]>
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 | |
2b82032c YZ |
45 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
46 | struct btrfs_root *root, | |
47 | struct btrfs_device *device); | |
48 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 49 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 50 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 51 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 52 | |
67a2c45e | 53 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc CM |
54 | static LIST_HEAD(fs_uuids); |
55 | ||
7d9eb12c CM |
56 | static void lock_chunks(struct btrfs_root *root) |
57 | { | |
7d9eb12c CM |
58 | mutex_lock(&root->fs_info->chunk_mutex); |
59 | } | |
60 | ||
61 | static void unlock_chunks(struct btrfs_root *root) | |
62 | { | |
7d9eb12c CM |
63 | mutex_unlock(&root->fs_info->chunk_mutex); |
64 | } | |
65 | ||
2208a378 ID |
66 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
67 | { | |
68 | struct btrfs_fs_devices *fs_devs; | |
69 | ||
70 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS); | |
71 | if (!fs_devs) | |
72 | return ERR_PTR(-ENOMEM); | |
73 | ||
74 | mutex_init(&fs_devs->device_list_mutex); | |
75 | ||
76 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 77 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
78 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
79 | INIT_LIST_HEAD(&fs_devs->list); | |
80 | ||
81 | return fs_devs; | |
82 | } | |
83 | ||
84 | /** | |
85 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
86 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
87 | * generated. | |
88 | * | |
89 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
90 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
91 | * can be destroyed with kfree() right away. | |
92 | */ | |
93 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
94 | { | |
95 | struct btrfs_fs_devices *fs_devs; | |
96 | ||
97 | fs_devs = __alloc_fs_devices(); | |
98 | if (IS_ERR(fs_devs)) | |
99 | return fs_devs; | |
100 | ||
101 | if (fsid) | |
102 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
103 | else | |
104 | generate_random_uuid(fs_devs->fsid); | |
105 | ||
106 | return fs_devs; | |
107 | } | |
108 | ||
e4404d6e YZ |
109 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
110 | { | |
111 | struct btrfs_device *device; | |
112 | WARN_ON(fs_devices->opened); | |
113 | while (!list_empty(&fs_devices->devices)) { | |
114 | device = list_entry(fs_devices->devices.next, | |
115 | struct btrfs_device, dev_list); | |
116 | list_del(&device->dev_list); | |
606686ee | 117 | rcu_string_free(device->name); |
e4404d6e YZ |
118 | kfree(device); |
119 | } | |
120 | kfree(fs_devices); | |
121 | } | |
122 | ||
b8b8ff59 LC |
123 | static void btrfs_kobject_uevent(struct block_device *bdev, |
124 | enum kobject_action action) | |
125 | { | |
126 | int ret; | |
127 | ||
128 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
129 | if (ret) | |
efe120a0 | 130 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
131 | action, |
132 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
133 | &disk_to_dev(bdev->bd_disk)->kobj); | |
134 | } | |
135 | ||
143bede5 | 136 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
137 | { |
138 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 139 | |
2b82032c YZ |
140 | while (!list_empty(&fs_uuids)) { |
141 | fs_devices = list_entry(fs_uuids.next, | |
142 | struct btrfs_fs_devices, list); | |
143 | list_del(&fs_devices->list); | |
e4404d6e | 144 | free_fs_devices(fs_devices); |
8a4b83cc | 145 | } |
8a4b83cc CM |
146 | } |
147 | ||
12bd2fc0 ID |
148 | static struct btrfs_device *__alloc_device(void) |
149 | { | |
150 | struct btrfs_device *dev; | |
151 | ||
152 | dev = kzalloc(sizeof(*dev), GFP_NOFS); | |
153 | if (!dev) | |
154 | return ERR_PTR(-ENOMEM); | |
155 | ||
156 | INIT_LIST_HEAD(&dev->dev_list); | |
157 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 158 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
159 | |
160 | spin_lock_init(&dev->io_lock); | |
161 | ||
162 | spin_lock_init(&dev->reada_lock); | |
163 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 164 | atomic_set(&dev->dev_stats_ccnt, 0); |
12bd2fc0 ID |
165 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT); |
166 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
167 | ||
168 | return dev; | |
169 | } | |
170 | ||
a1b32a59 CM |
171 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
172 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
173 | { |
174 | struct btrfs_device *dev; | |
8a4b83cc | 175 | |
c6e30871 | 176 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 177 | if (dev->devid == devid && |
8f18cf13 | 178 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 179 | return dev; |
a443755f | 180 | } |
8a4b83cc CM |
181 | } |
182 | return NULL; | |
183 | } | |
184 | ||
a1b32a59 | 185 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 186 | { |
8a4b83cc CM |
187 | struct btrfs_fs_devices *fs_devices; |
188 | ||
c6e30871 | 189 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
190 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
191 | return fs_devices; | |
192 | } | |
193 | return NULL; | |
194 | } | |
195 | ||
beaf8ab3 SB |
196 | static int |
197 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
198 | int flush, struct block_device **bdev, | |
199 | struct buffer_head **bh) | |
200 | { | |
201 | int ret; | |
202 | ||
203 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
204 | ||
205 | if (IS_ERR(*bdev)) { | |
206 | ret = PTR_ERR(*bdev); | |
efe120a0 | 207 | printk(KERN_INFO "BTRFS: open %s failed\n", device_path); |
beaf8ab3 SB |
208 | goto error; |
209 | } | |
210 | ||
211 | if (flush) | |
212 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
213 | ret = set_blocksize(*bdev, 4096); | |
214 | if (ret) { | |
215 | blkdev_put(*bdev, flags); | |
216 | goto error; | |
217 | } | |
218 | invalidate_bdev(*bdev); | |
219 | *bh = btrfs_read_dev_super(*bdev); | |
220 | if (!*bh) { | |
221 | ret = -EINVAL; | |
222 | blkdev_put(*bdev, flags); | |
223 | goto error; | |
224 | } | |
225 | ||
226 | return 0; | |
227 | ||
228 | error: | |
229 | *bdev = NULL; | |
230 | *bh = NULL; | |
231 | return ret; | |
232 | } | |
233 | ||
ffbd517d CM |
234 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
235 | struct bio *head, struct bio *tail) | |
236 | { | |
237 | ||
238 | struct bio *old_head; | |
239 | ||
240 | old_head = pending_bios->head; | |
241 | pending_bios->head = head; | |
242 | if (pending_bios->tail) | |
243 | tail->bi_next = old_head; | |
244 | else | |
245 | pending_bios->tail = tail; | |
246 | } | |
247 | ||
8b712842 CM |
248 | /* |
249 | * we try to collect pending bios for a device so we don't get a large | |
250 | * number of procs sending bios down to the same device. This greatly | |
251 | * improves the schedulers ability to collect and merge the bios. | |
252 | * | |
253 | * But, it also turns into a long list of bios to process and that is sure | |
254 | * to eventually make the worker thread block. The solution here is to | |
255 | * make some progress and then put this work struct back at the end of | |
256 | * the list if the block device is congested. This way, multiple devices | |
257 | * can make progress from a single worker thread. | |
258 | */ | |
143bede5 | 259 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
260 | { |
261 | struct bio *pending; | |
262 | struct backing_dev_info *bdi; | |
b64a2851 | 263 | struct btrfs_fs_info *fs_info; |
ffbd517d | 264 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
265 | struct bio *tail; |
266 | struct bio *cur; | |
267 | int again = 0; | |
ffbd517d | 268 | unsigned long num_run; |
d644d8a1 | 269 | unsigned long batch_run = 0; |
b64a2851 | 270 | unsigned long limit; |
b765ead5 | 271 | unsigned long last_waited = 0; |
d84275c9 | 272 | int force_reg = 0; |
0e588859 | 273 | int sync_pending = 0; |
211588ad CM |
274 | struct blk_plug plug; |
275 | ||
276 | /* | |
277 | * this function runs all the bios we've collected for | |
278 | * a particular device. We don't want to wander off to | |
279 | * another device without first sending all of these down. | |
280 | * So, setup a plug here and finish it off before we return | |
281 | */ | |
282 | blk_start_plug(&plug); | |
8b712842 | 283 | |
bedf762b | 284 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
285 | fs_info = device->dev_root->fs_info; |
286 | limit = btrfs_async_submit_limit(fs_info); | |
287 | limit = limit * 2 / 3; | |
288 | ||
8b712842 CM |
289 | loop: |
290 | spin_lock(&device->io_lock); | |
291 | ||
a6837051 | 292 | loop_lock: |
d84275c9 | 293 | num_run = 0; |
ffbd517d | 294 | |
8b712842 CM |
295 | /* take all the bios off the list at once and process them |
296 | * later on (without the lock held). But, remember the | |
297 | * tail and other pointers so the bios can be properly reinserted | |
298 | * into the list if we hit congestion | |
299 | */ | |
d84275c9 | 300 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 301 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
302 | force_reg = 1; |
303 | } else { | |
ffbd517d | 304 | pending_bios = &device->pending_bios; |
d84275c9 CM |
305 | force_reg = 0; |
306 | } | |
ffbd517d CM |
307 | |
308 | pending = pending_bios->head; | |
309 | tail = pending_bios->tail; | |
8b712842 | 310 | WARN_ON(pending && !tail); |
8b712842 CM |
311 | |
312 | /* | |
313 | * if pending was null this time around, no bios need processing | |
314 | * at all and we can stop. Otherwise it'll loop back up again | |
315 | * and do an additional check so no bios are missed. | |
316 | * | |
317 | * device->running_pending is used to synchronize with the | |
318 | * schedule_bio code. | |
319 | */ | |
ffbd517d CM |
320 | if (device->pending_sync_bios.head == NULL && |
321 | device->pending_bios.head == NULL) { | |
8b712842 CM |
322 | again = 0; |
323 | device->running_pending = 0; | |
ffbd517d CM |
324 | } else { |
325 | again = 1; | |
326 | device->running_pending = 1; | |
8b712842 | 327 | } |
ffbd517d CM |
328 | |
329 | pending_bios->head = NULL; | |
330 | pending_bios->tail = NULL; | |
331 | ||
8b712842 CM |
332 | spin_unlock(&device->io_lock); |
333 | ||
d397712b | 334 | while (pending) { |
ffbd517d CM |
335 | |
336 | rmb(); | |
d84275c9 CM |
337 | /* we want to work on both lists, but do more bios on the |
338 | * sync list than the regular list | |
339 | */ | |
340 | if ((num_run > 32 && | |
341 | pending_bios != &device->pending_sync_bios && | |
342 | device->pending_sync_bios.head) || | |
343 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
344 | device->pending_bios.head)) { | |
ffbd517d CM |
345 | spin_lock(&device->io_lock); |
346 | requeue_list(pending_bios, pending, tail); | |
347 | goto loop_lock; | |
348 | } | |
349 | ||
8b712842 CM |
350 | cur = pending; |
351 | pending = pending->bi_next; | |
352 | cur->bi_next = NULL; | |
b64a2851 | 353 | |
66657b31 | 354 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
355 | waitqueue_active(&fs_info->async_submit_wait)) |
356 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
357 | |
358 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 359 | |
2ab1ba68 CM |
360 | /* |
361 | * if we're doing the sync list, record that our | |
362 | * plug has some sync requests on it | |
363 | * | |
364 | * If we're doing the regular list and there are | |
365 | * sync requests sitting around, unplug before | |
366 | * we add more | |
367 | */ | |
368 | if (pending_bios == &device->pending_sync_bios) { | |
369 | sync_pending = 1; | |
370 | } else if (sync_pending) { | |
371 | blk_finish_plug(&plug); | |
372 | blk_start_plug(&plug); | |
373 | sync_pending = 0; | |
374 | } | |
375 | ||
21adbd5c | 376 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
377 | num_run++; |
378 | batch_run++; | |
7eaceacc | 379 | if (need_resched()) |
ffbd517d | 380 | cond_resched(); |
8b712842 CM |
381 | |
382 | /* | |
383 | * we made progress, there is more work to do and the bdi | |
384 | * is now congested. Back off and let other work structs | |
385 | * run instead | |
386 | */ | |
57fd5a5f | 387 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 388 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 389 | struct io_context *ioc; |
8b712842 | 390 | |
b765ead5 CM |
391 | ioc = current->io_context; |
392 | ||
393 | /* | |
394 | * the main goal here is that we don't want to | |
395 | * block if we're going to be able to submit | |
396 | * more requests without blocking. | |
397 | * | |
398 | * This code does two great things, it pokes into | |
399 | * the elevator code from a filesystem _and_ | |
400 | * it makes assumptions about how batching works. | |
401 | */ | |
402 | if (ioc && ioc->nr_batch_requests > 0 && | |
403 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
404 | (last_waited == 0 || | |
405 | ioc->last_waited == last_waited)) { | |
406 | /* | |
407 | * we want to go through our batch of | |
408 | * requests and stop. So, we copy out | |
409 | * the ioc->last_waited time and test | |
410 | * against it before looping | |
411 | */ | |
412 | last_waited = ioc->last_waited; | |
7eaceacc | 413 | if (need_resched()) |
ffbd517d | 414 | cond_resched(); |
b765ead5 CM |
415 | continue; |
416 | } | |
8b712842 | 417 | spin_lock(&device->io_lock); |
ffbd517d | 418 | requeue_list(pending_bios, pending, tail); |
a6837051 | 419 | device->running_pending = 1; |
8b712842 CM |
420 | |
421 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
422 | btrfs_queue_work(fs_info->submit_workers, |
423 | &device->work); | |
8b712842 CM |
424 | goto done; |
425 | } | |
d85c8a6f CM |
426 | /* unplug every 64 requests just for good measure */ |
427 | if (batch_run % 64 == 0) { | |
428 | blk_finish_plug(&plug); | |
429 | blk_start_plug(&plug); | |
430 | sync_pending = 0; | |
431 | } | |
8b712842 | 432 | } |
ffbd517d | 433 | |
51684082 CM |
434 | cond_resched(); |
435 | if (again) | |
436 | goto loop; | |
437 | ||
438 | spin_lock(&device->io_lock); | |
439 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
440 | goto loop_lock; | |
441 | spin_unlock(&device->io_lock); | |
442 | ||
8b712842 | 443 | done: |
211588ad | 444 | blk_finish_plug(&plug); |
8b712842 CM |
445 | } |
446 | ||
b2950863 | 447 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
448 | { |
449 | struct btrfs_device *device; | |
450 | ||
451 | device = container_of(work, struct btrfs_device, work); | |
452 | run_scheduled_bios(device); | |
453 | } | |
454 | ||
60999ca4 DS |
455 | /* |
456 | * Add new device to list of registered devices | |
457 | * | |
458 | * Returns: | |
459 | * 1 - first time device is seen | |
460 | * 0 - device already known | |
461 | * < 0 - error | |
462 | */ | |
a1b32a59 | 463 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
464 | struct btrfs_super_block *disk_super, |
465 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
466 | { | |
467 | struct btrfs_device *device; | |
468 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 469 | struct rcu_string *name; |
60999ca4 | 470 | int ret = 0; |
8a4b83cc CM |
471 | u64 found_transid = btrfs_super_generation(disk_super); |
472 | ||
473 | fs_devices = find_fsid(disk_super->fsid); | |
474 | if (!fs_devices) { | |
2208a378 ID |
475 | fs_devices = alloc_fs_devices(disk_super->fsid); |
476 | if (IS_ERR(fs_devices)) | |
477 | return PTR_ERR(fs_devices); | |
478 | ||
8a4b83cc | 479 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 480 | |
8a4b83cc CM |
481 | device = NULL; |
482 | } else { | |
a443755f CM |
483 | device = __find_device(&fs_devices->devices, devid, |
484 | disk_super->dev_item.uuid); | |
8a4b83cc | 485 | } |
443f24fe | 486 | |
8a4b83cc | 487 | if (!device) { |
2b82032c YZ |
488 | if (fs_devices->opened) |
489 | return -EBUSY; | |
490 | ||
12bd2fc0 ID |
491 | device = btrfs_alloc_device(NULL, &devid, |
492 | disk_super->dev_item.uuid); | |
493 | if (IS_ERR(device)) { | |
8a4b83cc | 494 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 495 | return PTR_ERR(device); |
8a4b83cc | 496 | } |
606686ee JB |
497 | |
498 | name = rcu_string_strdup(path, GFP_NOFS); | |
499 | if (!name) { | |
8a4b83cc CM |
500 | kfree(device); |
501 | return -ENOMEM; | |
502 | } | |
606686ee | 503 | rcu_assign_pointer(device->name, name); |
90519d66 | 504 | |
e5e9a520 | 505 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 506 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 507 | fs_devices->num_devices++; |
e5e9a520 CM |
508 | mutex_unlock(&fs_devices->device_list_mutex); |
509 | ||
60999ca4 | 510 | ret = 1; |
2b82032c | 511 | device->fs_devices = fs_devices; |
606686ee | 512 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
513 | /* |
514 | * When FS is already mounted. | |
515 | * 1. If you are here and if the device->name is NULL that | |
516 | * means this device was missing at time of FS mount. | |
517 | * 2. If you are here and if the device->name is different | |
518 | * from 'path' that means either | |
519 | * a. The same device disappeared and reappeared with | |
520 | * different name. or | |
521 | * b. The missing-disk-which-was-replaced, has | |
522 | * reappeared now. | |
523 | * | |
524 | * We must allow 1 and 2a above. But 2b would be a spurious | |
525 | * and unintentional. | |
526 | * | |
527 | * Further in case of 1 and 2a above, the disk at 'path' | |
528 | * would have missed some transaction when it was away and | |
529 | * in case of 2a the stale bdev has to be updated as well. | |
530 | * 2b must not be allowed at all time. | |
531 | */ | |
532 | ||
533 | /* | |
534 | * As of now don't allow update to btrfs_fs_device through | |
535 | * the btrfs dev scan cli, after FS has been mounted. | |
536 | */ | |
77bdae4d | 537 | if (fs_devices->opened) { |
b96de000 | 538 | return -EBUSY; |
77bdae4d AJ |
539 | } else { |
540 | /* | |
541 | * That is if the FS is _not_ mounted and if you | |
542 | * are here, that means there is more than one | |
543 | * disk with same uuid and devid.We keep the one | |
544 | * with larger generation number or the last-in if | |
545 | * generation are equal. | |
546 | */ | |
547 | if (found_transid < device->generation) | |
548 | return -EEXIST; | |
549 | } | |
b96de000 | 550 | |
606686ee | 551 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
552 | if (!name) |
553 | return -ENOMEM; | |
606686ee JB |
554 | rcu_string_free(device->name); |
555 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
556 | if (device->missing) { |
557 | fs_devices->missing_devices--; | |
558 | device->missing = 0; | |
559 | } | |
8a4b83cc CM |
560 | } |
561 | ||
77bdae4d AJ |
562 | /* |
563 | * Unmount does not free the btrfs_device struct but would zero | |
564 | * generation along with most of the other members. So just update | |
565 | * it back. We need it to pick the disk with largest generation | |
566 | * (as above). | |
567 | */ | |
568 | if (!fs_devices->opened) | |
569 | device->generation = found_transid; | |
570 | ||
8a4b83cc | 571 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
572 | |
573 | return ret; | |
8a4b83cc CM |
574 | } |
575 | ||
e4404d6e YZ |
576 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
577 | { | |
578 | struct btrfs_fs_devices *fs_devices; | |
579 | struct btrfs_device *device; | |
580 | struct btrfs_device *orig_dev; | |
581 | ||
2208a378 ID |
582 | fs_devices = alloc_fs_devices(orig->fsid); |
583 | if (IS_ERR(fs_devices)) | |
584 | return fs_devices; | |
e4404d6e | 585 | |
adbbb863 | 586 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 587 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 588 | |
46224705 | 589 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 590 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
591 | struct rcu_string *name; |
592 | ||
12bd2fc0 ID |
593 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
594 | orig_dev->uuid); | |
595 | if (IS_ERR(device)) | |
e4404d6e YZ |
596 | goto error; |
597 | ||
606686ee JB |
598 | /* |
599 | * This is ok to do without rcu read locked because we hold the | |
600 | * uuid mutex so nothing we touch in here is going to disappear. | |
601 | */ | |
e755f780 AJ |
602 | if (orig_dev->name) { |
603 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
604 | if (!name) { | |
605 | kfree(device); | |
606 | goto error; | |
607 | } | |
608 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 609 | } |
e4404d6e | 610 | |
e4404d6e YZ |
611 | list_add(&device->dev_list, &fs_devices->devices); |
612 | device->fs_devices = fs_devices; | |
613 | fs_devices->num_devices++; | |
614 | } | |
adbbb863 | 615 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
616 | return fs_devices; |
617 | error: | |
adbbb863 | 618 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
619 | free_fs_devices(fs_devices); |
620 | return ERR_PTR(-ENOMEM); | |
621 | } | |
622 | ||
8dabb742 SB |
623 | void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info, |
624 | struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 625 | { |
c6e30871 | 626 | struct btrfs_device *device, *next; |
443f24fe | 627 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 628 | |
dfe25020 CM |
629 | mutex_lock(&uuid_mutex); |
630 | again: | |
46224705 | 631 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 632 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 633 | if (device->in_fs_metadata) { |
63a212ab | 634 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
635 | (!latest_dev || |
636 | device->generation > latest_dev->generation)) { | |
637 | latest_dev = device; | |
a6b0d5c8 | 638 | } |
2b82032c | 639 | continue; |
a6b0d5c8 | 640 | } |
2b82032c | 641 | |
8dabb742 SB |
642 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
643 | /* | |
644 | * In the first step, keep the device which has | |
645 | * the correct fsid and the devid that is used | |
646 | * for the dev_replace procedure. | |
647 | * In the second step, the dev_replace state is | |
648 | * read from the device tree and it is known | |
649 | * whether the procedure is really active or | |
650 | * not, which means whether this device is | |
651 | * used or whether it should be removed. | |
652 | */ | |
653 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
654 | continue; | |
655 | } | |
656 | } | |
2b82032c | 657 | if (device->bdev) { |
d4d77629 | 658 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
659 | device->bdev = NULL; |
660 | fs_devices->open_devices--; | |
661 | } | |
662 | if (device->writeable) { | |
663 | list_del_init(&device->dev_alloc_list); | |
664 | device->writeable = 0; | |
8dabb742 SB |
665 | if (!device->is_tgtdev_for_dev_replace) |
666 | fs_devices->rw_devices--; | |
2b82032c | 667 | } |
e4404d6e YZ |
668 | list_del_init(&device->dev_list); |
669 | fs_devices->num_devices--; | |
606686ee | 670 | rcu_string_free(device->name); |
e4404d6e | 671 | kfree(device); |
dfe25020 | 672 | } |
2b82032c YZ |
673 | |
674 | if (fs_devices->seed) { | |
675 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
676 | goto again; |
677 | } | |
678 | ||
443f24fe | 679 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 680 | |
dfe25020 | 681 | mutex_unlock(&uuid_mutex); |
dfe25020 | 682 | } |
a0af469b | 683 | |
1f78160c XG |
684 | static void __free_device(struct work_struct *work) |
685 | { | |
686 | struct btrfs_device *device; | |
687 | ||
688 | device = container_of(work, struct btrfs_device, rcu_work); | |
689 | ||
690 | if (device->bdev) | |
691 | blkdev_put(device->bdev, device->mode); | |
692 | ||
606686ee | 693 | rcu_string_free(device->name); |
1f78160c XG |
694 | kfree(device); |
695 | } | |
696 | ||
697 | static void free_device(struct rcu_head *head) | |
698 | { | |
699 | struct btrfs_device *device; | |
700 | ||
701 | device = container_of(head, struct btrfs_device, rcu); | |
702 | ||
703 | INIT_WORK(&device->rcu_work, __free_device); | |
704 | schedule_work(&device->rcu_work); | |
705 | } | |
706 | ||
2b82032c | 707 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 708 | { |
8a4b83cc | 709 | struct btrfs_device *device; |
e4404d6e | 710 | |
2b82032c YZ |
711 | if (--fs_devices->opened > 0) |
712 | return 0; | |
