 | struct rcu_string *name; | |
12bd2fc0 | 2481 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2482 | int ret = 0; |
2483 | ||
2484 | *device_out = NULL; | |
1c43366d MX |
2485 | if (fs_info->fs_devices->seeding) { |
2486 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2487 | return -EINVAL; |
1c43366d | 2488 | } |
e93c89c1 SB |
2489 | |
2490 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2491 | fs_info->bdev_holder); | |
1c43366d MX |
2492 | if (IS_ERR(bdev)) { |
2493 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2494 | return PTR_ERR(bdev); |
1c43366d | 2495 | } |
e93c89c1 SB |
2496 | |
2497 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2498 | ||
2499 | devices = &fs_info->fs_devices->devices; | |
2500 | list_for_each_entry(device, devices, dev_list) { | |
2501 | if (device->bdev == bdev) { | |
1c43366d | 2502 | btrfs_err(fs_info, "target device is in the filesystem!"); |
e93c89c1 SB |
2503 | ret = -EEXIST; |
2504 | goto error; | |
2505 | } | |
2506 | } | |
2507 | ||
1c43366d | 2508 | |
7cc8e58d MX |
2509 | if (i_size_read(bdev->bd_inode) < |
2510 | btrfs_device_get_total_bytes(srcdev)) { | |
1c43366d MX |
2511 | btrfs_err(fs_info, "target device is smaller than source device!"); |
2512 | ret = -EINVAL; | |
2513 | goto error; | |
2514 | } | |
2515 | ||
2516 | ||
12bd2fc0 ID |
2517 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2518 | if (IS_ERR(device)) { | |
2519 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2520 | goto error; |
2521 | } | |
2522 | ||
2523 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2524 | if (!name) { | |
2525 | kfree(device); | |
2526 | ret = -ENOMEM; | |
2527 | goto error; | |
2528 | } | |
2529 | rcu_assign_pointer(device->name, name); | |
2530 | ||
2531 | q = bdev_get_queue(bdev); | |
2532 | if (blk_queue_discard(q)) | |
2533 | device->can_discard = 1; | |
2534 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2535 | device->writeable = 1; | |
e93c89c1 SB |
2536 | device->generation = 0; |
2537 | device->io_width = root->sectorsize; | |
2538 | device->io_align = root->sectorsize; | |
2539 | device->sector_size = root->sectorsize; | |
7cc8e58d MX |
2540 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2541 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2542 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2543 | ASSERT(list_empty(&srcdev->resized_list)); |
2544 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2545 | device->commit_bytes_used = device->bytes_used; |
e93c89c1 SB |
2546 | device->dev_root = fs_info->dev_root; |
2547 | device->bdev = bdev; | |
2548 | device->in_fs_metadata = 1; | |
2549 | device->is_tgtdev_for_dev_replace = 1; | |
2550 | device->mode = FMODE_EXCL; | |
27087f37 | 2551 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2552 | set_blocksize(device->bdev, 4096); |
2553 | device->fs_devices = fs_info->fs_devices; | |
2554 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2555 | fs_info->fs_devices->num_devices++; | |
2556 | fs_info->fs_devices->open_devices++; | |
e93c89c1 SB |
2557 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2558 | ||
2559 | *device_out = device; | |
2560 | return ret; | |
2561 | ||
2562 | error: | |
2563 | blkdev_put(bdev, FMODE_EXCL); | |
2564 | return ret; | |
2565 | } | |
2566 | ||
2567 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2568 | struct btrfs_device *tgtdev) | |
2569 | { | |
2570 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2571 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2572 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2573 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2574 | tgtdev->dev_root = fs_info->dev_root; | |
2575 | tgtdev->in_fs_metadata = 1; | |
2576 | } | |
2577 | ||
d397712b CM |
2578 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2579 | struct btrfs_device *device) | |
0b86a832 CM |
2580 | { |
2581 | int ret; | |
2582 | struct btrfs_path *path; | |
2583 | struct btrfs_root *root; | |
2584 | struct btrfs_dev_item *dev_item; | |
2585 | struct extent_buffer *leaf; | |
2586 | struct btrfs_key key; | |
2587 | ||
2588 | root = device->dev_root->fs_info->chunk_root; | |
2589 | ||
2590 | path = btrfs_alloc_path(); | |
2591 | if (!path) | |
2592 | return -ENOMEM; | |
2593 | ||
2594 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2595 | key.type = BTRFS_DEV_ITEM_KEY; | |
2596 | key.offset = device->devid; | |
2597 | ||
2598 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2599 | if (ret < 0) | |
2600 | goto out; | |
2601 | ||
2602 | if (ret > 0) { | |
2603 | ret = -ENOENT; | |
2604 | goto out; | |
2605 | } | |
2606 | ||
2607 | leaf = path->nodes[0]; | |
2608 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2609 | ||
2610 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2611 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2612 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2613 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2614 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2615 | btrfs_set_device_total_bytes(leaf, dev_item, |
2616 | btrfs_device_get_disk_total_bytes(device)); | |
2617 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2618 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2619 | btrfs_mark_buffer_dirty(leaf); |
2620 | ||
2621 | out: | |
2622 | btrfs_free_path(path); | |
2623 | return ret; | |
2624 | } | |
2625 | ||
2196d6e8 | 2626 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2627 | struct btrfs_device *device, u64 new_size) |
2628 | { | |
2629 | struct btrfs_super_block *super_copy = | |
6c41761f | 2630 | device->dev_root->fs_info->super_copy; |
935e5cc9 | 2631 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2632 | u64 old_total; |
2633 | u64 diff; | |
8f18cf13 | 2634 | |
2b82032c YZ |
2635 | if (!device->writeable) |
2636 | return -EACCES; | |
2196d6e8 MX |
2637 | |
2638 | lock_chunks(device->dev_root); | |
2639 | old_total = btrfs_super_total_bytes(super_copy); | |
2640 | diff = new_size - device->total_bytes; | |
2641 | ||
63a212ab | 2642 | if (new_size <= device->total_bytes || |
2196d6e8 MX |
2643 | device->is_tgtdev_for_dev_replace) { |
2644 | unlock_chunks(device->dev_root); | |
2b82032c | 2645 | return -EINVAL; |
2196d6e8 | 2646 | } |
2b82032c | 2647 | |
935e5cc9 | 2648 | fs_devices = device->dev_root->fs_info->fs_devices; |
2b82032c | 2649 | |
8f18cf13 | 2650 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2651 | device->fs_devices->total_rw_bytes += diff; |
2652 | ||
7cc8e58d MX |
2653 | btrfs_device_set_total_bytes(device, new_size); |
2654 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4184ea7f | 2655 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
935e5cc9 MX |
2656 | if (list_empty(&device->resized_list)) |
2657 | list_add_tail(&device->resized_list, | |
2658 | &fs_devices->resized_devices); | |
2196d6e8 | 2659 | unlock_chunks(device->dev_root); |
4184ea7f | 2660 | |
8f18cf13 CM |
2661 | return btrfs_update_device(trans, device); |
2662 | } | |
2663 | ||
2664 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
a688a04a | 2665 | struct btrfs_root *root, u64 chunk_objectid, |
8f18cf13 CM |
2666 | u64 chunk_offset) |
2667 | { | |
2668 | int ret; | |
2669 | struct btrfs_path *path; | |
2670 | struct btrfs_key key; | |
2671 | ||
2672 | root = root->fs_info->chunk_root; | |
2673 | path = btrfs_alloc_path(); | |
2674 | if (!path) | |
2675 | return -ENOMEM; | |
2676 | ||
2677 | key.objectid = chunk_objectid; | |
2678 | key.offset = chunk_offset; | |
2679 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2680 | ||
2681 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2682 | if (ret < 0) |
2683 | goto out; | |
2684 | else if (ret > 0) { /* Logic error or corruption */ | |
2685 | btrfs_error(root->fs_info, -ENOENT, | |
2686 | "Failed lookup while freeing chunk."); | |
2687 | ret = -ENOENT; | |
2688 | goto out; | |
2689 | } | |
8f18cf13 CM |
2690 | |
2691 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
2692 | if (ret < 0) |
2693 | btrfs_error(root->fs_info, ret, | |
2694 | "Failed to delete chunk item."); | |
2695 | out: | |
8f18cf13 | 2696 | btrfs_free_path(path); |
65a246c5 | 2697 | return ret; |
8f18cf13 CM |
2698 | } |
2699 | ||
b2950863 | 2700 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2701 | chunk_offset) |
2702 | { | |
6c41761f | 2703 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2704 | struct btrfs_disk_key *disk_key; |
2705 | struct btrfs_chunk *chunk; | |
2706 | u8 *ptr; | |
2707 | int ret = 0; | |
2708 | u32 num_stripes; | |
2709 | u32 array_size; | |
2710 | u32 len = 0; | |
2711 | u32 cur; | |
2712 | struct btrfs_key key; | |
2713 | ||
2196d6e8 | 2714 | lock_chunks(root); |
8f18cf13 CM |
2715 | array_size = btrfs_super_sys_array_size(super_copy); |
2716 | ||
2717 | ptr = super_copy->sys_chunk_array; | |
2718 | cur = 0; | |
2719 | ||
2720 | while (cur < array_size) { | |
2721 | disk_key = (struct btrfs_disk_key *)ptr; | |
2722 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2723 | ||
2724 | len = sizeof(*disk_key); | |
2725 | ||
2726 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2727 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2728 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2729 | len += btrfs_chunk_item_size(num_stripes); | |
2730 | } else { | |
2731 | ret = -EIO; | |
2732 | break; | |
2733 | } | |
2734 | if (key.objectid == chunk_objectid && | |
2735 | key.offset == chunk_offset) { | |
2736 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2737 | array_size -= len; | |
2738 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2739 | } else { | |
2740 | ptr += len; | |
2741 | cur += len; | |
2742 | } | |
2743 | } | |
2196d6e8 | 2744 | unlock_chunks(root); |
8f18cf13 CM |
2745 | return ret; |
2746 | } | |
2747 | ||
47ab2a6c JB |
2748 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
2749 | struct btrfs_root *root, u64 chunk_offset) | |
8f18cf13 CM |
2750 | { |
2751 | struct extent_map_tree *em_tree; | |
8f18cf13 | 2752 | struct extent_map *em; |
47ab2a6c | 2753 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
8f18cf13 | 2754 | struct map_lookup *map; |
2196d6e8 | 2755 | u64 dev_extent_len = 0; |
47ab2a6c | 2756 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2757 | int i, ret = 0; |
8f18cf13 | 2758 | |
47ab2a6c | 2759 | /* Just in case */ |
8f18cf13 | 2760 | root = root->fs_info->chunk_root; |
8f18cf13 CM |
2761 | em_tree = &root->fs_info->mapping_tree.map_tree; |
2762 | ||
890871be | 2763 | read_lock(&em_tree->lock); |
8f18cf13 | 2764 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2765 | read_unlock(&em_tree->lock); |
8f18cf13 | 2766 | |
47ab2a6c JB |
2767 | if (!em || em->start > chunk_offset || |
2768 | em->start + em->len < chunk_offset) { | |
2769 | /* | |
2770 | * This is a logic error, but we don't want to just rely on the | |
2771 | * user having built with ASSERT enabled, so if ASSERT doens't | |
2772 | * do anything we still error out. | |
2773 | */ | |
2774 | ASSERT(0); | |
2775 | if (em) | |
2776 | free_extent_map(em); | |
2777 | return -EINVAL; | |
2778 | } | |
8f18cf13 | 2779 | map = (struct map_lookup *)em->bdev; |
39c2d7fa | 2780 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 2781 | check_system_chunk(trans, extent_root, map->type); |
39c2d7fa | 2782 | unlock_chunks(root->fs_info->chunk_root); |
8f18cf13 CM |
2783 | |
2784 | for (i = 0; i < map->num_stripes; i++) { | |
47ab2a6c | 2785 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2786 | ret = btrfs_free_dev_extent(trans, device, |
2787 | map->stripes[i].physical, | |
2788 | &dev_extent_len); | |
47ab2a6c JB |
2789 | if (ret) { |
2790 | btrfs_abort_transaction(trans, root, ret); | |
2791 | goto out; | |
2792 | } | |
a061fc8d | 2793 | |
2196d6e8 MX |
2794 | if (device->bytes_used > 0) { |
2795 | lock_chunks(root); | |
2796 | btrfs_device_set_bytes_used(device, | |
2797 | device->bytes_used - dev_extent_len); | |
2798 | spin_lock(&root->fs_info->free_chunk_lock); | |
2799 | root->fs_info->free_chunk_space += dev_extent_len; | |
2800 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2801 | btrfs_clear_space_info_full(root->fs_info); | |
2802 | unlock_chunks(root); | |
2803 | } | |
a061fc8d | 2804 | |
dfe25020 CM |
2805 | if (map->stripes[i].dev) { |
2806 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c JB |
2807 | if (ret) { |
2808 | btrfs_abort_transaction(trans, root, ret); | |
2809 | goto out; | |
2810 | } | |
dfe25020 | 2811 | } |
8f18cf13 | 2812 | } |
a688a04a | 2813 | ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset); |
47ab2a6c JB |
2814 | if (ret) { |
2815 | btrfs_abort_transaction(trans, root, ret); | |
2816 | goto out; | |
2817 | } | |
8f18cf13 | 2818 | |
1abe9b8a | 2819 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2820 | ||
8f18cf13 CM |
2821 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2822 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
47ab2a6c JB |
2823 | if (ret) { |
2824 | btrfs_abort_transaction(trans, root, ret); | |
2825 | goto out; | |
2826 | } | |
8f18cf13 CM |
2827 | } |
2828 | ||
04216820 | 2829 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em); |
47ab2a6c JB |
2830 | if (ret) { |
2831 | btrfs_abort_transaction(trans, extent_root, ret); | |
2832 | goto out; | |
2833 | } | |
2b82032c | 2834 | |
47ab2a6c | 2835 | out: |
2b82032c YZ |
2836 | /* once for us */ |
2837 | free_extent_map(em); | |
47ab2a6c JB |
2838 | return ret; |
2839 | } | |
2b82032c | 2840 | |
dc2ee4e2 | 2841 | static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset) |
47ab2a6c JB |
2842 | { |
2843 | struct btrfs_root *extent_root; | |
2844 | struct btrfs_trans_handle *trans; | |
2845 | int ret; | |
2b82032c | 2846 | |
47ab2a6c JB |
2847 | root = root->fs_info->chunk_root; |
2848 | extent_root = root->fs_info->extent_root; | |
2849 | ||
67c5e7d4 FM |
2850 | /* |
2851 | * Prevent races with automatic removal of unused block groups. | |
2852 | * After we relocate and before we remove the chunk with offset | |
2853 | * chunk_offset, automatic removal of the block group can kick in, | |
2854 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2855 | * | |
2856 | * Make sure to acquire this mutex before doing a tree search (dev | |
2857 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2858 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2859 | * we release the path used to search the chunk/dev tree and before | |
2860 | * the current task acquires this mutex and calls us. | |
2861 | */ | |
2862 | ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex)); | |
2863 | ||
47ab2a6c JB |
2864 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2865 | if (ret) | |
2866 | return -ENOSPC; | |
2867 | ||
2868 | /* step one, relocate all the extents inside this chunk */ | |
55e3a601 | 2869 | btrfs_scrub_pause(root); |
47ab2a6c | 2870 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
55e3a601 | 2871 | btrfs_scrub_continue(root); |
47ab2a6c JB |
2872 | if (ret) |
2873 | return ret; | |
2874 | ||
2875 | trans = btrfs_start_transaction(root, 0); | |
2876 | if (IS_ERR(trans)) { | |
2877 | ret = PTR_ERR(trans); | |
2878 | btrfs_std_error(root->fs_info, ret); | |
2879 | return ret; | |
2880 | } | |
2881 | ||
2882 | /* | |
2883 | * step two, delete the device extents and the | |
2884 | * chunk tree entries | |
2885 | */ | |
2886 | ret = btrfs_remove_chunk(trans, root, chunk_offset); | |
2b82032c | 2887 | btrfs_end_transaction(trans, root); |
47ab2a6c | 2888 | return ret; |
2b82032c YZ |
2889 | } |
2890 | ||
2891 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2892 | { | |
2893 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2894 | struct btrfs_path *path; | |
2895 | struct extent_buffer *leaf; | |
2896 | struct btrfs_chunk *chunk; | |
2897 | struct btrfs_key key; | |
2898 | struct btrfs_key found_key; | |
2b82032c | 2899 | u64 chunk_type; |
ba1bf481 JB |
2900 | bool retried = false; |
2901 | int failed = 0; | |
2b82032c YZ |
2902 | int ret; |
2903 | ||
2904 | path = btrfs_alloc_path(); | |
2905 | if (!path) | |
2906 | return -ENOMEM; | |
2907 | ||
ba1bf481 | 2908 | again: |
2b82032c YZ |
2909 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2910 | key.offset = (u64)-1; | |
2911 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2912 | ||
2913 | while (1) { | |
67c5e7d4 | 2914 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
2b82032c | 2915 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
2916 | if (ret < 0) { |
2917 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c | 2918 | goto error; |
67c5e7d4 | 2919 | } |
79787eaa | 2920 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2921 | |
2922 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2923 | key.type); | |
67c5e7d4 FM |
2924 | if (ret) |
2925 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2b82032c YZ |
2926 | if (ret < 0) |
2927 | goto error; | |
2928 | if (ret > 0) | |
2929 | break; | |
1a40e23b | 2930 | |
2b82032c YZ |
2931 | leaf = path->nodes[0]; |
2932 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2933 | |
2b82032c YZ |
2934 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2935 | struct btrfs_chunk); | |
2936 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2937 | btrfs_release_path(path); |
8f18cf13 | 2938 | |
2b82032c | 2939 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
a688a04a | 2940 | ret = btrfs_relocate_chunk(chunk_root, |
2b82032c | 2941 | found_key.offset); |
ba1bf481 JB |
2942 | if (ret == -ENOSPC) |
2943 | failed++; | |
14586651 HS |
2944 | else |
2945 | BUG_ON(ret); | |
2b82032c | 2946 | } |
67c5e7d4 | 2947 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 2948 | |
2b82032c YZ |
2949 | if (found_key.offset == 0) |
2950 | break; | |
2951 | key.offset = found_key.offset - 1; | |
2952 | } | |
2953 | ret = 0; | |
ba1bf481 JB |
2954 | if (failed && !retried) { |
2955 | failed = 0; | |
2956 | retried = true; | |
2957 | goto again; | |
fae7f21c | 2958 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2959 | ret = -ENOSPC; |
2960 | } | |
2b82032c YZ |
2961 | error: |
2962 | btrfs_free_path(path); | |
2963 | return ret; | |
8f18cf13 CM |
2964 | } |
2965 | ||
0940ebf6 ID |
2966 | static int insert_balance_item(struct btrfs_root *root, |
2967 | struct btrfs_balance_control *bctl) | |
2968 | { | |
2969 | struct btrfs_trans_handle *trans; | |
2970 | struct btrfs_balance_item *item; | |
2971 | struct btrfs_disk_balance_args disk_bargs; | |
2972 | struct btrfs_path *path; | |
2973 | struct extent_buffer *leaf; | |
2974 | struct btrfs_key key; | |
2975 | int ret, err; | |
2976 | ||
2977 | path = btrfs_alloc_path(); | |
2978 | if (!path) | |
2979 | return -ENOMEM; | |
2980 | ||
2981 | trans = btrfs_start_transaction(root, 0); | |
2982 | if (IS_ERR(trans)) { | |
2983 | btrfs_free_path(path); | |
2984 | return PTR_ERR(trans); | |
2985 | } | |
2986 | ||
2987 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2988 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2989 | key.offset = 0; | |
2990 | ||
2991 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2992 | sizeof(*item)); | |
2993 | if (ret) | |
2994 | goto out; | |
2995 | ||
2996 | leaf = path->nodes[0]; | |
2997 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2998 | ||
2999 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
3000 | ||
3001 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3002 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3003 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3004 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3005 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3006 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3007 | ||
3008 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3009 | ||
3010 | btrfs_mark_buffer_dirty(leaf); | |
3011 | out: | |
3012 | btrfs_free_path(path); | |
3013 | err = btrfs_commit_transaction(trans, root); | |
3014 | if (err && !ret) | |
3015 | ret = err; | |
3016 | return ret; | |
3017 | } | |
3018 | ||
3019 | static int del_balance_item(struct btrfs_root *root) | |
3020 | { | |
3021 | struct btrfs_trans_handle *trans; | |
3022 | struct btrfs_path *path; | |
3023 | struct btrfs_key key; | |
3024 | int ret, err; | |
3025 | ||
3026 | path = btrfs_alloc_path(); | |
3027 | if (!path) | |
3028 | return -ENOMEM; | |
3029 | ||
3030 | trans = btrfs_start_transaction(root, 0); | |
3031 | if (IS_ERR(trans)) { | |
3032 | btrfs_free_path(path); | |
3033 | return PTR_ERR(trans); | |
3034 | } | |
3035 | ||
3036 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
3037 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3038 | key.offset = 0; | |
3039 | ||
3040 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3041 | if (ret < 0) | |
3042 | goto out; | |
3043 | if (ret > 0) { | |
3044 | ret = -ENOENT; | |
3045 | goto out; | |
3046 | } | |
3047 | ||
3048 | ret = btrfs_del_item(trans, root, path); | |
3049 | out: | |
3050 | btrfs_free_path(path); | |
3051 | err = btrfs_commit_transaction(trans, root); | |
3052 | if (err && !ret) | |
3053 | ret = err; | |
3054 | return ret; | |
3055 | } | |
3056 | ||
59641015 ID |
3057 | /* |
3058 | * This is a heuristic used to reduce the number of chunks balanced on | |
3059 | * resume after balance was interrupted. | |
3060 | */ | |
3061 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3062 | { | |
3063 | /* | |
3064 | * Turn on soft mode for chunk types that were being converted. | |
3065 | */ | |
3066 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3067 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3068 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3069 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3070 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3071 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3072 | ||
3073 | /* | |
3074 | * Turn on usage filter if is not already used. The idea is | |
3075 | * that chunks that we have already balanced should be | |
3076 | * reasonably full. Don't do it for chunks that are being | |
3077 | * converted - that will keep us from relocating unconverted | |
3078 | * (albeit full) chunks. | |
3079 | */ | |
3080 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3081 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
3082 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3083 | bctl->data.usage = 90; | |
3084 | } | |
3085 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3086 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
3087 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3088 | bctl->sys.usage = 90; | |
3089 | } | |
3090 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3091 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
3092 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3093 | bctl->meta.usage = 90; | |
3094 | } | |
3095 | } | |
3096 | ||
c9e9f97b ID |
3097 | /* |
3098 | * Should be called with both balance and volume mutexes held to | |
3099 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3100 | * restriper. Same goes for unset_balance_control. | |
3101 | */ | |
3102 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3103 | { | |
3104 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3105 | ||
3106 | BUG_ON(fs_info->balance_ctl); | |
3107 | ||
3108 | spin_lock(&fs_info->balance_lock); | |
3109 | fs_info->balance_ctl = bctl; | |
3110 | spin_unlock(&fs_info->balance_lock); | |
3111 | } | |
3112 | ||
3113 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3114 | { | |
3115 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3116 | ||
3117 | BUG_ON(!fs_info->balance_ctl); | |
3118 | ||
3119 | spin_lock(&fs_info->balance_lock); | |
3120 | fs_info->balance_ctl = NULL; | |
3121 | spin_unlock(&fs_info->balance_lock); | |
3122 | ||
3123 | kfree(bctl); | |
3124 | } | |
3125 | ||
ed25e9b2 ID |
3126 | /* |
3127 | * Balance filters. Return 1 if chunk should be filtered out | |
3128 | * (should not be balanced). | |
3129 | */ | |
899c81ea | 3130 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3131 | struct btrfs_balance_args *bargs) |
3132 | { | |
899c81ea ID |
3133 | chunk_type = chunk_to_extended(chunk_type) & |
3134 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3135 | |
899c81ea | 3136 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3137 | return 0; |
3138 | ||
3139 | return 1; | |
3140 | } | |
3141 | ||
5ce5b3c0 ID |
3142 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
3143 | struct btrfs_balance_args *bargs) | |
3144 | { | |
3145 | struct btrfs_block_group_cache *cache; | |
3146 | u64 chunk_used, user_thresh; | |
3147 | int ret = 1; | |
3148 | ||
3149 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3150 | chunk_used = btrfs_block_group_used(&cache->item); | |
3151 | ||
a105bb88 | 3152 | if (bargs->usage == 0) |
3e39cea6 | 3153 | user_thresh = 1; |
a105bb88 ID |
3154 | else if (bargs->usage > 100) |
3155 | user_thresh = cache->key.offset; | |
3156 | else | |
3157 | user_thresh = div_factor_fine(cache->key.offset, | |
3158 | bargs->usage); | |
3159 | ||
5ce5b3c0 ID |
3160 | if (chunk_used < user_thresh) |
3161 | ret = 0; | |
3162 | ||
3163 | btrfs_put_block_group(cache); | |
3164 | return ret; | |
3165 | } | |
3166 | ||
409d404b ID |
3167 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3168 | struct btrfs_chunk *chunk, | |
3169 | struct btrfs_balance_args *bargs) | |
3170 | { | |
3171 | struct btrfs_stripe *stripe; | |
3172 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3173 | int i; | |
3174 | ||
3175 | for (i = 0; i < num_stripes; i++) { | |
3176 | stripe = btrfs_stripe_nr(chunk, i); | |
3177 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3178 | return 0; | |
3179 | } | |
3180 | ||
3181 | return 1; | |
3182 | } | |
3183 | ||
94e60d5a ID |
3184 | /* [pstart, pend) */ |
3185 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3186 | struct btrfs_chunk *chunk, | |
3187 | u64 chunk_offset, | |
3188 | struct btrfs_balance_args *bargs) | |
3189 | { | |
3190 | struct btrfs_stripe *stripe; | |
3191 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3192 | u64 stripe_offset; | |
3193 | u64 stripe_length; | |
3194 | int factor; | |
3195 | int i; | |
3196 | ||
3197 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3198 | return 0; | |
3199 | ||
3200 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3201 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3202 | factor = num_stripes / 2; | |
3203 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3204 | factor = num_stripes - 1; | |
3205 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3206 | factor = num_stripes - 2; | |
3207 | } else { | |
3208 | factor = num_stripes; | |
3209 | } | |
94e60d5a ID |
3210 | |
3211 | for (i = 0; i < num_stripes; i++) { | |
3212 | stripe = btrfs_stripe_nr(chunk, i); | |
3213 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3214 | continue; | |
3215 | ||
3216 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3217 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3218 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3219 | |
3220 | if (stripe_offset < bargs->pend && | |
3221 | stripe_offset + stripe_length > bargs->pstart) | |
3222 | return 0; | |
3223 | } | |
3224 | ||
3225 | return 1; | |
3226 | } | |
3227 | ||
ea67176a ID |
3228 | /* [vstart, vend) */ |
3229 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3230 | struct btrfs_chunk *chunk, | |
3231 | u64 chunk_offset, | |
3232 | struct btrfs_balance_args *bargs) | |
3233 | { | |
3234 | if (chunk_offset < bargs->vend && | |
3235 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3236 | /* at least part of the chunk is inside this vrange */ | |
3237 | return 0; | |
3238 | ||
3239 | return 1; | |
3240 | } | |
3241 | ||
899c81ea | 3242 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3243 | struct btrfs_balance_args *bargs) |
3244 | { | |
3245 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3246 | return 0; | |
3247 | ||
899c81ea ID |
3248 | chunk_type = chunk_to_extended(chunk_type) & |
3249 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3250 | |
899c81ea | 3251 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3252 | return 1; |
3253 | ||
3254 | return 0; | |
3255 | } | |
3256 | ||
f43ffb60 ID |
3257 | static int should_balance_chunk(struct btrfs_root *root, |
3258 | struct extent_buffer *leaf, | |
3259 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3260 | { | |
3261 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3262 | struct btrfs_balance_args *bargs = NULL; | |
3263 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3264 | ||
3265 | /* type filter */ | |
3266 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3267 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3268 | return 0; | |
3269 | } | |
3270 | ||
3271 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3272 | bargs = &bctl->data; | |
3273 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3274 | bargs = &bctl->sys; | |
3275 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3276 | bargs = &bctl->meta; | |
3277 | ||
ed25e9b2 ID |
3278 | /* profiles filter */ |
3279 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3280 | chunk_profiles_filter(chunk_type, bargs)) { | |
3281 | return 0; | |
5ce5b3c0 ID |
3282 | } |
3283 | ||
3284 | /* usage filter */ | |
3285 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3286 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3287 | return 0; | |
409d404b ID |
3288 | } |
3289 | ||
3290 | /* devid filter */ | |
3291 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3292 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3293 | return 0; | |
94e60d5a ID |
3294 | } |
3295 | ||
3296 | /* drange filter, makes sense only with devid filter */ | |
3297 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3298 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3299 | return 0; | |
ea67176a ID |
3300 | } |
3301 | ||
3302 | /* vrange filter */ | |
3303 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3304 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3305 | return 0; | |
ed25e9b2 ID |
3306 | } |
3307 | ||
cfa4c961 ID |
3308 | /* soft profile changing mode */ |
3309 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3310 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3311 | return 0; | |
3312 | } | |
3313 | ||
7d824b6f DS |
3314 | /* |
3315 | * limited by count, must be the last filter | |
3316 | */ | |
3317 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3318 | if (bargs->limit == 0) | |
3319 | return 0; | |
3320 | else | |
3321 | bargs->limit--; | |
3322 | } | |
3323 | ||
f43ffb60 ID |
3324 | return 1; |
3325 | } | |
3326 | ||
c9e9f97b | 3327 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3328 | { |
19a39dce | 3329 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3330 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3331 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3332 | struct list_head *devices; | |
ec44a35c CM |
3333 | struct btrfs_device *device; |
3334 | u64 old_size; | |
3335 | u64 size_to_free; | |
f43ffb60 | 3336 | struct btrfs_chunk *chunk; |
ec44a35c CM |
3337 | struct btrfs_path *path; |
3338 | struct btrfs_key key; | |
ec44a35c | 3339 | struct btrfs_key found_key; |
c9e9f97b | 3340 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3341 | struct extent_buffer *leaf; |
3342 | int slot; | |
c9e9f97b ID |
3343 | int ret; |
3344 | int enospc_errors = 0; | |
19a39dce | 3345 | bool counting = true; |
7d824b6f DS |
3346 | u64 limit_data = bctl->data.limit; |
3347 | u64 limit_meta = bctl->meta.limit; | |
3348 | u64 limit_sys = bctl->sys.limit; | |
ec44a35c | 3349 | |
ec44a35c | 3350 | /* step one make some room on all the devices */ |
c9e9f97b | 3351 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3352 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3353 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c CM |
3354 | size_to_free = div_factor(old_size, 1); |
3355 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c | 3356 | if (!device->writeable || |
7cc8e58d MX |
3357 | btrfs_device_get_total_bytes(device) - |
3358 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3359 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3360 | continue; |
3361 | ||
3362 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3363 | if (ret == -ENOSPC) |
3364 | break; | |
ec44a35c CM |
3365 | BUG_ON(ret); |
3366 | ||
a22285a6 | 3367 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 3368 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
3369 | |
3370 | ret = btrfs_grow_device(trans, device, old_size); | |
3371 | BUG_ON(ret); | |
3372 | ||
3373 | btrfs_end_transaction(trans, dev_root); | |
3374 | } | |
3375 | ||
3376 | /* step two, relocate all the chunks */ | |
3377 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3378 | if (!path) { |
3379 | ret = -ENOMEM; | |
3380 | goto error; | |
3381 | } | |
19a39dce ID |
3382 | |
3383 | /* zero out stat counters */ | |
3384 | spin_lock(&fs_info->balance_lock); | |
3385 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3386 | spin_unlock(&fs_info->balance_lock); | |
3387 | again: | |
7d824b6f DS |
3388 | if (!counting) { |
3389 | bctl->data.limit = limit_data; | |
3390 | bctl->meta.limit = limit_meta; | |
3391 | bctl->sys.limit = limit_sys; | |
3392 | } | |
ec44a35c CM |
3393 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3394 | key.offset = (u64)-1; | |
3395 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3396 | ||
d397712b | 3397 | while (1) { |
19a39dce | 3398 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3399 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3400 | ret = -ECANCELED; |
3401 | goto error; | |
3402 | } | |
3403 | ||
67c5e7d4 | 3404 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3405 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3406 | if (ret < 0) { |
3407 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3408 | goto error; |
67c5e7d4 | 3409 | } |
ec44a35c CM |
3410 | |
3411 | /* | |
3412 | * this shouldn't happen, it means the last relocate | |
3413 | * failed | |
3414 | */ | |
3415 | if (ret == 0) | |
c9e9f97b | 3416 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3417 | |
3418 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3419 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3420 | if (ret) { |
67c5e7d4 | 3421 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3422 | ret = 0; |
ec44a35c | 3423 | break; |
c9e9f97b | 3424 | } |
7d9eb12c | 3425 | |
f43ffb60 ID |
3426 | leaf = path->nodes[0]; |
3427 | slot = path->slots[0]; | |
3428 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3429 | |
67c5e7d4 FM |
3430 | if (found_key.objectid != key.objectid) { |
3431 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3432 | break; |
67c5e7d4 | 3433 | } |
7d9eb12c | 3434 | |
f43ffb60 ID |
3435 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
3436 | ||
19a39dce ID |
3437 | if (!counting) { |
3438 | spin_lock(&fs_info->balance_lock); | |
3439 | bctl->stat.considered++; | |
3440 | spin_unlock(&fs_info->balance_lock); | |
3441 | } | |
3442 | ||
f43ffb60 ID |
3443 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3444 | found_key.offset); | |
b3b4aa74 | 3445 | btrfs_release_path(path); |
67c5e7d4 FM |
3446 | if (!ret) { |
3447 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3448 | goto loop; |
67c5e7d4 | 3449 | } |
f43ffb60 | 3450 | |
19a39dce | 3451 | if (counting) { |
67c5e7d4 | 3452 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3453 | spin_lock(&fs_info->balance_lock); |
3454 | bctl->stat.expected++; | |
3455 | spin_unlock(&fs_info->balance_lock); | |
3456 | goto loop; | |
3457 | } | |
3458 | ||
ec44a35c | 3459 | ret = btrfs_relocate_chunk(chunk_root, |
ec44a35c | 3460 | found_key.offset); |
67c5e7d4 | 3461 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3462 | if (ret && ret != -ENOSPC) |
3463 | goto error; | |
19a39dce | 3464 | if (ret == -ENOSPC) { |
c9e9f97b | 3465 | enospc_errors++; |
19a39dce ID |
3466 | } else { |
3467 | spin_lock(&fs_info->balance_lock); | |
3468 | bctl->stat.completed++; | |
3469 | spin_unlock(&fs_info->balance_lock); | |
3470 | } | |
f43ffb60 | 3471 | loop: |
795a3321 ID |
3472 | if (found_key.offset == 0) |
3473 | break; | |
ba1bf481 | 3474 | key.offset = found_key.offset - 1; |
ec44a35c | 3475 | } |
c9e9f97b | 3476 | |
19a39dce ID |
3477 | if (counting) { |
3478 | btrfs_release_path(path); | |
3479 | counting = false; | |
3480 | goto again; | |
3481 | } | |
ec44a35c CM |
3482 | error: |
3483 | btrfs_free_path(path); | |
c9e9f97b | 3484 | if (enospc_errors) { |
efe120a0 | 3485 | btrfs_info(fs_info, "%d enospc errors during balance", |
c9e9f97b ID |
3486 | enospc_errors); |
3487 | if (!ret) | |
3488 | ret = -ENOSPC; | |
3489 | } | |
3490 | ||
ec44a35c CM |
3491 | return ret; |
3492 | } | |
3493 | ||
0c460c0d ID |
3494 | /** |
3495 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3496 | * @flags: profile to validate | |
3497 | * @extended: if true @flags is treated as an extended profile | |
3498 | */ | |
3499 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3500 | { | |
3501 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3502 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3503 | ||
3504 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3505 | ||
3506 | /* 1) check that all other bits are zeroed */ | |
3507 | if (flags & ~mask) | |
3508 | return 0; | |
3509 | ||
3510 | /* 2) see if profile is reduced */ | |
3511 | if (flags == 0) | |
3512 | return !extended; /* "0" is valid for usual profiles */ | |
3513 | ||
3514 | /* true if exactly one bit set */ | |
3515 | return (flags & (flags - 1)) == 0; | |
3516 | } | |
3517 | ||
837d5b6e ID |
3518 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3519 | { | |
a7e99c69 ID |
3520 | /* cancel requested || normal exit path */ |
3521 | return atomic_read(&fs_info->balance_cancel_req) || | |
3522 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3523 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3524 | } |
3525 | ||
c9e9f97b ID |
3526 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3527 | { | |
0940ebf6 ID |
3528 | int ret; |
3529 | ||
c9e9f97b | 3530 | unset_balance_control(fs_info); |
0940ebf6 | 3531 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 LB |
3532 | if (ret) |
3533 | btrfs_std_error(fs_info, ret); | |
ed0fb78f ID |
3534 | |
3535 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3536 | } |
3537 | ||
c9e9f97b ID |
3538 | /* |
3539 | * Should be called with both balance and volume mutexes held | |
3540 | */ | |
3541 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3542 | struct btrfs_ioctl_balance_args *bargs) | |
3543 | { | |
3544 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3545 | u64 allowed; |
e4837f8f | 3546 | int mixed = 0; |
c9e9f97b | 3547 | int ret; |
8dabb742 | 3548 | u64 num_devices; |
de98ced9 | 3549 | unsigned seq; |
c9e9f97b | 3550 | |
837d5b6e | 3551 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3552 | atomic_read(&fs_info->balance_pause_req) || |
3553 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3554 | ret = -EINVAL; |
3555 | goto out; | |
3556 | } | |
3557 | ||
e4837f8f ID |
3558 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3559 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3560 | mixed = 1; | |
3561 | ||
f43ffb60 ID |
3562 | /* |
3563 | * In case of mixed groups both data and meta should be picked, | |
3564 | * and identical options should be given for both of them. | |
3565 | */ | |
e4837f8f ID |
3566 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3567 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3568 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3569 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3570 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
efe120a0 FH |
3571 | btrfs_err(fs_info, "with mixed groups data and " |
3572 | "metadata balance options must be the same"); | |
f43ffb60 ID |
3573 | ret = -EINVAL; |
3574 | goto out; | |
3575 | } | |
3576 | } | |
3577 | ||
8dabb742 SB |
3578 | num_devices = fs_info->fs_devices->num_devices; |
3579 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
3580 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3581 | BUG_ON(num_devices < 1); | |
3582 | num_devices--; | |
3583 | } | |
3584 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
e4d8ec0f | 3585 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3586 | if (num_devices == 1) |
e4d8ec0f | 3587 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8250dabe | 3588 | else if (num_devices > 1) |
e4d8ec0f | 3589 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3590 | if (num_devices > 2) |
3591 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3592 | if (num_devices > 3) | |
3593 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3594 | BTRFS_BLOCK_GROUP_RAID6); | |
6728b198 ID |
3595 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3596 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
3597 | (bctl->data.target & ~allowed))) { | |
efe120a0 FH |
3598 | btrfs_err(fs_info, "unable to start balance with target " |