8a4b83cc | 713 | |
c9513edb | 714 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 715 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c | 716 | struct btrfs_device *new_device; |
606686ee | 717 | struct rcu_string *name; |
1f78160c XG |
718 | |
719 | if (device->bdev) | |
a0af469b | 720 | fs_devices->open_devices--; |
1f78160c | 721 | |
f747cab7 ID |
722 | if (device->writeable && |
723 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
724 | list_del_init(&device->dev_alloc_list); |
725 | fs_devices->rw_devices--; | |
726 | } | |
727 | ||
726551eb JB |
728 | if (device->missing) |
729 | fs_devices->missing_devices--; | |
d5e2003c | 730 | |
a1e8780a ID |
731 | new_device = btrfs_alloc_device(NULL, &device->devid, |
732 | device->uuid); | |
733 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
606686ee JB |
734 | |
735 | /* Safe because we are under uuid_mutex */ | |
99f5944b JB |
736 | if (device->name) { |
737 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
a1e8780a | 738 | BUG_ON(!name); /* -ENOMEM */ |
99f5944b JB |
739 | rcu_assign_pointer(new_device->name, name); |
740 | } | |
a1e8780a | 741 | |
1f78160c | 742 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
a1e8780a | 743 | new_device->fs_devices = device->fs_devices; |
1f78160c XG |
744 | |
745 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 746 | } |
c9513edb XG |
747 | mutex_unlock(&fs_devices->device_list_mutex); |
748 | ||
e4404d6e YZ |
749 | WARN_ON(fs_devices->open_devices); |
750 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
751 | fs_devices->opened = 0; |
752 | fs_devices->seeding = 0; | |
2b82032c | 753 | |
8a4b83cc CM |
754 | return 0; |
755 | } | |
756 | ||
2b82032c YZ |
757 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
758 | { | |
e4404d6e | 759 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
760 | int ret; |
761 | ||
762 | mutex_lock(&uuid_mutex); | |
763 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
764 | if (!fs_devices->opened) { |
765 | seed_devices = fs_devices->seed; | |
766 | fs_devices->seed = NULL; | |
767 | } | |
2b82032c | 768 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
769 | |
770 | while (seed_devices) { | |
771 | fs_devices = seed_devices; | |
772 | seed_devices = fs_devices->seed; | |
773 | __btrfs_close_devices(fs_devices); | |
774 | free_fs_devices(fs_devices); | |
775 | } | |
bc178622 ES |
776 | /* |
777 | * Wait for rcu kworkers under __btrfs_close_devices | |
778 | * to finish all blkdev_puts so device is really | |
779 | * free when umount is done. | |
780 | */ | |
781 | rcu_barrier(); | |
2b82032c YZ |
782 | return ret; |
783 | } | |
784 | ||
e4404d6e YZ |
785 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
786 | fmode_t flags, void *holder) | |
8a4b83cc | 787 | { |
d5e2003c | 788 | struct request_queue *q; |
8a4b83cc CM |
789 | struct block_device *bdev; |
790 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 791 | struct btrfs_device *device; |
443f24fe | 792 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
793 | struct buffer_head *bh; |
794 | struct btrfs_super_block *disk_super; | |
a0af469b | 795 | u64 devid; |
2b82032c | 796 | int seeding = 1; |
a0af469b | 797 | int ret = 0; |
8a4b83cc | 798 | |
d4d77629 TH |
799 | flags |= FMODE_EXCL; |
800 | ||
c6e30871 | 801 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
802 | if (device->bdev) |
803 | continue; | |
dfe25020 CM |
804 | if (!device->name) |
805 | continue; | |
806 | ||
f63e0cca ES |
807 | /* Just open everything we can; ignore failures here */ |
808 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
809 | &bdev, &bh)) | |
beaf8ab3 | 810 | continue; |
a0af469b CM |
811 | |
812 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 813 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
814 | if (devid != device->devid) |
815 | goto error_brelse; | |
816 | ||
2b82032c YZ |
817 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
818 | BTRFS_UUID_SIZE)) | |
819 | goto error_brelse; | |
820 | ||
821 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
822 | if (!latest_dev || |
823 | device->generation > latest_dev->generation) | |
824 | latest_dev = device; | |
a0af469b | 825 | |
2b82032c YZ |
826 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
827 | device->writeable = 0; | |
828 | } else { | |
829 | device->writeable = !bdev_read_only(bdev); | |
830 | seeding = 0; | |
831 | } | |
832 | ||
d5e2003c | 833 | q = bdev_get_queue(bdev); |
90180da4 | 834 | if (blk_queue_discard(q)) |
d5e2003c | 835 | device->can_discard = 1; |
d5e2003c | 836 | |
8a4b83cc | 837 | device->bdev = bdev; |
dfe25020 | 838 | device->in_fs_metadata = 0; |
15916de8 CM |
839 | device->mode = flags; |
840 | ||
c289811c CM |
841 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
842 | fs_devices->rotating = 1; | |
843 | ||
a0af469b | 844 | fs_devices->open_devices++; |
55e50e45 ID |
845 | if (device->writeable && |
846 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
847 | fs_devices->rw_devices++; |
848 | list_add(&device->dev_alloc_list, | |
849 | &fs_devices->alloc_list); | |
850 | } | |
4f6c9328 | 851 | brelse(bh); |
a0af469b | 852 | continue; |
a061fc8d | 853 | |
a0af469b CM |
854 | error_brelse: |
855 | brelse(bh); | |
d4d77629 | 856 | blkdev_put(bdev, flags); |
a0af469b | 857 | continue; |
8a4b83cc | 858 | } |
a0af469b | 859 | if (fs_devices->open_devices == 0) { |
20bcd649 | 860 | ret = -EINVAL; |
a0af469b CM |
861 | goto out; |
862 | } | |
2b82032c YZ |
863 | fs_devices->seeding = seeding; |
864 | fs_devices->opened = 1; | |
443f24fe | 865 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 866 | fs_devices->total_rw_bytes = 0; |
a0af469b | 867 | out: |
2b82032c YZ |
868 | return ret; |
869 | } | |
870 | ||
871 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 872 | fmode_t flags, void *holder) |
2b82032c YZ |
873 | { |
874 | int ret; | |
875 | ||
876 | mutex_lock(&uuid_mutex); | |
877 | if (fs_devices->opened) { | |
e4404d6e YZ |
878 | fs_devices->opened++; |
879 | ret = 0; | |
2b82032c | 880 | } else { |
15916de8 | 881 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 882 | } |
8a4b83cc | 883 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
884 | return ret; |
885 | } | |
886 | ||
6f60cbd3 DS |
887 | /* |
888 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
889 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
890 | * is read via pagecache | |
891 | */ | |
97288f2c | 892 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
893 | struct btrfs_fs_devices **fs_devices_ret) |
894 | { | |
895 | struct btrfs_super_block *disk_super; | |
896 | struct block_device *bdev; | |
6f60cbd3 DS |
897 | struct page *page; |
898 | void *p; | |
899 | int ret = -EINVAL; | |
8a4b83cc | 900 | u64 devid; |
f2984462 | 901 | u64 transid; |
02db0844 | 902 | u64 total_devices; |
6f60cbd3 DS |
903 | u64 bytenr; |
904 | pgoff_t index; | |
8a4b83cc | 905 | |
6f60cbd3 DS |
906 | /* |
907 | * we would like to check all the supers, but that would make | |
908 | * a btrfs mount succeed after a mkfs from a different FS. | |
909 | * So, we need to add a special mount option to scan for | |
910 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
911 | */ | |
912 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 913 | flags |= FMODE_EXCL; |
10f6327b | 914 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
915 | |
916 | bdev = blkdev_get_by_path(path, flags, holder); | |
917 | ||
918 | if (IS_ERR(bdev)) { | |
919 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 920 | goto error; |
6f60cbd3 DS |
921 | } |
922 | ||
923 | /* make sure our super fits in the device */ | |
924 | if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode)) | |
925 | goto error_bdev_put; | |
926 | ||
927 | /* make sure our super fits in the page */ | |
928 | if (sizeof(*disk_super) > PAGE_CACHE_SIZE) | |
929 | goto error_bdev_put; | |
930 | ||
931 | /* make sure our super doesn't straddle pages on disk */ | |
932 | index = bytenr >> PAGE_CACHE_SHIFT; | |
933 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index) | |
934 | goto error_bdev_put; | |
935 | ||
936 | /* pull in the page with our super */ | |
937 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
938 | index, GFP_NOFS); | |
939 | ||
940 | if (IS_ERR_OR_NULL(page)) | |
941 | goto error_bdev_put; | |
942 | ||
943 | p = kmap(page); | |
944 | ||
945 | /* align our pointer to the offset of the super block */ | |
946 | disk_super = p + (bytenr & ~PAGE_CACHE_MASK); | |
947 | ||
948 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
3cae210f | 949 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) |
6f60cbd3 DS |
950 | goto error_unmap; |
951 | ||
a343832f | 952 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 953 | transid = btrfs_super_generation(disk_super); |
02db0844 | 954 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 955 | |
8a4b83cc | 956 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
957 | if (ret > 0) { |
958 | if (disk_super->label[0]) { | |
959 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
960 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
961 | printk(KERN_INFO "BTRFS: device label %s ", disk_super->label); | |
962 | } else { | |
963 | printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid); | |
964 | } | |
965 | ||
966 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); | |
967 | ret = 0; | |
968 | } | |
02db0844 JB |
969 | if (!ret && fs_devices_ret) |
970 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 DS |
971 | |
972 | error_unmap: | |
973 | kunmap(page); | |
974 | page_cache_release(page); | |
975 | ||
976 | error_bdev_put: | |
d4d77629 | 977 | blkdev_put(bdev, flags); |
8a4b83cc | 978 | error: |
beaf8ab3 | 979 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
980 | return ret; |
981 | } | |
0b86a832 | 982 | |
6d07bcec MX |
983 | /* helper to account the used device space in the range */ |
984 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
985 | u64 end, u64 *length) | |
986 | { | |
987 | struct btrfs_key key; | |
988 | struct btrfs_root *root = device->dev_root; | |
989 | struct btrfs_dev_extent *dev_extent; | |
990 | struct btrfs_path *path; | |
991 | u64 extent_end; | |
992 | int ret; | |
993 | int slot; | |
994 | struct extent_buffer *l; | |
995 | ||
996 | *length = 0; | |
997 | ||
63a212ab | 998 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
999 | return 0; |
1000 | ||
1001 | path = btrfs_alloc_path(); | |
1002 | if (!path) | |
1003 | return -ENOMEM; | |
1004 | path->reada = 2; | |
1005 | ||
1006 | key.objectid = device->devid; | |
1007 | key.offset = start; | |
1008 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1009 | ||
1010 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1011 | if (ret < 0) | |
1012 | goto out; | |
1013 | if (ret > 0) { | |
1014 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1015 | if (ret < 0) | |
1016 | goto out; | |
1017 | } | |
1018 | ||
1019 | while (1) { | |
1020 | l = path->nodes[0]; | |
1021 | slot = path->slots[0]; | |
1022 | if (slot >= btrfs_header_nritems(l)) { | |
1023 | ret = btrfs_next_leaf(root, path); | |
1024 | if (ret == 0) | |
1025 | continue; | |
1026 | if (ret < 0) | |
1027 | goto out; | |
1028 | ||
1029 | break; | |
1030 | } | |
1031 | btrfs_item_key_to_cpu(l, &key, slot); | |
1032 | ||
1033 | if (key.objectid < device->devid) | |
1034 | goto next; | |
1035 | ||
1036 | if (key.objectid > device->devid) | |
1037 | break; | |
1038 | ||
962a298f | 1039 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1040 | goto next; |
1041 | ||
1042 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1043 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1044 | dev_extent); | |
1045 | if (key.offset <= start && extent_end > end) { | |
1046 | *length = end - start + 1; | |
1047 | break; | |
1048 | } else if (key.offset <= start && extent_end > start) | |
1049 | *length += extent_end - start; | |
1050 | else if (key.offset > start && extent_end <= end) | |
1051 | *length += extent_end - key.offset; | |
1052 | else if (key.offset > start && key.offset <= end) { | |
1053 | *length += end - key.offset + 1; | |
1054 | break; | |
1055 | } else if (key.offset > end) | |
1056 | break; | |
1057 | ||
1058 | next: | |
1059 | path->slots[0]++; | |
1060 | } | |
1061 | ret = 0; | |
1062 | out: | |
1063 | btrfs_free_path(path); | |
1064 | return ret; | |
1065 | } | |
1066 | ||
6df9a95e JB |
1067 | static int contains_pending_extent(struct btrfs_trans_handle *trans, |
1068 | struct btrfs_device *device, | |
1069 | u64 *start, u64 len) | |
1070 | { | |
1071 | struct extent_map *em; | |
1072 | int ret = 0; | |
1073 | ||
1074 | list_for_each_entry(em, &trans->transaction->pending_chunks, list) { | |
1075 | struct map_lookup *map; | |
1076 | int i; | |
1077 | ||
1078 | map = (struct map_lookup *)em->bdev; | |
1079 | for (i = 0; i < map->num_stripes; i++) { | |
1080 | if (map->stripes[i].dev != device) | |
1081 | continue; | |
1082 | if (map->stripes[i].physical >= *start + len || | |
1083 | map->stripes[i].physical + em->orig_block_len <= | |
1084 | *start) | |
1085 | continue; | |
1086 | *start = map->stripes[i].physical + | |
1087 | em->orig_block_len; | |
1088 | ret = 1; | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | return ret; | |
1093 | } | |
1094 | ||
1095 | ||
0b86a832 | 1096 | /* |
7bfc837d | 1097 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
1098 | * @device: the device which we search the free space in |
1099 | * @num_bytes: the size of the free space that we need | |
1100 | * @start: store the start of the free space. | |
1101 | * @len: the size of the free space. that we find, or the size of the max | |
1102 | * free space if we don't find suitable free space | |
1103 | * | |
0b86a832 CM |
1104 | * this uses a pretty simple search, the expectation is that it is |
1105 | * called very infrequently and that a given device has a small number | |
1106 | * of extents | |
7bfc837d MX |
1107 | * |
1108 | * @start is used to store the start of the free space if we find. But if we | |
1109 | * don't find suitable free space, it will be used to store the start position | |
1110 | * of the max free space. | |
1111 | * | |
1112 | * @len is used to store the size of the free space that we find. | |
1113 | * But if we don't find suitable free space, it is used to store the size of | |
1114 | * the max free space. | |
0b86a832 | 1115 | */ |
6df9a95e JB |
1116 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1117 | struct btrfs_device *device, u64 num_bytes, | |
7bfc837d | 1118 | u64 *start, u64 *len) |
0b86a832 CM |
1119 | { |
1120 | struct btrfs_key key; | |
1121 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1122 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1123 | struct btrfs_path *path; |
7bfc837d MX |
1124 | u64 hole_size; |
1125 | u64 max_hole_start; | |
1126 | u64 max_hole_size; | |
1127 | u64 extent_end; | |
1128 | u64 search_start; | |
0b86a832 CM |
1129 | u64 search_end = device->total_bytes; |
1130 | int ret; | |
7bfc837d | 1131 | int slot; |
0b86a832 CM |
1132 | struct extent_buffer *l; |
1133 | ||
0b86a832 CM |
1134 | /* FIXME use last free of some kind */ |
1135 | ||
8a4b83cc CM |
1136 | /* we don't want to overwrite the superblock on the drive, |
1137 | * so we make sure to start at an offset of at least 1MB | |
1138 | */ | |
a9c9bf68 | 1139 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 1140 | |
6df9a95e JB |
1141 | path = btrfs_alloc_path(); |
1142 | if (!path) | |
1143 | return -ENOMEM; | |
1144 | again: | |
7bfc837d MX |
1145 | max_hole_start = search_start; |
1146 | max_hole_size = 0; | |
38c01b96 | 1147 | hole_size = 0; |
7bfc837d | 1148 | |
63a212ab | 1149 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1150 | ret = -ENOSPC; |
6df9a95e | 1151 | goto out; |
7bfc837d MX |
1152 | } |
1153 | ||
7bfc837d | 1154 | path->reada = 2; |
6df9a95e JB |
1155 | path->search_commit_root = 1; |
1156 | path->skip_locking = 1; | |
7bfc837d | 1157 | |
0b86a832 CM |
1158 | key.objectid = device->devid; |
1159 | key.offset = search_start; | |
1160 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1161 | |
125ccb0a | 1162 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1163 | if (ret < 0) |
7bfc837d | 1164 | goto out; |
1fcbac58 YZ |
1165 | if (ret > 0) { |
1166 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1167 | if (ret < 0) | |
7bfc837d | 1168 | goto out; |
1fcbac58 | 1169 | } |
7bfc837d | 1170 | |
0b86a832 CM |
1171 | while (1) { |
1172 | l = path->nodes[0]; | |
1173 | slot = path->slots[0]; | |
1174 | if (slot >= btrfs_header_nritems(l)) { | |
1175 | ret = btrfs_next_leaf(root, path); | |
1176 | if (ret == 0) | |
1177 | continue; | |
1178 | if (ret < 0) | |
7bfc837d MX |
1179 | goto out; |
1180 | ||
1181 | break; | |
0b86a832 CM |
1182 | } |
1183 | btrfs_item_key_to_cpu(l, &key, slot); | |
1184 | ||
1185 | if (key.objectid < device->devid) | |
1186 | goto next; | |
1187 | ||
1188 | if (key.objectid > device->devid) | |
7bfc837d | 1189 | break; |
0b86a832 | 1190 | |
962a298f | 1191 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1192 | goto next; |
9779b72f | 1193 | |
7bfc837d MX |
1194 | if (key.offset > search_start) { |
1195 | hole_size = key.offset - search_start; | |
9779b72f | 1196 | |
6df9a95e JB |
1197 | /* |
1198 | * Have to check before we set max_hole_start, otherwise | |
1199 | * we could end up sending back this offset anyway. | |
1200 | */ | |
1201 | if (contains_pending_extent(trans, device, | |
1202 | &search_start, | |
1203 | hole_size)) | |
1204 | hole_size = 0; | |
1205 | ||
7bfc837d MX |
1206 | if (hole_size > max_hole_size) { |
1207 | max_hole_start = search_start; | |
1208 | max_hole_size = hole_size; | |
1209 | } | |
9779b72f | 1210 | |
7bfc837d MX |
1211 | /* |
1212 | * If this free space is greater than which we need, | |
1213 | * it must be the max free space that we have found | |
1214 | * until now, so max_hole_start must point to the start | |
1215 | * of this free space and the length of this free space | |
1216 | * is stored in max_hole_size. Thus, we return | |
1217 | * max_hole_start and max_hole_size and go back to the | |
1218 | * caller. | |
1219 | */ | |
1220 | if (hole_size >= num_bytes) { | |
1221 | ret = 0; | |
1222 | goto out; | |
0b86a832 CM |
1223 | } |
1224 | } | |
0b86a832 | 1225 | |
0b86a832 | 1226 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1227 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1228 | dev_extent); | |
1229 | if (extent_end > search_start) | |
1230 | search_start = extent_end; | |
0b86a832 CM |
1231 | next: |
1232 | path->slots[0]++; | |
1233 | cond_resched(); | |
1234 | } | |
0b86a832 | 1235 | |
38c01b96 | 1236 | /* |
1237 | * At this point, search_start should be the end of | |
1238 | * allocated dev extents, and when shrinking the device, | |
1239 | * search_end may be smaller than search_start. | |
1240 | */ | |
1241 | if (search_end > search_start) | |
1242 | hole_size = search_end - search_start; | |
1243 | ||
7bfc837d MX |
1244 | if (hole_size > max_hole_size) { |
1245 | max_hole_start = search_start; | |
1246 | max_hole_size = hole_size; | |
0b86a832 | 1247 | } |
0b86a832 | 1248 | |
6df9a95e JB |
1249 | if (contains_pending_extent(trans, device, &search_start, hole_size)) { |
1250 | btrfs_release_path(path); | |
1251 | goto again; | |
1252 | } | |
1253 | ||
7bfc837d MX |
1254 | /* See above. */ |
1255 | if (hole_size < num_bytes) | |
1256 | ret = -ENOSPC; | |
1257 | else | |
1258 | ret = 0; | |
1259 | ||
1260 | out: | |
2b82032c | 1261 | btrfs_free_path(path); |
7bfc837d | 1262 | *start = max_hole_start; |
b2117a39 | 1263 | if (len) |
7bfc837d | 1264 | *len = max_hole_size; |
0b86a832 CM |
1265 | return ret; |
1266 | } | |
1267 | ||
b2950863 | 1268 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1269 | struct btrfs_device *device, |
2196d6e8 | 1270 | u64 start, u64 *dev_extent_len) |
8f18cf13 CM |
1271 | { |
1272 | int ret; | |
1273 | struct btrfs_path *path; | |
1274 | struct btrfs_root *root = device->dev_root; | |
1275 | struct btrfs_key key; | |
a061fc8d CM |
1276 | struct btrfs_key found_key; |
1277 | struct extent_buffer *leaf = NULL; | |
1278 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1279 | |
1280 | path = btrfs_alloc_path(); | |
1281 | if (!path) | |
1282 | return -ENOMEM; | |
1283 | ||
1284 | key.objectid = device->devid; | |
1285 | key.offset = start; | |
1286 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1287 | again: |
8f18cf13 | 1288 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1289 | if (ret > 0) { |
1290 | ret = btrfs_previous_item(root, path, key.objectid, | |
1291 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1292 | if (ret) |
1293 | goto out; | |
a061fc8d CM |
1294 | leaf = path->nodes[0]; |
1295 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1296 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1297 | struct btrfs_dev_extent); | |
1298 | BUG_ON(found_key.offset > start || found_key.offset + | |
1299 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1300 | key = found_key; |
1301 | btrfs_release_path(path); | |
1302 | goto again; | |
a061fc8d CM |
1303 | } else if (ret == 0) { |
1304 | leaf = path->nodes[0]; | |
1305 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1306 | struct btrfs_dev_extent); | |
79787eaa JM |
1307 | } else { |
1308 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1309 | goto out; | |
a061fc8d | 1310 | } |
8f18cf13 | 1311 | |
2196d6e8 MX |
1312 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1313 | ||
8f18cf13 | 1314 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1315 | if (ret) { |
1316 | btrfs_error(root->fs_info, ret, | |
1317 | "Failed to remove dev extent item"); | |
1318 | } | |
b0b802d7 | 1319 | out: |
8f18cf13 CM |
1320 | btrfs_free_path(path); |
1321 | return ret; | |
1322 | } | |
1323 | ||
48a3b636 ES |
1324 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1325 | struct btrfs_device *device, | |
1326 | u64 chunk_tree, u64 chunk_objectid, | |
1327 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1328 | { |
1329 | int ret; | |
1330 | struct btrfs_path *path; | |
1331 | struct btrfs_root *root = device->dev_root; | |
1332 | struct btrfs_dev_extent *extent; | |
1333 | struct extent_buffer *leaf; | |
1334 | struct btrfs_key key; | |
1335 | ||
dfe25020 | 1336 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1337 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1338 | path = btrfs_alloc_path(); |
1339 | if (!path) | |
1340 | return -ENOMEM; | |
1341 | ||
0b86a832 | 1342 | key.objectid = device->devid; |
2b82032c | 1343 | key.offset = start; |
0b86a832 CM |
1344 | key.type = BTRFS_DEV_EXTENT_KEY; |
1345 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1346 | sizeof(*extent)); | |
2cdcecbc MF |
1347 | if (ret) |
1348 | goto out; | |
0b86a832 CM |
1349 | |
1350 | leaf = path->nodes[0]; | |
1351 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1352 | struct btrfs_dev_extent); | |
e17cade2 CM |
1353 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1354 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1355 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1356 | ||
1357 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1358 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1359 | |
0b86a832 CM |
1360 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1361 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1362 | out: |
0b86a832 CM |
1363 | btrfs_free_path(path); |
1364 | return ret; | |
1365 | } | |
1366 | ||
6df9a95e | 1367 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1368 | { |
6df9a95e JB |
1369 | struct extent_map_tree *em_tree; |
1370 | struct extent_map *em; | |
1371 | struct rb_node *n; | |
1372 | u64 ret = 0; | |
0b86a832 | 1373 | |
6df9a95e JB |
1374 | em_tree = &fs_info->mapping_tree.map_tree; |
1375 | read_lock(&em_tree->lock); | |
1376 | n = rb_last(&em_tree->map); | |
1377 | if (n) { | |
1378 | em = rb_entry(n, struct extent_map, rb_node); | |
1379 | ret = em->start + em->len; | |
0b86a832 | 1380 | } |
6df9a95e JB |
1381 | read_unlock(&em_tree->lock); |