3599 | "data profile %llu", | |
c1c9ff7c | 3600 | bctl->data.target); |
e4d8ec0f ID |
3601 | ret = -EINVAL; |
3602 | goto out; | |
3603 | } | |
6728b198 ID |
3604 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3605 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
3606 | (bctl->meta.target & ~allowed))) { | |
efe120a0 FH |
3607 | btrfs_err(fs_info, |
3608 | "unable to start balance with target metadata profile %llu", | |
c1c9ff7c | 3609 | bctl->meta.target); |
e4d8ec0f ID |
3610 | ret = -EINVAL; |
3611 | goto out; | |
3612 | } | |
6728b198 ID |
3613 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3614 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
3615 | (bctl->sys.target & ~allowed))) { | |
efe120a0 FH |
3616 | btrfs_err(fs_info, |
3617 | "unable to start balance with target system profile %llu", | |
c1c9ff7c | 3618 | bctl->sys.target); |
e4d8ec0f ID |
3619 | ret = -EINVAL; |
3620 | goto out; | |
3621 | } | |
3622 | ||
e4837f8f ID |
3623 | /* allow dup'ed data chunks only in mixed mode */ |
3624 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 3625 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
efe120a0 | 3626 | btrfs_err(fs_info, "dup for data is not allowed"); |
e4d8ec0f ID |
3627 | ret = -EINVAL; |
3628 | goto out; | |
3629 | } | |
3630 | ||
3631 | /* allow to reduce meta or sys integrity only if force set */ | |
3632 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3633 | BTRFS_BLOCK_GROUP_RAID10 | |
3634 | BTRFS_BLOCK_GROUP_RAID5 | | |
3635 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3636 | do { |
3637 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3638 | ||
3639 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3640 | (fs_info->avail_system_alloc_bits & allowed) && | |
3641 | !(bctl->sys.target & allowed)) || | |
3642 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3643 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3644 | !(bctl->meta.target & allowed))) { | |
3645 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
efe120a0 | 3646 | btrfs_info(fs_info, "force reducing metadata integrity"); |
de98ced9 | 3647 | } else { |
efe120a0 FH |
3648 | btrfs_err(fs_info, "balance will reduce metadata " |
3649 | "integrity, use force if you want this"); | |
de98ced9 MX |
3650 | ret = -EINVAL; |
3651 | goto out; | |
3652 | } | |
e4d8ec0f | 3653 | } |
de98ced9 | 3654 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3655 | |
5af3e8cc | 3656 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
943c6e99 ZL |
3657 | fs_info->num_tolerated_disk_barrier_failures = min( |
3658 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3659 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3660 | bctl->sys.target)); | |
5af3e8cc SB |
3661 | } |
3662 | ||
0940ebf6 | 3663 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3664 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3665 | goto out; |
3666 | ||
59641015 ID |
3667 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3668 | BUG_ON(ret == -EEXIST); | |
3669 | set_balance_control(bctl); | |
3670 | } else { | |
3671 | BUG_ON(ret != -EEXIST); | |
3672 | spin_lock(&fs_info->balance_lock); | |
3673 | update_balance_args(bctl); | |
3674 | spin_unlock(&fs_info->balance_lock); | |
3675 | } | |
c9e9f97b | 3676 | |
837d5b6e | 3677 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3678 | mutex_unlock(&fs_info->balance_mutex); |
3679 | ||
3680 | ret = __btrfs_balance(fs_info); | |
3681 | ||
3682 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3683 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3684 | |
bf023ecf ID |
3685 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3686 | fs_info->num_tolerated_disk_barrier_failures = | |
3687 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3688 | } | |
3689 | ||
c9e9f97b ID |
3690 | if (bargs) { |
3691 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3692 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3693 | } |
3694 | ||
3a01aa7a ID |
3695 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3696 | balance_need_close(fs_info)) { | |
3697 | __cancel_balance(fs_info); | |
3698 | } | |
3699 | ||
837d5b6e | 3700 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3701 | |
3702 | return ret; | |
3703 | out: | |
59641015 ID |
3704 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3705 | __cancel_balance(fs_info); | |
ed0fb78f | 3706 | else { |
59641015 | 3707 | kfree(bctl); |
ed0fb78f ID |
3708 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3709 | } | |
59641015 ID |
3710 | return ret; |
3711 | } | |
3712 | ||
3713 | static int balance_kthread(void *data) | |
3714 | { | |
2b6ba629 | 3715 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3716 | int ret = 0; |
59641015 ID |
3717 | |
3718 | mutex_lock(&fs_info->volume_mutex); | |
3719 | mutex_lock(&fs_info->balance_mutex); | |
3720 | ||
2b6ba629 | 3721 | if (fs_info->balance_ctl) { |
efe120a0 | 3722 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3723 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3724 | } |
59641015 ID |
3725 | |
3726 | mutex_unlock(&fs_info->balance_mutex); | |
3727 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3728 | |
59641015 ID |
3729 | return ret; |
3730 | } | |
3731 | ||
2b6ba629 ID |
3732 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3733 | { | |
3734 | struct task_struct *tsk; | |
3735 | ||
3736 | spin_lock(&fs_info->balance_lock); | |
3737 | if (!fs_info->balance_ctl) { | |
3738 | spin_unlock(&fs_info->balance_lock); | |
3739 | return 0; | |
3740 | } | |
3741 | spin_unlock(&fs_info->balance_lock); | |
3742 | ||
3743 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
efe120a0 | 3744 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3745 | return 0; |
3746 | } | |
3747 | ||
3748 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3749 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3750 | } |
3751 | ||
68310a5e | 3752 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3753 | { |
59641015 ID |
3754 | struct btrfs_balance_control *bctl; |
3755 | struct btrfs_balance_item *item; | |
3756 | struct btrfs_disk_balance_args disk_bargs; | |
3757 | struct btrfs_path *path; | |
3758 | struct extent_buffer *leaf; | |
3759 | struct btrfs_key key; | |
3760 | int ret; | |
3761 | ||
3762 | path = btrfs_alloc_path(); | |
3763 | if (!path) | |
3764 | return -ENOMEM; | |
3765 | ||
59641015 ID |
3766 | key.objectid = BTRFS_BALANCE_OBJECTID; |
3767 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3768 | key.offset = 0; | |
3769 | ||
68310a5e | 3770 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3771 | if (ret < 0) |
68310a5e | 3772 | goto out; |
59641015 ID |
3773 | if (ret > 0) { /* ret = -ENOENT; */ |
3774 | ret = 0; | |
68310a5e ID |
3775 | goto out; |
3776 | } | |
3777 | ||
3778 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3779 | if (!bctl) { | |
3780 | ret = -ENOMEM; | |
3781 | goto out; | |
59641015 ID |
3782 | } |
3783 | ||
3784 | leaf = path->nodes[0]; | |
3785 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3786 | ||
68310a5e ID |
3787 | bctl->fs_info = fs_info; |
3788 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3789 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3790 | |
3791 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3792 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3793 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3794 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3795 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3796 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3797 | ||
ed0fb78f ID |
3798 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3799 | ||
68310a5e ID |
3800 | mutex_lock(&fs_info->volume_mutex); |
3801 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3802 | |
68310a5e ID |
3803 | set_balance_control(bctl); |
3804 | ||
3805 | mutex_unlock(&fs_info->balance_mutex); | |
3806 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3807 | out: |
3808 | btrfs_free_path(path); | |
ec44a35c CM |
3809 | return ret; |
3810 | } | |
3811 | ||
837d5b6e ID |
3812 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3813 | { | |
3814 | int ret = 0; | |
3815 | ||
3816 | mutex_lock(&fs_info->balance_mutex); | |
3817 | if (!fs_info->balance_ctl) { | |
3818 | mutex_unlock(&fs_info->balance_mutex); | |
3819 | return -ENOTCONN; | |
3820 | } | |
3821 | ||
3822 | if (atomic_read(&fs_info->balance_running)) { | |
3823 | atomic_inc(&fs_info->balance_pause_req); | |
3824 | mutex_unlock(&fs_info->balance_mutex); | |
3825 | ||
3826 | wait_event(fs_info->balance_wait_q, | |
3827 | atomic_read(&fs_info->balance_running) == 0); | |
3828 | ||
3829 | mutex_lock(&fs_info->balance_mutex); | |
3830 | /* we are good with balance_ctl ripped off from under us */ | |
3831 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3832 | atomic_dec(&fs_info->balance_pause_req); | |
3833 | } else { | |
3834 | ret = -ENOTCONN; | |
3835 | } | |
3836 | ||
3837 | mutex_unlock(&fs_info->balance_mutex); | |
3838 | return ret; | |
3839 | } | |
3840 | ||
a7e99c69 ID |
3841 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3842 | { | |
e649e587 ID |
3843 | if (fs_info->sb->s_flags & MS_RDONLY) |
3844 | return -EROFS; | |
3845 | ||
a7e99c69 ID |
3846 | mutex_lock(&fs_info->balance_mutex); |
3847 | if (!fs_info->balance_ctl) { | |
3848 | mutex_unlock(&fs_info->balance_mutex); | |
3849 | return -ENOTCONN; | |
3850 | } | |
3851 | ||
3852 | atomic_inc(&fs_info->balance_cancel_req); | |
3853 | /* | |
3854 | * if we are running just wait and return, balance item is | |
3855 | * deleted in btrfs_balance in this case | |
3856 | */ | |
3857 | if (atomic_read(&fs_info->balance_running)) { | |
3858 | mutex_unlock(&fs_info->balance_mutex); | |
3859 | wait_event(fs_info->balance_wait_q, | |
3860 | atomic_read(&fs_info->balance_running) == 0); | |
3861 | mutex_lock(&fs_info->balance_mutex); | |
3862 | } else { | |
3863 | /* __cancel_balance needs volume_mutex */ | |
3864 | mutex_unlock(&fs_info->balance_mutex); | |
3865 | mutex_lock(&fs_info->volume_mutex); | |
3866 | mutex_lock(&fs_info->balance_mutex); | |
3867 | ||
3868 | if (fs_info->balance_ctl) | |
3869 | __cancel_balance(fs_info); | |
3870 | ||
3871 | mutex_unlock(&fs_info->volume_mutex); | |
3872 | } | |
3873 | ||
3874 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3875 | atomic_dec(&fs_info->balance_cancel_req); | |
3876 | mutex_unlock(&fs_info->balance_mutex); | |
3877 | return 0; | |
3878 | } | |
3879 | ||
803b2f54 SB |
3880 | static int btrfs_uuid_scan_kthread(void *data) |
3881 | { | |
3882 | struct btrfs_fs_info *fs_info = data; | |
3883 | struct btrfs_root *root = fs_info->tree_root; | |
3884 | struct btrfs_key key; | |
3885 | struct btrfs_key max_key; | |
3886 | struct btrfs_path *path = NULL; | |
3887 | int ret = 0; | |
3888 | struct extent_buffer *eb; | |
3889 | int slot; | |
3890 | struct btrfs_root_item root_item; | |
3891 | u32 item_size; | |
f45388f3 | 3892 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
3893 | |
3894 | path = btrfs_alloc_path(); | |
3895 | if (!path) { | |
3896 | ret = -ENOMEM; | |
3897 | goto out; | |
3898 | } | |
3899 | ||
3900 | key.objectid = 0; | |
3901 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3902 | key.offset = 0; | |
3903 | ||
3904 | max_key.objectid = (u64)-1; | |
3905 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
3906 | max_key.offset = (u64)-1; | |
3907 | ||
803b2f54 | 3908 | while (1) { |
6174d3cb | 3909 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
3910 | if (ret) { |
3911 | if (ret > 0) | |
3912 | ret = 0; | |
3913 | break; | |
3914 | } | |
3915 | ||
3916 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
3917 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
3918 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
3919 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
3920 | goto skip; | |
3921 | ||
3922 | eb = path->nodes[0]; | |
3923 | slot = path->slots[0]; | |
3924 | item_size = btrfs_item_size_nr(eb, slot); | |
3925 | if (item_size < sizeof(root_item)) | |
3926 | goto skip; | |
3927 | ||
803b2f54 SB |
3928 | read_extent_buffer(eb, &root_item, |
3929 | btrfs_item_ptr_offset(eb, slot), | |
3930 | (int)sizeof(root_item)); | |
3931 | if (btrfs_root_refs(&root_item) == 0) | |
3932 | goto skip; | |
f45388f3 FDBM |
3933 | |
3934 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
3935 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
3936 | if (trans) | |
3937 | goto update_tree; | |
3938 | ||
3939 | btrfs_release_path(path); | |
803b2f54 SB |
3940 | /* |
3941 | * 1 - subvol uuid item | |
3942 | * 1 - received_subvol uuid item | |
3943 | */ | |
3944 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
3945 | if (IS_ERR(trans)) { | |
3946 | ret = PTR_ERR(trans); | |
3947 | break; | |
3948 | } | |
f45388f3 FDBM |
3949 | continue; |
3950 | } else { | |
3951 | goto skip; | |
3952 | } | |
3953 | update_tree: | |
3954 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
803b2f54 SB |
3955 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3956 | root_item.uuid, | |
3957 | BTRFS_UUID_KEY_SUBVOL, | |
3958 | key.objectid); | |
3959 | if (ret < 0) { | |
efe120a0 | 3960 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3961 | ret); |
803b2f54 SB |
3962 | break; |
3963 | } | |
3964 | } | |
3965 | ||
3966 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
803b2f54 SB |
3967 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3968 | root_item.received_uuid, | |
3969 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
3970 | key.objectid); | |
3971 | if (ret < 0) { | |
efe120a0 | 3972 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3973 | ret); |
803b2f54 SB |
3974 | break; |
3975 | } | |
3976 | } | |
3977 | ||
f45388f3 | 3978 | skip: |
803b2f54 SB |
3979 | if (trans) { |
3980 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 3981 | trans = NULL; |
803b2f54 SB |
3982 | if (ret) |
3983 | break; | |
3984 | } | |
3985 | ||
803b2f54 SB |
3986 | btrfs_release_path(path); |
3987 | if (key.offset < (u64)-1) { | |
3988 | key.offset++; | |
3989 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
3990 | key.offset = 0; | |
3991 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3992 | } else if (key.objectid < (u64)-1) { | |
3993 | key.offset = 0; | |
3994 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3995 | key.objectid++; | |
3996 | } else { | |
3997 | break; | |
3998 | } | |
3999 | cond_resched(); | |
4000 | } | |
4001 | ||
4002 | out: | |
4003 | btrfs_free_path(path); | |
f45388f3 FDBM |
4004 | if (trans && !IS_ERR(trans)) |
4005 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 4006 | if (ret) |
efe120a0 | 4007 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 SB |
4008 | else |
4009 | fs_info->update_uuid_tree_gen = 1; | |
803b2f54 SB |
4010 | up(&fs_info->uuid_tree_rescan_sem); |
4011 | return 0; | |
4012 | } | |
4013 | ||
70f80175 SB |
4014 | /* |
4015 | * Callback for btrfs_uuid_tree_iterate(). | |
4016 | * returns: | |
4017 | * 0 check succeeded, the entry is not outdated. | |
4018 | * < 0 if an error occured. | |
4019 | * > 0 if the check failed, which means the caller shall remove the entry. | |
4020 | */ | |
4021 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4022 | u8 *uuid, u8 type, u64 subid) | |
4023 | { | |
4024 | struct btrfs_key key; | |
4025 | int ret = 0; | |
4026 | struct btrfs_root *subvol_root; | |
4027 | ||
4028 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4029 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4030 | goto out; | |
4031 | ||
4032 | key.objectid = subid; | |
4033 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4034 | key.offset = (u64)-1; | |
4035 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4036 | if (IS_ERR(subvol_root)) { | |
4037 | ret = PTR_ERR(subvol_root); | |
4038 | if (ret == -ENOENT) | |
4039 | ret = 1; | |
4040 | goto out; | |
4041 | } | |
4042 | ||
4043 | switch (type) { | |
4044 | case BTRFS_UUID_KEY_SUBVOL: | |
4045 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4046 | ret = 1; | |
4047 | break; | |
4048 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4049 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4050 | BTRFS_UUID_SIZE)) | |
4051 | ret = 1; | |
4052 | break; | |
4053 | } | |
4054 | ||
4055 | out: | |
4056 | return ret; | |
4057 | } | |
4058 | ||
4059 | static int btrfs_uuid_rescan_kthread(void *data) | |
4060 | { | |
4061 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4062 | int ret; | |
4063 | ||
4064 | /* | |
4065 | * 1st step is to iterate through the existing UUID tree and | |
4066 | * to delete all entries that contain outdated data. | |
4067 | * 2nd step is to add all missing entries to the UUID tree. | |
4068 | */ | |
4069 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4070 | if (ret < 0) { | |
efe120a0 | 4071 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4072 | up(&fs_info->uuid_tree_rescan_sem); |
4073 | return ret; | |
4074 | } | |
4075 | return btrfs_uuid_scan_kthread(data); | |
4076 | } | |
4077 | ||
f7a81ea4 SB |
4078 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4079 | { | |
4080 | struct btrfs_trans_handle *trans; | |
4081 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4082 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4083 | struct task_struct *task; |
4084 | int ret; | |
f7a81ea4 SB |
4085 | |
4086 | /* | |
4087 | * 1 - root node | |
4088 | * 1 - root item | |
4089 | */ | |
4090 | trans = btrfs_start_transaction(tree_root, 2); | |
4091 | if (IS_ERR(trans)) | |
4092 | return PTR_ERR(trans); | |
4093 | ||
4094 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4095 | BTRFS_UUID_TREE_OBJECTID); | |
4096 | if (IS_ERR(uuid_root)) { | |
6d13f549 DS |
4097 | ret = PTR_ERR(uuid_root); |
4098 | btrfs_abort_transaction(trans, tree_root, ret); | |
4099 | return ret; | |
f7a81ea4 SB |
4100 | } |
4101 | ||
4102 | fs_info->uuid_root = uuid_root; | |
4103 | ||
803b2f54 SB |
4104 | ret = btrfs_commit_transaction(trans, tree_root); |
4105 | if (ret) | |
4106 | return ret; | |
4107 | ||
4108 | down(&fs_info->uuid_tree_rescan_sem); | |
4109 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4110 | if (IS_ERR(task)) { | |
70f80175 | 4111 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4112 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4113 | up(&fs_info->uuid_tree_rescan_sem); |
4114 | return PTR_ERR(task); | |
4115 | } | |
4116 | ||
4117 | return 0; | |
f7a81ea4 | 4118 | } |
803b2f54 | 4119 | |
70f80175 SB |
4120 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4121 | { | |
4122 | struct task_struct *task; | |
4123 | ||
4124 | down(&fs_info->uuid_tree_rescan_sem); | |
4125 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4126 | if (IS_ERR(task)) { | |
4127 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4128 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4129 | up(&fs_info->uuid_tree_rescan_sem); |
4130 | return PTR_ERR(task); | |
4131 | } | |
4132 | ||
4133 | return 0; | |
4134 | } | |
4135 | ||
8f18cf13 CM |
4136 | /* |
4137 | * shrinking a device means finding all of the device extents past | |
4138 | * the new size, and then following the back refs to the chunks. | |
4139 | * The chunk relocation code actually frees the device extent | |
4140 | */ | |
4141 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4142 | { | |
4143 | struct btrfs_trans_handle *trans; | |
4144 | struct btrfs_root *root = device->dev_root; | |
4145 | struct btrfs_dev_extent *dev_extent = NULL; | |
4146 | struct btrfs_path *path; | |
4147 | u64 length; | |
8f18cf13 CM |
4148 | u64 chunk_offset; |
4149 | int ret; | |
4150 | int slot; | |
ba1bf481 JB |
4151 | int failed = 0; |
4152 | bool retried = false; | |
53e489bc | 4153 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4154 | struct extent_buffer *l; |
4155 | struct btrfs_key key; | |
6c41761f | 4156 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 4157 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4158 | u64 old_size = btrfs_device_get_total_bytes(device); |
4159 | u64 diff = old_size - new_size; | |
8f18cf13 | 4160 | |
63a212ab SB |
4161 | if (device->is_tgtdev_for_dev_replace) |
4162 | return -EINVAL; | |
4163 | ||
8f18cf13 CM |
4164 | path = btrfs_alloc_path(); |
4165 | if (!path) | |
4166 | return -ENOMEM; | |
4167 | ||
8f18cf13 CM |
4168 | path->reada = 2; |
4169 | ||
7d9eb12c CM |
4170 | lock_chunks(root); |
4171 | ||
7cc8e58d | 4172 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4173 | if (device->writeable) { |
2b82032c | 4174 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
4175 | spin_lock(&root->fs_info->free_chunk_lock); |
4176 | root->fs_info->free_chunk_space -= diff; | |
4177 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4178 | } | |
7d9eb12c | 4179 | unlock_chunks(root); |
8f18cf13 | 4180 | |
ba1bf481 | 4181 | again: |
8f18cf13 CM |
4182 | key.objectid = device->devid; |
4183 | key.offset = (u64)-1; | |
4184 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4185 | ||
213e64da | 4186 | do { |
67c5e7d4 | 4187 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4188 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 FM |
4189 | if (ret < 0) { |
4190 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 | 4191 | goto done; |
67c5e7d4 | 4192 | } |
8f18cf13 CM |
4193 | |
4194 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 FM |
4195 | if (ret) |
4196 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
8f18cf13 CM |
4197 | if (ret < 0) |
4198 | goto done; | |
4199 | if (ret) { | |
4200 | ret = 0; | |
b3b4aa74 | 4201 | btrfs_release_path(path); |
bf1fb512 | 4202 | break; |
8f18cf13 CM |
4203 | } |
4204 | ||
4205 | l = path->nodes[0]; | |
4206 | slot = path->slots[0]; | |
4207 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4208 | ||
ba1bf481 | 4209 | if (key.objectid != device->devid) { |
67c5e7d4 | 4210 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4211 | btrfs_release_path(path); |
bf1fb512 | 4212 | break; |
ba1bf481 | 4213 | } |
8f18cf13 CM |
4214 | |
4215 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4216 | length = btrfs_dev_extent_length(l, dev_extent); | |
4217 | ||
ba1bf481 | 4218 | if (key.offset + length <= new_size) { |
67c5e7d4 | 4219 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4220 | btrfs_release_path(path); |
d6397bae | 4221 | break; |
ba1bf481 | 4222 | } |
8f18cf13 | 4223 | |
8f18cf13 | 4224 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4225 | btrfs_release_path(path); |
8f18cf13 | 4226 | |
dc2ee4e2 | 4227 | ret = btrfs_relocate_chunk(root, chunk_offset); |
67c5e7d4 | 4228 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); |
ba1bf481 | 4229 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4230 | goto done; |
ba1bf481 JB |
4231 | if (ret == -ENOSPC) |
4232 | failed++; | |
213e64da | 4233 | } while (key.offset-- > 0); |
ba1bf481 JB |
4234 | |
4235 | if (failed && !retried) { | |
4236 | failed = 0; | |
4237 | retried = true; | |
4238 | goto again; | |
4239 | } else if (failed && retried) { | |
4240 | ret = -ENOSPC; | |
ba1bf481 | 4241 | goto done; |
8f18cf13 CM |
4242 | } |
4243 | ||
d6397bae | 4244 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4245 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4246 | if (IS_ERR(trans)) { |
4247 | ret = PTR_ERR(trans); | |
4248 | goto done; | |
4249 | } | |
4250 | ||
d6397bae | 4251 | lock_chunks(root); |
53e489bc FM |
4252 | |
4253 | /* | |
4254 | * We checked in the above loop all device extents that were already in | |
4255 | * the device tree. However before we have updated the device's | |
4256 | * total_bytes to the new size, we might have had chunk allocations that | |
4257 | * have not complete yet (new block groups attached to transaction | |
4258 | * handles), and therefore their device extents were not yet in the | |
4259 | * device tree and we missed them in the loop above. So if we have any | |
4260 | * pending chunk using a device extent that overlaps the device range | |
4261 | * that we can not use anymore, commit the current transaction and | |
4262 | * repeat the search on the device tree - this way we guarantee we will | |
4263 | * not have chunks using device extents that end beyond 'new_size'. | |
4264 | */ | |
4265 | if (!checked_pending_chunks) { | |
4266 | u64 start = new_size; | |
4267 | u64 len = old_size - new_size; | |
4268 | ||
499f377f JM |
4269 | if (contains_pending_extent(trans->transaction, device, |
4270 | &start, len)) { | |
53e489bc FM |
4271 | unlock_chunks(root); |
4272 | checked_pending_chunks = true; | |
4273 | failed = 0; | |
4274 | retried = false; | |
4275 | ret = btrfs_commit_transaction(trans, root); | |
4276 | if (ret) | |
4277 | goto done; | |
4278 | goto again; | |
4279 | } | |
4280 | } | |
4281 | ||
7cc8e58d | 4282 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4283 | if (list_empty(&device->resized_list)) |
4284 | list_add_tail(&device->resized_list, | |
4285 | &root->fs_info->fs_devices->resized_devices); | |
d6397bae | 4286 | |
d6397bae CB |
4287 | WARN_ON(diff > old_total); |
4288 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4289 | unlock_chunks(root); | |
2196d6e8 MX |
4290 | |
4291 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4292 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4293 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4294 | done: |
4295 | btrfs_free_path(path); | |
53e489bc FM |
4296 | if (ret) { |
4297 | lock_chunks(root); | |
4298 | btrfs_device_set_total_bytes(device, old_size); | |
4299 | if (device->writeable) | |
4300 | device->fs_devices->total_rw_bytes += diff; | |
4301 | spin_lock(&root->fs_info->free_chunk_lock); | |
4302 | root->fs_info->free_chunk_space += diff; | |
4303 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4304 | unlock_chunks(root); | |
4305 | } | |
8f18cf13 CM |
4306 | return ret; |
4307 | } | |
4308 | ||
125ccb0a | 4309 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4310 | struct btrfs_key *key, |
4311 | struct btrfs_chunk *chunk, int item_size) | |
4312 | { | |
6c41761f | 4313 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
4314 | struct btrfs_disk_key disk_key; |
4315 | u32 array_size; | |
4316 | u8 *ptr; | |
4317 | ||
fe48a5c0 | 4318 | lock_chunks(root); |
0b86a832 | 4319 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4320 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 MX |
4321 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
4322 | unlock_chunks(root); | |
0b86a832 | 4323 | return -EFBIG; |
fe48a5c0 | 4324 | } |
0b86a832 CM |
4325 | |
4326 | ptr = super_copy->sys_chunk_array + array_size; | |
4327 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4328 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4329 | ptr += sizeof(disk_key); | |
4330 | memcpy(ptr, chunk, item_size); | |
4331 | item_size += sizeof(disk_key); | |
4332 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 MX |
4333 | unlock_chunks(root); |
4334 | ||
0b86a832 CM |
4335 | return 0; |
4336 | } | |
4337 | ||
73c5de00 AJ |
4338 | /* |
4339 | * sort the devices in descending order by max_avail, total_avail | |
4340 | */ | |
4341 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4342 | { |
73c5de00 AJ |
4343 | const struct btrfs_device_info *di_a = a; |
4344 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4345 | |
73c5de00 | 4346 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4347 | return -1; |
73c5de00 | 4348 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4349 | return 1; |
73c5de00 AJ |
4350 | if (di_a->total_avail > di_b->total_avail) |
4351 | return -1; | |
4352 | if (di_a->total_avail < di_b->total_avail) | |
4353 | return 1; | |
4354 | return 0; | |
b2117a39 | 4355 | } |
0b86a832 | 4356 | |
53b381b3 DW |
4357 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4358 | { | |
4359 | /* TODO allow them to set a preferred stripe size */ | |
4360 | return 64 * 1024; | |
4361 | } | |
4362 | ||
4363 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4364 | { | |
ffe2d203 | 4365 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4366 | return; |
4367 | ||
ceda0864 | 4368 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4369 | } |
4370 | ||
23f8f9b7 GH |
4371 | #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \ |
4372 | - sizeof(struct btrfs_item) \ | |
4373 | - sizeof(struct btrfs_chunk)) \ | |
4374 | / sizeof(struct btrfs_stripe) + 1) | |
4375 | ||
4376 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4377 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4378 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4379 | / sizeof(struct btrfs_stripe) + 1) | |
4380 | ||
73c5de00 | 4381 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4382 | struct btrfs_root *extent_root, u64 start, |
4383 | u64 type) | |
b2117a39 | 4384 | { |
73c5de00 AJ |
4385 | struct btrfs_fs_info *info = extent_root->fs_info; |
4386 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4387 | struct list_head *cur; | |
4388 | struct map_lookup *map = NULL; | |
4389 | struct extent_map_tree *em_tree; | |
4390 | struct extent_map *em; | |
4391 | struct btrfs_device_info *devices_info = NULL; | |
4392 | u64 total_avail; | |
4393 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4394 | int data_stripes; /* number of stripes that count for |
4395 | block group size */ | |
73c5de00 AJ |
4396 | int sub_stripes; /* sub_stripes info for map */ |
4397 | int dev_stripes; /* stripes per dev */ | |
4398 | int devs_max; /* max devs to use */ | |
4399 | int devs_min; /* min devs needed */ | |
4400 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4401 | int ncopies; /* how many copies to data has */ | |
4402 | int ret; | |
4403 | u64 max_stripe_size; | |
4404 | u64 max_chunk_size; | |
4405 | u64 stripe_size; | |
4406 | u64 num_bytes; | |
53b381b3 | 4407 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4408 | int ndevs; |
4409 | int i; | |
4410 | int j; | |
31e50229 | 4411 | int index; |
593060d7 | 4412 | |
0c460c0d | 4413 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4414 | |
73c5de00 AJ |
4415 | if (list_empty(&fs_devices->alloc_list)) |
4416 | return -ENOSPC; | |
b2117a39 | 4417 | |
31e50229 | 4418 | index = __get_raid_index(type); |
73c5de00 | 4419 | |
31e50229 LB |
4420 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4421 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4422 | devs_max = btrfs_raid_array[index].devs_max; | |
4423 | devs_min = btrfs_raid_array[index].devs_min; | |
4424 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4425 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4426 | |
9b3f68b9 | 4427 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
4428 | max_stripe_size = 1024 * 1024 * 1024; |
4429 | max_chunk_size = 10 * max_stripe_size; | |
23f8f9b7 GH |
4430 | if (!devs_max) |
4431 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4432 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
4433 | /* for larger filesystems, use larger metadata chunks */ |
4434 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
4435 | max_stripe_size = 1024 * 1024 * 1024; | |
4436 | else | |
4437 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 4438 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4439 | if (!devs_max) |
4440 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4441 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 4442 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 | 4443 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4444 | if (!devs_max) |
4445 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4446 | } else { |
351fd353 | 4447 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4448 | type); |
4449 | BUG_ON(1); | |
9b3f68b9 CM |
4450 | } |
4451 | ||
2b82032c YZ |
4452 | /* we don't want a chunk larger than 10% of writeable space */ |
4453 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4454 | max_chunk_size); | |
9b3f68b9 | 4455 | |
31e818fe | 4456 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4457 | GFP_NOFS); |
4458 | if (!devices_info) | |
4459 | return -ENOMEM; | |
0cad8a11 | 4460 | |
73c5de00 | 4461 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4462 | |
9f680ce0 | 4463 | /* |
73c5de00 AJ |
4464 | * in the first pass through the devices list, we gather information |
4465 | * about the available holes on each device. | |
9f680ce0 | 4466 | */ |
73c5de00 AJ |
4467 | ndevs = 0; |
4468 | while (cur != &fs_devices->alloc_list) { | |
4469 | struct btrfs_device *device; | |
4470 | u64 max_avail; | |
4471 | u64 dev_offset; | |
b2117a39 | 4472 | |
73c5de00 | 4473 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4474 | |
73c5de00 | 4475 | cur = cur->next; |
b2117a39 | 4476 | |
73c5de00 | 4477 | if (!device->writeable) { |
31b1a2bd | 4478 | WARN(1, KERN_ERR |
efe120a0 | 4479 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4480 | continue; |
4481 | } | |
b2117a39 | 4482 | |
63a212ab SB |
4483 | if (!device->in_fs_metadata || |
4484 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4485 | continue; |
b2117a39 | 4486 | |
73c5de00 AJ |
4487 | if (device->total_bytes > device->bytes_used) |
4488 | total_avail = device->total_bytes - device->bytes_used; | |
4489 | else | |
4490 | total_avail = 0; | |
38c01b96 | 4491 | |
4492 | /* If there is no space on this device, skip it. */ | |
4493 | if (total_avail == 0) | |
4494 | continue; | |
b2117a39 | 4495 | |
6df9a95e | 4496 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4497 | max_stripe_size * dev_stripes, |
4498 | &dev_offset, &max_avail); | |
4499 | if (ret && ret != -ENOSPC) | |
4500 | goto error; | |
b2117a39 | 4501 | |
73c5de00 AJ |
4502 | if (ret == 0) |
4503 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4504 | |
73c5de00 AJ |
4505 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4506 | continue; | |
b2117a39 | 4507 | |
063d006f ES |
4508 | if (ndevs == fs_devices->rw_devices) { |
4509 | WARN(1, "%s: found more than %llu devices\n", | |
4510 | __func__, fs_devices->rw_devices); | |
4511 | break; | |
4512 | } | |
73c5de00 AJ |
4513 | devices_info[ndevs].dev_offset = dev_offset; |
4514 | devices_info[ndevs].max_avail = max_avail; | |
4515 | devices_info[ndevs].total_avail = total_avail; | |
4516 | devices_info[ndevs].dev = device; | |
4517 | ++ndevs; | |
4518 | } | |
b2117a39 | 4519 | |
73c5de00 AJ |
4520 | /* |
4521 | * now sort the devices by hole size / available space | |
4522 | */ | |
4523 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4524 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4525 | |
73c5de00 AJ |
4526 | /* round down to number of usable stripes */ |
4527 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4528 | |
73c5de00 AJ |
4529 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4530 | ret = -ENOSPC; | |
4531 | goto error; | |
b2117a39 | 4532 | } |
9f680ce0 | 4533 | |
73c5de00 AJ |
4534 | if (devs_max && ndevs > devs_max) |
4535 | ndevs = devs_max; | |
4536 | /* | |
4537 | * the primary goal is to maximize the number of stripes, so use as many | |
4538 | * devices as possible, even if the stripes are not maximum sized. | |
4539 | */ | |
4540 | stripe_size = devices_info[ndevs-1].max_avail; | |
4541 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4542 | |
53b381b3 DW |
4543 | /* |
4544 | * this will have to be fixed for RAID1 and RAID10 over | |
4545 | * more drives | |
4546 | */ | |
4547 | data_stripes = num_stripes / ncopies; | |
4548 | ||
53b381b3 DW |
4549 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4550 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4551 | btrfs_super_stripesize(info->super_copy)); | |
4552 | data_stripes = num_stripes - 1; | |
4553 | } | |
4554 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4555 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4556 | btrfs_super_stripesize(info->super_copy)); | |
4557 | data_stripes = num_stripes - 2; | |
4558 | } | |
86db2578 CM |
4559 | |
4560 | /* | |
4561 | * Use the number of data stripes to figure out how big this chunk | |
4562 | * is really going to be in terms of logical address space, | |
4563 | * and compare that answer with the max chunk size | |
4564 | */ | |
4565 | if (stripe_size * data_stripes > max_chunk_size) { | |
4566 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4567 | |
4568 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4569 | |
4570 | /* bump the answer up to a 16MB boundary */ | |
4571 | stripe_size = (stripe_size + mask) & ~mask; | |
4572 | ||
4573 | /* but don't go higher than the limits we found | |
4574 | * while searching for free extents | |
4575 | */ | |
4576 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4577 | stripe_size = devices_info[ndevs-1].max_avail; | |
4578 | } | |
4579 | ||
b8b93add | 4580 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4581 | |
4582 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4583 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4584 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4585 | |
4586 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4587 | if (!map) { | |
4588 | ret = -ENOMEM; | |
4589 | goto error; | |
4590 | } | |
4591 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4592 | |
73c5de00 AJ |
4593 | for (i = 0; i < ndevs; ++i) { |
4594 | for (j = 0; j < dev_stripes; ++j) { | |
4595 | int s = i * dev_stripes + j; | |
4596 | map->stripes[s].dev = devices_info[i].dev; | |
4597 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4598 | j * stripe_size; | |
6324fbf3 | 4599 | } |
6324fbf3 | 4600 | } |
2b82032c | 4601 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4602 | map->stripe_len = raid_stripe_len; |
4603 | map->io_align = raid_stripe_len; | |
4604 | map->io_width = raid_stripe_len; | |
2b82032c | 4605 | map->type = type; |
2b82032c | 4606 | map->sub_stripes = sub_stripes; |
0b86a832 | 4607 | |
53b381b3 | 4608 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4609 | |
73c5de00 | 4610 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4611 | |
172ddd60 | 4612 | em = alloc_extent_map(); |
2b82032c | 4613 | if (!em) { |
298a8f9c | 4614 | kfree(map); |
b2117a39 MX |
4615 | ret = -ENOMEM; |
4616 | goto error; | |
593060d7 | 4617 | } |
298a8f9c | 4618 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
2b82032c YZ |
4619 | em->bdev = (struct block_device *)map; |
4620 | em->start = start; | |
73c5de00 | 4621 | em->len = num_bytes; |
2b82032c YZ |
4622 | em->block_start = 0; |
4623 | em->block_len = em->len; | |
6df9a95e | 4624 | em->orig_block_len = stripe_size; |
593060d7 | 4625 | |
2b82032c | 4626 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4627 | write_lock(&em_tree->lock); |
09a2a8f9 | 4628 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4629 | if (!ret) { |
4630 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4631 | atomic_inc(&em->refs); | |
4632 | } | |
890871be | 4633 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4634 | if (ret) { |
4635 | free_extent_map(em); | |
1dd4602f | 4636 | goto error; |
0f5d42b2 | 4637 | } |
0b86a832 | 4638 | |
04487488 JB |
4639 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4640 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4641 | start, num_bytes); | |
6df9a95e JB |
4642 | if (ret) |
4643 | goto error_del_extent; | |
2b82032c | 4644 | |
7cc8e58d MX |
4645 | for (i = 0; i < map->num_stripes; i++) { |
4646 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4647 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4648 | } | |
43530c46 | 4649 | |
1c116187 MX |
4650 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4651 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4652 | map->num_stripes); | |
4653 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4654 | ||
0f5d42b2 | 4655 | free_extent_map(em); |
53b381b3 DW |
4656 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4657 | ||
b2117a39 | 4658 | kfree(devices_info); |
2b82032c | 4659 | return 0; |
b2117a39 | 4660 | |
6df9a95e | 4661 | error_del_extent: |
0f5d42b2 JB |
4662 | write_lock(&em_tree->lock); |
4663 | remove_extent_mapping(em_tree, em); | |
4664 | write_unlock(&em_tree->lock); | |
4665 | ||
4666 | /* One for our allocation */ | |
4667 | free_extent_map(em); | |
4668 | /* One for the tree reference */ | |
4669 | free_extent_map(em); | |
495e64f4 FM |
4670 | /* One for the pending_chunks list reference */ |
4671 | free_extent_map(em); | |
b2117a39 | 4672 | error: |
b2117a39 MX |
4673 | kfree(devices_info); |
4674 | return ret; | |
2b82032c YZ |
4675 | } |
4676 | ||
6df9a95e | 4677 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4678 | struct btrfs_root *extent_root, |
6df9a95e | 4679 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4680 | { |
2b82032c YZ |
4681 | struct btrfs_key key; |
4682 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4683 | struct btrfs_device *device; | |
4684 | struct btrfs_chunk *chunk; | |
4685 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4686 | struct extent_map_tree *em_tree; |
4687 | struct extent_map *em; | |
4688 | struct map_lookup *map; | |
4689 | size_t item_size; | |