1382 | ||
0b86a832 CM |
1383 | return ret; |
1384 | } | |
1385 | ||
53f10659 ID |
1386 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1387 | u64 *devid_ret) | |
0b86a832 CM |
1388 | { |
1389 | int ret; | |
1390 | struct btrfs_key key; | |
1391 | struct btrfs_key found_key; | |
2b82032c YZ |
1392 | struct btrfs_path *path; |
1393 | ||
2b82032c YZ |
1394 | path = btrfs_alloc_path(); |
1395 | if (!path) | |
1396 | return -ENOMEM; | |
0b86a832 CM |
1397 | |
1398 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1399 | key.type = BTRFS_DEV_ITEM_KEY; | |
1400 | key.offset = (u64)-1; | |
1401 | ||
53f10659 | 1402 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1403 | if (ret < 0) |
1404 | goto error; | |
1405 | ||
79787eaa | 1406 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1407 | |
53f10659 ID |
1408 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1409 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1410 | BTRFS_DEV_ITEM_KEY); |
1411 | if (ret) { | |
53f10659 | 1412 | *devid_ret = 1; |
0b86a832 CM |
1413 | } else { |
1414 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1415 | path->slots[0]); | |
53f10659 | 1416 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1417 | } |
1418 | ret = 0; | |
1419 | error: | |
2b82032c | 1420 | btrfs_free_path(path); |
0b86a832 CM |
1421 | return ret; |
1422 | } | |
1423 | ||
1424 | /* | |
1425 | * the device information is stored in the chunk root | |
1426 | * the btrfs_device struct should be fully filled in | |
1427 | */ | |
48a3b636 ES |
1428 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1429 | struct btrfs_root *root, | |
1430 | struct btrfs_device *device) | |
0b86a832 CM |
1431 | { |
1432 | int ret; | |
1433 | struct btrfs_path *path; | |
1434 | struct btrfs_dev_item *dev_item; | |
1435 | struct extent_buffer *leaf; | |
1436 | struct btrfs_key key; | |
1437 | unsigned long ptr; | |
0b86a832 CM |
1438 | |
1439 | root = root->fs_info->chunk_root; | |
1440 | ||
1441 | path = btrfs_alloc_path(); | |
1442 | if (!path) | |
1443 | return -ENOMEM; | |
1444 | ||
0b86a832 CM |
1445 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1446 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1447 | key.offset = device->devid; |
0b86a832 CM |
1448 | |
1449 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1450 | sizeof(*dev_item)); |
0b86a832 CM |
1451 | if (ret) |
1452 | goto out; | |
1453 | ||
1454 | leaf = path->nodes[0]; | |
1455 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1456 | ||
1457 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1458 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1459 | btrfs_set_device_type(leaf, dev_item, device->type); |
1460 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1461 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1462 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1463 | btrfs_set_device_total_bytes(leaf, dev_item, |
1464 | btrfs_device_get_disk_total_bytes(device)); | |
1465 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1466 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1467 | btrfs_set_device_group(leaf, dev_item, 0); |
1468 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1469 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1470 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1471 | |
410ba3a2 | 1472 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1473 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1474 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1475 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1476 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1477 | |
2b82032c | 1478 | ret = 0; |
0b86a832 CM |
1479 | out: |
1480 | btrfs_free_path(path); | |
1481 | return ret; | |
1482 | } | |
8f18cf13 | 1483 | |
5a1972bd QW |
1484 | /* |
1485 | * Function to update ctime/mtime for a given device path. | |
1486 | * Mainly used for ctime/mtime based probe like libblkid. | |
1487 | */ | |
1488 | static void update_dev_time(char *path_name) | |
1489 | { | |
1490 | struct file *filp; | |
1491 | ||
1492 | filp = filp_open(path_name, O_RDWR, 0); | |
1493 | if (!filp) | |
1494 | return; | |
1495 | file_update_time(filp); | |
1496 | filp_close(filp, NULL); | |
1497 | return; | |
1498 | } | |
1499 | ||
a061fc8d CM |
1500 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1501 | struct btrfs_device *device) | |
1502 | { | |
1503 | int ret; | |
1504 | struct btrfs_path *path; | |
a061fc8d | 1505 | struct btrfs_key key; |
a061fc8d CM |
1506 | struct btrfs_trans_handle *trans; |
1507 | ||
1508 | root = root->fs_info->chunk_root; | |
1509 | ||
1510 | path = btrfs_alloc_path(); | |
1511 | if (!path) | |
1512 | return -ENOMEM; | |
1513 | ||
a22285a6 | 1514 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1515 | if (IS_ERR(trans)) { |
1516 | btrfs_free_path(path); | |
1517 | return PTR_ERR(trans); | |
1518 | } | |
a061fc8d CM |
1519 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1520 | key.type = BTRFS_DEV_ITEM_KEY; | |
1521 | key.offset = device->devid; | |
1522 | ||
1523 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1524 | if (ret < 0) | |
1525 | goto out; | |
1526 | ||
1527 | if (ret > 0) { | |
1528 | ret = -ENOENT; | |
1529 | goto out; | |
1530 | } | |
1531 | ||
1532 | ret = btrfs_del_item(trans, root, path); | |
1533 | if (ret) | |
1534 | goto out; | |
a061fc8d CM |
1535 | out: |
1536 | btrfs_free_path(path); | |
1537 | btrfs_commit_transaction(trans, root); | |
1538 | return ret; | |
1539 | } | |
1540 | ||
1541 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1542 | { | |
1543 | struct btrfs_device *device; | |
2b82032c | 1544 | struct btrfs_device *next_device; |
a061fc8d | 1545 | struct block_device *bdev; |
dfe25020 | 1546 | struct buffer_head *bh = NULL; |
a061fc8d | 1547 | struct btrfs_super_block *disk_super; |
1f78160c | 1548 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1549 | u64 all_avail; |
1550 | u64 devid; | |
2b82032c YZ |
1551 | u64 num_devices; |
1552 | u8 *dev_uuid; | |
de98ced9 | 1553 | unsigned seq; |
a061fc8d | 1554 | int ret = 0; |
1f78160c | 1555 | bool clear_super = false; |
a061fc8d | 1556 | |
a061fc8d CM |
1557 | mutex_lock(&uuid_mutex); |
1558 | ||
de98ced9 MX |
1559 | do { |
1560 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
1561 | ||
1562 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1563 | root->fs_info->avail_system_alloc_bits | | |
1564 | root->fs_info->avail_metadata_alloc_bits; | |
1565 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
a061fc8d | 1566 | |
8dabb742 SB |
1567 | num_devices = root->fs_info->fs_devices->num_devices; |
1568 | btrfs_dev_replace_lock(&root->fs_info->dev_replace); | |
1569 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1570 | WARN_ON(num_devices < 1); | |
1571 | num_devices--; | |
1572 | } | |
1573 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace); | |
1574 | ||
1575 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) { | |
183860f6 | 1576 | ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET; |
a061fc8d CM |
1577 | goto out; |
1578 | } | |
1579 | ||
8dabb742 | 1580 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) { |
183860f6 | 1581 | ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET; |
a061fc8d CM |
1582 | goto out; |
1583 | } | |
1584 | ||
53b381b3 DW |
1585 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && |
1586 | root->fs_info->fs_devices->rw_devices <= 2) { | |
183860f6 | 1587 | ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET; |
53b381b3 DW |
1588 | goto out; |
1589 | } | |
1590 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1591 | root->fs_info->fs_devices->rw_devices <= 3) { | |
183860f6 | 1592 | ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET; |
53b381b3 DW |
1593 | goto out; |
1594 | } | |
1595 | ||
dfe25020 | 1596 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1597 | struct list_head *devices; |
1598 | struct btrfs_device *tmp; | |
a061fc8d | 1599 | |
dfe25020 CM |
1600 | device = NULL; |
1601 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1602 | /* |
1603 | * It is safe to read the devices since the volume_mutex | |
1604 | * is held. | |
1605 | */ | |
c6e30871 | 1606 | list_for_each_entry(tmp, devices, dev_list) { |
63a212ab SB |
1607 | if (tmp->in_fs_metadata && |
1608 | !tmp->is_tgtdev_for_dev_replace && | |
1609 | !tmp->bdev) { | |
dfe25020 CM |
1610 | device = tmp; |
1611 | break; | |
1612 | } | |
1613 | } | |
1614 | bdev = NULL; | |
1615 | bh = NULL; | |
1616 | disk_super = NULL; | |
1617 | if (!device) { | |
183860f6 | 1618 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
dfe25020 CM |
1619 | goto out; |
1620 | } | |
dfe25020 | 1621 | } else { |
beaf8ab3 | 1622 | ret = btrfs_get_bdev_and_sb(device_path, |
cc975eb4 | 1623 | FMODE_WRITE | FMODE_EXCL, |
beaf8ab3 SB |
1624 | root->fs_info->bdev_holder, 0, |
1625 | &bdev, &bh); | |
1626 | if (ret) | |
dfe25020 | 1627 | goto out; |
dfe25020 | 1628 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1629 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c | 1630 | dev_uuid = disk_super->dev_item.uuid; |
aa1b8cd4 | 1631 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2b82032c | 1632 | disk_super->fsid); |
dfe25020 CM |
1633 | if (!device) { |
1634 | ret = -ENOENT; | |
1635 | goto error_brelse; | |
1636 | } | |
2b82032c | 1637 | } |
dfe25020 | 1638 | |
63a212ab | 1639 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1640 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
63a212ab SB |
1641 | goto error_brelse; |
1642 | } | |
1643 | ||
2b82032c | 1644 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1645 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
2b82032c YZ |
1646 | goto error_brelse; |
1647 | } | |
1648 | ||
1649 | if (device->writeable) { | |
0c1daee0 | 1650 | lock_chunks(root); |
2b82032c | 1651 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1652 | device->fs_devices->rw_devices--; |
0c1daee0 | 1653 | unlock_chunks(root); |
1f78160c | 1654 | clear_super = true; |
dfe25020 | 1655 | } |
a061fc8d | 1656 | |
d7901554 | 1657 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1658 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1659 | mutex_lock(&uuid_mutex); |
a061fc8d | 1660 | if (ret) |
9b3517e9 | 1661 | goto error_undo; |
a061fc8d | 1662 | |
63a212ab SB |
1663 | /* |
1664 | * TODO: the superblock still includes this device in its num_devices | |
1665 | * counter although write_all_supers() is not locked out. This | |
1666 | * could give a filesystem state which requires a degraded mount. | |
1667 | */ | |
a061fc8d CM |
1668 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1669 | if (ret) | |
9b3517e9 | 1670 | goto error_undo; |
a061fc8d | 1671 | |
2b82032c | 1672 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1673 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1674 | |
1675 | /* | |
1676 | * the device list mutex makes sure that we don't change | |
1677 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1678 | * the device supers. Whoever is writing all supers, should |
1679 | * lock the device list mutex before getting the number of | |
1680 | * devices in the super block (super_copy). Conversely, | |
1681 | * whoever updates the number of devices in the super block | |
1682 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1683 | */ |
1f78160c XG |
1684 | |
1685 | cur_devices = device->fs_devices; | |
e5e9a520 | 1686 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1687 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1688 | |
e4404d6e | 1689 | device->fs_devices->num_devices--; |
02db0844 | 1690 | device->fs_devices->total_devices--; |
2b82032c | 1691 | |
cd02dca5 | 1692 | if (device->missing) |
3a7d55c8 | 1693 | device->fs_devices->missing_devices--; |
cd02dca5 | 1694 | |
2b82032c YZ |
1695 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1696 | struct btrfs_device, dev_list); | |
1697 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1698 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1699 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1700 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1701 | ||
0bfaa9c5 | 1702 | if (device->bdev) { |
e4404d6e | 1703 | device->fs_devices->open_devices--; |
0bfaa9c5 ES |
1704 | /* remove sysfs entry */ |
1705 | btrfs_kobj_rm_device(root->fs_info, device); | |
1706 | } | |
99994cde | 1707 | |
1f78160c | 1708 | call_rcu(&device->rcu, free_device); |
e4404d6e | 1709 | |
6c41761f DS |
1710 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1711 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
d7306801 | 1712 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2b82032c | 1713 | |
1f78160c | 1714 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1715 | struct btrfs_fs_devices *fs_devices; |
1716 | fs_devices = root->fs_info->fs_devices; | |
1717 | while (fs_devices) { | |
8321cf25 RS |
1718 | if (fs_devices->seed == cur_devices) { |
1719 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1720 | break; |
8321cf25 | 1721 | } |
e4404d6e | 1722 | fs_devices = fs_devices->seed; |
2b82032c | 1723 | } |
1f78160c | 1724 | cur_devices->seed = NULL; |
1f78160c | 1725 | __btrfs_close_devices(cur_devices); |
1f78160c | 1726 | free_fs_devices(cur_devices); |
2b82032c YZ |
1727 | } |
1728 | ||
5af3e8cc SB |
1729 | root->fs_info->num_tolerated_disk_barrier_failures = |
1730 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1731 | ||
2b82032c YZ |
1732 | /* |
1733 | * at this point, the device is zero sized. We want to | |
1734 | * remove it from the devices list and zero out the old super | |
1735 | */ | |
aa1b8cd4 | 1736 | if (clear_super && disk_super) { |
4d90d28b AJ |
1737 | u64 bytenr; |
1738 | int i; | |
1739 | ||
dfe25020 CM |
1740 | /* make sure this device isn't detected as part of |
1741 | * the FS anymore | |
1742 | */ | |
1743 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1744 | set_buffer_dirty(bh); | |
1745 | sync_dirty_buffer(bh); | |
4d90d28b AJ |
1746 | |
1747 | /* clear the mirror copies of super block on the disk | |
1748 | * being removed, 0th copy is been taken care above and | |
1749 | * the below would take of the rest | |
1750 | */ | |
1751 | for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1752 | bytenr = btrfs_sb_offset(i); | |
1753 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
1754 | i_size_read(bdev->bd_inode)) | |
1755 | break; | |
1756 | ||
1757 | brelse(bh); | |
1758 | bh = __bread(bdev, bytenr / 4096, | |
1759 | BTRFS_SUPER_INFO_SIZE); | |
1760 | if (!bh) | |
1761 | continue; | |
1762 | ||
1763 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1764 | ||
1765 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
1766 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { | |
1767 | continue; | |
1768 | } | |
1769 | memset(&disk_super->magic, 0, | |
1770 | sizeof(disk_super->magic)); | |
1771 | set_buffer_dirty(bh); | |
1772 | sync_dirty_buffer(bh); | |
1773 | } | |
dfe25020 | 1774 | } |
a061fc8d | 1775 | |
a061fc8d | 1776 | ret = 0; |
a061fc8d | 1777 | |
5a1972bd QW |
1778 | if (bdev) { |
1779 | /* Notify udev that device has changed */ | |
3c911608 | 1780 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); |
b8b8ff59 | 1781 | |
5a1972bd QW |
1782 | /* Update ctime/mtime for device path for libblkid */ |
1783 | update_dev_time(device_path); | |
1784 | } | |
1785 | ||
a061fc8d CM |
1786 | error_brelse: |
1787 | brelse(bh); | |
dfe25020 | 1788 | if (bdev) |
e525fd89 | 1789 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1790 | out: |
1791 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1792 | return ret; |
9b3517e9 ID |
1793 | error_undo: |
1794 | if (device->writeable) { | |
0c1daee0 | 1795 | lock_chunks(root); |
9b3517e9 ID |
1796 | list_add(&device->dev_alloc_list, |
1797 | &root->fs_info->fs_devices->alloc_list); | |
c3929c36 | 1798 | device->fs_devices->rw_devices++; |
0c1daee0 | 1799 | unlock_chunks(root); |
9b3517e9 ID |
1800 | } |
1801 | goto error_brelse; | |
a061fc8d CM |
1802 | } |
1803 | ||
e93c89c1 SB |
1804 | void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info, |
1805 | struct btrfs_device *srcdev) | |
1806 | { | |
d51908ce AJ |
1807 | struct btrfs_fs_devices *fs_devices; |
1808 | ||
e93c89c1 | 1809 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 1810 | |
25e8e911 AJ |
1811 | /* |
1812 | * in case of fs with no seed, srcdev->fs_devices will point | |
1813 | * to fs_devices of fs_info. However when the dev being replaced is | |
1814 | * a seed dev it will point to the seed's local fs_devices. In short | |
1815 | * srcdev will have its correct fs_devices in both the cases. | |
1816 | */ | |
1817 | fs_devices = srcdev->fs_devices; | |
d51908ce | 1818 | |
e93c89c1 SB |
1819 | list_del_rcu(&srcdev->dev_list); |
1820 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 1821 | fs_devices->num_devices--; |
82372bc8 | 1822 | if (srcdev->missing) |
d51908ce | 1823 | fs_devices->missing_devices--; |
82372bc8 MX |
1824 | |
1825 | if (srcdev->writeable) { | |
1826 | fs_devices->rw_devices--; | |
1827 | /* zero out the old super if it is writable */ | |
1828 | btrfs_scratch_superblock(srcdev); | |
e93c89c1 | 1829 | } |
90180da4 | 1830 | |
82372bc8 | 1831 | if (srcdev->bdev) |
d51908ce | 1832 | fs_devices->open_devices--; |
e93c89c1 SB |
1833 | |
1834 | call_rcu(&srcdev->rcu, free_device); | |
94d5f0c2 AJ |
1835 | |
1836 | /* | |
1837 | * unless fs_devices is seed fs, num_devices shouldn't go | |
1838 | * zero | |
1839 | */ | |
1840 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
1841 | ||
1842 | /* if this is no devs we rather delete the fs_devices */ | |
1843 | if (!fs_devices->num_devices) { | |
1844 | struct btrfs_fs_devices *tmp_fs_devices; | |
1845 | ||
1846 | tmp_fs_devices = fs_info->fs_devices; | |
1847 | while (tmp_fs_devices) { | |
1848 | if (tmp_fs_devices->seed == fs_devices) { | |
1849 | tmp_fs_devices->seed = fs_devices->seed; | |
1850 | break; | |
1851 | } | |
1852 | tmp_fs_devices = tmp_fs_devices->seed; | |
1853 | } | |
1854 | fs_devices->seed = NULL; | |
8bef8401 AJ |
1855 | __btrfs_close_devices(fs_devices); |
1856 | free_fs_devices(fs_devices); | |
94d5f0c2 | 1857 | } |
e93c89c1 SB |
1858 | } |
1859 | ||
1860 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
1861 | struct btrfs_device *tgtdev) | |
1862 | { | |
1863 | struct btrfs_device *next_device; | |
1864 | ||
67a2c45e | 1865 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
1866 | WARN_ON(!tgtdev); |
1867 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
1868 | if (tgtdev->bdev) { | |
1869 | btrfs_scratch_superblock(tgtdev); | |
1870 | fs_info->fs_devices->open_devices--; | |
1871 | } | |
1872 | fs_info->fs_devices->num_devices--; | |
e93c89c1 SB |
1873 | |
1874 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
1875 | struct btrfs_device, dev_list); | |
1876 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
1877 | fs_info->sb->s_bdev = next_device->bdev; | |
1878 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
1879 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1880 | list_del_rcu(&tgtdev->dev_list); | |
1881 | ||
1882 | call_rcu(&tgtdev->rcu, free_device); | |
1883 | ||
1884 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 1885 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
1886 | } |
1887 | ||
48a3b636 ES |
1888 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
1889 | struct btrfs_device **device) | |
7ba15b7d SB |
1890 | { |
1891 | int ret = 0; | |
1892 | struct btrfs_super_block *disk_super; | |
1893 | u64 devid; | |
1894 | u8 *dev_uuid; | |
1895 | struct block_device *bdev; | |
1896 | struct buffer_head *bh; | |
1897 | ||
1898 | *device = NULL; | |
1899 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
1900 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
1901 | if (ret) | |
1902 | return ret; | |
1903 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1904 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
1905 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 1906 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
1907 | disk_super->fsid); |
1908 | brelse(bh); | |
1909 | if (!*device) | |
1910 | ret = -ENOENT; | |
1911 | blkdev_put(bdev, FMODE_READ); | |
1912 | return ret; | |
1913 | } | |
1914 | ||
1915 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
1916 | char *device_path, | |
1917 | struct btrfs_device **device) | |
1918 | { | |
1919 | *device = NULL; | |
1920 | if (strcmp(device_path, "missing") == 0) { | |
1921 | struct list_head *devices; | |
1922 | struct btrfs_device *tmp; | |
1923 | ||
1924 | devices = &root->fs_info->fs_devices->devices; | |
1925 | /* | |
1926 | * It is safe to read the devices since the volume_mutex | |
1927 | * is held by the caller. | |
1928 | */ | |
1929 | list_for_each_entry(tmp, devices, dev_list) { | |
1930 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
1931 | *device = tmp; | |
1932 | break; | |
1933 | } | |
1934 | } | |
1935 | ||
1936 | if (!*device) { | |
efe120a0 | 1937 | btrfs_err(root->fs_info, "no missing device found"); |
7ba15b7d SB |
1938 | return -ENOENT; |
1939 | } | |
1940 | ||
1941 | return 0; | |
1942 | } else { | |
1943 | return btrfs_find_device_by_path(root, device_path, device); | |
1944 | } | |
1945 | } | |
1946 | ||
2b82032c YZ |
1947 | /* |
1948 | * does all the dirty work required for changing file system's UUID. | |
1949 | */ | |
125ccb0a | 1950 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1951 | { |
1952 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1953 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1954 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1955 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1956 | struct btrfs_device *device; |
1957 | u64 super_flags; | |
1958 | ||
1959 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1960 | if (!fs_devices->seeding) |
2b82032c YZ |
1961 | return -EINVAL; |
1962 | ||
2208a378 ID |
1963 | seed_devices = __alloc_fs_devices(); |
1964 | if (IS_ERR(seed_devices)) | |
1965 | return PTR_ERR(seed_devices); | |
2b82032c | 1966 | |
e4404d6e YZ |
1967 | old_devices = clone_fs_devices(fs_devices); |
1968 | if (IS_ERR(old_devices)) { | |
1969 | kfree(seed_devices); | |
1970 | return PTR_ERR(old_devices); | |
2b82032c | 1971 | } |
e4404d6e | 1972 | |
2b82032c YZ |
1973 | list_add(&old_devices->list, &fs_uuids); |
1974 | ||
e4404d6e YZ |
1975 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1976 | seed_devices->opened = 1; | |
1977 | INIT_LIST_HEAD(&seed_devices->devices); | |
1978 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1979 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1980 | |
1981 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1982 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1983 | synchronize_rcu); | |
2196d6e8 MX |
1984 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
1985 | device->fs_devices = seed_devices; | |
c9513edb | 1986 | |
2196d6e8 | 1987 | lock_chunks(root); |
e4404d6e | 1988 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
2196d6e8 | 1989 | unlock_chunks(root); |
e4404d6e | 1990 | |
2b82032c YZ |
1991 | fs_devices->seeding = 0; |
1992 | fs_devices->num_devices = 0; | |
1993 | fs_devices->open_devices = 0; | |
69611ac8 | 1994 | fs_devices->missing_devices = 0; |
69611ac8 | 1995 | fs_devices->rotating = 0; |
e4404d6e | 1996 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1997 | |
1998 | generate_random_uuid(fs_devices->fsid); | |
1999 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2000 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
f7171750 FDBM |
2001 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2002 | ||
2b82032c YZ |
2003 | super_flags = btrfs_super_flags(disk_super) & |
2004 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2005 | btrfs_set_super_flags(disk_super, super_flags); | |
2006 | ||
2007 | return 0; | |
2008 | } | |
2009 | ||
2010 | /* | |
2011 | * strore the expected generation for seed devices in device items. | |
2012 | */ | |
2013 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2014 | struct btrfs_root *root) | |