4690 | u64 dev_offset; | |
4691 | u64 stripe_size; | |
4692 | int i = 0; | |
2b82032c YZ |
4693 | int ret; |
4694 | ||
6df9a95e JB |
4695 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4696 | read_lock(&em_tree->lock); | |
4697 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4698 | read_unlock(&em_tree->lock); | |
4699 | ||
4700 | if (!em) { | |
4701 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4702 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4703 | return -EINVAL; | |
4704 | } | |
4705 | ||
4706 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4707 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
351fd353 | 4708 | " %Lu-%Lu, found %Lu-%Lu", chunk_offset, |
6df9a95e JB |
4709 | chunk_size, em->start, em->len); |
4710 | free_extent_map(em); | |
4711 | return -EINVAL; | |
4712 | } | |
4713 | ||
4714 | map = (struct map_lookup *)em->bdev; | |
4715 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
4716 | stripe_size = em->orig_block_len; | |
4717 | ||
2b82032c | 4718 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4719 | if (!chunk) { |
4720 | ret = -ENOMEM; | |
4721 | goto out; | |
4722 | } | |
4723 | ||
4724 | for (i = 0; i < map->num_stripes; i++) { | |
4725 | device = map->stripes[i].dev; | |
4726 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4727 | |
0b86a832 | 4728 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4729 | if (ret) |
6df9a95e JB |
4730 | goto out; |
4731 | ret = btrfs_alloc_dev_extent(trans, device, | |
4732 | chunk_root->root_key.objectid, | |
4733 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4734 | chunk_offset, dev_offset, | |
4735 | stripe_size); | |
4736 | if (ret) | |
4737 | goto out; | |
2b82032c YZ |
4738 | } |
4739 | ||
2b82032c | 4740 | stripe = &chunk->stripe; |
6df9a95e JB |
4741 | for (i = 0; i < map->num_stripes; i++) { |
4742 | device = map->stripes[i].dev; | |
4743 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4744 | |
e17cade2 CM |
4745 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4746 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4747 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4748 | stripe++; |
0b86a832 CM |
4749 | } |
4750 | ||
2b82032c | 4751 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4752 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4753 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4754 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4755 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4756 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4757 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4758 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4759 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4760 | |
2b82032c YZ |
4761 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4762 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4763 | key.offset = chunk_offset; | |
0b86a832 | 4764 | |
2b82032c | 4765 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4766 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4767 | /* | |
4768 | * TODO: Cleanup of inserted chunk root in case of | |
4769 | * failure. | |
4770 | */ | |
125ccb0a | 4771 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4772 | item_size); |
8f18cf13 | 4773 | } |
1abe9b8a | 4774 | |
6df9a95e | 4775 | out: |
0b86a832 | 4776 | kfree(chunk); |
6df9a95e | 4777 | free_extent_map(em); |
4ed1d16e | 4778 | return ret; |
2b82032c | 4779 | } |
0b86a832 | 4780 | |
2b82032c YZ |
4781 | /* |
4782 | * Chunk allocation falls into two parts. The first part does works | |
4783 | * that make the new allocated chunk useable, but not do any operation | |
4784 | * that modifies the chunk tree. The second part does the works that | |
4785 | * require modifying the chunk tree. This division is important for the | |
4786 | * bootstrap process of adding storage to a seed btrfs. | |
4787 | */ | |
4788 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4789 | struct btrfs_root *extent_root, u64 type) | |
4790 | { | |
4791 | u64 chunk_offset; | |
2b82032c | 4792 | |
a9629596 | 4793 | ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex)); |
6df9a95e JB |
4794 | chunk_offset = find_next_chunk(extent_root->fs_info); |
4795 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
4796 | } |
4797 | ||
d397712b | 4798 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4799 | struct btrfs_root *root, |
4800 | struct btrfs_device *device) | |
4801 | { | |
4802 | u64 chunk_offset; | |
4803 | u64 sys_chunk_offset; | |
2b82032c | 4804 | u64 alloc_profile; |
2b82032c YZ |
4805 | struct btrfs_fs_info *fs_info = root->fs_info; |
4806 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4807 | int ret; | |
4808 | ||
6df9a95e | 4809 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 4810 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
4811 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
4812 | alloc_profile); | |
79787eaa JM |
4813 | if (ret) |
4814 | return ret; | |
2b82032c | 4815 | |
6df9a95e | 4816 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 4817 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
4818 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
4819 | alloc_profile); | |
79787eaa | 4820 | return ret; |
2b82032c YZ |
4821 | } |
4822 | ||
d20983b4 MX |
4823 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
4824 | { | |
4825 | int max_errors; | |
4826 | ||
4827 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4828 | BTRFS_BLOCK_GROUP_RAID10 | | |
4829 | BTRFS_BLOCK_GROUP_RAID5 | | |
4830 | BTRFS_BLOCK_GROUP_DUP)) { | |
4831 | max_errors = 1; | |
4832 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4833 | max_errors = 2; | |
4834 | } else { | |
4835 | max_errors = 0; | |
005d6427 | 4836 | } |
2b82032c | 4837 | |
d20983b4 | 4838 | return max_errors; |
2b82032c YZ |
4839 | } |
4840 | ||
4841 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4842 | { | |
4843 | struct extent_map *em; | |
4844 | struct map_lookup *map; | |
4845 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4846 | int readonly = 0; | |
d20983b4 | 4847 | int miss_ndevs = 0; |
2b82032c YZ |
4848 | int i; |
4849 | ||
890871be | 4850 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4851 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4852 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4853 | if (!em) |
4854 | return 1; | |
4855 | ||
4856 | map = (struct map_lookup *)em->bdev; | |
4857 | for (i = 0; i < map->num_stripes; i++) { | |
d20983b4 MX |
4858 | if (map->stripes[i].dev->missing) { |
4859 | miss_ndevs++; | |
4860 | continue; | |
4861 | } | |
4862 | ||
2b82032c YZ |
4863 | if (!map->stripes[i].dev->writeable) { |
4864 | readonly = 1; | |
d20983b4 | 4865 | goto end; |
2b82032c YZ |
4866 | } |
4867 | } | |
d20983b4 MX |
4868 | |
4869 | /* | |
4870 | * If the number of missing devices is larger than max errors, | |
4871 | * we can not write the data into that chunk successfully, so | |
4872 | * set it readonly. | |
4873 | */ | |
4874 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
4875 | readonly = 1; | |
4876 | end: | |
0b86a832 | 4877 | free_extent_map(em); |
2b82032c | 4878 | return readonly; |
0b86a832 CM |
4879 | } |
4880 | ||
4881 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4882 | { | |
a8067e02 | 4883 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
4884 | } |
4885 | ||
4886 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4887 | { | |
4888 | struct extent_map *em; | |
4889 | ||
d397712b | 4890 | while (1) { |
890871be | 4891 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
4892 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
4893 | if (em) | |
4894 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 4895 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
4896 | if (!em) |
4897 | break; | |
0b86a832 CM |
4898 | /* once for us */ |
4899 | free_extent_map(em); | |
4900 | /* once for the tree */ | |
4901 | free_extent_map(em); | |
4902 | } | |
4903 | } | |
4904 | ||
5d964051 | 4905 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 4906 | { |
5d964051 | 4907 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
4908 | struct extent_map *em; |
4909 | struct map_lookup *map; | |
4910 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4911 | int ret; | |
4912 | ||
890871be | 4913 | read_lock(&em_tree->lock); |
f188591e | 4914 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 4915 | read_unlock(&em_tree->lock); |
f188591e | 4916 | |
fb7669b5 JB |
4917 | /* |
4918 | * We could return errors for these cases, but that could get ugly and | |
4919 | * we'd probably do the same thing which is just not do anything else | |
4920 | * and exit, so return 1 so the callers don't try to use other copies. | |
4921 | */ | |
4922 | if (!em) { | |
351fd353 | 4923 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
4924 | logical+len); |
4925 | return 1; | |
4926 | } | |
4927 | ||
4928 | if (em->start > logical || em->start + em->len < logical) { | |
ccf39f92 | 4929 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " |
351fd353 | 4930 | "%Lu-%Lu", logical, logical+len, em->start, |
fb7669b5 | 4931 | em->start + em->len); |
7d3d1744 | 4932 | free_extent_map(em); |
fb7669b5 JB |
4933 | return 1; |
4934 | } | |
4935 | ||
f188591e CM |
4936 | map = (struct map_lookup *)em->bdev; |
4937 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
4938 | ret = map->num_stripes; | |
321aecc6 CM |
4939 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
4940 | ret = map->sub_stripes; | |
53b381b3 DW |
4941 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
4942 | ret = 2; | |
4943 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4944 | ret = 3; | |
f188591e CM |
4945 | else |
4946 | ret = 1; | |
4947 | free_extent_map(em); | |
ad6d620e SB |
4948 | |
4949 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
4950 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
4951 | ret++; | |
4952 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
4953 | ||
f188591e CM |
4954 | return ret; |
4955 | } | |
4956 | ||
53b381b3 DW |
4957 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
4958 | struct btrfs_mapping_tree *map_tree, | |
4959 | u64 logical) | |
4960 | { | |
4961 | struct extent_map *em; | |
4962 | struct map_lookup *map; | |
4963 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4964 | unsigned long len = root->sectorsize; | |
4965 | ||
4966 | read_lock(&em_tree->lock); | |
4967 | em = lookup_extent_mapping(em_tree, logical, len); | |
4968 | read_unlock(&em_tree->lock); | |
4969 | BUG_ON(!em); | |
4970 | ||
4971 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4972 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 4973 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 4974 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
4975 | free_extent_map(em); |
4976 | return len; | |
4977 | } | |
4978 | ||
4979 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
4980 | u64 logical, u64 len, int mirror_num) | |
4981 | { | |
4982 | struct extent_map *em; | |
4983 | struct map_lookup *map; | |
4984 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4985 | int ret = 0; | |
4986 | ||
4987 | read_lock(&em_tree->lock); | |
4988 | em = lookup_extent_mapping(em_tree, logical, len); | |
4989 | read_unlock(&em_tree->lock); | |
4990 | BUG_ON(!em); | |
4991 | ||
4992 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4993 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 4994 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
4995 | ret = 1; |
4996 | free_extent_map(em); | |
4997 | return ret; | |
4998 | } | |
4999 | ||
30d9861f SB |
5000 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5001 | struct map_lookup *map, int first, int num, | |
5002 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5003 | { |
5004 | int i; | |
30d9861f SB |
5005 | int tolerance; |
5006 | struct btrfs_device *srcdev; | |
5007 | ||
5008 | if (dev_replace_is_ongoing && | |
5009 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5010 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5011 | srcdev = fs_info->dev_replace.srcdev; | |
5012 | else | |
5013 | srcdev = NULL; | |
5014 | ||
5015 | /* | |
5016 | * try to avoid the drive that is the source drive for a | |
5017 | * dev-replace procedure, only choose it if no other non-missing | |
5018 | * mirror is available | |
5019 | */ | |
5020 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5021 | if (map->stripes[optimal].dev->bdev && | |
5022 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5023 | return optimal; | |
5024 | for (i = first; i < first + num; i++) { | |
5025 | if (map->stripes[i].dev->bdev && | |
5026 | (tolerance || map->stripes[i].dev != srcdev)) | |
5027 | return i; | |
5028 | } | |
dfe25020 | 5029 | } |
30d9861f | 5030 | |
dfe25020 CM |
5031 | /* we couldn't find one that doesn't fail. Just return something |
5032 | * and the io error handling code will clean up eventually | |
5033 | */ | |
5034 | return optimal; | |
5035 | } | |
5036 | ||
53b381b3 DW |
5037 | static inline int parity_smaller(u64 a, u64 b) |
5038 | { | |
5039 | return a > b; | |
5040 | } | |
5041 | ||
5042 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5043 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5044 | { |
5045 | struct btrfs_bio_stripe s; | |
5046 | int i; | |
5047 | u64 l; | |
5048 | int again = 1; | |
5049 | ||
5050 | while (again) { | |
5051 | again = 0; | |
cc7539ed | 5052 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5053 | if (parity_smaller(bbio->raid_map[i], |
5054 | bbio->raid_map[i+1])) { | |
53b381b3 | 5055 | s = bbio->stripes[i]; |
8e5cfb55 | 5056 | l = bbio->raid_map[i]; |
53b381b3 | 5057 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5058 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5059 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5060 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5061 | |
53b381b3 DW |
5062 | again = 1; |
5063 | } | |
5064 | } | |
5065 | } | |
5066 | } | |
5067 | ||
6e9606d2 ZL |
5068 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5069 | { | |
5070 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5071 | /* the size of the btrfs_bio */ |
6e9606d2 | 5072 | sizeof(struct btrfs_bio) + |
e57cf21e | 5073 | /* plus the variable array for the stripes */ |
6e9606d2 | 5074 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5075 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5076 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5077 | /* |
5078 | * plus the raid_map, which includes both the tgt dev | |
5079 | * and the stripes | |
5080 | */ | |
5081 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5082 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5083 | |
5084 | atomic_set(&bbio->error, 0); | |
5085 | atomic_set(&bbio->refs, 1); | |
5086 | ||
5087 | return bbio; | |
5088 | } | |
5089 | ||
5090 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5091 | { | |
5092 | WARN_ON(!atomic_read(&bbio->refs)); | |
5093 | atomic_inc(&bbio->refs); | |
5094 | } | |
5095 | ||
5096 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5097 | { | |
5098 | if (!bbio) | |
5099 | return; | |
5100 | if (atomic_dec_and_test(&bbio->refs)) | |
5101 | kfree(bbio); | |
5102 | } | |
5103 | ||
3ec706c8 | 5104 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5105 | u64 logical, u64 *length, |
a1d3c478 | 5106 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5107 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5108 | { |
5109 | struct extent_map *em; | |
5110 | struct map_lookup *map; | |
3ec706c8 | 5111 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
5112 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5113 | u64 offset; | |
593060d7 | 5114 | u64 stripe_offset; |
fce3bb9a | 5115 | u64 stripe_end_offset; |
593060d7 | 5116 | u64 stripe_nr; |
fce3bb9a LD |
5117 | u64 stripe_nr_orig; |
5118 | u64 stripe_nr_end; | |
53b381b3 | 5119 | u64 stripe_len; |
9d644a62 | 5120 | u32 stripe_index; |
cea9e445 | 5121 | int i; |
de11cc12 | 5122 | int ret = 0; |
f2d8d74d | 5123 | int num_stripes; |
a236aed1 | 5124 | int max_errors = 0; |
2c8cdd6e | 5125 | int tgtdev_indexes = 0; |
a1d3c478 | 5126 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5127 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5128 | int dev_replace_is_ongoing = 0; | |
5129 | int num_alloc_stripes; | |
ad6d620e SB |
5130 | int patch_the_first_stripe_for_dev_replace = 0; |
5131 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5132 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5133 | |
890871be | 5134 | read_lock(&em_tree->lock); |
0b86a832 | 5135 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 5136 | read_unlock(&em_tree->lock); |
f2d8d74d | 5137 | |
3b951516 | 5138 | if (!em) { |
c2cf52eb | 5139 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 5140 | logical, *length); |
9bb91873 JB |
5141 | return -EINVAL; |
5142 | } | |
5143 | ||
5144 | if (em->start > logical || em->start + em->len < logical) { | |
5145 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
351fd353 | 5146 | "found %Lu-%Lu", logical, em->start, |
9bb91873 | 5147 | em->start + em->len); |
7d3d1744 | 5148 | free_extent_map(em); |
9bb91873 | 5149 | return -EINVAL; |
3b951516 | 5150 | } |
0b86a832 | 5151 | |
0b86a832 CM |
5152 | map = (struct map_lookup *)em->bdev; |
5153 | offset = logical - em->start; | |
593060d7 | 5154 | |
53b381b3 | 5155 | stripe_len = map->stripe_len; |
593060d7 CM |
5156 | stripe_nr = offset; |
5157 | /* | |
5158 | * stripe_nr counts the total number of stripes we have to stride | |
5159 | * to get to this block | |
5160 | */ | |
47c5713f | 5161 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5162 | |
53b381b3 | 5163 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
5164 | BUG_ON(offset < stripe_offset); |
5165 | ||
5166 | /* stripe_offset is the offset of this block in its stripe*/ | |
5167 | stripe_offset = offset - stripe_offset; | |
5168 | ||
53b381b3 | 5169 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5170 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5171 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5172 | raid56_full_stripe_start = offset; | |
5173 | ||
5174 | /* allow a write of a full stripe, but make sure we don't | |
5175 | * allow straddling of stripes | |
5176 | */ | |
47c5713f DS |
5177 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5178 | full_stripe_len); | |
53b381b3 DW |
5179 | raid56_full_stripe_start *= full_stripe_len; |
5180 | } | |
5181 | ||
5182 | if (rw & REQ_DISCARD) { | |
5183 | /* we don't discard raid56 yet */ | |
ffe2d203 | 5184 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5185 | ret = -EOPNOTSUPP; |
5186 | goto out; | |
5187 | } | |
fce3bb9a | 5188 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
5189 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
5190 | u64 max_len; | |
5191 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5192 | For other RAID types and for RAID[56] reads, just allow a single | |
5193 | stripe (on a single disk). */ | |
ffe2d203 | 5194 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
53b381b3 DW |
5195 | (rw & REQ_WRITE)) { |
5196 | max_len = stripe_len * nr_data_stripes(map) - | |
5197 | (offset - raid56_full_stripe_start); | |
5198 | } else { | |
5199 | /* we limit the length of each bio to what fits in a stripe */ | |
5200 | max_len = stripe_len - stripe_offset; | |
5201 | } | |
5202 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5203 | } else { |
5204 | *length = em->len - offset; | |
5205 | } | |
f2d8d74d | 5206 | |
53b381b3 DW |
5207 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5208 | it cares about is the length */ | |
a1d3c478 | 5209 | if (!bbio_ret) |
cea9e445 CM |
5210 | goto out; |
5211 | ||
472262f3 SB |
5212 | btrfs_dev_replace_lock(dev_replace); |
5213 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
5214 | if (!dev_replace_is_ongoing) | |
5215 | btrfs_dev_replace_unlock(dev_replace); | |
5216 | ||
ad6d620e SB |
5217 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
5218 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
5219 | dev_replace->tgtdev != NULL) { | |
5220 | /* | |
5221 | * in dev-replace case, for repair case (that's the only | |
5222 | * case where the mirror is selected explicitly when | |
5223 | * calling btrfs_map_block), blocks left of the left cursor | |
5224 | * can also be read from the target drive. | |
5225 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5226 | * the last one to the array of stripes. For READ, it also | |