2015 | { | |
2016 | struct btrfs_path *path; | |
2017 | struct extent_buffer *leaf; | |
2018 | struct btrfs_dev_item *dev_item; | |
2019 | struct btrfs_device *device; | |
2020 | struct btrfs_key key; | |
2021 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2022 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2023 | u64 devid; | |
2024 | int ret; | |
2025 | ||
2026 | path = btrfs_alloc_path(); | |
2027 | if (!path) | |
2028 | return -ENOMEM; | |
2029 | ||
2030 | root = root->fs_info->chunk_root; | |
2031 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2032 | key.offset = 0; | |
2033 | key.type = BTRFS_DEV_ITEM_KEY; | |
2034 | ||
2035 | while (1) { | |
2036 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2037 | if (ret < 0) | |
2038 | goto error; | |
2039 | ||
2040 | leaf = path->nodes[0]; | |
2041 | next_slot: | |
2042 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2043 | ret = btrfs_next_leaf(root, path); | |
2044 | if (ret > 0) | |
2045 | break; | |
2046 | if (ret < 0) | |
2047 | goto error; | |
2048 | leaf = path->nodes[0]; | |
2049 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2050 | btrfs_release_path(path); |
2b82032c YZ |
2051 | continue; |
2052 | } | |
2053 | ||
2054 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2055 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2056 | key.type != BTRFS_DEV_ITEM_KEY) | |
2057 | break; | |
2058 | ||
2059 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2060 | struct btrfs_dev_item); | |
2061 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2062 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2063 | BTRFS_UUID_SIZE); |
1473b24e | 2064 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2065 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
2066 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2067 | fs_uuid); | |
79787eaa | 2068 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2069 | |
2070 | if (device->fs_devices->seeding) { | |
2071 | btrfs_set_device_generation(leaf, dev_item, | |
2072 | device->generation); | |
2073 | btrfs_mark_buffer_dirty(leaf); | |
2074 | } | |
2075 | ||
2076 | path->slots[0]++; | |
2077 | goto next_slot; | |
2078 | } | |
2079 | ret = 0; | |
2080 | error: | |
2081 | btrfs_free_path(path); | |
2082 | return ret; | |
2083 | } | |
2084 | ||
788f20eb CM |
2085 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
2086 | { | |
d5e2003c | 2087 | struct request_queue *q; |
788f20eb CM |
2088 | struct btrfs_trans_handle *trans; |
2089 | struct btrfs_device *device; | |
2090 | struct block_device *bdev; | |
788f20eb | 2091 | struct list_head *devices; |
2b82032c | 2092 | struct super_block *sb = root->fs_info->sb; |
606686ee | 2093 | struct rcu_string *name; |
3c1dbdf5 | 2094 | u64 tmp; |
2b82032c | 2095 | int seeding_dev = 0; |
788f20eb CM |
2096 | int ret = 0; |
2097 | ||
2b82032c | 2098 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 2099 | return -EROFS; |
788f20eb | 2100 | |
a5d16333 | 2101 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 2102 | root->fs_info->bdev_holder); |
7f59203a JB |
2103 | if (IS_ERR(bdev)) |
2104 | return PTR_ERR(bdev); | |
a2135011 | 2105 | |
2b82032c YZ |
2106 | if (root->fs_info->fs_devices->seeding) { |
2107 | seeding_dev = 1; | |
2108 | down_write(&sb->s_umount); | |
2109 | mutex_lock(&uuid_mutex); | |
2110 | } | |
2111 | ||
8c8bee1d | 2112 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2113 | |
788f20eb | 2114 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
2115 | |
2116 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 2117 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2118 | if (device->bdev == bdev) { |
2119 | ret = -EEXIST; | |
d25628bd LB |
2120 | mutex_unlock( |
2121 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2122 | goto error; |
788f20eb CM |
2123 | } |
2124 | } | |
d25628bd | 2125 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2126 | |
12bd2fc0 ID |
2127 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
2128 | if (IS_ERR(device)) { | |
788f20eb | 2129 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2130 | ret = PTR_ERR(device); |
2b82032c | 2131 | goto error; |
788f20eb CM |
2132 | } |
2133 | ||
606686ee JB |
2134 | name = rcu_string_strdup(device_path, GFP_NOFS); |
2135 | if (!name) { | |
788f20eb | 2136 | kfree(device); |
2b82032c YZ |
2137 | ret = -ENOMEM; |
2138 | goto error; | |
788f20eb | 2139 | } |
606686ee | 2140 | rcu_assign_pointer(device->name, name); |
2b82032c | 2141 | |
a22285a6 | 2142 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2143 | if (IS_ERR(trans)) { |
606686ee | 2144 | rcu_string_free(device->name); |
98d5dc13 TI |
2145 | kfree(device); |
2146 | ret = PTR_ERR(trans); | |
2147 | goto error; | |
2148 | } | |
2149 | ||
d5e2003c JB |
2150 | q = bdev_get_queue(bdev); |
2151 | if (blk_queue_discard(q)) | |
2152 | device->can_discard = 1; | |
2b82032c | 2153 | device->writeable = 1; |
2b82032c | 2154 | device->generation = trans->transid; |
788f20eb CM |
2155 | device->io_width = root->sectorsize; |
2156 | device->io_align = root->sectorsize; | |
2157 | device->sector_size = root->sectorsize; | |
2158 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2159 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2160 | device->commit_total_bytes = device->total_bytes; |
788f20eb CM |
2161 | device->dev_root = root->fs_info->dev_root; |
2162 | device->bdev = bdev; | |
dfe25020 | 2163 | device->in_fs_metadata = 1; |
63a212ab | 2164 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2165 | device->mode = FMODE_EXCL; |
27087f37 | 2166 | device->dev_stats_valid = 1; |
2b82032c | 2167 | set_blocksize(device->bdev, 4096); |
788f20eb | 2168 | |
2b82032c YZ |
2169 | if (seeding_dev) { |
2170 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2171 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2172 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2173 | } |
788f20eb | 2174 | |
2b82032c | 2175 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2176 | |
e5e9a520 | 2177 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
2196d6e8 | 2178 | lock_chunks(root); |
1f78160c | 2179 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2180 | list_add(&device->dev_alloc_list, |
2181 | &root->fs_info->fs_devices->alloc_list); | |
2182 | root->fs_info->fs_devices->num_devices++; | |
2183 | root->fs_info->fs_devices->open_devices++; | |
2184 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2185 | root->fs_info->fs_devices->total_devices++; |
2b82032c | 2186 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2187 | |
2bf64758 JB |
2188 | spin_lock(&root->fs_info->free_chunk_lock); |
2189 | root->fs_info->free_chunk_space += device->total_bytes; | |
2190 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2191 | ||
c289811c CM |
2192 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2193 | root->fs_info->fs_devices->rotating = 1; | |
2194 | ||
3c1dbdf5 | 2195 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); |
6c41761f | 2196 | btrfs_set_super_total_bytes(root->fs_info->super_copy, |
3c1dbdf5 | 2197 | tmp + device->total_bytes); |
788f20eb | 2198 | |
3c1dbdf5 | 2199 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); |
6c41761f | 2200 | btrfs_set_super_num_devices(root->fs_info->super_copy, |
3c1dbdf5 | 2201 | tmp + 1); |
0d39376a AJ |
2202 | |
2203 | /* add sysfs device entry */ | |
2204 | btrfs_kobj_add_device(root->fs_info, device); | |
2205 | ||
2196d6e8 MX |
2206 | /* |
2207 | * we've got more storage, clear any full flags on the space | |
2208 | * infos | |
2209 | */ | |
2210 | btrfs_clear_space_info_full(root->fs_info); | |
2211 | ||
2212 | unlock_chunks(root); | |
e5e9a520 | 2213 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2214 | |
2b82032c | 2215 | if (seeding_dev) { |
2196d6e8 | 2216 | lock_chunks(root); |
2b82032c | 2217 | ret = init_first_rw_device(trans, root, device); |
2196d6e8 | 2218 | unlock_chunks(root); |
005d6427 DS |
2219 | if (ret) { |
2220 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2221 | goto error_trans; |
005d6427 | 2222 | } |
2196d6e8 MX |
2223 | } |
2224 | ||
2225 | ret = btrfs_add_device(trans, root, device); | |
2226 | if (ret) { | |
2227 | btrfs_abort_transaction(trans, root, ret); | |
2228 | goto error_trans; | |
2229 | } | |
2230 | ||
2231 | if (seeding_dev) { | |
2232 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2233 | ||
2b82032c | 2234 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2235 | if (ret) { |
2236 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2237 | goto error_trans; |
005d6427 | 2238 | } |
b2373f25 AJ |
2239 | |
2240 | /* Sprouting would change fsid of the mounted root, | |
2241 | * so rename the fsid on the sysfs | |
2242 | */ | |
2243 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2244 | root->fs_info->fsid); | |
2245 | if (kobject_rename(&root->fs_info->super_kobj, fsid_buf)) | |
2246 | goto error_trans; | |
2b82032c YZ |
2247 | } |
2248 | ||
5af3e8cc SB |
2249 | root->fs_info->num_tolerated_disk_barrier_failures = |
2250 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2251 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2252 | |
2b82032c YZ |
2253 | if (seeding_dev) { |
2254 | mutex_unlock(&uuid_mutex); | |
2255 | up_write(&sb->s_umount); | |
788f20eb | 2256 | |
79787eaa JM |
2257 | if (ret) /* transaction commit */ |
2258 | return ret; | |
2259 | ||
2b82032c | 2260 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
2261 | if (ret < 0) |
2262 | btrfs_error(root->fs_info, ret, | |
2263 | "Failed to relocate sys chunks after " | |
2264 | "device initialization. This can be fixed " | |
2265 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2266 | trans = btrfs_attach_transaction(root); |
2267 | if (IS_ERR(trans)) { | |
2268 | if (PTR_ERR(trans) == -ENOENT) | |
2269 | return 0; | |
2270 | return PTR_ERR(trans); | |
2271 | } | |
2272 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2273 | } |
c9e9f97b | 2274 | |
5a1972bd QW |
2275 | /* Update ctime/mtime for libblkid */ |
2276 | update_dev_time(device_path); | |
2b82032c | 2277 | return ret; |
79787eaa JM |
2278 | |
2279 | error_trans: | |
79787eaa | 2280 | btrfs_end_transaction(trans, root); |
606686ee | 2281 | rcu_string_free(device->name); |
0d39376a | 2282 | btrfs_kobj_rm_device(root->fs_info, device); |
79787eaa | 2283 | kfree(device); |
2b82032c | 2284 | error: |
e525fd89 | 2285 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2286 | if (seeding_dev) { |
2287 | mutex_unlock(&uuid_mutex); | |
2288 | up_write(&sb->s_umount); | |
2289 | } | |
c9e9f97b | 2290 | return ret; |
788f20eb CM |
2291 | } |
2292 | ||
e93c89c1 | 2293 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2294 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2295 | struct btrfs_device **device_out) |
2296 | { | |
2297 | struct request_queue *q; | |
2298 | struct btrfs_device *device; | |
2299 | struct block_device *bdev; | |
2300 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2301 | struct list_head *devices; | |
2302 | struct rcu_string *name; | |
12bd2fc0 | 2303 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2304 | int ret = 0; |
2305 | ||
2306 | *device_out = NULL; | |
1c43366d MX |
2307 | if (fs_info->fs_devices->seeding) { |
2308 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2309 | return -EINVAL; |
1c43366d | 2310 | } |
e93c89c1 SB |
2311 | |
2312 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2313 | fs_info->bdev_holder); | |
1c43366d MX |
2314 | if (IS_ERR(bdev)) { |
2315 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2316 | return PTR_ERR(bdev); |
1c43366d | 2317 | } |
e93c89c1 SB |
2318 | |
2319 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2320 | ||
2321 | devices = &fs_info->fs_devices->devices; | |
2322 | list_for_each_entry(device, devices, dev_list) { | |
2323 | if (device->bdev == bdev) { | |
1c43366d | 2324 | btrfs_err(fs_info, "target device is in the filesystem!"); |
e93c89c1 SB |
2325 | ret = -EEXIST; |
2326 | goto error; | |
2327 | } | |
2328 | } | |
2329 | ||
1c43366d | 2330 | |
7cc8e58d MX |
2331 | if (i_size_read(bdev->bd_inode) < |
2332 | btrfs_device_get_total_bytes(srcdev)) { | |
1c43366d MX |
2333 | btrfs_err(fs_info, "target device is smaller than source device!"); |
2334 | ret = -EINVAL; | |
2335 | goto error; | |
2336 | } | |
2337 | ||
2338 | ||
12bd2fc0 ID |
2339 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2340 | if (IS_ERR(device)) { | |
2341 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2342 | goto error; |
2343 | } | |
2344 | ||
2345 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2346 | if (!name) { | |
2347 | kfree(device); | |
2348 | ret = -ENOMEM; | |
2349 | goto error; | |
2350 | } | |
2351 | rcu_assign_pointer(device->name, name); | |
2352 | ||
2353 | q = bdev_get_queue(bdev); | |
2354 | if (blk_queue_discard(q)) | |
2355 | device->can_discard = 1; | |
2356 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2357 | device->writeable = 1; | |
e93c89c1 SB |
2358 | device->generation = 0; |
2359 | device->io_width = root->sectorsize; | |
2360 | device->io_align = root->sectorsize; | |
2361 | device->sector_size = root->sectorsize; | |
7cc8e58d MX |
2362 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2363 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2364 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2365 | ASSERT(list_empty(&srcdev->resized_list)); |
2366 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2367 | device->commit_bytes_used = device->bytes_used; |
e93c89c1 SB |
2368 | device->dev_root = fs_info->dev_root; |
2369 | device->bdev = bdev; | |
2370 | device->in_fs_metadata = 1; | |
2371 | device->is_tgtdev_for_dev_replace = 1; | |
2372 | device->mode = FMODE_EXCL; | |
27087f37 | 2373 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2374 | set_blocksize(device->bdev, 4096); |
2375 | device->fs_devices = fs_info->fs_devices; | |
2376 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2377 | fs_info->fs_devices->num_devices++; | |
2378 | fs_info->fs_devices->open_devices++; | |
e93c89c1 SB |
2379 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2380 | ||
2381 | *device_out = device; | |
2382 | return ret; | |
2383 | ||
2384 | error: | |
2385 | blkdev_put(bdev, FMODE_EXCL); | |
2386 | return ret; | |
2387 | } | |
2388 | ||
2389 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2390 | struct btrfs_device *tgtdev) | |
2391 | { | |
2392 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2393 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2394 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2395 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2396 | tgtdev->dev_root = fs_info->dev_root; | |
2397 | tgtdev->in_fs_metadata = 1; | |
2398 | } | |
2399 | ||
d397712b CM |
2400 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2401 | struct btrfs_device *device) | |
0b86a832 CM |
2402 | { |
2403 | int ret; | |
2404 | struct btrfs_path *path; | |
2405 | struct btrfs_root *root; | |
2406 | struct btrfs_dev_item *dev_item; | |
2407 | struct extent_buffer *leaf; | |
2408 | struct btrfs_key key; | |
2409 | ||
2410 | root = device->dev_root->fs_info->chunk_root; | |
2411 | ||
2412 | path = btrfs_alloc_path(); | |
2413 | if (!path) | |
2414 | return -ENOMEM; | |
2415 | ||
2416 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2417 | key.type = BTRFS_DEV_ITEM_KEY; | |
2418 | key.offset = device->devid; | |
2419 | ||
2420 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2421 | if (ret < 0) | |
2422 | goto out; | |
2423 | ||
2424 | if (ret > 0) { | |
2425 | ret = -ENOENT; | |
2426 | goto out; | |
2427 | } | |
2428 | ||
2429 | leaf = path->nodes[0]; | |
2430 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2431 | ||
2432 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2433 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2434 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2435 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2436 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2437 | btrfs_set_device_total_bytes(leaf, dev_item, |
2438 | btrfs_device_get_disk_total_bytes(device)); | |
2439 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2440 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2441 | btrfs_mark_buffer_dirty(leaf); |
2442 | ||
2443 | out: | |
2444 | btrfs_free_path(path); | |
2445 | return ret; | |
2446 | } | |
2447 | ||
2196d6e8 | 2448 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2449 | struct btrfs_device *device, u64 new_size) |
2450 | { | |
2451 | struct btrfs_super_block *super_copy = | |
6c41761f | 2452 | device->dev_root->fs_info->super_copy; |
935e5cc9 | 2453 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2454 | u64 old_total; |
2455 | u64 diff; | |
8f18cf13 | 2456 | |
2b82032c YZ |
2457 | if (!device->writeable) |
2458 | return -EACCES; | |
2196d6e8 MX |
2459 | |
2460 | lock_chunks(device->dev_root); | |
2461 | old_total = btrfs_super_total_bytes(super_copy); | |
2462 | diff = new_size - device->total_bytes; | |
2463 | ||
63a212ab | 2464 | if (new_size <= device->total_bytes || |
2196d6e8 MX |
2465 | device->is_tgtdev_for_dev_replace) { |
2466 | unlock_chunks(device->dev_root); | |
2b82032c | 2467 | return -EINVAL; |
2196d6e8 | 2468 | } |
2b82032c | 2469 | |
935e5cc9 MX |
2470 | fs_devices = device->dev_root->fs_info->fs_devices; |
2471 | ||
8f18cf13 | 2472 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2473 | device->fs_devices->total_rw_bytes += diff; |
2474 | ||
7cc8e58d MX |
2475 | btrfs_device_set_total_bytes(device, new_size); |
2476 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4184ea7f | 2477 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
935e5cc9 MX |
2478 | if (list_empty(&device->resized_list)) |
2479 | list_add_tail(&device->resized_list, | |
2480 | &fs_devices->resized_devices); | |
2196d6e8 | 2481 | unlock_chunks(device->dev_root); |
4184ea7f | 2482 | |
8f18cf13 CM |
2483 | return btrfs_update_device(trans, device); |
2484 | } | |
2485 | ||
2486 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
2487 | struct btrfs_root *root, | |
2488 | u64 chunk_tree, u64 chunk_objectid, | |
2489 | u64 chunk_offset) | |
2490 | { | |
2491 | int ret; | |
2492 | struct btrfs_path *path; | |
2493 | struct btrfs_key key; | |
2494 | ||
2495 | root = root->fs_info->chunk_root; | |
2496 | path = btrfs_alloc_path(); | |
2497 | if (!path) | |
2498 | return -ENOMEM; | |
2499 | ||
2500 | key.objectid = chunk_objectid; | |
2501 | key.offset = chunk_offset; | |
2502 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2503 | ||
2504 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2505 | if (ret < 0) |
2506 | goto out; | |
2507 | else if (ret > 0) { /* Logic error or corruption */ | |
2508 | btrfs_error(root->fs_info, -ENOENT, | |
2509 | "Failed lookup while freeing chunk."); | |
2510 | ret = -ENOENT; | |
2511 | goto out; | |
2512 | } | |
8f18cf13 CM |
2513 | |
2514 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
2515 | if (ret < 0) |
2516 | btrfs_error(root->fs_info, ret, | |
2517 | "Failed to delete chunk item."); | |
2518 | out: | |
8f18cf13 | 2519 | btrfs_free_path(path); |
65a246c5 | 2520 | return ret; |
8f18cf13 CM |
2521 | } |
2522 | ||
b2950863 | 2523 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2524 | chunk_offset) |
2525 | { | |
6c41761f | 2526 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2527 | struct btrfs_disk_key *disk_key; |
2528 | struct btrfs_chunk *chunk; | |
2529 | u8 *ptr; | |
2530 | int ret = 0; | |
2531 | u32 num_stripes; | |
2532 | u32 array_size; | |
2533 | u32 len = 0; | |
2534 | u32 cur; | |
2535 | struct btrfs_key key; | |
2536 | ||
2196d6e8 | 2537 | lock_chunks(root); |
8f18cf13 CM |
2538 | array_size = btrfs_super_sys_array_size(super_copy); |
2539 | ||
2540 | ptr = super_copy->sys_chunk_array; | |
2541 | cur = 0; | |
2542 | ||
2543 | while (cur < array_size) { | |
2544 | disk_key = (struct btrfs_disk_key *)ptr; | |
2545 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2546 | ||
2547 | len = sizeof(*disk_key); | |
2548 | ||
2549 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2550 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2551 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2552 | len += btrfs_chunk_item_size(num_stripes); | |
2553 | } else { | |
2554 | ret = -EIO; | |
2555 | break; | |
2556 | } | |
2557 | if (key.objectid == chunk_objectid && | |
2558 | key.offset == chunk_offset) { | |
2559 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2560 | array_size -= len; | |
2561 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2562 | } else { | |
2563 | ptr += len; | |
2564 | cur += len; | |
2565 | } | |
2566 | } | |
2196d6e8 | 2567 | unlock_chunks(root); |
8f18cf13 CM |
2568 | return ret; |
2569 | } | |
2570 | ||
b2950863 | 2571 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
2572 | u64 chunk_tree, u64 chunk_objectid, |
2573 | u64 chunk_offset) | |
2574 | { | |
2575 | struct extent_map_tree *em_tree; | |
2576 | struct btrfs_root *extent_root; | |
2577 | struct btrfs_trans_handle *trans; | |
2196d6e8 | 2578 | struct btrfs_device *device; |
8f18cf13 CM |
2579 | struct extent_map *em; |
2580 | struct map_lookup *map; | |
2196d6e8 | 2581 | u64 dev_extent_len = 0; |
8f18cf13 CM |
2582 | int ret; |
2583 | int i; | |
2584 | ||
2585 | root = root->fs_info->chunk_root; | |
2586 | extent_root = root->fs_info->extent_root; | |
2587 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
2588 | ||
ba1bf481 JB |
2589 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2590 | if (ret) | |
2591 | return -ENOSPC; | |
2592 | ||
8f18cf13 | 2593 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 2594 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
2595 | if (ret) |
2596 | return ret; | |
8f18cf13 | 2597 | |
a22285a6 | 2598 | trans = btrfs_start_transaction(root, 0); |
0f788c58 LB |
2599 | if (IS_ERR(trans)) { |
2600 | ret = PTR_ERR(trans); | |
2601 | btrfs_std_error(root->fs_info, ret); | |
2602 | return ret; | |
2603 | } | |
8f18cf13 CM |
2604 | |
2605 | /* | |
2606 | * step two, delete the device extents and the | |
2607 | * chunk tree entries | |
2608 | */ | |
890871be | 2609 | read_lock(&em_tree->lock); |
8f18cf13 | 2610 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2611 | read_unlock(&em_tree->lock); |
8f18cf13 | 2612 | |
285190d9 | 2613 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 2614 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
2615 | map = (struct map_lookup *)em->bdev; |
2616 | ||
2617 | for (i = 0; i < map->num_stripes; i++) { | |
2196d6e8 MX |
2618 | device = map->stripes[i].dev; |
2619 | ret = btrfs_free_dev_extent(trans, device, | |
2620 | map->stripes[i].physical, | |
2621 | &dev_extent_len); | |
8f18cf13 | 2622 | BUG_ON(ret); |
a061fc8d | 2623 | |
2196d6e8 MX |
2624 | if (device->bytes_used > 0) { |
2625 | lock_chunks(root); | |
2626 | btrfs_device_set_bytes_used(device, | |
2627 | device->bytes_used - dev_extent_len); | |
2628 | spin_lock(&root->fs_info->free_chunk_lock); | |
2629 | root->fs_info->free_chunk_space += dev_extent_len; | |
2630 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2631 | btrfs_clear_space_info_full(root->fs_info); | |
2632 | unlock_chunks(root); | |
2633 | } | |
2634 | ||
dfe25020 CM |
2635 | if (map->stripes[i].dev) { |
2636 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2637 | BUG_ON(ret); | |
2638 | } | |
8f18cf13 CM |
2639 | } |
2640 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
2641 | chunk_offset); | |
2642 | ||
2643 | BUG_ON(ret); | |
2644 | ||
1abe9b8a | 2645 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2646 | ||
8f18cf13 CM |
2647 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2648 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
2649 | BUG_ON(ret); | |
8f18cf13 CM |
2650 | } |
2651 | ||
2b82032c YZ |
2652 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
2653 | BUG_ON(ret); | |
2654 | ||
890871be | 2655 | write_lock(&em_tree->lock); |
2b82032c | 2656 | remove_extent_mapping(em_tree, em); |
890871be | 2657 | write_unlock(&em_tree->lock); |