5227 | * needs to be supported using the same mirror number. | |
5228 | * If the requested block is not left of the left cursor, | |
5229 | * EIO is returned. This can happen because btrfs_num_copies() | |
5230 | * returns one more in the dev-replace case. | |
5231 | */ | |
5232 | u64 tmp_length = *length; | |
5233 | struct btrfs_bio *tmp_bbio = NULL; | |
5234 | int tmp_num_stripes; | |
5235 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5236 | int index_srcdev = 0; | |
5237 | int found = 0; | |
5238 | u64 physical_of_found = 0; | |
5239 | ||
5240 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
8e5cfb55 | 5241 | logical, &tmp_length, &tmp_bbio, 0, 0); |
ad6d620e SB |
5242 | if (ret) { |
5243 | WARN_ON(tmp_bbio != NULL); | |
5244 | goto out; | |
5245 | } | |
5246 | ||
5247 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5248 | if (mirror_num > tmp_num_stripes) { | |
5249 | /* | |
5250 | * REQ_GET_READ_MIRRORS does not contain this | |
5251 | * mirror, that means that the requested area | |
5252 | * is not left of the left cursor | |
5253 | */ | |
5254 | ret = -EIO; | |
6e9606d2 | 5255 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5256 | goto out; |
5257 | } | |
5258 | ||
5259 | /* | |
5260 | * process the rest of the function using the mirror_num | |
5261 | * of the source drive. Therefore look it up first. | |
5262 | * At the end, patch the device pointer to the one of the | |
5263 | * target drive. | |
5264 | */ | |
5265 | for (i = 0; i < tmp_num_stripes; i++) { | |
5266 | if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) { | |
5267 | /* | |
5268 | * In case of DUP, in order to keep it | |
5269 | * simple, only add the mirror with the | |
5270 | * lowest physical address | |
5271 | */ | |
5272 | if (found && | |
5273 | physical_of_found <= | |
5274 | tmp_bbio->stripes[i].physical) | |
5275 | continue; | |
5276 | index_srcdev = i; | |
5277 | found = 1; | |
5278 | physical_of_found = | |
5279 | tmp_bbio->stripes[i].physical; | |
5280 | } | |
5281 | } | |
5282 | ||
5283 | if (found) { | |
5284 | mirror_num = index_srcdev + 1; | |
5285 | patch_the_first_stripe_for_dev_replace = 1; | |
5286 | physical_to_patch_in_first_stripe = physical_of_found; | |
5287 | } else { | |
5288 | WARN_ON(1); | |
5289 | ret = -EIO; | |
6e9606d2 | 5290 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5291 | goto out; |
5292 | } | |
5293 | ||
6e9606d2 | 5294 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5295 | } else if (mirror_num > map->num_stripes) { |
5296 | mirror_num = 0; | |
5297 | } | |
5298 | ||
f2d8d74d | 5299 | num_stripes = 1; |
cea9e445 | 5300 | stripe_index = 0; |
fce3bb9a | 5301 | stripe_nr_orig = stripe_nr; |
fda2832f | 5302 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
b8b93add | 5303 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); |
fce3bb9a LD |
5304 | stripe_end_offset = stripe_nr_end * map->stripe_len - |
5305 | (offset + *length); | |
53b381b3 | 5306 | |
fce3bb9a LD |
5307 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5308 | if (rw & REQ_DISCARD) | |
5309 | num_stripes = min_t(u64, map->num_stripes, | |
5310 | stripe_nr_end - stripe_nr_orig); | |
47c5713f DS |
5311 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5312 | &stripe_index); | |
28e1cc7d MX |
5313 | if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))) |
5314 | mirror_num = 1; | |
fce3bb9a | 5315 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
29a8d9a0 | 5316 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5317 | num_stripes = map->num_stripes; |
2fff734f | 5318 | else if (mirror_num) |
f188591e | 5319 | stripe_index = mirror_num - 1; |
dfe25020 | 5320 | else { |
30d9861f | 5321 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5322 | map->num_stripes, |
30d9861f SB |
5323 | current->pid % map->num_stripes, |
5324 | dev_replace_is_ongoing); | |
a1d3c478 | 5325 | mirror_num = stripe_index + 1; |
dfe25020 | 5326 | } |
2fff734f | 5327 | |
611f0e00 | 5328 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 5329 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 5330 | num_stripes = map->num_stripes; |
a1d3c478 | 5331 | } else if (mirror_num) { |
f188591e | 5332 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5333 | } else { |
5334 | mirror_num = 1; | |
5335 | } | |
2fff734f | 5336 | |
321aecc6 | 5337 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5338 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5339 | |
47c5713f | 5340 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5341 | stripe_index *= map->sub_stripes; |
5342 | ||
29a8d9a0 | 5343 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5344 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
5345 | else if (rw & REQ_DISCARD) |
5346 | num_stripes = min_t(u64, map->sub_stripes * | |
5347 | (stripe_nr_end - stripe_nr_orig), | |
5348 | map->num_stripes); | |
321aecc6 CM |
5349 | else if (mirror_num) |
5350 | stripe_index += mirror_num - 1; | |
dfe25020 | 5351 | else { |
3e74317a | 5352 | int old_stripe_index = stripe_index; |
30d9861f SB |
5353 | stripe_index = find_live_mirror(fs_info, map, |
5354 | stripe_index, | |
dfe25020 | 5355 | map->sub_stripes, stripe_index + |
30d9861f SB |
5356 | current->pid % map->sub_stripes, |
5357 | dev_replace_is_ongoing); | |
3e74317a | 5358 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5359 | } |
53b381b3 | 5360 | |
ffe2d203 | 5361 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5362 | if (need_raid_map && |
af8e2d1d MX |
5363 | ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5364 | mirror_num > 1)) { | |
53b381b3 | 5365 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5366 | stripe_nr = div_u64(raid56_full_stripe_start, |
5367 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5368 | |
5369 | /* RAID[56] write or recovery. Return all stripes */ | |
5370 | num_stripes = map->num_stripes; | |
5371 | max_errors = nr_parity_stripes(map); | |
5372 | ||
53b381b3 DW |
5373 | *length = map->stripe_len; |
5374 | stripe_index = 0; | |
5375 | stripe_offset = 0; | |
5376 | } else { | |
5377 | /* | |
5378 | * Mirror #0 or #1 means the original data block. | |
5379 | * Mirror #2 is RAID5 parity block. | |
5380 | * Mirror #3 is RAID6 Q block. | |
5381 | */ | |
47c5713f DS |
5382 | stripe_nr = div_u64_rem(stripe_nr, |
5383 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5384 | if (mirror_num > 1) |
5385 | stripe_index = nr_data_stripes(map) + | |
5386 | mirror_num - 2; | |
5387 | ||
5388 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5389 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5390 | &stripe_index); | |
28e1cc7d MX |
5391 | if (!(rw & (REQ_WRITE | REQ_DISCARD | |
5392 | REQ_GET_READ_MIRRORS)) && mirror_num <= 1) | |
5393 | mirror_num = 1; | |
53b381b3 | 5394 | } |
8790d502 CM |
5395 | } else { |
5396 | /* | |
47c5713f DS |
5397 | * after this, stripe_nr is the number of stripes on this |
5398 | * device we have to walk to find the data, and stripe_index is | |
5399 | * the number of our device in the stripe array | |
8790d502 | 5400 | */ |
47c5713f DS |
5401 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5402 | &stripe_index); | |
a1d3c478 | 5403 | mirror_num = stripe_index + 1; |
8790d502 | 5404 | } |
593060d7 | 5405 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 5406 | |
472262f3 | 5407 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
5408 | if (dev_replace_is_ongoing) { |
5409 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
5410 | num_alloc_stripes <<= 1; | |
5411 | if (rw & REQ_GET_READ_MIRRORS) | |
5412 | num_alloc_stripes++; | |
2c8cdd6e | 5413 | tgtdev_indexes = num_stripes; |
ad6d620e | 5414 | } |
2c8cdd6e | 5415 | |
6e9606d2 | 5416 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5417 | if (!bbio) { |
5418 | ret = -ENOMEM; | |
5419 | goto out; | |
5420 | } | |
2c8cdd6e MX |
5421 | if (dev_replace_is_ongoing) |
5422 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
de11cc12 | 5423 | |
8e5cfb55 | 5424 | /* build raid_map */ |
ffe2d203 | 5425 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
8e5cfb55 ZL |
5426 | need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5427 | mirror_num > 1)) { | |
5428 | u64 tmp; | |
9d644a62 | 5429 | unsigned rot; |
8e5cfb55 ZL |
5430 | |
5431 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5432 | sizeof(struct btrfs_bio_stripe) * | |
5433 | num_alloc_stripes + | |
5434 | sizeof(int) * tgtdev_indexes); | |
5435 | ||
5436 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5437 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5438 | |
5439 | /* Fill in the logical address of each stripe */ | |
5440 | tmp = stripe_nr * nr_data_stripes(map); | |
5441 | for (i = 0; i < nr_data_stripes(map); i++) | |
5442 | bbio->raid_map[(i+rot) % num_stripes] = | |
5443 | em->start + (tmp + i) * map->stripe_len; | |
5444 | ||
5445 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5446 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5447 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5448 | RAID6_Q_STRIPE; | |
5449 | } | |
5450 | ||
fce3bb9a | 5451 | if (rw & REQ_DISCARD) { |
9d644a62 DS |
5452 | u32 factor = 0; |
5453 | u32 sub_stripes = 0; | |
ec9ef7a1 LZ |
5454 | u64 stripes_per_dev = 0; |
5455 | u32 remaining_stripes = 0; | |
b89203f7 | 5456 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5457 | |
5458 | if (map->type & | |
5459 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5460 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5461 | sub_stripes = 1; | |
5462 | else | |
5463 | sub_stripes = map->sub_stripes; | |
5464 | ||
5465 | factor = map->num_stripes / sub_stripes; | |
5466 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5467 | stripe_nr_orig, | |
5468 | factor, | |
5469 | &remaining_stripes); | |
b89203f7 LB |
5470 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5471 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5472 | } |
5473 | ||
fce3bb9a | 5474 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5475 | bbio->stripes[i].physical = |
f2d8d74d CM |
5476 | map->stripes[stripe_index].physical + |
5477 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5478 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5479 | |
ec9ef7a1 LZ |
5480 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5481 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5482 | bbio->stripes[i].length = stripes_per_dev * | |
5483 | map->stripe_len; | |
b89203f7 | 5484 | |
ec9ef7a1 LZ |
5485 | if (i / sub_stripes < remaining_stripes) |
5486 | bbio->stripes[i].length += | |
5487 | map->stripe_len; | |
b89203f7 LB |
5488 | |
5489 | /* | |
5490 | * Special for the first stripe and | |
5491 | * the last stripe: | |
5492 | * | |
5493 | * |-------|...|-------| | |
5494 | * |----------| | |
5495 | * off end_off | |
5496 | */ | |
ec9ef7a1 | 5497 | if (i < sub_stripes) |
a1d3c478 | 5498 | bbio->stripes[i].length -= |
fce3bb9a | 5499 | stripe_offset; |
b89203f7 LB |
5500 | |
5501 | if (stripe_index >= last_stripe && | |
5502 | stripe_index <= (last_stripe + | |
5503 | sub_stripes - 1)) | |
a1d3c478 | 5504 | bbio->stripes[i].length -= |
fce3bb9a | 5505 | stripe_end_offset; |
b89203f7 | 5506 | |
ec9ef7a1 LZ |
5507 | if (i == sub_stripes - 1) |
5508 | stripe_offset = 0; | |
fce3bb9a | 5509 | } else |
a1d3c478 | 5510 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5511 | |
5512 | stripe_index++; | |
5513 | if (stripe_index == map->num_stripes) { | |
5514 | /* This could only happen for RAID0/10 */ | |
5515 | stripe_index = 0; | |
5516 | stripe_nr++; | |
5517 | } | |
5518 | } | |
5519 | } else { | |
5520 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5521 | bbio->stripes[i].physical = |
212a17ab LT |
5522 | map->stripes[stripe_index].physical + |
5523 | stripe_offset + | |
5524 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5525 | bbio->stripes[i].dev = |
212a17ab | 5526 | map->stripes[stripe_index].dev; |
fce3bb9a | 5527 | stripe_index++; |
f2d8d74d | 5528 | } |
593060d7 | 5529 | } |
de11cc12 | 5530 | |
d20983b4 MX |
5531 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
5532 | max_errors = btrfs_chunk_max_errors(map); | |
de11cc12 | 5533 | |
8e5cfb55 ZL |
5534 | if (bbio->raid_map) |
5535 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5536 | |
2c8cdd6e | 5537 | tgtdev_indexes = 0; |
472262f3 SB |
5538 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
5539 | dev_replace->tgtdev != NULL) { | |
5540 | int index_where_to_add; | |
5541 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5542 | ||
5543 | /* | |
5544 | * duplicate the write operations while the dev replace | |
5545 | * procedure is running. Since the copying of the old disk | |
5546 | * to the new disk takes place at run time while the | |
5547 | * filesystem is mounted writable, the regular write | |
5548 | * operations to the old disk have to be duplicated to go | |
5549 | * to the new disk as well. | |
5550 | * Note that device->missing is handled by the caller, and | |
5551 | * that the write to the old disk is already set up in the | |
5552 | * stripes array. | |
5553 | */ | |
5554 | index_where_to_add = num_stripes; | |
5555 | for (i = 0; i < num_stripes; i++) { | |
5556 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5557 | /* write to new disk, too */ | |
5558 | struct btrfs_bio_stripe *new = | |
5559 | bbio->stripes + index_where_to_add; | |
5560 | struct btrfs_bio_stripe *old = | |
5561 | bbio->stripes + i; | |
5562 | ||
5563 | new->physical = old->physical; | |
5564 | new->length = old->length; | |
5565 | new->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5566 | bbio->tgtdev_map[i] = index_where_to_add; |
472262f3 SB |
5567 | index_where_to_add++; |
5568 | max_errors++; | |
2c8cdd6e | 5569 | tgtdev_indexes++; |
472262f3 SB |
5570 | } |
5571 | } | |
5572 | num_stripes = index_where_to_add; | |
ad6d620e SB |
5573 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
5574 | dev_replace->tgtdev != NULL) { | |
5575 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5576 | int index_srcdev = 0; | |
5577 | int found = 0; | |
5578 | u64 physical_of_found = 0; | |
5579 | ||
5580 | /* | |
5581 | * During the dev-replace procedure, the target drive can | |
5582 | * also be used to read data in case it is needed to repair | |
5583 | * a corrupt block elsewhere. This is possible if the | |
5584 | * requested area is left of the left cursor. In this area, | |
5585 | * the target drive is a full copy of the source drive. | |
5586 | */ | |
5587 | for (i = 0; i < num_stripes; i++) { | |
5588 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5589 | /* | |
5590 | * In case of DUP, in order to keep it | |
5591 | * simple, only add the mirror with the | |
5592 | * lowest physical address | |
5593 | */ | |
5594 | if (found && | |
5595 | physical_of_found <= | |
5596 | bbio->stripes[i].physical) | |
5597 | continue; | |
5598 | index_srcdev = i; | |
5599 | found = 1; | |
5600 | physical_of_found = bbio->stripes[i].physical; | |
5601 | } | |
5602 | } | |
5603 | if (found) { | |
258ece02 | 5604 | if (physical_of_found + map->stripe_len <= |
ad6d620e SB |
5605 | dev_replace->cursor_left) { |
5606 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5607 | bbio->stripes + num_stripes; | |
5608 | ||
5609 | tgtdev_stripe->physical = physical_of_found; | |
5610 | tgtdev_stripe->length = | |
5611 | bbio->stripes[index_srcdev].length; | |
5612 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5613 | bbio->tgtdev_map[index_srcdev] = num_stripes; |
ad6d620e | 5614 | |
2c8cdd6e | 5615 | tgtdev_indexes++; |
ad6d620e SB |
5616 | num_stripes++; |
5617 | } | |
5618 | } | |
472262f3 SB |
5619 | } |
5620 | ||
de11cc12 | 5621 | *bbio_ret = bbio; |
10f11900 | 5622 | bbio->map_type = map->type; |
de11cc12 LZ |
5623 | bbio->num_stripes = num_stripes; |
5624 | bbio->max_errors = max_errors; | |
5625 | bbio->mirror_num = mirror_num; | |
2c8cdd6e | 5626 | bbio->num_tgtdevs = tgtdev_indexes; |
ad6d620e SB |
5627 | |
5628 | /* | |
5629 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5630 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5631 | * available as a mirror | |
5632 | */ | |
5633 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5634 | WARN_ON(num_stripes > 1); | |
5635 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5636 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5637 | bbio->mirror_num = map->num_stripes + 1; | |
5638 | } | |
cea9e445 | 5639 | out: |
472262f3 SB |
5640 | if (dev_replace_is_ongoing) |
5641 | btrfs_dev_replace_unlock(dev_replace); | |
0b86a832 | 5642 | free_extent_map(em); |
de11cc12 | 5643 | return ret; |
0b86a832 CM |
5644 | } |
5645 | ||
3ec706c8 | 5646 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5647 | u64 logical, u64 *length, |
a1d3c478 | 5648 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5649 | { |
3ec706c8 | 5650 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
8e5cfb55 | 5651 | mirror_num, 0); |
f2d8d74d CM |
5652 | } |
5653 | ||
af8e2d1d MX |
5654 | /* For Scrub/replace */ |
5655 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw, | |
5656 | u64 logical, u64 *length, | |
5657 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5658 | int need_raid_map) |
af8e2d1d MX |
5659 | { |
5660 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, | |
8e5cfb55 | 5661 | mirror_num, need_raid_map); |
af8e2d1d MX |
5662 | } |
5663 | ||
a512bbf8 YZ |
5664 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
5665 | u64 chunk_start, u64 physical, u64 devid, | |
5666 | u64 **logical, int *naddrs, int *stripe_len) | |
5667 | { | |
5668 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5669 | struct extent_map *em; | |
5670 | struct map_lookup *map; | |
5671 | u64 *buf; | |
5672 | u64 bytenr; | |
5673 | u64 length; | |
5674 | u64 stripe_nr; | |
53b381b3 | 5675 | u64 rmap_len; |
a512bbf8 YZ |
5676 | int i, j, nr = 0; |
5677 | ||
890871be | 5678 | read_lock(&em_tree->lock); |
a512bbf8 | 5679 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5680 | read_unlock(&em_tree->lock); |
a512bbf8 | 5681 | |
835d974f | 5682 | if (!em) { |
efe120a0 | 5683 | printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n", |
835d974f JB |
5684 | chunk_start); |
5685 | return -EIO; | |
5686 | } | |
5687 | ||
5688 | if (em->start != chunk_start) { | |
efe120a0 | 5689 | printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n", |
835d974f JB |
5690 | em->start, chunk_start); |
5691 | free_extent_map(em); | |
5692 | return -EIO; | |
5693 | } | |
a512bbf8 YZ |
5694 | map = (struct map_lookup *)em->bdev; |
5695 | ||
5696 | length = em->len; | |
53b381b3 DW |
5697 | rmap_len = map->stripe_len; |
5698 | ||
a512bbf8 | 5699 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5700 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5701 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5702 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5703 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5704 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5705 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5706 | } | |
a512bbf8 | 5707 | |
31e818fe | 5708 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5709 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5710 | |
5711 | for (i = 0; i < map->num_stripes; i++) { | |
5712 | if (devid && map->stripes[i].dev->devid != devid) | |
5713 | continue; | |
5714 | if (map->stripes[i].physical > physical || | |
5715 | map->stripes[i].physical + length <= physical) | |
5716 | continue; | |
5717 | ||
5718 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 5719 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5720 | |
5721 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5722 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5723 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5724 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5725 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5726 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5727 | * Alternatively, just use rmap_len below instead of | |
5728 | * map->stripe_len */ | |
5729 | ||
5730 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5731 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5732 | for (j = 0; j < nr; j++) { |
5733 | if (buf[j] == bytenr) | |
5734 | break; | |
5735 | } | |
934d375b CM |
5736 | if (j == nr) { |
5737 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5738 | buf[nr++] = bytenr; |
934d375b | 5739 | } |
a512bbf8 YZ |
5740 | } |