2b82032c | 2658 | |
2b82032c YZ |
2659 | /* once for the tree */ |
2660 | free_extent_map(em); | |
2661 | /* once for us */ | |
2662 | free_extent_map(em); | |
2663 | ||
2b82032c YZ |
2664 | btrfs_end_transaction(trans, root); |
2665 | return 0; | |
2666 | } | |
2667 | ||
2668 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2669 | { | |
2670 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2671 | struct btrfs_path *path; | |
2672 | struct extent_buffer *leaf; | |
2673 | struct btrfs_chunk *chunk; | |
2674 | struct btrfs_key key; | |
2675 | struct btrfs_key found_key; | |
2676 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2677 | u64 chunk_type; | |
ba1bf481 JB |
2678 | bool retried = false; |
2679 | int failed = 0; | |
2b82032c YZ |
2680 | int ret; |
2681 | ||
2682 | path = btrfs_alloc_path(); | |
2683 | if (!path) | |
2684 | return -ENOMEM; | |
2685 | ||
ba1bf481 | 2686 | again: |
2b82032c YZ |
2687 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2688 | key.offset = (u64)-1; | |
2689 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2690 | ||
2691 | while (1) { | |
2692 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2693 | if (ret < 0) | |
2694 | goto error; | |
79787eaa | 2695 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2696 | |
2697 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2698 | key.type); | |
2699 | if (ret < 0) | |
2700 | goto error; | |
2701 | if (ret > 0) | |
2702 | break; | |
1a40e23b | 2703 | |
2b82032c YZ |
2704 | leaf = path->nodes[0]; |
2705 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2706 | |
2b82032c YZ |
2707 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2708 | struct btrfs_chunk); | |
2709 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2710 | btrfs_release_path(path); |
8f18cf13 | 2711 | |
2b82032c YZ |
2712 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2713 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2714 | found_key.objectid, | |
2715 | found_key.offset); | |
ba1bf481 JB |
2716 | if (ret == -ENOSPC) |
2717 | failed++; | |
14586651 HS |
2718 | else |
2719 | BUG_ON(ret); | |
2b82032c | 2720 | } |
8f18cf13 | 2721 | |
2b82032c YZ |
2722 | if (found_key.offset == 0) |
2723 | break; | |
2724 | key.offset = found_key.offset - 1; | |
2725 | } | |
2726 | ret = 0; | |
ba1bf481 JB |
2727 | if (failed && !retried) { |
2728 | failed = 0; | |
2729 | retried = true; | |
2730 | goto again; | |
fae7f21c | 2731 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2732 | ret = -ENOSPC; |
2733 | } | |
2b82032c YZ |
2734 | error: |
2735 | btrfs_free_path(path); | |
2736 | return ret; | |
8f18cf13 CM |
2737 | } |
2738 | ||
0940ebf6 ID |
2739 | static int insert_balance_item(struct btrfs_root *root, |
2740 | struct btrfs_balance_control *bctl) | |
2741 | { | |
2742 | struct btrfs_trans_handle *trans; | |
2743 | struct btrfs_balance_item *item; | |
2744 | struct btrfs_disk_balance_args disk_bargs; | |
2745 | struct btrfs_path *path; | |
2746 | struct extent_buffer *leaf; | |
2747 | struct btrfs_key key; | |
2748 | int ret, err; | |
2749 | ||
2750 | path = btrfs_alloc_path(); | |
2751 | if (!path) | |
2752 | return -ENOMEM; | |
2753 | ||
2754 | trans = btrfs_start_transaction(root, 0); | |
2755 | if (IS_ERR(trans)) { | |
2756 | btrfs_free_path(path); | |
2757 | return PTR_ERR(trans); | |
2758 | } | |
2759 | ||
2760 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2761 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2762 | key.offset = 0; | |
2763 | ||
2764 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2765 | sizeof(*item)); | |
2766 | if (ret) | |
2767 | goto out; | |
2768 | ||
2769 | leaf = path->nodes[0]; | |
2770 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2771 | ||
2772 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2773 | ||
2774 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2775 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2776 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2777 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2778 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2779 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2780 | ||
2781 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2782 | ||
2783 | btrfs_mark_buffer_dirty(leaf); | |
2784 | out: | |
2785 | btrfs_free_path(path); | |
2786 | err = btrfs_commit_transaction(trans, root); | |
2787 | if (err && !ret) | |
2788 | ret = err; | |
2789 | return ret; | |
2790 | } | |
2791 | ||
2792 | static int del_balance_item(struct btrfs_root *root) | |
2793 | { | |
2794 | struct btrfs_trans_handle *trans; | |
2795 | struct btrfs_path *path; | |
2796 | struct btrfs_key key; | |
2797 | int ret, err; | |
2798 | ||
2799 | path = btrfs_alloc_path(); | |
2800 | if (!path) | |
2801 | return -ENOMEM; | |
2802 | ||
2803 | trans = btrfs_start_transaction(root, 0); | |
2804 | if (IS_ERR(trans)) { | |
2805 | btrfs_free_path(path); | |
2806 | return PTR_ERR(trans); | |
2807 | } | |
2808 | ||
2809 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2810 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2811 | key.offset = 0; | |
2812 | ||
2813 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2814 | if (ret < 0) | |
2815 | goto out; | |
2816 | if (ret > 0) { | |
2817 | ret = -ENOENT; | |
2818 | goto out; | |
2819 | } | |
2820 | ||
2821 | ret = btrfs_del_item(trans, root, path); | |
2822 | out: | |
2823 | btrfs_free_path(path); | |
2824 | err = btrfs_commit_transaction(trans, root); | |
2825 | if (err && !ret) | |
2826 | ret = err; | |
2827 | return ret; | |
2828 | } | |
2829 | ||
59641015 ID |
2830 | /* |
2831 | * This is a heuristic used to reduce the number of chunks balanced on | |
2832 | * resume after balance was interrupted. | |
2833 | */ | |
2834 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2835 | { | |
2836 | /* | |
2837 | * Turn on soft mode for chunk types that were being converted. | |
2838 | */ | |
2839 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2840 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2841 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2842 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2843 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2844 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2845 | ||
2846 | /* | |
2847 | * Turn on usage filter if is not already used. The idea is | |
2848 | * that chunks that we have already balanced should be | |
2849 | * reasonably full. Don't do it for chunks that are being | |
2850 | * converted - that will keep us from relocating unconverted | |
2851 | * (albeit full) chunks. | |
2852 | */ | |
2853 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2854 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2855 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2856 | bctl->data.usage = 90; | |
2857 | } | |
2858 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2859 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2860 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2861 | bctl->sys.usage = 90; | |
2862 | } | |
2863 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2864 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2865 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2866 | bctl->meta.usage = 90; | |
2867 | } | |
2868 | } | |
2869 | ||
c9e9f97b ID |
2870 | /* |
2871 | * Should be called with both balance and volume mutexes held to | |
2872 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2873 | * restriper. Same goes for unset_balance_control. | |
2874 | */ | |
2875 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2876 | { | |
2877 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2878 | ||
2879 | BUG_ON(fs_info->balance_ctl); | |
2880 | ||
2881 | spin_lock(&fs_info->balance_lock); | |
2882 | fs_info->balance_ctl = bctl; | |
2883 | spin_unlock(&fs_info->balance_lock); | |
2884 | } | |
2885 | ||
2886 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2887 | { | |
2888 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2889 | ||
2890 | BUG_ON(!fs_info->balance_ctl); | |
2891 | ||
2892 | spin_lock(&fs_info->balance_lock); | |
2893 | fs_info->balance_ctl = NULL; | |
2894 | spin_unlock(&fs_info->balance_lock); | |
2895 | ||
2896 | kfree(bctl); | |
2897 | } | |
2898 | ||
ed25e9b2 ID |
2899 | /* |
2900 | * Balance filters. Return 1 if chunk should be filtered out | |
2901 | * (should not be balanced). | |
2902 | */ | |
899c81ea | 2903 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2904 | struct btrfs_balance_args *bargs) |
2905 | { | |
899c81ea ID |
2906 | chunk_type = chunk_to_extended(chunk_type) & |
2907 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2908 | |
899c81ea | 2909 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2910 | return 0; |
2911 | ||
2912 | return 1; | |
2913 | } | |
2914 | ||
5ce5b3c0 ID |
2915 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
2916 | struct btrfs_balance_args *bargs) | |
2917 | { | |
2918 | struct btrfs_block_group_cache *cache; | |
2919 | u64 chunk_used, user_thresh; | |
2920 | int ret = 1; | |
2921 | ||
2922 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2923 | chunk_used = btrfs_block_group_used(&cache->item); | |
2924 | ||
a105bb88 | 2925 | if (bargs->usage == 0) |
3e39cea6 | 2926 | user_thresh = 1; |
a105bb88 ID |
2927 | else if (bargs->usage > 100) |
2928 | user_thresh = cache->key.offset; | |
2929 | else | |
2930 | user_thresh = div_factor_fine(cache->key.offset, | |
2931 | bargs->usage); | |
2932 | ||
5ce5b3c0 ID |
2933 | if (chunk_used < user_thresh) |
2934 | ret = 0; | |
2935 | ||
2936 | btrfs_put_block_group(cache); | |
2937 | return ret; | |
2938 | } | |
2939 | ||
409d404b ID |
2940 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2941 | struct btrfs_chunk *chunk, | |
2942 | struct btrfs_balance_args *bargs) | |
2943 | { | |
2944 | struct btrfs_stripe *stripe; | |
2945 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2946 | int i; | |
2947 | ||
2948 | for (i = 0; i < num_stripes; i++) { | |
2949 | stripe = btrfs_stripe_nr(chunk, i); | |
2950 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2951 | return 0; | |
2952 | } | |
2953 | ||
2954 | return 1; | |
2955 | } | |
2956 | ||
94e60d5a ID |
2957 | /* [pstart, pend) */ |
2958 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2959 | struct btrfs_chunk *chunk, | |
2960 | u64 chunk_offset, | |
2961 | struct btrfs_balance_args *bargs) | |
2962 | { | |
2963 | struct btrfs_stripe *stripe; | |
2964 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2965 | u64 stripe_offset; | |
2966 | u64 stripe_length; | |
2967 | int factor; | |
2968 | int i; | |
2969 | ||
2970 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2971 | return 0; | |
2972 | ||
2973 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
2974 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
2975 | factor = num_stripes / 2; | |
2976 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
2977 | factor = num_stripes - 1; | |
2978 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
2979 | factor = num_stripes - 2; | |
2980 | } else { | |
2981 | factor = num_stripes; | |
2982 | } | |
94e60d5a ID |
2983 | |
2984 | for (i = 0; i < num_stripes; i++) { | |
2985 | stripe = btrfs_stripe_nr(chunk, i); | |
2986 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2987 | continue; | |
2988 | ||
2989 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2990 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2991 | do_div(stripe_length, factor); | |
2992 | ||
2993 | if (stripe_offset < bargs->pend && | |
2994 | stripe_offset + stripe_length > bargs->pstart) | |
2995 | return 0; | |
2996 | } | |
2997 | ||
2998 | return 1; | |
2999 | } | |
3000 | ||
ea67176a ID |
3001 | /* [vstart, vend) */ |
3002 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3003 | struct btrfs_chunk *chunk, | |
3004 | u64 chunk_offset, | |
3005 | struct btrfs_balance_args *bargs) | |
3006 | { | |
3007 | if (chunk_offset < bargs->vend && | |
3008 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3009 | /* at least part of the chunk is inside this vrange */ | |
3010 | return 0; | |
3011 | ||
3012 | return 1; | |
3013 | } | |
3014 | ||
899c81ea | 3015 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3016 | struct btrfs_balance_args *bargs) |
3017 | { | |
3018 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3019 | return 0; | |
3020 | ||
899c81ea ID |
3021 | chunk_type = chunk_to_extended(chunk_type) & |
3022 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3023 | |
899c81ea | 3024 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3025 | return 1; |
3026 | ||
3027 | return 0; | |
3028 | } | |
3029 | ||
f43ffb60 ID |
3030 | static int should_balance_chunk(struct btrfs_root *root, |
3031 | struct extent_buffer *leaf, | |
3032 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3033 | { | |
3034 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3035 | struct btrfs_balance_args *bargs = NULL; | |
3036 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3037 | ||
3038 | /* type filter */ | |
3039 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3040 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3041 | return 0; | |
3042 | } | |
3043 | ||
3044 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3045 | bargs = &bctl->data; | |
3046 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3047 | bargs = &bctl->sys; | |
3048 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3049 | bargs = &bctl->meta; | |
3050 | ||
ed25e9b2 ID |
3051 | /* profiles filter */ |
3052 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3053 | chunk_profiles_filter(chunk_type, bargs)) { | |
3054 | return 0; | |
5ce5b3c0 ID |
3055 | } |
3056 | ||
3057 | /* usage filter */ | |
3058 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3059 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3060 | return 0; | |
409d404b ID |
3061 | } |
3062 | ||
3063 | /* devid filter */ | |
3064 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3065 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3066 | return 0; | |
94e60d5a ID |
3067 | } |
3068 | ||
3069 | /* drange filter, makes sense only with devid filter */ | |
3070 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3071 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3072 | return 0; | |
ea67176a ID |
3073 | } |
3074 | ||
3075 | /* vrange filter */ | |
3076 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3077 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3078 | return 0; | |
ed25e9b2 ID |
3079 | } |
3080 | ||
cfa4c961 ID |
3081 | /* soft profile changing mode */ |
3082 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3083 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3084 | return 0; | |
3085 | } | |
3086 | ||
7d824b6f DS |
3087 | /* |
3088 | * limited by count, must be the last filter | |
3089 | */ | |
3090 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3091 | if (bargs->limit == 0) | |
3092 | return 0; | |
3093 | else | |
3094 | bargs->limit--; | |
3095 | } | |
3096 | ||
f43ffb60 ID |
3097 | return 1; |
3098 | } | |
3099 | ||
c9e9f97b | 3100 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3101 | { |
19a39dce | 3102 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3103 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3104 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3105 | struct list_head *devices; | |
ec44a35c CM |
3106 | struct btrfs_device *device; |
3107 | u64 old_size; | |
3108 | u64 size_to_free; | |
f43ffb60 | 3109 | struct btrfs_chunk *chunk; |
ec44a35c CM |
3110 | struct btrfs_path *path; |
3111 | struct btrfs_key key; | |
ec44a35c | 3112 | struct btrfs_key found_key; |
c9e9f97b | 3113 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3114 | struct extent_buffer *leaf; |
3115 | int slot; | |
c9e9f97b ID |
3116 | int ret; |
3117 | int enospc_errors = 0; | |
19a39dce | 3118 | bool counting = true; |
7d824b6f DS |
3119 | u64 limit_data = bctl->data.limit; |
3120 | u64 limit_meta = bctl->meta.limit; | |
3121 | u64 limit_sys = bctl->sys.limit; | |
ec44a35c | 3122 | |
ec44a35c | 3123 | /* step one make some room on all the devices */ |
c9e9f97b | 3124 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3125 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3126 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c CM |
3127 | size_to_free = div_factor(old_size, 1); |
3128 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c | 3129 | if (!device->writeable || |
7cc8e58d MX |
3130 | btrfs_device_get_total_bytes(device) - |
3131 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3132 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3133 | continue; |
3134 | ||
3135 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3136 | if (ret == -ENOSPC) |
3137 | break; | |
ec44a35c CM |
3138 | BUG_ON(ret); |
3139 | ||
a22285a6 | 3140 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 3141 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
3142 | |
3143 | ret = btrfs_grow_device(trans, device, old_size); | |
3144 | BUG_ON(ret); | |
3145 | ||
3146 | btrfs_end_transaction(trans, dev_root); | |
3147 | } | |
3148 | ||
3149 | /* step two, relocate all the chunks */ | |
3150 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3151 | if (!path) { |
3152 | ret = -ENOMEM; | |
3153 | goto error; | |
3154 | } | |
19a39dce ID |
3155 | |
3156 | /* zero out stat counters */ | |
3157 | spin_lock(&fs_info->balance_lock); | |
3158 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3159 | spin_unlock(&fs_info->balance_lock); | |
3160 | again: | |
7d824b6f DS |
3161 | if (!counting) { |
3162 | bctl->data.limit = limit_data; | |
3163 | bctl->meta.limit = limit_meta; | |
3164 | bctl->sys.limit = limit_sys; | |
3165 | } | |
ec44a35c CM |
3166 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3167 | key.offset = (u64)-1; | |
3168 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3169 | ||
d397712b | 3170 | while (1) { |
19a39dce | 3171 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3172 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3173 | ret = -ECANCELED; |
3174 | goto error; | |
3175 | } | |
3176 | ||
ec44a35c CM |
3177 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
3178 | if (ret < 0) | |
3179 | goto error; | |
3180 | ||
3181 | /* | |
3182 | * this shouldn't happen, it means the last relocate | |
3183 | * failed | |
3184 | */ | |
3185 | if (ret == 0) | |
c9e9f97b | 3186 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3187 | |
3188 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3189 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
3190 | if (ret) { |
3191 | ret = 0; | |
ec44a35c | 3192 | break; |
c9e9f97b | 3193 | } |
7d9eb12c | 3194 | |
f43ffb60 ID |
3195 | leaf = path->nodes[0]; |
3196 | slot = path->slots[0]; | |
3197 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3198 | |
ec44a35c CM |
3199 | if (found_key.objectid != key.objectid) |
3200 | break; | |
7d9eb12c | 3201 | |
f43ffb60 ID |
3202 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
3203 | ||
19a39dce ID |
3204 | if (!counting) { |
3205 | spin_lock(&fs_info->balance_lock); | |
3206 | bctl->stat.considered++; | |
3207 | spin_unlock(&fs_info->balance_lock); | |
3208 | } | |
3209 | ||
f43ffb60 ID |
3210 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3211 | found_key.offset); | |
b3b4aa74 | 3212 | btrfs_release_path(path); |
f43ffb60 ID |
3213 | if (!ret) |
3214 | goto loop; | |
3215 | ||
19a39dce ID |
3216 | if (counting) { |
3217 | spin_lock(&fs_info->balance_lock); | |
3218 | bctl->stat.expected++; | |
3219 | spin_unlock(&fs_info->balance_lock); | |
3220 | goto loop; | |
3221 | } | |
3222 | ||
ec44a35c CM |
3223 | ret = btrfs_relocate_chunk(chunk_root, |
3224 | chunk_root->root_key.objectid, | |
3225 | found_key.objectid, | |
3226 | found_key.offset); | |
508794eb JB |
3227 | if (ret && ret != -ENOSPC) |
3228 | goto error; | |
19a39dce | 3229 | if (ret == -ENOSPC) { |
c9e9f97b | 3230 | enospc_errors++; |
19a39dce ID |
3231 | } else { |
3232 | spin_lock(&fs_info->balance_lock); | |
3233 | bctl->stat.completed++; | |
3234 | spin_unlock(&fs_info->balance_lock); | |
3235 | } | |
f43ffb60 | 3236 | loop: |
795a3321 ID |
3237 | if (found_key.offset == 0) |
3238 | break; | |
ba1bf481 | 3239 | key.offset = found_key.offset - 1; |
ec44a35c | 3240 | } |
c9e9f97b | 3241 | |
19a39dce ID |
3242 | if (counting) { |
3243 | btrfs_release_path(path); | |
3244 | counting = false; | |
3245 | goto again; | |
3246 | } | |
ec44a35c CM |
3247 | error: |
3248 | btrfs_free_path(path); | |
c9e9f97b | 3249 | if (enospc_errors) { |
efe120a0 | 3250 | btrfs_info(fs_info, "%d enospc errors during balance", |
c9e9f97b ID |
3251 | enospc_errors); |
3252 | if (!ret) | |
3253 | ret = -ENOSPC; | |
3254 | } | |
3255 | ||
ec44a35c CM |
3256 | return ret; |
3257 | } | |
3258 | ||
0c460c0d ID |
3259 | /** |
3260 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3261 | * @flags: profile to validate | |
3262 | * @extended: if true @flags is treated as an extended profile | |
3263 | */ | |
3264 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3265 | { | |
3266 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3267 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3268 | ||
3269 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3270 | ||
3271 | /* 1) check that all other bits are zeroed */ | |
3272 | if (flags & ~mask) | |
3273 | return 0; | |
3274 | ||
3275 | /* 2) see if profile is reduced */ | |
3276 | if (flags == 0) | |
3277 | return !extended; /* "0" is valid for usual profiles */ | |
3278 | ||
3279 | /* true if exactly one bit set */ | |
3280 | return (flags & (flags - 1)) == 0; | |
3281 | } | |
3282 | ||
837d5b6e ID |
3283 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3284 | { | |
a7e99c69 ID |
3285 | /* cancel requested || normal exit path */ |
3286 | return atomic_read(&fs_info->balance_cancel_req) || | |
3287 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3288 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3289 | } |
3290 | ||
c9e9f97b ID |
3291 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3292 | { | |
0940ebf6 ID |
3293 | int ret; |
3294 | ||
c9e9f97b | 3295 | unset_balance_control(fs_info); |
0940ebf6 | 3296 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 LB |
3297 | if (ret) |
3298 | btrfs_std_error(fs_info, ret); | |
ed0fb78f ID |
3299 | |
3300 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3301 | } |
3302 | ||
c9e9f97b ID |
3303 | /* |
3304 | * Should be called with both balance and volume mutexes held | |
3305 | */ | |
3306 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3307 | struct btrfs_ioctl_balance_args *bargs) | |
3308 | { | |
3309 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3310 | u64 allowed; |
e4837f8f | 3311 | int mixed = 0; |
c9e9f97b | 3312 | int ret; |
8dabb742 | 3313 | u64 num_devices; |
de98ced9 | 3314 | unsigned seq; |
c9e9f97b | 3315 | |
837d5b6e | 3316 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3317 | atomic_read(&fs_info->balance_pause_req) || |
3318 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3319 | ret = -EINVAL; |
3320 | goto out; | |
3321 | } | |
3322 | ||
e4837f8f ID |
3323 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3324 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3325 | mixed = 1; | |
3326 | ||
f43ffb60 ID |
3327 | /* |
3328 | * In case of mixed groups both data and meta should be picked, | |
3329 | * and identical options should be given for both of them. | |
3330 | */ | |
e4837f8f ID |
3331 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3332 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3333 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3334 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3335 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
efe120a0 FH |
3336 | btrfs_err(fs_info, "with mixed groups data and " |
3337 | "metadata balance options must be the same"); | |
f43ffb60 ID |
3338 | ret = -EINVAL; |
3339 | goto out; | |
3340 | } | |
3341 | } | |
3342 | ||
8dabb742 SB |
3343 | num_devices = fs_info->fs_devices->num_devices; |
3344 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
3345 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3346 | BUG_ON(num_devices < 1); | |
3347 | num_devices--; | |
3348 | } | |
3349 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
e4d8ec0f | 3350 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3351 | if (num_devices == 1) |
e4d8ec0f | 3352 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8250dabe | 3353 | else if (num_devices > 1) |
e4d8ec0f | 3354 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3355 | if (num_devices > 2) |
3356 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3357 | if (num_devices > 3) | |
3358 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3359 | BTRFS_BLOCK_GROUP_RAID6); | |
6728b198 ID |
3360 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3361 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
3362 | (bctl->data.target & ~allowed))) { | |
efe120a0 FH |
3363 | btrfs_err(fs_info, "unable to start balance with target " |
3364 | "data profile %llu", | |
c1c9ff7c | 3365 | bctl->data.target); |
e4d8ec0f ID |
3366 | ret = -EINVAL; |
3367 | goto out; | |
3368 | } | |
6728b198 ID |
3369 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3370 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
3371 | (bctl->meta.target & ~allowed))) { | |
efe120a0 FH |
3372 | btrfs_err(fs_info, |
3373 | "unable to start balance with target metadata profile %llu", | |
c1c9ff7c | 3374 | bctl->meta.target); |
e4d8ec0f ID |
3375 | ret = -EINVAL; |
3376 | goto out; | |
3377 | } | |
6728b198 ID |
3378 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3379 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
3380 | (bctl->sys.target & ~allowed))) { | |
efe120a0 FH |
3381 | btrfs_err(fs_info, |
3382 | "unable to start balance with target system profile %llu", | |
c1c9ff7c | 3383 | bctl->sys.target); |
e4d8ec0f ID |
3384 | ret = -EINVAL; |
3385 | goto out; | |
3386 | } | |
3387 | ||
e4837f8f ID |
3388 | /* allow dup'ed data chunks only in mixed mode */ |
3389 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 3390 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
efe120a0 | 3391 | btrfs_err(fs_info, "dup for data is not allowed"); |
e4d8ec0f ID |
3392 | ret = -EINVAL; |
3393 | goto out; | |
3394 | } | |
3395 | ||
3396 | /* allow to reduce meta or sys integrity only if force set */ | |
3397 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3398 | BTRFS_BLOCK_GROUP_RAID10 | |
3399 | BTRFS_BLOCK_GROUP_RAID5 | | |
3400 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3401 | do { |
3402 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3403 | ||
3404 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3405 | (fs_info->avail_system_alloc_bits & allowed) && | |
3406 | !(bctl->sys.target & allowed)) || | |
3407 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3408 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3409 | !(bctl->meta.target & allowed))) { | |
3410 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
efe120a0 | 3411 | btrfs_info(fs_info, "force reducing metadata integrity"); |
de98ced9 | 3412 | } else { |
efe120a0 FH |
3413 | btrfs_err(fs_info, "balance will reduce metadata " |
3414 | "integrity, use force if you want this"); | |
de98ced9 MX |
3415 | ret = -EINVAL; |
3416 | goto out; | |
3417 | } | |
e4d8ec0f | 3418 | } |
de98ced9 | 3419 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3420 | |
5af3e8cc SB |
3421 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3422 | int num_tolerated_disk_barrier_failures; | |
3423 | u64 target = bctl->sys.target; | |
3424 | ||
3425 | num_tolerated_disk_barrier_failures = | |
3426 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3427 | if (num_tolerated_disk_barrier_failures > 0 && | |
3428 | (target & | |
3429 | (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | | |
3430 | BTRFS_AVAIL_ALLOC_BIT_SINGLE))) | |
3431 | num_tolerated_disk_barrier_failures = 0; | |
3432 | else if (num_tolerated_disk_barrier_failures > 1 && | |
3433 | (target & | |
3434 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))) | |
3435 | num_tolerated_disk_barrier_failures = 1; | |
3436 | ||
3437 | fs_info->num_tolerated_disk_barrier_failures = | |
3438 | num_tolerated_disk_barrier_failures; | |
3439 | } | |
3440 | ||
0940ebf6 | 3441 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3442 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3443 | goto out; |
3444 | ||
59641015 ID |
3445 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3446 | BUG_ON(ret == -EEXIST); | |
3447 | set_balance_control(bctl); | |
3448 | } else { | |
3449 | BUG_ON(ret != -EEXIST); | |
3450 | spin_lock(&fs_info->balance_lock); | |
3451 | update_balance_args(bctl); | |
3452 | spin_unlock(&fs_info->balance_lock); | |
3453 | } | |
c9e9f97b | 3454 | |
837d5b6e | 3455 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3456 | mutex_unlock(&fs_info->balance_mutex); |
3457 | ||
3458 | ret = __btrfs_balance(fs_info); | |
3459 | ||
3460 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3461 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3462 | |
bf023ecf ID |
3463 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3464 | fs_info->num_tolerated_disk_barrier_failures = | |
3465 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3466 | } | |
3467 | ||
c9e9f97b ID |
3468 | if (bargs) { |
3469 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3470 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3471 | } |
3472 | ||
3a01aa7a ID |
3473 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3474 | balance_need_close(fs_info)) { | |
3475 | __cancel_balance(fs_info); | |
3476 | } | |
3477 | ||
837d5b6e | 3478 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3479 | |
3480 | return ret; | |
3481 | out: | |
59641015 ID |
3482 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3483 | __cancel_balance(fs_info); | |
ed0fb78f | 3484 | else { |
59641015 | 3485 | kfree(bctl); |
ed0fb78f ID |
3486 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3487 | } | |
59641015 ID |
3488 | return ret; |
3489 | } | |
3490 | ||
3491 | static int balance_kthread(void *data) | |
3492 | { | |
2b6ba629 | 3493 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3494 | int ret = 0; |
59641015 ID |
3495 | |
3496 | mutex_lock(&fs_info->volume_mutex); | |
3497 | mutex_lock(&fs_info->balance_mutex); | |
3498 | ||
2b6ba629 | 3499 | if (fs_info->balance_ctl) { |
efe120a0 | 3500 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3501 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3502 | } |
59641015 ID |
3503 | |
3504 | mutex_unlock(&fs_info->balance_mutex); | |
3505 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3506 | |
59641015 ID |
3507 | return ret; |
3508 | } | |
3509 | ||
2b6ba629 ID |
3510 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3511 | { | |
3512 | struct task_struct *tsk; | |
3513 | ||
3514 | spin_lock(&fs_info->balance_lock); | |
3515 | if (!fs_info->balance_ctl) { | |
3516 | spin_unlock(&fs_info->balance_lock); | |
3517 | return 0; | |
3518 | } | |
3519 | spin_unlock(&fs_info->balance_lock); | |
3520 | ||
3521 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
efe120a0 | 3522 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3523 | return 0; |
3524 | } | |
3525 | ||
3526 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3527 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3528 | } |
3529 | ||
68310a5e | 3530 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3531 | { |
59641015 ID |
3532 | struct btrfs_balance_control *bctl; |
3533 | struct btrfs_balance_item *item; | |
3534 | struct btrfs_disk_balance_args disk_bargs; | |
3535 | struct btrfs_path *path; | |
3536 | struct extent_buffer *leaf; | |
3537 | struct btrfs_key key; | |
3538 | int ret; | |
3539 | ||
3540 | path = btrfs_alloc_path(); | |
3541 | if (!path) | |
3542 | return -ENOMEM; | |
3543 | ||
59641015 ID |
3544 | key.objectid = BTRFS_BALANCE_OBJECTID; |
3545 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3546 | key.offset = 0; | |
3547 | ||
68310a5e | 3548 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3549 | if (ret < 0) |
68310a5e | 3550 | goto out; |
59641015 ID |
3551 | if (ret > 0) { /* ret = -ENOENT; */ |
3552 | ret = 0; | |
68310a5e ID |
3553 | goto out; |
3554 | } | |
3555 | ||
3556 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3557 | if (!bctl) { | |
3558 | ret = -ENOMEM; | |
3559 | goto out; | |
59641015 ID |
3560 | } |
3561 | ||
3562 | leaf = path->nodes[0]; | |
3563 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3564 | ||
68310a5e ID |
3565 | bctl->fs_info = fs_info; |
3566 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3567 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3568 | |
3569 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3570 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3571 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3572 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3573 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3574 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3575 | ||
ed0fb78f ID |
3576 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3577 | ||
68310a5e ID |
3578 | mutex_lock(&fs_info->volume_mutex); |
3579 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3580 | |
68310a5e ID |
3581 | set_balance_control(bctl); |
3582 | ||
3583 | mutex_unlock(&fs_info->balance_mutex); | |
3584 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3585 | out: |
3586 | btrfs_free_path(path); | |
ec44a35c CM |
3587 | return ret; |
3588 | } | |
3589 | ||
837d5b6e ID |
3590 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3591 | { | |
3592 | int ret = 0; | |
3593 | ||
3594 | mutex_lock(&fs_info->balance_mutex); | |
3595 | if (!fs_info->balance_ctl) { | |
3596 | mutex_unlock(&fs_info->balance_mutex); | |
3597 | return -ENOTCONN; | |
3598 | } | |
3599 | ||
3600 | if (atomic_read(&fs_info->balance_running)) { | |
3601 | atomic_inc(&fs_info->balance_pause_req); | |
3602 | mutex_unlock(&fs_info->balance_mutex); | |
3603 | ||
3604 | wait_event(fs_info->balance_wait_q, | |
3605 | atomic_read(&fs_info->balance_running) == 0); | |
3606 | ||
3607 | mutex_lock(&fs_info->balance_mutex); | |
3608 | /* we are good with balance_ctl ripped off from under us */ | |
3609 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3610 | atomic_dec(&fs_info->balance_pause_req); | |
3611 | } else { | |
3612 | ret = -ENOTCONN; | |
3613 | } | |
3614 | ||
3615 | mutex_unlock(&fs_info->balance_mutex); | |
3616 | return ret; | |
3617 | } | |
3618 | ||
a7e99c69 ID |
3619 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3620 | { | |
e649e587 ID |
3621 | if (fs_info->sb->s_flags & MS_RDONLY) |
3622 | return -EROFS; | |
3623 | ||
a7e99c69 ID |
3624 | mutex_lock(&fs_info->balance_mutex); |
3625 | if (!fs_info->balance_ctl) { | |
3626 | mutex_unlock(&fs_info->balance_mutex); | |
3627 | return -ENOTCONN; | |
3628 | } | |
3629 | ||
3630 | atomic_inc(&fs_info->balance_cancel_req); | |
3631 | /* | |
3632 | * if we are running just wait and return, balance item is | |
3633 | * deleted in btrfs_balance in this case | |
3634 | */ | |
3635 | if (atomic_read(&fs_info->balance_running)) { | |
3636 | mutex_unlock(&fs_info->balance_mutex); | |
3637 | wait_event(fs_info->balance_wait_q, | |
3638 | atomic_read(&fs_info->balance_running) == 0); | |
3639 | mutex_lock(&fs_info->balance_mutex); | |
3640 | } else { | |
3641 | /* __cancel_balance needs volume_mutex */ | |
3642 | mutex_unlock(&fs_info->balance_mutex); | |
3643 | mutex_lock(&fs_info->volume_mutex); | |
3644 | mutex_lock(&fs_info->balance_mutex); | |
3645 | ||
3646 | if (fs_info->balance_ctl) | |
3647 | __cancel_balance(fs_info); | |
3648 | ||
3649 | mutex_unlock(&fs_info->volume_mutex); | |
3650 | } | |
3651 | ||
3652 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3653 | atomic_dec(&fs_info->balance_cancel_req); | |
3654 | mutex_unlock(&fs_info->balance_mutex); | |
3655 | return 0; | |
3656 | } | |
3657 | ||
803b2f54 SB |
3658 | static int btrfs_uuid_scan_kthread(void *data) |
3659 | { | |
3660 | struct btrfs_fs_info *fs_info = data; | |
3661 | struct btrfs_root *root = fs_info->tree_root; | |
3662 | struct btrfs_key key; | |
3663 | struct btrfs_key max_key; | |
3664 | struct btrfs_path *path = NULL; | |
3665 | int ret = 0; | |
3666 | struct extent_buffer *eb; | |
3667 | int slot; | |
3668 | struct btrfs_root_item root_item; | |
3669 | u32 item_size; | |
f45388f3 | 3670 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
3671 | |
3672 | path = btrfs_alloc_path(); | |
3673 | if (!path) { | |
3674 | ret = -ENOMEM; | |
3675 | goto out; | |
3676 | } | |
3677 | ||
3678 | key.objectid = 0; | |
3679 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3680 | key.offset = 0; | |
3681 | ||
3682 | max_key.objectid = (u64)-1; | |
3683 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
3684 | max_key.offset = (u64)-1; | |
3685 | ||
803b2f54 | 3686 | while (1) { |
6174d3cb | 3687 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
3688 | if (ret) { |
3689 | if (ret > 0) | |
3690 | ret = 0; | |
3691 | break; | |
3692 | } | |
3693 | ||
3694 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
3695 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
3696 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
3697 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
3698 | goto skip; | |
3699 | ||
3700 | eb = path->nodes[0]; | |
3701 | slot = path->slots[0]; | |
3702 | item_size = btrfs_item_size_nr(eb, slot); | |
3703 | if (item_size < sizeof(root_item)) | |
3704 | goto skip; | |
3705 | ||
803b2f54 SB |
3706 | read_extent_buffer(eb, &root_item, |
3707 | btrfs_item_ptr_offset(eb, slot), | |
3708 | (int)sizeof(root_item)); | |
3709 | if (btrfs_root_refs(&root_item) == 0) | |
3710 | goto skip; | |
f45388f3 FDBM |
3711 | |
3712 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
3713 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
3714 | if (trans) | |
3715 | goto update_tree; | |
3716 | ||
3717 | btrfs_release_path(path); | |
803b2f54 SB |
3718 | /* |
3719 | * 1 - subvol uuid item | |
3720 | * 1 - received_subvol uuid item | |
3721 | */ | |
3722 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
3723 | if (IS_ERR(trans)) { | |
3724 | ret = PTR_ERR(trans); | |
3725 | break; | |
3726 | } | |
f45388f3 FDBM |
3727 | continue; |
3728 | } else { | |
3729 | goto skip; | |
3730 | } | |
3731 | update_tree: | |
3732 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
803b2f54 SB |
3733 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3734 | root_item.uuid, | |
3735 | BTRFS_UUID_KEY_SUBVOL, | |
3736 | key.objectid); | |
3737 | if (ret < 0) { | |
efe120a0 | 3738 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3739 | ret); |
803b2f54 SB |
3740 | break; |
3741 | } | |
3742 | } | |
3743 | ||
3744 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
803b2f54 SB |
3745 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3746 | root_item.received_uuid, | |
3747 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
3748 | key.objectid); | |
3749 | if (ret < 0) { | |
efe120a0 | 3750 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3751 | ret); |
803b2f54 SB |
3752 | break; |
3753 | } | |
3754 | } | |
3755 | ||
f45388f3 | 3756 | skip: |
803b2f54 SB |
3757 | if (trans) { |
3758 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 3759 | trans = NULL; |
803b2f54 SB |
3760 | if (ret) |
3761 | break; | |
3762 | } | |
3763 | ||
803b2f54 SB |
3764 | btrfs_release_path(path); |
3765 | if (key.offset < (u64)-1) { | |
3766 | key.offset++; | |
3767 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
3768 | key.offset = 0; | |
3769 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3770 | } else if (key.objectid < (u64)-1) { | |
3771 | key.offset = 0; | |
3772 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3773 | key.objectid++; | |
3774 | } else { | |
3775 | break; | |
3776 | } | |
3777 | cond_resched(); | |
3778 | } | |
3779 | ||
3780 | out: | |
3781 | btrfs_free_path(path); | |
f45388f3 FDBM |
3782 | if (trans && !IS_ERR(trans)) |
3783 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 3784 | if (ret) |
efe120a0 | 3785 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 SB |
3786 | else |
3787 | fs_info->update_uuid_tree_gen = 1; | |
803b2f54 SB |
3788 | up(&fs_info->uuid_tree_rescan_sem); |
3789 | return 0; | |
3790 | } | |
3791 | ||
70f80175 SB |
3792 | /* |
3793 | * Callback for btrfs_uuid_tree_iterate(). | |
3794 | * returns: | |
3795 | * 0 check succeeded, the entry is not outdated. | |
3796 | * < 0 if an error occured. | |
3797 | * > 0 if the check failed, which means the caller shall remove the entry. | |
3798 | */ | |
3799 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
3800 | u8 *uuid, u8 type, u64 subid) | |
3801 | { | |
3802 | struct btrfs_key key; | |
3803 | int ret = 0; | |
3804 | struct btrfs_root *subvol_root; | |
3805 | ||
3806 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
3807 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
3808 | goto out; | |
3809 | ||
3810 | key.objectid = subid; | |
3811 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3812 | key.offset = (u64)-1; | |
3813 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
3814 | if (IS_ERR(subvol_root)) { | |
3815 | ret = PTR_ERR(subvol_root); | |
3816 | if (ret == -ENOENT) | |
3817 | ret = 1; | |
3818 | goto out; | |
3819 | } | |
3820 | ||
3821 | switch (type) { | |
3822 | case BTRFS_UUID_KEY_SUBVOL: | |
3823 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
3824 | ret = 1; | |
3825 | break; | |
3826 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
3827 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
3828 | BTRFS_UUID_SIZE)) | |
3829 | ret = 1; | |
3830 | break; | |
3831 | } | |
3832 | ||
3833 | out: | |
3834 | return ret; | |
3835 | } | |
3836 | ||
3837 | static int btrfs_uuid_rescan_kthread(void *data) | |
3838 | { | |
3839 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
3840 | int ret; | |
3841 | ||
3842 | /* | |
3843 | * 1st step is to iterate through the existing UUID tree and | |
3844 | * to delete all entries that contain outdated data. | |
3845 | * 2nd step is to add all missing entries to the UUID tree. | |
3846 | */ | |
3847 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
3848 | if (ret < 0) { | |
efe120a0 | 3849 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
3850 | up(&fs_info->uuid_tree_rescan_sem); |
3851 | return ret; | |
3852 | } | |
3853 | return btrfs_uuid_scan_kthread(data); | |
3854 | } | |
3855 | ||
f7a81ea4 SB |
3856 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
3857 | { | |
3858 | struct btrfs_trans_handle *trans; | |
3859 | struct btrfs_root *tree_root = fs_info->tree_root; | |
3860 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
3861 | struct task_struct *task; |
3862 | int ret; | |
f7a81ea4 SB |
3863 | |
3864 | /* | |
3865 | * 1 - root node | |
3866 | * 1 - root item | |
3867 | */ | |
3868 | trans = btrfs_start_transaction(tree_root, 2); | |
3869 | if (IS_ERR(trans)) | |
3870 | return PTR_ERR(trans); | |
3871 | ||
3872 | uuid_root = btrfs_create_tree(trans, fs_info, | |
3873 | BTRFS_UUID_TREE_OBJECTID); | |
3874 | if (IS_ERR(uuid_root)) { | |
3875 | btrfs_abort_transaction(trans, tree_root, | |
3876 | PTR_ERR(uuid_root)); | |
3877 | return PTR_ERR(uuid_root); | |
3878 | } | |
3879 | ||
3880 | fs_info->uuid_root = uuid_root; | |
3881 | ||
803b2f54 SB |
3882 | ret = btrfs_commit_transaction(trans, tree_root); |
3883 | if (ret) | |
3884 | return ret; | |
3885 | ||
3886 | down(&fs_info->uuid_tree_rescan_sem); | |
3887 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
3888 | if (IS_ERR(task)) { | |
70f80175 | 3889 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 3890 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
3891 | up(&fs_info->uuid_tree_rescan_sem); |
3892 | return PTR_ERR(task); | |
3893 | } | |
3894 | ||
3895 | return 0; | |
f7a81ea4 | 3896 | } |
803b2f54 | 3897 | |
70f80175 SB |
3898 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
3899 | { | |
3900 | struct task_struct *task; | |
3901 | ||
3902 | down(&fs_info->uuid_tree_rescan_sem); | |
3903 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
3904 | if (IS_ERR(task)) { | |
3905 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 3906 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
3907 | up(&fs_info->uuid_tree_rescan_sem); |
3908 | return PTR_ERR(task); | |
3909 | } | |
3910 | ||
3911 | return 0; | |
3912 | } | |
3913 | ||
8f18cf13 CM |
3914 | /* |
3915 | * shrinking a device means finding all of the device extents past | |
3916 | * the new size, and then following the back refs to the chunks. | |
3917 | * The chunk relocation code actually frees the device extent | |
3918 | */ | |
3919 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
3920 | { | |
3921 | struct btrfs_trans_handle *trans; | |
3922 | struct btrfs_root *root = device->dev_root; | |
3923 | struct btrfs_dev_extent *dev_extent = NULL; | |
3924 | struct btrfs_path *path; | |
3925 | u64 length; | |
3926 | u64 chunk_tree; | |
3927 | u64 chunk_objectid; | |
3928 | u64 chunk_offset; | |
3929 | int ret; | |
3930 | int slot; | |
ba1bf481 JB |
3931 | int failed = 0; |
3932 | bool retried = false; | |
8f18cf13 CM |
3933 | struct extent_buffer *l; |
3934 | struct btrfs_key key; | |
6c41761f | 3935 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 3936 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
3937 | u64 old_size = btrfs_device_get_total_bytes(device); |
3938 | u64 diff = old_size - new_size; | |
8f18cf13 | 3939 | |
63a212ab SB |
3940 | if (device->is_tgtdev_for_dev_replace) |
3941 | return -EINVAL; | |
3942 | ||
8f18cf13 CM |
3943 | path = btrfs_alloc_path(); |
3944 | if (!path) | |
3945 | return -ENOMEM; | |
3946 | ||
8f18cf13 CM |
3947 | path->reada = 2; |
3948 | ||
7d9eb12c CM |
3949 | lock_chunks(root); |
3950 | ||
7cc8e58d | 3951 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 3952 | if (device->writeable) { |
2b82032c | 3953 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
3954 | spin_lock(&root->fs_info->free_chunk_lock); |
3955 | root->fs_info->free_chunk_space -= diff; | |
3956 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3957 | } | |
7d9eb12c | 3958 | unlock_chunks(root); |
8f18cf13 | 3959 | |
ba1bf481 | 3960 | again: |
8f18cf13 CM |
3961 | key.objectid = device->devid; |
3962 | key.offset = (u64)-1; | |
3963 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3964 | ||
213e64da | 3965 | do { |
8f18cf13 CM |
3966 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3967 | if (ret < 0) | |
3968 | goto done; | |
3969 | ||
3970 | ret = btrfs_previous_item(root, path, 0, key.type); | |
3971 | if (ret < 0) | |
3972 | goto done; | |
3973 | if (ret) { | |
3974 | ret = 0; | |
b3b4aa74 | 3975 | btrfs_release_path(path); |
bf1fb512 | 3976 | break; |
8f18cf13 CM |
3977 | } |
3978 | ||
3979 | l = path->nodes[0]; | |
3980 | slot = path->slots[0]; | |
3981 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
3982 | ||
ba1bf481 | 3983 | if (key.objectid != device->devid) { |
b3b4aa74 | 3984 | btrfs_release_path(path); |
bf1fb512 | 3985 | break; |
ba1bf481 | 3986 | } |
8f18cf13 CM |
3987 | |
3988 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3989 | length = btrfs_dev_extent_length(l, dev_extent); | |
3990 | ||
ba1bf481 | 3991 | if (key.offset + length <= new_size) { |
b3b4aa74 | 3992 | btrfs_release_path(path); |
d6397bae | 3993 | break; |
ba1bf481 | 3994 | } |
8f18cf13 CM |
3995 | |
3996 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3997 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3998 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3999 | btrfs_release_path(path); |
8f18cf13 CM |
4000 | |
4001 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
4002 | chunk_offset); | |
ba1bf481 | 4003 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4004 | goto done; |
ba1bf481 JB |
4005 | if (ret == -ENOSPC) |
4006 | failed++; | |
213e64da | 4007 | } while (key.offset-- > 0); |