5741 | ||
a512bbf8 YZ |
5742 | *logical = buf; |
5743 | *naddrs = nr; | |
53b381b3 | 5744 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5745 | |
5746 | free_extent_map(em); | |
5747 | return 0; | |
f2d8d74d CM |
5748 | } |
5749 | ||
4246a0b6 | 5750 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 5751 | { |
326e1dbb MS |
5752 | bio->bi_private = bbio->private; |
5753 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 5754 | bio_endio(bio); |
326e1dbb | 5755 | |
6e9606d2 | 5756 | btrfs_put_bbio(bbio); |
8408c716 MX |
5757 | } |
5758 | ||
4246a0b6 | 5759 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 5760 | { |
9be3395b | 5761 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 5762 | int is_orig_bio = 0; |
8790d502 | 5763 | |
4246a0b6 | 5764 | if (bio->bi_error) { |
a1d3c478 | 5765 | atomic_inc(&bbio->error); |
4246a0b6 | 5766 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { |
442a4f63 | 5767 | unsigned int stripe_index = |
9be3395b | 5768 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 5769 | struct btrfs_device *dev; |
442a4f63 SB |
5770 | |
5771 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5772 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5773 | if (dev->bdev) { |
5774 | if (bio->bi_rw & WRITE) | |
5775 | btrfs_dev_stat_inc(dev, | |
5776 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5777 | else | |
5778 | btrfs_dev_stat_inc(dev, | |
5779 | BTRFS_DEV_STAT_READ_ERRS); | |
5780 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5781 | btrfs_dev_stat_inc(dev, | |
5782 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5783 | btrfs_dev_stat_print_on_error(dev); | |
5784 | } | |
442a4f63 SB |
5785 | } |
5786 | } | |
8790d502 | 5787 | |
a1d3c478 | 5788 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5789 | is_orig_bio = 1; |
5790 | ||
c404e0dc MX |
5791 | btrfs_bio_counter_dec(bbio->fs_info); |
5792 | ||
a1d3c478 | 5793 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5794 | if (!is_orig_bio) { |
5795 | bio_put(bio); | |
a1d3c478 | 5796 | bio = bbio->orig_bio; |
7d2b4daa | 5797 | } |
c7b22bb1 | 5798 | |
9be3395b | 5799 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 5800 | /* only send an error to the higher layers if it is |
53b381b3 | 5801 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5802 | */ |
a1d3c478 | 5803 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4246a0b6 | 5804 | bio->bi_error = -EIO; |
5dbc8fca | 5805 | } else { |
1259ab75 CM |
5806 | /* |
5807 | * this bio is actually up to date, we didn't | |
5808 | * go over the max number of errors | |
5809 | */ | |
4246a0b6 | 5810 | bio->bi_error = 0; |
1259ab75 | 5811 | } |
c55f1396 | 5812 | |
4246a0b6 | 5813 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 5814 | } else if (!is_orig_bio) { |
8790d502 CM |
5815 | bio_put(bio); |
5816 | } | |
8790d502 CM |
5817 | } |
5818 | ||
8b712842 CM |
5819 | /* |
5820 | * see run_scheduled_bios for a description of why bios are collected for | |
5821 | * async submit. | |
5822 | * | |
5823 | * This will add one bio to the pending list for a device and make sure | |
5824 | * the work struct is scheduled. | |
5825 | */ | |
48a3b636 ES |
5826 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
5827 | struct btrfs_device *device, | |
5828 | int rw, struct bio *bio) | |
8b712842 CM |
5829 | { |
5830 | int should_queue = 1; | |
ffbd517d | 5831 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5832 | |
53b381b3 | 5833 | if (device->missing || !device->bdev) { |
4246a0b6 | 5834 | bio_io_error(bio); |
53b381b3 DW |
5835 | return; |
5836 | } | |
5837 | ||
8b712842 | 5838 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5839 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5840 | bio_get(bio); |
21adbd5c | 5841 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5842 | bio_put(bio); |
143bede5 | 5843 | return; |
8b712842 CM |
5844 | } |
5845 | ||
5846 | /* | |
0986fe9e | 5847 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5848 | * higher layers. Otherwise, the async bio makes it appear we have |
5849 | * made progress against dirty pages when we've really just put it | |
5850 | * on a queue for later | |
5851 | */ | |
0986fe9e | 5852 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5853 | WARN_ON(bio->bi_next); |
8b712842 CM |
5854 | bio->bi_next = NULL; |
5855 | bio->bi_rw |= rw; | |
5856 | ||
5857 | spin_lock(&device->io_lock); | |
7b6d91da | 5858 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5859 | pending_bios = &device->pending_sync_bios; |
5860 | else | |
5861 | pending_bios = &device->pending_bios; | |
8b712842 | 5862 | |
ffbd517d CM |
5863 | if (pending_bios->tail) |
5864 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5865 | |
ffbd517d CM |
5866 | pending_bios->tail = bio; |
5867 | if (!pending_bios->head) | |
5868 | pending_bios->head = bio; | |
8b712842 CM |
5869 | if (device->running_pending) |
5870 | should_queue = 0; | |
5871 | ||
5872 | spin_unlock(&device->io_lock); | |
5873 | ||
5874 | if (should_queue) | |
a8c93d4e QW |
5875 | btrfs_queue_work(root->fs_info->submit_workers, |
5876 | &device->work); | |
8b712842 CM |
5877 | } |
5878 | ||
de1ee92a JB |
5879 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, |
5880 | struct bio *bio, u64 physical, int dev_nr, | |
5881 | int rw, int async) | |
5882 | { | |
5883 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5884 | ||
5885 | bio->bi_private = bbio; | |
9be3395b | 5886 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 5887 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 5888 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
5889 | #ifdef DEBUG |
5890 | { | |
5891 | struct rcu_string *name; | |
5892 | ||
5893 | rcu_read_lock(); | |
5894 | name = rcu_dereference(dev->name); | |
d1423248 | 5895 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a | 5896 | "(%s id %llu), size=%u\n", rw, |
1b6e4469 FF |
5897 | (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev, |
5898 | name->str, dev->devid, bio->bi_iter.bi_size); | |
de1ee92a JB |
5899 | rcu_read_unlock(); |
5900 | } | |
5901 | #endif | |
5902 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
5903 | |
5904 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
5905 | ||
de1ee92a | 5906 | if (async) |
53b381b3 | 5907 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
5908 | else |
5909 | btrfsic_submit_bio(rw, bio); | |
5910 | } | |
5911 | ||
de1ee92a JB |
5912 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
5913 | { | |
5914 | atomic_inc(&bbio->error); | |
5915 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
8408c716 MX |
5916 | /* Shoud be the original bio. */ |
5917 | WARN_ON(bio != bbio->orig_bio); | |
5918 | ||
9be3395b | 5919 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 5920 | bio->bi_iter.bi_sector = logical >> 9; |
4246a0b6 CH |
5921 | bio->bi_error = -EIO; |
5922 | btrfs_end_bbio(bbio, bio); | |
de1ee92a JB |
5923 | } |
5924 | } | |
5925 | ||
f188591e | 5926 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 5927 | int mirror_num, int async_submit) |
0b86a832 | 5928 | { |
0b86a832 | 5929 | struct btrfs_device *dev; |
8790d502 | 5930 | struct bio *first_bio = bio; |
4f024f37 | 5931 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
5932 | u64 length = 0; |
5933 | u64 map_length; | |
0b86a832 | 5934 | int ret; |
08da757d ZL |
5935 | int dev_nr; |
5936 | int total_devs; | |
a1d3c478 | 5937 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 5938 | |
4f024f37 | 5939 | length = bio->bi_iter.bi_size; |
0b86a832 | 5940 | map_length = length; |
cea9e445 | 5941 | |
c404e0dc | 5942 | btrfs_bio_counter_inc_blocked(root->fs_info); |
53b381b3 | 5943 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
8e5cfb55 | 5944 | mirror_num, 1); |
c404e0dc MX |
5945 | if (ret) { |
5946 | btrfs_bio_counter_dec(root->fs_info); | |
79787eaa | 5947 | return ret; |
c404e0dc | 5948 | } |
cea9e445 | 5949 | |
a1d3c478 | 5950 | total_devs = bbio->num_stripes; |
53b381b3 DW |
5951 | bbio->orig_bio = first_bio; |
5952 | bbio->private = first_bio->bi_private; | |
5953 | bbio->end_io = first_bio->bi_end_io; | |
c404e0dc | 5954 | bbio->fs_info = root->fs_info; |
53b381b3 DW |
5955 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
5956 | ||
8e5cfb55 | 5957 | if (bbio->raid_map) { |
53b381b3 DW |
5958 | /* In this case, map_length has been set to the length of |
5959 | a single stripe; not the whole write */ | |
5960 | if (rw & WRITE) { | |
8e5cfb55 | 5961 | ret = raid56_parity_write(root, bio, bbio, map_length); |
53b381b3 | 5962 | } else { |
8e5cfb55 | 5963 | ret = raid56_parity_recover(root, bio, bbio, map_length, |
4245215d | 5964 | mirror_num, 1); |
53b381b3 | 5965 | } |
4245215d | 5966 | |
c404e0dc MX |
5967 | btrfs_bio_counter_dec(root->fs_info); |
5968 | return ret; | |
53b381b3 DW |
5969 | } |
5970 | ||
cea9e445 | 5971 | if (map_length < length) { |
c2cf52eb | 5972 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", |
c1c9ff7c | 5973 | logical, length, map_length); |
cea9e445 CM |
5974 | BUG(); |
5975 | } | |
a1d3c478 | 5976 | |
08da757d | 5977 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a JB |
5978 | dev = bbio->stripes[dev_nr].dev; |
5979 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
5980 | bbio_error(bbio, first_bio, logical); | |
de1ee92a JB |
5981 | continue; |
5982 | } | |
5983 | ||
a1d3c478 | 5984 | if (dev_nr < total_devs - 1) { |
9be3395b | 5985 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 5986 | BUG_ON(!bio); /* -ENOMEM */ |
326e1dbb | 5987 | } else |
a1d3c478 | 5988 | bio = first_bio; |
de1ee92a JB |
5989 | |
5990 | submit_stripe_bio(root, bbio, bio, | |
5991 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
5992 | async_submit); | |
8790d502 | 5993 | } |
c404e0dc | 5994 | btrfs_bio_counter_dec(root->fs_info); |
0b86a832 CM |
5995 | return 0; |
5996 | } | |
5997 | ||
aa1b8cd4 | 5998 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 5999 | u8 *uuid, u8 *fsid) |
0b86a832 | 6000 | { |
2b82032c YZ |
6001 | struct btrfs_device *device; |
6002 | struct btrfs_fs_devices *cur_devices; | |
6003 | ||
aa1b8cd4 | 6004 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6005 | while (cur_devices) { |
6006 | if (!fsid || | |
6007 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6008 | device = __find_device(&cur_devices->devices, | |
6009 | devid, uuid); | |
6010 | if (device) | |
6011 | return device; | |
6012 | } | |
6013 | cur_devices = cur_devices->seed; | |
6014 | } | |
6015 | return NULL; | |
0b86a832 CM |
6016 | } |
6017 | ||
dfe25020 | 6018 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 6019 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6020 | u64 devid, u8 *dev_uuid) |
6021 | { | |
6022 | struct btrfs_device *device; | |
dfe25020 | 6023 | |
12bd2fc0 ID |
6024 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6025 | if (IS_ERR(device)) | |
7cbd8a83 | 6026 | return NULL; |
12bd2fc0 ID |
6027 | |
6028 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6029 | device->fs_devices = fs_devices; |
dfe25020 | 6030 | fs_devices->num_devices++; |
12bd2fc0 ID |
6031 | |
6032 | device->missing = 1; | |
cd02dca5 | 6033 | fs_devices->missing_devices++; |
12bd2fc0 | 6034 | |
dfe25020 CM |
6035 | return device; |
6036 | } | |
6037 | ||
12bd2fc0 ID |
6038 | /** |
6039 | * btrfs_alloc_device - allocate struct btrfs_device | |
6040 | * @fs_info: used only for generating a new devid, can be NULL if | |
6041 | * devid is provided (i.e. @devid != NULL). | |
6042 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6043 | * is generated. | |
6044 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6045 | * is generated. | |
6046 | * | |
6047 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6048 | * on error. Returned struct is not linked onto any lists and can be | |
6049 | * destroyed with kfree() right away. | |
6050 | */ | |
6051 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6052 | const u64 *devid, | |
6053 | const u8 *uuid) | |
6054 | { | |
6055 | struct btrfs_device *dev; | |
6056 | u64 tmp; | |
6057 | ||
fae7f21c | 6058 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6059 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6060 | |
6061 | dev = __alloc_device(); | |
6062 | if (IS_ERR(dev)) | |
6063 | return dev; | |
6064 | ||
6065 | if (devid) | |
6066 | tmp = *devid; | |
6067 | else { | |
6068 | int ret; | |
6069 | ||
6070 | ret = find_next_devid(fs_info, &tmp); | |
6071 | if (ret) { | |
6072 | kfree(dev); | |
6073 | return ERR_PTR(ret); | |
6074 | } | |
6075 | } | |
6076 | dev->devid = tmp; | |
6077 | ||
6078 | if (uuid) | |
6079 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6080 | else | |
6081 | generate_random_uuid(dev->uuid); | |
6082 | ||
9e0af237 LB |
6083 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6084 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6085 | |
6086 | return dev; | |
6087 | } | |
6088 | ||
0b86a832 CM |
6089 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
6090 | struct extent_buffer *leaf, | |
6091 | struct btrfs_chunk *chunk) | |
6092 | { | |
6093 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6094 | struct map_lookup *map; | |
6095 | struct extent_map *em; | |
6096 | u64 logical; | |
6097 | u64 length; | |
6098 | u64 devid; | |
a443755f | 6099 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 6100 | int num_stripes; |
0b86a832 | 6101 | int ret; |
593060d7 | 6102 | int i; |
0b86a832 | 6103 | |
e17cade2 CM |
6104 | logical = key->offset; |
6105 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 6106 | |
890871be | 6107 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6108 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6109 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6110 | |
6111 | /* already mapped? */ | |
6112 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6113 | free_extent_map(em); | |
0b86a832 CM |
6114 | return 0; |
6115 | } else if (em) { | |
6116 | free_extent_map(em); | |
6117 | } | |
0b86a832 | 6118 | |
172ddd60 | 6119 | em = alloc_extent_map(); |
0b86a832 CM |
6120 | if (!em) |
6121 | return -ENOMEM; | |
593060d7 CM |
6122 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6123 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
6124 | if (!map) { |
6125 | free_extent_map(em); | |
6126 | return -ENOMEM; | |
6127 | } | |
6128 | ||
298a8f9c | 6129 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
0b86a832 CM |
6130 | em->bdev = (struct block_device *)map; |
6131 | em->start = logical; | |
6132 | em->len = length; | |
70c8a91c | 6133 | em->orig_start = 0; |
0b86a832 | 6134 | em->block_start = 0; |
c8b97818 | 6135 | em->block_len = em->len; |
0b86a832 | 6136 | |
593060d7 CM |
6137 | map->num_stripes = num_stripes; |
6138 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6139 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6140 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6141 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6142 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6143 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6144 | for (i = 0; i < num_stripes; i++) { |
6145 | map->stripes[i].physical = | |
6146 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6147 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6148 | read_extent_buffer(leaf, uuid, (unsigned long) |
6149 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6150 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
6151 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
6152 | uuid, NULL); | |
dfe25020 | 6153 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
6154 | free_extent_map(em); |
6155 | return -EIO; | |
6156 | } | |
dfe25020 CM |
6157 | if (!map->stripes[i].dev) { |
6158 | map->stripes[i].dev = | |
5f375835 MX |
6159 | add_missing_dev(root, root->fs_info->fs_devices, |
6160 | devid, uuid); | |
dfe25020 | 6161 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6162 | free_extent_map(em); |
6163 | return -EIO; | |
6164 | } | |
816fcebe AJ |
6165 | btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing", |
6166 | devid, uuid); | |
dfe25020 CM |
6167 | } |
6168 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6169 | } |
6170 | ||
890871be | 6171 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6172 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6173 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6174 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6175 | free_extent_map(em); |
6176 | ||
6177 | return 0; | |
6178 | } | |
6179 | ||
143bede5 | 6180 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6181 | struct btrfs_dev_item *dev_item, |
6182 | struct btrfs_device *device) | |
6183 | { | |
6184 | unsigned long ptr; | |
0b86a832 CM |
6185 | |
6186 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6187 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6188 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6189 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6190 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6191 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6192 | device->type = btrfs_device_type(leaf, dev_item); |
6193 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6194 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6195 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6196 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6197 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6198 | |
410ba3a2 | 6199 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6200 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6201 | } |
6202 | ||
5f375835 MX |
6203 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
6204 | u8 *fsid) | |
2b82032c YZ |
6205 | { |
6206 | struct btrfs_fs_devices *fs_devices; | |
6207 | int ret; | |
6208 | ||
b367e47f | 6209 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
6210 | |
6211 | fs_devices = root->fs_info->fs_devices->seed; | |
6212 | while (fs_devices) { | |
5f375835 MX |
6213 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6214 | return fs_devices; | |
6215 | ||
2b82032c YZ |
6216 | fs_devices = fs_devices->seed; |
6217 | } | |
6218 | ||
6219 | fs_devices = find_fsid(fsid); | |
6220 | if (!fs_devices) { | |
5f375835 MX |
6221 | if (!btrfs_test_opt(root, DEGRADED)) |
6222 | return ERR_PTR(-ENOENT); | |
6223 | ||
6224 | fs_devices = alloc_fs_devices(fsid); | |
6225 | if (IS_ERR(fs_devices)) | |
6226 | return fs_devices; | |
6227 | ||
6228 | fs_devices->seeding = 1; | |
6229 | fs_devices->opened = 1; | |
6230 | return fs_devices; | |
2b82032c | 6231 | } |
e4404d6e YZ |
6232 | |
6233 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6234 | if (IS_ERR(fs_devices)) |
6235 | return fs_devices; | |
2b82032c | 6236 | |
97288f2c | 6237 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 6238 | root->fs_info->bdev_holder); |
48d28232 JL |
6239 | if (ret) { |
6240 | free_fs_devices(fs_devices); | |
5f375835 | 6241 | fs_devices = ERR_PTR(ret); |
2b82032c | 6242 | goto out; |
48d28232 | 6243 | } |
2b82032c YZ |
6244 | |
6245 | if (!fs_devices->seeding) { | |
6246 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6247 | free_fs_devices(fs_devices); |
5f375835 | 6248 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6249 | goto out; |
6250 | } | |
6251 | ||
6252 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6253 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6254 | out: |
5f375835 | 6255 | return fs_devices; |
2b82032c YZ |
6256 | } |
6257 | ||
0d81ba5d | 6258 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6259 | struct extent_buffer *leaf, |
6260 | struct btrfs_dev_item *dev_item) | |
6261 | { | |
5f375835 | 6262 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
0b86a832 CM |
6263 | struct btrfs_device *device; |
6264 | u64 devid; | |
6265 | int ret; | |
2b82032c | 6266 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6267 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6268 | ||
0b86a832 | 6269 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6270 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6271 | BTRFS_UUID_SIZE); |
1473b24e | 6272 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6273 | BTRFS_UUID_SIZE); |
6274 | ||
6275 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5f375835 MX |
6276 | fs_devices = open_seed_devices(root, fs_uuid); |
6277 | if (IS_ERR(fs_devices)) | |
6278 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6279 | } |
6280 | ||
aa1b8cd4 | 6281 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6282 | if (!device) { |
e4404d6e | 6283 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
6284 | return -EIO; |
6285 | ||
5f375835 MX |
6286 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); |
6287 | if (!device) | |
6288 | return -ENOMEM; | |
33b97e43 AJ |
6289 | btrfs_warn(root->fs_info, "devid %llu uuid %pU missing", |
6290 | devid, dev_uuid); | |
5f375835 MX |
6291 | } else { |
6292 | if (!device->bdev && !btrfs_test_opt(root, DEGRADED)) | |
6293 | return -EIO; | |
6294 | ||
6295 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6296 | /* |
6297 | * this happens when a device that was properly setup | |
6298 | * in the device info lists suddenly goes bad. | |
6299 | * device->bdev is NULL, and so we have to set | |
6300 | * device->missing to one here | |
6301 | */ | |
5f375835 | 6302 | device->fs_devices->missing_devices++; |
cd02dca5 | 6303 | device->missing = 1; |
2b82032c | 6304 | } |
5f375835 MX |
6305 | |
6306 | /* Move the device to its own fs_devices */ | |
6307 | if (device->fs_devices != fs_devices) { | |
6308 | ASSERT(device->missing); | |
6309 | ||
6310 | list_move(&device->dev_list, &fs_devices->devices); | |
6311 | device->fs_devices->num_devices--; | |
6312 | fs_devices->num_devices++; | |
6313 | ||
6314 | device->fs_devices->missing_devices--; | |
6315 | fs_devices->missing_devices++; | |
6316 | ||
6317 | device->fs_devices = fs_devices; | |
6318 | } | |
2b82032c YZ |
6319 | } |
6320 | ||
6321 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6322 | BUG_ON(device->writeable); | |
6323 | if (device->generation != | |
6324 | btrfs_device_generation(leaf, dev_item)) | |
6325 | return -EINVAL; | |
6324fbf3 | 6326 | } |
0b86a832 CM |
6327 | |
6328 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6329 | device->in_fs_metadata = 1; |
63a212ab | 6330 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6331 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
6332 | spin_lock(&root->fs_info->free_chunk_lock); |
6333 | root->fs_info->free_chunk_space += device->total_bytes - | |
6334 | device->bytes_used; | |
6335 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6336 | } | |
0b86a832 | 6337 | ret = 0; |
0b86a832 CM |
6338 | return ret; |
6339 | } | |
6340 | ||
e4404d6e | 6341 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 6342 | { |
6c41761f | 6343 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 6344 | struct extent_buffer *sb; |
0b86a832 | 6345 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6346 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6347 | u8 *array_ptr; |
6348 | unsigned long sb_array_offset; | |
84eed90f | 6349 | int ret = 0; |
0b86a832 CM |
6350 | u32 num_stripes; |
6351 | u32 array_size; | |
6352 | u32 len = 0; | |
1ffb22cf | 6353 | u32 cur_offset; |
84eed90f | 6354 | struct btrfs_key key; |
0b86a832 | 6355 | |
a83fffb7 DS |
6356 | ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize); |
6357 | /* | |
6358 | * This will create extent buffer of nodesize, superblock size is | |
6359 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6360 | * overallocate but we can keep it as-is, only the first page is used. | |
6361 | */ | |
6362 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
a061fc8d CM |
6363 | if (!sb) |
6364 | return -ENOMEM; | |
6365 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 6366 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
6367 | /* |
6368 | * The sb extent buffer is artifical and just used to read the system array. | |
6369 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
6370 | * pages up-to-date when the page is larger: extent does not cover the | |
6371 | * whole page and consequently check_page_uptodate does not find all | |
6372 | * the page's extents up-to-date (the hole beyond sb), | |
6373 | * write_extent_buffer then triggers a WARN_ON. | |
6374 | * | |
6375 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6376 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6377 | * to silence the warning eg. on PowerPC 64. | |
6378 | */ | |
6379 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 6380 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6381 | |
a061fc8d | 6382 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6383 | array_size = btrfs_super_sys_array_size(super_copy); |
6384 | ||
1ffb22cf DS |
6385 | array_ptr = super_copy->sys_chunk_array; |
6386 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6387 | cur_offset = 0; | |
0b86a832 | 6388 | |
1ffb22cf DS |
6389 | while (cur_offset < array_size) { |
6390 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6391 | len = sizeof(*disk_key); |
6392 | if (cur_offset + len > array_size) | |
6393 | goto out_short_read; | |
6394 | ||
0b86a832 CM |
6395 | btrfs_disk_key_to_cpu(&key, disk_key); |
6396 | ||
1ffb22cf DS |
6397 | array_ptr += len; |
6398 | sb_array_offset += len; | |
6399 | cur_offset += len; | |
0b86a832 | 6400 | |
0d81ba5d | 6401 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6402 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6403 | /* |
6404 | * At least one btrfs_chunk with one stripe must be | |
6405 | * present, exact stripe count check comes afterwards | |
6406 | */ | |
6407 | len = btrfs_chunk_item_size(1); | |
6408 | if (cur_offset + len > array_size) | |
6409 | goto out_short_read; | |
6410 | ||
6411 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
6412 | len = btrfs_chunk_item_size(num_stripes); | |
6413 | if (cur_offset + len > array_size) | |
6414 | goto out_short_read; | |
6415 | ||
0d81ba5d | 6416 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6417 | if (ret) |
6418 | break; | |
0b86a832 | 6419 | } else { |
84eed90f CM |
6420 | ret = -EIO; |
6421 | break; | |
0b86a832 | 6422 | } |
1ffb22cf DS |
6423 | array_ptr += len; |
6424 | sb_array_offset += len; | |
6425 | cur_offset += len; | |
0b86a832 | 6426 | } |
a061fc8d | 6427 | free_extent_buffer(sb); |
84eed90f | 6428 | return ret; |
e3540eab DS |
6429 | |
6430 | out_short_read: | |
6431 | printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n", | |
6432 | len, cur_offset); | |
6433 | free_extent_buffer(sb); | |
6434 | return -EIO; | |
0b86a832 CM |
6435 | } |
6436 | ||
6437 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6438 | { | |
6439 | struct btrfs_path *path; | |
6440 | struct extent_buffer *leaf; | |
6441 | struct btrfs_key key; | |
6442 | struct btrfs_key found_key; | |
6443 | int ret; | |
6444 | int slot; | |
6445 | ||
6446 | root = root->fs_info->chunk_root; | |
6447 | ||
6448 | path = btrfs_alloc_path(); | |
6449 | if (!path) | |
6450 | return -ENOMEM; | |
6451 | ||
b367e47f LZ |
6452 | mutex_lock(&uuid_mutex); |
6453 | lock_chunks(root); | |
6454 | ||
395927a9 FDBM |
6455 | /* |
6456 | * Read all device items, and then all the chunk items. All | |
6457 | * device items are found before any chunk item (their object id | |
6458 | * is smaller than the lowest possible object id for a chunk | |
6459 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6460 | */ |
6461 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6462 | key.offset = 0; | |
6463 | key.type = 0; | |
0b86a832 | 6464 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6465 | if (ret < 0) |
6466 | goto error; | |
d397712b | 6467 | while (1) { |
0b86a832 CM |
6468 | leaf = path->nodes[0]; |
6469 | slot = path->slots[0]; | |
6470 | if (slot >= btrfs_header_nritems(leaf)) { | |
6471 | ret = btrfs_next_leaf(root, path); | |
6472 | if (ret == 0) | |
6473 | continue; | |
6474 | if (ret < 0) | |
6475 | goto error; | |
6476 | break; | |
6477 | } | |
6478 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6479 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6480 | struct btrfs_dev_item *dev_item; | |
6481 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6482 | struct btrfs_dev_item); |
395927a9 FDBM |
6483 | ret = read_one_dev(root, leaf, dev_item); |
6484 | if (ret) | |
6485 | goto error; | |
0b86a832 CM |
6486 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6487 | struct btrfs_chunk *chunk; | |
6488 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6489 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6490 | if (ret) |
6491 | goto error; | |
0b86a832 CM |
6492 | } |
6493 | path->slots[0]++; | |
6494 | } | |
0b86a832 CM |
6495 | ret = 0; |
6496 | error: | |
b367e47f LZ |
6497 | unlock_chunks(root); |
6498 | mutex_unlock(&uuid_mutex); | |
6499 | ||
2b82032c | 6500 | btrfs_free_path(path); |
0b86a832 CM |
6501 | return ret; |
6502 | } | |
442a4f63 | 6503 | |
cb517eab MX |
6504 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6505 | { | |
6506 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6507 | struct btrfs_device *device; | |
6508 | ||
29cc83f6 LB |
6509 | while (fs_devices) { |
6510 | mutex_lock(&fs_devices->device_list_mutex); | |
6511 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6512 | device->dev_root = fs_info->dev_root; | |
6513 | mutex_unlock(&fs_devices->device_list_mutex); | |
6514 | ||
6515 | fs_devices = fs_devices->seed; | |
6516 | } | |
cb517eab MX |
6517 | } |
6518 | ||
733f4fbb SB |
6519 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6520 | { | |
6521 | int i; | |
6522 | ||
6523 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6524 | btrfs_dev_stat_reset(dev, i); | |
6525 | } | |
6526 | ||
6527 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6528 | { | |
6529 | struct btrfs_key key; | |
6530 | struct btrfs_key found_key; | |
6531 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6532 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6533 | struct extent_buffer *eb; | |
6534 | int slot; | |
6535 | int ret = 0; | |
6536 | struct btrfs_device *device; | |
6537 | struct btrfs_path *path = NULL; | |
6538 | int i; | |
6539 | ||
6540 | path = btrfs_alloc_path(); | |
6541 | if (!path) { | |
6542 | ret = -ENOMEM; | |
6543 | goto out; | |
6544 | } | |
6545 | ||
6546 | mutex_lock(&fs_devices->device_list_mutex); | |
6547 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6548 | int item_size; | |
6549 | struct btrfs_dev_stats_item *ptr; | |
6550 | ||
6551 | key.objectid = 0; | |
6552 | key.type = BTRFS_DEV_STATS_KEY; | |
6553 | key.offset = device->devid; | |
6554 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6555 | if (ret) { | |
733f4fbb SB |
6556 | __btrfs_reset_dev_stats(device); |
6557 | device->dev_stats_valid = 1; | |
6558 | btrfs_release_path(path); | |
6559 | continue; | |
6560 | } | |
6561 | slot = path->slots[0]; | |
6562 | eb = path->nodes[0]; | |
6563 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6564 | item_size = btrfs_item_size_nr(eb, slot); | |
6565 | ||
6566 | ptr = btrfs_item_ptr(eb, slot, | |
6567 | struct btrfs_dev_stats_item); | |
6568 | ||
6569 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6570 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6571 | btrfs_dev_stat_set(device, i, | |
6572 | btrfs_dev_stats_value(eb, ptr, i)); | |
6573 | else | |
6574 | btrfs_dev_stat_reset(device, i); | |
6575 | } | |
6576 | ||
6577 | device->dev_stats_valid = 1; | |
6578 | btrfs_dev_stat_print_on_load(device); | |
6579 | btrfs_release_path(path); | |
6580 | } | |
6581 | mutex_unlock(&fs_devices->device_list_mutex); | |
6582 | ||
6583 | out: | |
6584 | btrfs_free_path(path); | |
6585 | return ret < 0 ? ret : 0; | |
6586 | } | |
6587 | ||
6588 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6589 | struct btrfs_root *dev_root, | |
6590 | struct btrfs_device *device) | |
6591 | { | |
6592 | struct btrfs_path *path; | |
6593 | struct btrfs_key key; | |
6594 | struct extent_buffer *eb; | |
6595 | struct btrfs_dev_stats_item *ptr; | |
6596 | int ret; | |
6597 | int i; | |
6598 | ||
6599 | key.objectid = 0; | |
6600 | key.type = BTRFS_DEV_STATS_KEY; | |
6601 | key.offset = device->devid; | |
6602 | ||
6603 | path = btrfs_alloc_path(); | |
6604 | BUG_ON(!path); | |
6605 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6606 | if (ret < 0) { | |
efe120a0 FH |
6607 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6608 | "error %d while searching for dev_stats item for device %s!\n", | |
606686ee | 6609 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6610 | goto out; |
6611 | } | |
6612 | ||
6613 | if (ret == 0 && | |
6614 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6615 | /* need to delete old one and insert a new one */ | |
6616 | ret = btrfs_del_item(trans, dev_root, path); | |
6617 | if (ret != 0) { | |
efe120a0 FH |
6618 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6619 | "delete too small dev_stats item for device %s failed %d!\n", | |
606686ee | 6620 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6621 | goto out; |
6622 | } | |
6623 | ret = 1; | |
6624 | } | |
6625 | ||
6626 | if (ret == 1) { | |
6627 | /* need to insert a new item */ | |
6628 | btrfs_release_path(path); | |
6629 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6630 | &key, sizeof(*ptr)); | |
6631 | if (ret < 0) { | |
efe120a0 FH |
6632 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6633 | "insert dev_stats item for device %s failed %d!\n", | |
606686ee | 6634 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6635 | goto out; |
6636 | } | |
6637 | } | |
6638 | ||
6639 | eb = path->nodes[0]; | |
6640 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6641 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6642 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6643 | btrfs_dev_stat_read(device, i)); | |
6644 | btrfs_mark_buffer_dirty(eb); | |
6645 | ||
6646 | out: | |
6647 | btrfs_free_path(path); | |
6648 | return ret; | |
6649 | } | |
6650 | ||
6651 | /* | |
6652 | * called from commit_transaction. Writes all changed device stats to disk. | |
6653 | */ | |
6654 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6655 | struct btrfs_fs_info *fs_info) | |
6656 | { | |
6657 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6658 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6659 | struct btrfs_device *device; | |
addc3fa7 | 6660 | int stats_cnt; |
733f4fbb SB |
6661 | int ret = 0; |
6662 | ||
6663 | mutex_lock(&fs_devices->device_list_mutex); | |
6664 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 6665 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
6666 | continue; |
6667 | ||
addc3fa7 | 6668 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
733f4fbb SB |
6669 | ret = update_dev_stat_item(trans, dev_root, device); |
6670 | if (!ret) | |
addc3fa7 | 6671 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
6672 | } |
6673 | mutex_unlock(&fs_devices->device_list_mutex); | |
6674 | ||
6675 | return ret; | |
6676 | } | |
6677 | ||
442a4f63 SB |
6678 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
6679 | { | |
6680 | btrfs_dev_stat_inc(dev, index); | |
6681 | btrfs_dev_stat_print_on_error(dev); | |
6682 | } | |
6683 | ||
48a3b636 | 6684 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 6685 | { |
733f4fbb SB |
6686 | if (!dev->dev_stats_valid) |
6687 | return; | |
efe120a0 FH |
6688 | printk_ratelimited_in_rcu(KERN_ERR "BTRFS: " |
6689 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6690 | rcu_str_deref(dev->name), |
442a4f63 SB |
6691 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6692 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6693 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
6694 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
6695 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 6696 | } |
c11d2c23 | 6697 | |
733f4fbb SB |
6698 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
6699 | { | |
a98cdb85 SB |
6700 | int i; |
6701 | ||
6702 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6703 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6704 | break; | |
6705 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6706 | return; /* all values == 0, suppress message */ | |
6707 | ||
efe120a0 FH |
6708 | printk_in_rcu(KERN_INFO "BTRFS: " |
6709 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6710 | rcu_str_deref(dev->name), |
733f4fbb SB |
6711 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6712 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6713 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6714 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6715 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6716 | } | |
6717 | ||
c11d2c23 | 6718 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 6719 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
6720 | { |
6721 | struct btrfs_device *dev; | |
6722 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6723 | int i; | |
6724 | ||
6725 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 6726 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
6727 | mutex_unlock(&fs_devices->device_list_mutex); |
6728 | ||
6729 | if (!dev) { | |
efe120a0 | 6730 | btrfs_warn(root->fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 6731 | return -ENODEV; |
733f4fbb | 6732 | } else if (!dev->dev_stats_valid) { |
efe120a0 | 6733 | btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 6734 | return -ENODEV; |
b27f7c0c | 6735 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
6736 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
6737 | if (stats->nr_items > i) | |
6738 | stats->values[i] = | |
6739 | btrfs_dev_stat_read_and_reset(dev, i); | |
6740 | else | |
6741 | btrfs_dev_stat_reset(dev, i); | |
6742 | } | |
6743 | } else { | |
6744 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6745 | if (stats->nr_items > i) | |
6746 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6747 | } | |
6748 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6749 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6750 | return 0; | |
6751 | } | |
a8a6dab7 SB |
6752 | |
6753 | int btrfs_scratch_superblock(struct btrfs_device *device) | |
6754 | { | |
6755 | struct buffer_head *bh; | |
6756 | struct btrfs_super_block *disk_super; | |
6757 | ||
6758 | bh = btrfs_read_dev_super(device->bdev); | |
6759 | if (!bh) | |
6760 | return -EINVAL; | |
6761 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6762 | ||
6763 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6764 | set_buffer_dirty(bh); | |
6765 | sync_dirty_buffer(bh); | |
6766 | brelse(bh); | |
6767 | ||
6768 | return 0; | |
6769 | } | |
935e5cc9 MX |
6770 | |
6771 | /* | |
6772 | * Update the size of all devices, which is used for writing out the | |
6773 | * super blocks. | |
6774 | */ | |
6775 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
6776 | { | |
6777 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6778 | struct btrfs_device *curr, *next; | |
6779 | ||
6780 | if (list_empty(&fs_devices->resized_devices)) | |
6781 | return; | |
6782 | ||
6783 | mutex_lock(&fs_devices->device_list_mutex); | |
6784 | lock_chunks(fs_info->dev_root); | |
6785 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
6786 | resized_list) { | |
6787 | list_del_init(&curr->resized_list); | |
6788 | curr->commit_total_bytes = curr->disk_total_bytes; | |
6789 | } | |
6790 | unlock_chunks(fs_info->dev_root); | |
6791 | mutex_unlock(&fs_devices->device_list_mutex); | |
6792 | } | |
ce7213c7 MX |
6793 | |
6794 | /* Must be invoked during the transaction commit */ | |
6795 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
6796 | struct btrfs_transaction *transaction) | |
6797 | { | |
6798 | struct extent_map *em; | |
6799 | struct map_lookup *map; | |
6800 | struct btrfs_device *dev; | |
6801 | int i; | |
6802 | ||
6803 | if (list_empty(&transaction->pending_chunks)) | |
6804 | return; | |
6805 | ||
6806 | /* In order to kick the device replace finish process */ | |
6807 | lock_chunks(root); | |
6808 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
6809 | map = (struct map_lookup *)em->bdev; | |
6810 | ||
6811 | for (i = 0; i < map->num_stripes; i++) { | |
6812 | dev = map->stripes[i].dev; | |
6813 | dev->commit_bytes_used = dev->bytes_used; | |
6814 | } | |
6815 | } | |
6816 | unlock_chunks(root); | |
6817 | } | |
5a13f430 AJ |
6818 | |
6819 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6820 | { | |
6821 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6822 | while (fs_devices) { | |
6823 | fs_devices->fs_info = fs_info; | |
6824 | fs_devices = fs_devices->seed; | |
6825 | } | |
6826 | } | |
6827 | ||
6828 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6829 | { | |
6830 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6831 | while (fs_devices) { | |
6832 | fs_devices->fs_info = NULL; | |
6833 | fs_devices = fs_devices->seed; | |
6834 | } | |
6835 | } |