ba1bf481 JB |
4008 | |
4009 | if (failed && !retried) { | |
4010 | failed = 0; | |
4011 | retried = true; | |
4012 | goto again; | |
4013 | } else if (failed && retried) { | |
4014 | ret = -ENOSPC; | |
4015 | lock_chunks(root); | |
4016 | ||
7cc8e58d | 4017 | btrfs_device_set_total_bytes(device, old_size); |
ba1bf481 JB |
4018 | if (device->writeable) |
4019 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
4020 | spin_lock(&root->fs_info->free_chunk_lock); |
4021 | root->fs_info->free_chunk_space += diff; | |
4022 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
4023 | unlock_chunks(root); |
4024 | goto done; | |
8f18cf13 CM |
4025 | } |
4026 | ||
d6397bae | 4027 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4028 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4029 | if (IS_ERR(trans)) { |
4030 | ret = PTR_ERR(trans); | |
4031 | goto done; | |
4032 | } | |
4033 | ||
d6397bae | 4034 | lock_chunks(root); |
7cc8e58d | 4035 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4036 | if (list_empty(&device->resized_list)) |
4037 | list_add_tail(&device->resized_list, | |
4038 | &root->fs_info->fs_devices->resized_devices); | |
4039 | ||
d6397bae CB |
4040 | WARN_ON(diff > old_total); |
4041 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4042 | unlock_chunks(root); | |
2196d6e8 MX |
4043 | |
4044 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4045 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4046 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4047 | done: |
4048 | btrfs_free_path(path); | |
4049 | return ret; | |
4050 | } | |
4051 | ||
125ccb0a | 4052 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4053 | struct btrfs_key *key, |
4054 | struct btrfs_chunk *chunk, int item_size) | |
4055 | { | |
6c41761f | 4056 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
4057 | struct btrfs_disk_key disk_key; |
4058 | u32 array_size; | |
4059 | u8 *ptr; | |
4060 | ||
fe48a5c0 | 4061 | lock_chunks(root); |
0b86a832 | 4062 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4063 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 MX |
4064 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
4065 | unlock_chunks(root); | |
0b86a832 | 4066 | return -EFBIG; |
fe48a5c0 | 4067 | } |
0b86a832 CM |
4068 | |
4069 | ptr = super_copy->sys_chunk_array + array_size; | |
4070 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4071 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4072 | ptr += sizeof(disk_key); | |
4073 | memcpy(ptr, chunk, item_size); | |
4074 | item_size += sizeof(disk_key); | |
4075 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 MX |
4076 | unlock_chunks(root); |
4077 | ||
0b86a832 CM |
4078 | return 0; |
4079 | } | |
4080 | ||
73c5de00 AJ |
4081 | /* |
4082 | * sort the devices in descending order by max_avail, total_avail | |
4083 | */ | |
4084 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4085 | { |
73c5de00 AJ |
4086 | const struct btrfs_device_info *di_a = a; |
4087 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4088 | |
73c5de00 | 4089 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4090 | return -1; |
73c5de00 | 4091 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4092 | return 1; |
73c5de00 AJ |
4093 | if (di_a->total_avail > di_b->total_avail) |
4094 | return -1; | |
4095 | if (di_a->total_avail < di_b->total_avail) | |
4096 | return 1; | |
4097 | return 0; | |
b2117a39 | 4098 | } |
0b86a832 | 4099 | |
48a3b636 | 4100 | static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
e6ec716f MX |
4101 | [BTRFS_RAID_RAID10] = { |
4102 | .sub_stripes = 2, | |
4103 | .dev_stripes = 1, | |
4104 | .devs_max = 0, /* 0 == as many as possible */ | |
4105 | .devs_min = 4, | |
4106 | .devs_increment = 2, | |
4107 | .ncopies = 2, | |
4108 | }, | |
4109 | [BTRFS_RAID_RAID1] = { | |
4110 | .sub_stripes = 1, | |
4111 | .dev_stripes = 1, | |
4112 | .devs_max = 2, | |
4113 | .devs_min = 2, | |
4114 | .devs_increment = 2, | |
4115 | .ncopies = 2, | |
4116 | }, | |
4117 | [BTRFS_RAID_DUP] = { | |
4118 | .sub_stripes = 1, | |
4119 | .dev_stripes = 2, | |
4120 | .devs_max = 1, | |
4121 | .devs_min = 1, | |
4122 | .devs_increment = 1, | |
4123 | .ncopies = 2, | |
4124 | }, | |
4125 | [BTRFS_RAID_RAID0] = { | |
4126 | .sub_stripes = 1, | |
4127 | .dev_stripes = 1, | |
4128 | .devs_max = 0, | |
4129 | .devs_min = 2, | |
4130 | .devs_increment = 1, | |
4131 | .ncopies = 1, | |
4132 | }, | |
4133 | [BTRFS_RAID_SINGLE] = { | |
4134 | .sub_stripes = 1, | |
4135 | .dev_stripes = 1, | |
4136 | .devs_max = 1, | |
4137 | .devs_min = 1, | |
4138 | .devs_increment = 1, | |
4139 | .ncopies = 1, | |
4140 | }, | |
e942f883 CM |
4141 | [BTRFS_RAID_RAID5] = { |
4142 | .sub_stripes = 1, | |
4143 | .dev_stripes = 1, | |
4144 | .devs_max = 0, | |
4145 | .devs_min = 2, | |
4146 | .devs_increment = 1, | |
4147 | .ncopies = 2, | |
4148 | }, | |
4149 | [BTRFS_RAID_RAID6] = { | |
4150 | .sub_stripes = 1, | |
4151 | .dev_stripes = 1, | |
4152 | .devs_max = 0, | |
4153 | .devs_min = 3, | |
4154 | .devs_increment = 1, | |
4155 | .ncopies = 3, | |
4156 | }, | |
31e50229 LB |
4157 | }; |
4158 | ||
53b381b3 DW |
4159 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4160 | { | |
4161 | /* TODO allow them to set a preferred stripe size */ | |
4162 | return 64 * 1024; | |
4163 | } | |
4164 | ||
4165 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4166 | { | |
53b381b3 DW |
4167 | if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))) |
4168 | return; | |
4169 | ||
ceda0864 | 4170 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4171 | } |
4172 | ||
23f8f9b7 GH |
4173 | #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \ |
4174 | - sizeof(struct btrfs_item) \ | |
4175 | - sizeof(struct btrfs_chunk)) \ | |
4176 | / sizeof(struct btrfs_stripe) + 1) | |
4177 | ||
4178 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4179 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4180 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4181 | / sizeof(struct btrfs_stripe) + 1) | |
4182 | ||
73c5de00 | 4183 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4184 | struct btrfs_root *extent_root, u64 start, |
4185 | u64 type) | |
b2117a39 | 4186 | { |
73c5de00 AJ |
4187 | struct btrfs_fs_info *info = extent_root->fs_info; |
4188 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4189 | struct list_head *cur; | |
4190 | struct map_lookup *map = NULL; | |
4191 | struct extent_map_tree *em_tree; | |
4192 | struct extent_map *em; | |
4193 | struct btrfs_device_info *devices_info = NULL; | |
4194 | u64 total_avail; | |
4195 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4196 | int data_stripes; /* number of stripes that count for |
4197 | block group size */ | |
73c5de00 AJ |
4198 | int sub_stripes; /* sub_stripes info for map */ |
4199 | int dev_stripes; /* stripes per dev */ | |
4200 | int devs_max; /* max devs to use */ | |
4201 | int devs_min; /* min devs needed */ | |
4202 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4203 | int ncopies; /* how many copies to data has */ | |
4204 | int ret; | |
4205 | u64 max_stripe_size; | |
4206 | u64 max_chunk_size; | |
4207 | u64 stripe_size; | |
4208 | u64 num_bytes; | |
53b381b3 | 4209 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4210 | int ndevs; |
4211 | int i; | |
4212 | int j; | |
31e50229 | 4213 | int index; |
593060d7 | 4214 | |
0c460c0d | 4215 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4216 | |
73c5de00 AJ |
4217 | if (list_empty(&fs_devices->alloc_list)) |
4218 | return -ENOSPC; | |
b2117a39 | 4219 | |
31e50229 | 4220 | index = __get_raid_index(type); |
73c5de00 | 4221 | |
31e50229 LB |
4222 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4223 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4224 | devs_max = btrfs_raid_array[index].devs_max; | |
4225 | devs_min = btrfs_raid_array[index].devs_min; | |
4226 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4227 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4228 | |
9b3f68b9 | 4229 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
4230 | max_stripe_size = 1024 * 1024 * 1024; |
4231 | max_chunk_size = 10 * max_stripe_size; | |
23f8f9b7 GH |
4232 | if (!devs_max) |
4233 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4234 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
4235 | /* for larger filesystems, use larger metadata chunks */ |
4236 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
4237 | max_stripe_size = 1024 * 1024 * 1024; | |
4238 | else | |
4239 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 4240 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4241 | if (!devs_max) |
4242 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4243 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 4244 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 | 4245 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4246 | if (!devs_max) |
4247 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4248 | } else { |
351fd353 | 4249 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4250 | type); |
4251 | BUG_ON(1); | |
9b3f68b9 CM |
4252 | } |
4253 | ||
2b82032c YZ |
4254 | /* we don't want a chunk larger than 10% of writeable space */ |
4255 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4256 | max_chunk_size); | |
9b3f68b9 | 4257 | |
73c5de00 AJ |
4258 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
4259 | GFP_NOFS); | |
4260 | if (!devices_info) | |
4261 | return -ENOMEM; | |
0cad8a11 | 4262 | |
73c5de00 | 4263 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4264 | |
9f680ce0 | 4265 | /* |
73c5de00 AJ |
4266 | * in the first pass through the devices list, we gather information |
4267 | * about the available holes on each device. | |
9f680ce0 | 4268 | */ |
73c5de00 AJ |
4269 | ndevs = 0; |
4270 | while (cur != &fs_devices->alloc_list) { | |
4271 | struct btrfs_device *device; | |
4272 | u64 max_avail; | |
4273 | u64 dev_offset; | |
b2117a39 | 4274 | |
73c5de00 | 4275 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4276 | |
73c5de00 | 4277 | cur = cur->next; |
b2117a39 | 4278 | |
73c5de00 | 4279 | if (!device->writeable) { |
31b1a2bd | 4280 | WARN(1, KERN_ERR |
efe120a0 | 4281 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4282 | continue; |
4283 | } | |
b2117a39 | 4284 | |
63a212ab SB |
4285 | if (!device->in_fs_metadata || |
4286 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4287 | continue; |
b2117a39 | 4288 | |
73c5de00 AJ |
4289 | if (device->total_bytes > device->bytes_used) |
4290 | total_avail = device->total_bytes - device->bytes_used; | |
4291 | else | |
4292 | total_avail = 0; | |
38c01b96 | 4293 | |
4294 | /* If there is no space on this device, skip it. */ | |
4295 | if (total_avail == 0) | |
4296 | continue; | |
b2117a39 | 4297 | |
6df9a95e | 4298 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4299 | max_stripe_size * dev_stripes, |
4300 | &dev_offset, &max_avail); | |
4301 | if (ret && ret != -ENOSPC) | |
4302 | goto error; | |
b2117a39 | 4303 | |
73c5de00 AJ |
4304 | if (ret == 0) |
4305 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4306 | |
73c5de00 AJ |
4307 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4308 | continue; | |
b2117a39 | 4309 | |
063d006f ES |
4310 | if (ndevs == fs_devices->rw_devices) { |
4311 | WARN(1, "%s: found more than %llu devices\n", | |
4312 | __func__, fs_devices->rw_devices); | |
4313 | break; | |
4314 | } | |
73c5de00 AJ |
4315 | devices_info[ndevs].dev_offset = dev_offset; |
4316 | devices_info[ndevs].max_avail = max_avail; | |
4317 | devices_info[ndevs].total_avail = total_avail; | |
4318 | devices_info[ndevs].dev = device; | |
4319 | ++ndevs; | |
4320 | } | |
b2117a39 | 4321 | |
73c5de00 AJ |
4322 | /* |
4323 | * now sort the devices by hole size / available space | |
4324 | */ | |
4325 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4326 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4327 | |
73c5de00 AJ |
4328 | /* round down to number of usable stripes */ |
4329 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4330 | |
73c5de00 AJ |
4331 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4332 | ret = -ENOSPC; | |
4333 | goto error; | |
b2117a39 | 4334 | } |
9f680ce0 | 4335 | |
73c5de00 AJ |
4336 | if (devs_max && ndevs > devs_max) |
4337 | ndevs = devs_max; | |
4338 | /* | |
4339 | * the primary goal is to maximize the number of stripes, so use as many | |
4340 | * devices as possible, even if the stripes are not maximum sized. | |
4341 | */ | |
4342 | stripe_size = devices_info[ndevs-1].max_avail; | |
4343 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4344 | |
53b381b3 DW |
4345 | /* |
4346 | * this will have to be fixed for RAID1 and RAID10 over | |
4347 | * more drives | |
4348 | */ | |
4349 | data_stripes = num_stripes / ncopies; | |
4350 | ||
53b381b3 DW |
4351 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4352 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4353 | btrfs_super_stripesize(info->super_copy)); | |
4354 | data_stripes = num_stripes - 1; | |
4355 | } | |
4356 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4357 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4358 | btrfs_super_stripesize(info->super_copy)); | |
4359 | data_stripes = num_stripes - 2; | |
4360 | } | |
86db2578 CM |
4361 | |
4362 | /* | |
4363 | * Use the number of data stripes to figure out how big this chunk | |
4364 | * is really going to be in terms of logical address space, | |
4365 | * and compare that answer with the max chunk size | |
4366 | */ | |
4367 | if (stripe_size * data_stripes > max_chunk_size) { | |
4368 | u64 mask = (1ULL << 24) - 1; | |
4369 | stripe_size = max_chunk_size; | |
4370 | do_div(stripe_size, data_stripes); | |
4371 | ||
4372 | /* bump the answer up to a 16MB boundary */ | |
4373 | stripe_size = (stripe_size + mask) & ~mask; | |
4374 | ||
4375 | /* but don't go higher than the limits we found | |
4376 | * while searching for free extents | |
4377 | */ | |
4378 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4379 | stripe_size = devices_info[ndevs-1].max_avail; | |
4380 | } | |
4381 | ||
73c5de00 | 4382 | do_div(stripe_size, dev_stripes); |
37db63a4 ID |
4383 | |
4384 | /* align to BTRFS_STRIPE_LEN */ | |
53b381b3 DW |
4385 | do_div(stripe_size, raid_stripe_len); |
4386 | stripe_size *= raid_stripe_len; | |
b2117a39 MX |
4387 | |
4388 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4389 | if (!map) { | |
4390 | ret = -ENOMEM; | |
4391 | goto error; | |
4392 | } | |
4393 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4394 | |
73c5de00 AJ |
4395 | for (i = 0; i < ndevs; ++i) { |
4396 | for (j = 0; j < dev_stripes; ++j) { | |
4397 | int s = i * dev_stripes + j; | |
4398 | map->stripes[s].dev = devices_info[i].dev; | |
4399 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4400 | j * stripe_size; | |
6324fbf3 | 4401 | } |
6324fbf3 | 4402 | } |
2b82032c | 4403 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4404 | map->stripe_len = raid_stripe_len; |
4405 | map->io_align = raid_stripe_len; | |
4406 | map->io_width = raid_stripe_len; | |
2b82032c | 4407 | map->type = type; |
2b82032c | 4408 | map->sub_stripes = sub_stripes; |
0b86a832 | 4409 | |
53b381b3 | 4410 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4411 | |
73c5de00 | 4412 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4413 | |
172ddd60 | 4414 | em = alloc_extent_map(); |
2b82032c | 4415 | if (!em) { |
298a8f9c | 4416 | kfree(map); |
b2117a39 MX |
4417 | ret = -ENOMEM; |
4418 | goto error; | |
593060d7 | 4419 | } |
298a8f9c | 4420 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
2b82032c YZ |
4421 | em->bdev = (struct block_device *)map; |
4422 | em->start = start; | |
73c5de00 | 4423 | em->len = num_bytes; |
2b82032c YZ |
4424 | em->block_start = 0; |
4425 | em->block_len = em->len; | |
6df9a95e | 4426 | em->orig_block_len = stripe_size; |
593060d7 | 4427 | |
2b82032c | 4428 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4429 | write_lock(&em_tree->lock); |
09a2a8f9 | 4430 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4431 | if (!ret) { |
4432 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4433 | atomic_inc(&em->refs); | |
4434 | } | |
890871be | 4435 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4436 | if (ret) { |
4437 | free_extent_map(em); | |
1dd4602f | 4438 | goto error; |
0f5d42b2 | 4439 | } |
0b86a832 | 4440 | |
04487488 JB |
4441 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4442 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4443 | start, num_bytes); | |
6df9a95e JB |
4444 | if (ret) |
4445 | goto error_del_extent; | |
2b82032c | 4446 | |
7cc8e58d MX |
4447 | for (i = 0; i < map->num_stripes; i++) { |
4448 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4449 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4450 | } | |
43530c46 | 4451 | |
1c116187 MX |
4452 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4453 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4454 | map->num_stripes); | |
4455 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4456 | ||
0f5d42b2 | 4457 | free_extent_map(em); |
53b381b3 DW |
4458 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4459 | ||
b2117a39 | 4460 | kfree(devices_info); |
2b82032c | 4461 | return 0; |
b2117a39 | 4462 | |
6df9a95e | 4463 | error_del_extent: |
0f5d42b2 JB |
4464 | write_lock(&em_tree->lock); |
4465 | remove_extent_mapping(em_tree, em); | |
4466 | write_unlock(&em_tree->lock); | |
4467 | ||
4468 | /* One for our allocation */ | |
4469 | free_extent_map(em); | |
4470 | /* One for the tree reference */ | |
4471 | free_extent_map(em); | |
b2117a39 | 4472 | error: |
b2117a39 MX |
4473 | kfree(devices_info); |
4474 | return ret; | |
2b82032c YZ |
4475 | } |
4476 | ||
6df9a95e | 4477 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4478 | struct btrfs_root *extent_root, |
6df9a95e | 4479 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4480 | { |
2b82032c YZ |
4481 | struct btrfs_key key; |
4482 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4483 | struct btrfs_device *device; | |
4484 | struct btrfs_chunk *chunk; | |
4485 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4486 | struct extent_map_tree *em_tree; |
4487 | struct extent_map *em; | |
4488 | struct map_lookup *map; | |
4489 | size_t item_size; | |
4490 | u64 dev_offset; | |
4491 | u64 stripe_size; | |
4492 | int i = 0; | |
2b82032c YZ |
4493 | int ret; |
4494 | ||
6df9a95e JB |
4495 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4496 | read_lock(&em_tree->lock); | |
4497 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4498 | read_unlock(&em_tree->lock); | |
4499 | ||
4500 | if (!em) { | |
4501 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4502 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4503 | return -EINVAL; | |
4504 | } | |
4505 | ||
4506 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4507 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
351fd353 | 4508 | " %Lu-%Lu, found %Lu-%Lu", chunk_offset, |
6df9a95e JB |
4509 | chunk_size, em->start, em->len); |
4510 | free_extent_map(em); | |
4511 | return -EINVAL; | |
4512 | } | |
4513 | ||
4514 | map = (struct map_lookup *)em->bdev; | |
4515 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
4516 | stripe_size = em->orig_block_len; | |
4517 | ||
2b82032c | 4518 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4519 | if (!chunk) { |
4520 | ret = -ENOMEM; | |
4521 | goto out; | |
4522 | } | |
4523 | ||
4524 | for (i = 0; i < map->num_stripes; i++) { | |
4525 | device = map->stripes[i].dev; | |
4526 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4527 | |
0b86a832 | 4528 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4529 | if (ret) |
6df9a95e JB |
4530 | goto out; |
4531 | ret = btrfs_alloc_dev_extent(trans, device, | |
4532 | chunk_root->root_key.objectid, | |
4533 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4534 | chunk_offset, dev_offset, | |
4535 | stripe_size); | |
4536 | if (ret) | |
4537 | goto out; | |
2b82032c YZ |
4538 | } |
4539 | ||
2b82032c | 4540 | stripe = &chunk->stripe; |
6df9a95e JB |
4541 | for (i = 0; i < map->num_stripes; i++) { |
4542 | device = map->stripes[i].dev; | |
4543 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4544 | |
e17cade2 CM |
4545 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4546 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4547 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4548 | stripe++; |
0b86a832 CM |
4549 | } |
4550 | ||
2b82032c | 4551 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4552 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4553 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4554 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4555 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4556 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4557 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4558 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4559 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4560 | |
2b82032c YZ |
4561 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4562 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4563 | key.offset = chunk_offset; | |
0b86a832 | 4564 | |
2b82032c | 4565 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4566 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4567 | /* | |
4568 | * TODO: Cleanup of inserted chunk root in case of | |
4569 | * failure. | |
4570 | */ | |
125ccb0a | 4571 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4572 | item_size); |
8f18cf13 | 4573 | } |
1abe9b8a | 4574 | |
6df9a95e | 4575 | out: |
0b86a832 | 4576 | kfree(chunk); |
6df9a95e | 4577 | free_extent_map(em); |
4ed1d16e | 4578 | return ret; |
2b82032c | 4579 | } |
0b86a832 | 4580 | |
2b82032c YZ |
4581 | /* |
4582 | * Chunk allocation falls into two parts. The first part does works | |
4583 | * that make the new allocated chunk useable, but not do any operation | |
4584 | * that modifies the chunk tree. The second part does the works that | |
4585 | * require modifying the chunk tree. This division is important for the | |
4586 | * bootstrap process of adding storage to a seed btrfs. | |
4587 | */ | |
4588 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4589 | struct btrfs_root *extent_root, u64 type) | |
4590 | { | |
4591 | u64 chunk_offset; | |
2b82032c | 4592 | |
6df9a95e JB |
4593 | chunk_offset = find_next_chunk(extent_root->fs_info); |
4594 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
4595 | } |
4596 | ||
d397712b | 4597 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4598 | struct btrfs_root *root, |
4599 | struct btrfs_device *device) | |
4600 | { | |
4601 | u64 chunk_offset; | |
4602 | u64 sys_chunk_offset; | |
2b82032c | 4603 | u64 alloc_profile; |
2b82032c YZ |
4604 | struct btrfs_fs_info *fs_info = root->fs_info; |
4605 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4606 | int ret; | |
4607 | ||
6df9a95e | 4608 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 4609 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
4610 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
4611 | alloc_profile); | |
79787eaa JM |
4612 | if (ret) |
4613 | return ret; | |
2b82032c | 4614 | |
6df9a95e | 4615 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 4616 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
4617 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
4618 | alloc_profile); | |
79787eaa | 4619 | return ret; |
2b82032c YZ |
4620 | } |
4621 | ||
d20983b4 MX |
4622 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
4623 | { | |
4624 | int max_errors; | |
4625 | ||
4626 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4627 | BTRFS_BLOCK_GROUP_RAID10 | | |
4628 | BTRFS_BLOCK_GROUP_RAID5 | | |
4629 | BTRFS_BLOCK_GROUP_DUP)) { | |
4630 | max_errors = 1; | |
4631 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4632 | max_errors = 2; | |
4633 | } else { | |
4634 | max_errors = 0; | |
4635 | } | |
4636 | ||
4637 | return max_errors; | |
4638 | } | |
4639 | ||
2b82032c YZ |
4640 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) |
4641 | { | |
4642 | struct extent_map *em; | |
4643 | struct map_lookup *map; | |
4644 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4645 | int readonly = 0; | |
d20983b4 | 4646 | int miss_ndevs = 0; |
2b82032c YZ |
4647 | int i; |
4648 | ||
890871be | 4649 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4650 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4651 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4652 | if (!em) |
4653 | return 1; | |
4654 | ||
4655 | map = (struct map_lookup *)em->bdev; | |
4656 | for (i = 0; i < map->num_stripes; i++) { | |
d20983b4 MX |
4657 | if (map->stripes[i].dev->missing) { |
4658 | miss_ndevs++; | |
4659 | continue; | |
4660 | } | |
4661 | ||
2b82032c YZ |
4662 | if (!map->stripes[i].dev->writeable) { |
4663 | readonly = 1; | |
d20983b4 | 4664 | goto end; |
2b82032c YZ |
4665 | } |
4666 | } | |
d20983b4 MX |
4667 | |
4668 | /* | |
4669 | * If the number of missing devices is larger than max errors, | |
4670 | * we can not write the data into that chunk successfully, so | |
4671 | * set it readonly. | |
4672 | */ | |
4673 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
4674 | readonly = 1; | |
4675 | end: | |
0b86a832 | 4676 | free_extent_map(em); |
2b82032c | 4677 | return readonly; |
0b86a832 CM |
4678 | } |
4679 | ||
4680 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4681 | { | |
a8067e02 | 4682 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
4683 | } |
4684 | ||
4685 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4686 | { | |
4687 | struct extent_map *em; | |
4688 | ||
d397712b | 4689 | while (1) { |
890871be | 4690 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
4691 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
4692 | if (em) | |
4693 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 4694 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
4695 | if (!em) |
4696 | break; | |
0b86a832 CM |
4697 | /* once for us */ |
4698 | free_extent_map(em); | |
4699 | /* once for the tree */ | |
4700 | free_extent_map(em); | |
4701 | } | |
4702 | } | |
4703 | ||
5d964051 | 4704 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 4705 | { |
5d964051 | 4706 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
4707 | struct extent_map *em; |
4708 | struct map_lookup *map; | |
4709 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4710 | int ret; | |
4711 | ||
890871be | 4712 | read_lock(&em_tree->lock); |
f188591e | 4713 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 4714 | read_unlock(&em_tree->lock); |
f188591e | 4715 | |
fb7669b5 JB |
4716 | /* |
4717 | * We could return errors for these cases, but that could get ugly and | |
4718 | * we'd probably do the same thing which is just not do anything else | |
4719 | * and exit, so return 1 so the callers don't try to use other copies. | |
4720 | */ | |
4721 | if (!em) { | |
351fd353 | 4722 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
4723 | logical+len); |
4724 | return 1; | |
4725 | } | |
4726 | ||
4727 | if (em->start > logical || em->start + em->len < logical) { | |
ccf39f92 | 4728 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " |
351fd353 | 4729 | "%Lu-%Lu", logical, logical+len, em->start, |
fb7669b5 | 4730 | em->start + em->len); |
7d3d1744 | 4731 | free_extent_map(em); |
fb7669b5 JB |
4732 | return 1; |
4733 | } | |
4734 | ||
f188591e CM |
4735 | map = (struct map_lookup *)em->bdev; |
4736 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
4737 | ret = map->num_stripes; | |
321aecc6 CM |
4738 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
4739 | ret = map->sub_stripes; | |
53b381b3 DW |
4740 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
4741 | ret = 2; | |
4742 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4743 | ret = 3; | |
f188591e CM |
4744 | else |
4745 | ret = 1; | |
4746 | free_extent_map(em); | |
ad6d620e SB |
4747 | |
4748 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
4749 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
4750 | ret++; | |
4751 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
4752 | ||
f188591e CM |
4753 | return ret; |
4754 | } | |
4755 | ||
53b381b3 DW |
4756 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
4757 | struct btrfs_mapping_tree *map_tree, | |
4758 | u64 logical) | |
4759 | { | |
4760 | struct extent_map *em; | |
4761 | struct map_lookup *map; | |
4762 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4763 | unsigned long len = root->sectorsize; | |
4764 | ||
4765 | read_lock(&em_tree->lock); | |
4766 | em = lookup_extent_mapping(em_tree, logical, len); | |
4767 | read_unlock(&em_tree->lock); | |
4768 | BUG_ON(!em); | |
4769 | ||
4770 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4771 | map = (struct map_lookup *)em->bdev; | |
4772 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4773 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4774 | len = map->stripe_len * nr_data_stripes(map); | |
4775 | } | |
4776 | free_extent_map(em); | |
4777 | return len; | |
4778 | } | |
4779 | ||
4780 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
4781 | u64 logical, u64 len, int mirror_num) | |
4782 | { | |
4783 | struct extent_map *em; | |
4784 | struct map_lookup *map; | |
4785 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4786 | int ret = 0; | |
4787 | ||
4788 | read_lock(&em_tree->lock); | |
4789 | em = lookup_extent_mapping(em_tree, logical, len); | |
4790 | read_unlock(&em_tree->lock); | |
4791 | BUG_ON(!em); | |
4792 | ||
4793 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4794 | map = (struct map_lookup *)em->bdev; | |
4795 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4796 | BTRFS_BLOCK_GROUP_RAID6)) | |
4797 | ret = 1; | |
4798 | free_extent_map(em); | |
4799 | return ret; | |
4800 | } | |
4801 | ||
30d9861f SB |
4802 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
4803 | struct map_lookup *map, int first, int num, | |
4804 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
4805 | { |
4806 | int i; | |
30d9861f SB |
4807 | int tolerance; |
4808 | struct btrfs_device *srcdev; | |
4809 | ||
4810 | if (dev_replace_is_ongoing && | |
4811 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
4812 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
4813 | srcdev = fs_info->dev_replace.srcdev; | |
4814 | else | |
4815 | srcdev = NULL; | |
4816 | ||
4817 | /* | |
4818 | * try to avoid the drive that is the source drive for a | |
4819 | * dev-replace procedure, only choose it if no other non-missing | |
4820 | * mirror is available | |
4821 | */ | |
4822 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
4823 | if (map->stripes[optimal].dev->bdev && | |
4824 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
4825 | return optimal; | |
4826 | for (i = first; i < first + num; i++) { | |
4827 | if (map->stripes[i].dev->bdev && | |
4828 | (tolerance || map->stripes[i].dev != srcdev)) | |
4829 | return i; | |
4830 | } | |
dfe25020 | 4831 | } |
30d9861f | 4832 | |
dfe25020 CM |
4833 | /* we couldn't find one that doesn't fail. Just return something |
4834 | * and the io error handling code will clean up eventually | |
4835 | */ | |
4836 | return optimal; | |
4837 | } | |
4838 | ||
53b381b3 DW |
4839 | static inline int parity_smaller(u64 a, u64 b) |
4840 | { | |
4841 | return a > b; | |
4842 | } | |
4843 | ||
4844 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
4845 | static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map) | |
4846 | { | |
4847 | struct btrfs_bio_stripe s; | |
4848 | int i; | |
4849 | u64 l; | |
4850 | int again = 1; | |
4851 | ||
4852 | while (again) { | |
4853 | again = 0; | |
4854 | for (i = 0; i < bbio->num_stripes - 1; i++) { | |
4855 | if (parity_smaller(raid_map[i], raid_map[i+1])) { | |
4856 | s = bbio->stripes[i]; | |
4857 | l = raid_map[i]; | |
4858 | bbio->stripes[i] = bbio->stripes[i+1]; | |
4859 | raid_map[i] = raid_map[i+1]; | |
4860 | bbio->stripes[i+1] = s; | |
4861 | raid_map[i+1] = l; | |
4862 | again = 1; | |
4863 | } | |
4864 | } | |
4865 | } | |
4866 | } | |
4867 | ||
3ec706c8 | 4868 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 4869 | u64 logical, u64 *length, |
a1d3c478 | 4870 | struct btrfs_bio **bbio_ret, |
53b381b3 | 4871 | int mirror_num, u64 **raid_map_ret) |
0b86a832 CM |
4872 | { |
4873 | struct extent_map *em; | |
4874 | struct map_lookup *map; | |
3ec706c8 | 4875 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
4876 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
4877 | u64 offset; | |
593060d7 | 4878 | u64 stripe_offset; |
fce3bb9a | 4879 | u64 stripe_end_offset; |
593060d7 | 4880 | u64 stripe_nr; |
fce3bb9a LD |
4881 | u64 stripe_nr_orig; |
4882 | u64 stripe_nr_end; | |
53b381b3 DW |
4883 | u64 stripe_len; |
4884 | u64 *raid_map = NULL; | |
593060d7 | 4885 | int stripe_index; |
cea9e445 | 4886 | int i; |
de11cc12 | 4887 | int ret = 0; |
f2d8d74d | 4888 | int num_stripes; |
a236aed1 | 4889 | int max_errors = 0; |
a1d3c478 | 4890 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
4891 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
4892 | int dev_replace_is_ongoing = 0; | |
4893 | int num_alloc_stripes; | |
ad6d620e SB |
4894 | int patch_the_first_stripe_for_dev_replace = 0; |
4895 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 4896 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 4897 | |
890871be | 4898 | read_lock(&em_tree->lock); |
0b86a832 | 4899 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 4900 | read_unlock(&em_tree->lock); |
f2d8d74d | 4901 | |
3b951516 | 4902 | if (!em) { |
c2cf52eb | 4903 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 4904 | logical, *length); |
9bb91873 JB |
4905 | return -EINVAL; |
4906 | } | |
4907 | ||
4908 | if (em->start > logical || em->start + em->len < logical) { | |
4909 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
351fd353 | 4910 | "found %Lu-%Lu", logical, em->start, |
9bb91873 | 4911 | em->start + em->len); |
7d3d1744 | 4912 | free_extent_map(em); |
9bb91873 | 4913 | return -EINVAL; |
3b951516 | 4914 | } |
0b86a832 | 4915 | |
0b86a832 CM |
4916 | map = (struct map_lookup *)em->bdev; |
4917 | offset = logical - em->start; | |
593060d7 | 4918 | |
53b381b3 | 4919 | stripe_len = map->stripe_len; |
593060d7 CM |
4920 | stripe_nr = offset; |
4921 | /* | |
4922 | * stripe_nr counts the total number of stripes we have to stride | |
4923 | * to get to this block | |
4924 | */ | |
53b381b3 | 4925 | do_div(stripe_nr, stripe_len); |
593060d7 | 4926 | |
53b381b3 | 4927 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
4928 | BUG_ON(offset < stripe_offset); |
4929 | ||
4930 | /* stripe_offset is the offset of this block in its stripe*/ | |
4931 | stripe_offset = offset - stripe_offset; | |
4932 | ||
53b381b3 DW |
4933 | /* if we're here for raid56, we need to know the stripe aligned start */ |
4934 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4935 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); | |
4936 | raid56_full_stripe_start = offset; | |
4937 | ||
4938 | /* allow a write of a full stripe, but make sure we don't | |
4939 | * allow straddling of stripes | |
4940 | */ | |
4941 | do_div(raid56_full_stripe_start, full_stripe_len); | |
4942 | raid56_full_stripe_start *= full_stripe_len; | |
4943 | } | |
4944 | ||
4945 | if (rw & REQ_DISCARD) { | |
4946 | /* we don't discard raid56 yet */ | |
4947 | if (map->type & | |
4948 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4949 | ret = -EOPNOTSUPP; | |
4950 | goto out; | |
4951 | } | |
fce3bb9a | 4952 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
4953 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
4954 | u64 max_len; | |
4955 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
4956 | For other RAID types and for RAID[56] reads, just allow a single | |
4957 | stripe (on a single disk). */ | |
4958 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) && | |
4959 | (rw & REQ_WRITE)) { | |
4960 | max_len = stripe_len * nr_data_stripes(map) - | |
4961 | (offset - raid56_full_stripe_start); | |
4962 | } else { | |
4963 | /* we limit the length of each bio to what fits in a stripe */ | |
4964 | max_len = stripe_len - stripe_offset; | |
4965 | } | |
4966 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
4967 | } else { |
4968 | *length = em->len - offset; | |
4969 | } | |
f2d8d74d | 4970 | |
53b381b3 DW |
4971 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
4972 | it cares about is the length */ | |
a1d3c478 | 4973 | if (!bbio_ret) |
cea9e445 CM |
4974 | goto out; |
4975 | ||
472262f3 SB |
4976 | btrfs_dev_replace_lock(dev_replace); |
4977 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
4978 | if (!dev_replace_is_ongoing) | |
4979 | btrfs_dev_replace_unlock(dev_replace); | |
4980 | ||
ad6d620e SB |
4981 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
4982 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
4983 | dev_replace->tgtdev != NULL) { | |
4984 | /* | |
4985 | * in dev-replace case, for repair case (that's the only | |
4986 | * case where the mirror is selected explicitly when | |
4987 | * calling btrfs_map_block), blocks left of the left cursor | |
4988 | * can also be read from the target drive. | |
4989 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
4990 | * the last one to the array of stripes. For READ, it also | |
4991 | * needs to be supported using the same mirror number. | |
4992 | * If the requested block is not left of the left cursor, | |
4993 | * EIO is returned. This can happen because btrfs_num_copies() | |
4994 | * returns one more in the dev-replace case. | |
4995 | */ | |
4996 | u64 tmp_length = *length; | |
4997 | struct btrfs_bio *tmp_bbio = NULL; | |
4998 | int tmp_num_stripes; | |
4999 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5000 | int index_srcdev = 0; | |
5001 | int found = 0; | |
5002 | u64 physical_of_found = 0; | |
5003 | ||
5004 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
53b381b3 | 5005 | logical, &tmp_length, &tmp_bbio, 0, NULL); |
ad6d620e SB |
5006 | if (ret) { |
5007 | WARN_ON(tmp_bbio != NULL); | |
5008 | goto out; | |
5009 | } | |
5010 | ||
5011 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5012 | if (mirror_num > tmp_num_stripes) { | |
5013 | /* | |
5014 | * REQ_GET_READ_MIRRORS does not contain this | |
5015 | * mirror, that means that the requested area | |
5016 | * is not left of the left cursor | |
5017 | */ | |
5018 | ret = -EIO; | |
5019 | kfree(tmp_bbio); | |
5020 | goto out; | |
5021 | } | |
5022 | ||
5023 | /* | |
5024 | * process the rest of the function using the mirror_num | |
5025 | * of the source drive. Therefore look it up first. | |
5026 | * At the end, patch the device pointer to the one of the | |
5027 | * target drive. | |
5028 | */ | |
5029 | for (i = 0; i < tmp_num_stripes; i++) { | |
5030 | if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) { | |
5031 | /* | |
5032 | * In case of DUP, in order to keep it | |
5033 | * simple, only add the mirror with the | |
5034 | * lowest physical address | |
5035 | */ | |
5036 | if (found && | |
5037 | physical_of_found <= | |
5038 | tmp_bbio->stripes[i].physical) | |
5039 | continue; | |
5040 | index_srcdev = i; | |
5041 | found = 1; | |
5042 | physical_of_found = | |
5043 | tmp_bbio->stripes[i].physical; | |
5044 | } | |
5045 | } | |
5046 | ||
5047 | if (found) { | |
5048 | mirror_num = index_srcdev + 1; | |
5049 | patch_the_first_stripe_for_dev_replace = 1; | |
5050 | physical_to_patch_in_first_stripe = physical_of_found; | |
5051 | } else { | |
5052 | WARN_ON(1); | |
5053 | ret = -EIO; | |
5054 | kfree(tmp_bbio); | |
5055 | goto out; | |
5056 | } | |
5057 | ||
5058 | kfree(tmp_bbio); | |
5059 | } else if (mirror_num > map->num_stripes) { | |
5060 | mirror_num = 0; | |
5061 | } | |
5062 | ||
f2d8d74d | 5063 | num_stripes = 1; |
cea9e445 | 5064 | stripe_index = 0; |
fce3bb9a | 5065 | stripe_nr_orig = stripe_nr; |
fda2832f | 5066 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
fce3bb9a LD |
5067 | do_div(stripe_nr_end, map->stripe_len); |
5068 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5069 | (offset + *length); | |
53b381b3 | 5070 | |
fce3bb9a LD |
5071 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5072 | if (rw & REQ_DISCARD) | |
5073 | num_stripes = min_t(u64, map->num_stripes, | |
5074 | stripe_nr_end - stripe_nr_orig); | |
5075 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
5076 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
29a8d9a0 | 5077 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5078 | num_stripes = map->num_stripes; |
2fff734f | 5079 | else if (mirror_num) |
f188591e | 5080 | stripe_index = mirror_num - 1; |
dfe25020 | 5081 | else { |
30d9861f | 5082 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5083 | map->num_stripes, |
30d9861f SB |
5084 | current->pid % map->num_stripes, |
5085 | dev_replace_is_ongoing); | |
a1d3c478 | 5086 | mirror_num = stripe_index + 1; |
dfe25020 | 5087 | } |
2fff734f | 5088 | |
611f0e00 | 5089 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 5090 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 5091 | num_stripes = map->num_stripes; |
a1d3c478 | 5092 | } else if (mirror_num) { |
f188591e | 5093 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5094 | } else { |
5095 | mirror_num = 1; | |
5096 | } | |
2fff734f | 5097 | |
321aecc6 CM |
5098 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
5099 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
5100 | |
5101 | stripe_index = do_div(stripe_nr, factor); | |
5102 | stripe_index *= map->sub_stripes; | |
5103 | ||
29a8d9a0 | 5104 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5105 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
5106 | else if (rw & REQ_DISCARD) |
5107 | num_stripes = min_t(u64, map->sub_stripes * | |
5108 | (stripe_nr_end - stripe_nr_orig), | |
5109 | map->num_stripes); | |
321aecc6 CM |
5110 | else if (mirror_num) |
5111 | stripe_index += mirror_num - 1; | |
dfe25020 | 5112 | else { |
3e74317a | 5113 | int old_stripe_index = stripe_index; |
30d9861f SB |
5114 | stripe_index = find_live_mirror(fs_info, map, |
5115 | stripe_index, | |
dfe25020 | 5116 | map->sub_stripes, stripe_index + |
30d9861f SB |
5117 | current->pid % map->sub_stripes, |
5118 | dev_replace_is_ongoing); | |
3e74317a | 5119 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5120 | } |
53b381b3 DW |
5121 | |
5122 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
5123 | BTRFS_BLOCK_GROUP_RAID6)) { | |
5124 | u64 tmp; | |
5125 | ||
5126 | if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1) | |
5127 | && raid_map_ret) { | |
5128 | int i, rot; | |
5129 | ||
5130 | /* push stripe_nr back to the start of the full stripe */ | |
5131 | stripe_nr = raid56_full_stripe_start; | |
5132 | do_div(stripe_nr, stripe_len); | |
5133 | ||
5134 | stripe_index = do_div(stripe_nr, nr_data_stripes(map)); | |
5135 | ||
5136 | /* RAID[56] write or recovery. Return all stripes */ | |
5137 | num_stripes = map->num_stripes; | |
5138 | max_errors = nr_parity_stripes(map); | |
5139 | ||
d9b0d9ba | 5140 | raid_map = kmalloc_array(num_stripes, sizeof(u64), |
53b381b3 DW |
5141 | GFP_NOFS); |
5142 | if (!raid_map) { | |
5143 | ret = -ENOMEM; | |
5144 | goto out; | |
5145 | } | |
5146 | ||
5147 | /* Work out the disk rotation on this stripe-set */ | |
5148 | tmp = stripe_nr; | |
5149 | rot = do_div(tmp, num_stripes); | |
5150 | ||
5151 | /* Fill in the logical address of each stripe */ | |
5152 | tmp = stripe_nr * nr_data_stripes(map); | |
5153 | for (i = 0; i < nr_data_stripes(map); i++) | |
5154 | raid_map[(i+rot) % num_stripes] = | |
5155 | em->start + (tmp + i) * map->stripe_len; | |
5156 | ||
5157 | raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5158 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5159 | raid_map[(i+rot+1) % num_stripes] = | |
5160 | RAID6_Q_STRIPE; | |
5161 | ||
5162 | *length = map->stripe_len; | |
5163 | stripe_index = 0; | |
5164 | stripe_offset = 0; | |
5165 | } else { | |
5166 | /* | |
5167 | * Mirror #0 or #1 means the original data block. | |
5168 | * Mirror #2 is RAID5 parity block. | |
5169 | * Mirror #3 is RAID6 Q block. | |
5170 | */ | |
5171 | stripe_index = do_div(stripe_nr, nr_data_stripes(map)); | |
5172 | if (mirror_num > 1) | |
5173 | stripe_index = nr_data_stripes(map) + | |
5174 | mirror_num - 2; | |
5175 | ||
5176 | /* We distribute the parity blocks across stripes */ | |
5177 | tmp = stripe_nr + stripe_index; | |
5178 | stripe_index = do_div(tmp, map->num_stripes); | |
5179 | } | |
8790d502 CM |
5180 | } else { |
5181 | /* | |
5182 | * after this do_div call, stripe_nr is the number of stripes | |
5183 | * on this device we have to walk to find the data, and | |
5184 | * stripe_index is the number of our device in the stripe array | |
5185 | */ | |
5186 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 5187 | mirror_num = stripe_index + 1; |
8790d502 | 5188 | } |
593060d7 | 5189 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 5190 | |
472262f3 | 5191 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
5192 | if (dev_replace_is_ongoing) { |
5193 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
5194 | num_alloc_stripes <<= 1; | |
5195 | if (rw & REQ_GET_READ_MIRRORS) | |
5196 | num_alloc_stripes++; | |
5197 | } | |
472262f3 | 5198 | bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS); |
de11cc12 | 5199 | if (!bbio) { |
eb2067f7 | 5200 | kfree(raid_map); |
de11cc12 LZ |
5201 | ret = -ENOMEM; |
5202 | goto out; | |
5203 | } | |
5204 | atomic_set(&bbio->error, 0); | |
5205 | ||
fce3bb9a | 5206 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
5207 | int factor = 0; |
5208 | int sub_stripes = 0; | |
5209 | u64 stripes_per_dev = 0; | |
5210 | u32 remaining_stripes = 0; | |
b89203f7 | 5211 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5212 | |
5213 | if (map->type & | |
5214 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5215 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5216 | sub_stripes = 1; | |
5217 | else | |
5218 | sub_stripes = map->sub_stripes; | |
5219 | ||
5220 | factor = map->num_stripes / sub_stripes; | |
5221 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5222 | stripe_nr_orig, | |
5223 | factor, | |
5224 | &remaining_stripes); | |
b89203f7 LB |
5225 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5226 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5227 | } |
5228 | ||
fce3bb9a | 5229 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5230 | bbio->stripes[i].physical = |
f2d8d74d CM |
5231 | map->stripes[stripe_index].physical + |
5232 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5233 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5234 | |
ec9ef7a1 LZ |
5235 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5236 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5237 | bbio->stripes[i].length = stripes_per_dev * | |
5238 | map->stripe_len; | |
b89203f7 | 5239 | |
ec9ef7a1 LZ |
5240 | if (i / sub_stripes < remaining_stripes) |
5241 | bbio->stripes[i].length += | |
5242 | map->stripe_len; | |
b89203f7 LB |
5243 | |
5244 | /* | |
5245 | * Special for the first stripe and | |
5246 | * the last stripe: | |
5247 | * | |
5248 | * |-------|...|-------| | |
5249 | * |----------| | |
5250 | * off end_off | |
5251 | */ | |
ec9ef7a1 | 5252 | if (i < sub_stripes) |
a1d3c478 | 5253 | bbio->stripes[i].length -= |
fce3bb9a | 5254 | stripe_offset; |
b89203f7 LB |
5255 | |
5256 | if (stripe_index >= last_stripe && | |
5257 | stripe_index <= (last_stripe + | |
5258 | sub_stripes - 1)) | |
a1d3c478 | 5259 | bbio->stripes[i].length -= |
fce3bb9a | 5260 | stripe_end_offset; |
b89203f7 | 5261 | |
ec9ef7a1 LZ |
5262 | if (i == sub_stripes - 1) |
5263 | stripe_offset = 0; | |
fce3bb9a | 5264 | } else |
a1d3c478 | 5265 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5266 | |
5267 | stripe_index++; | |
5268 | if (stripe_index == map->num_stripes) { | |
5269 | /* This could only happen for RAID0/10 */ | |
5270 | stripe_index = 0; | |
5271 | stripe_nr++; | |
5272 | } | |
5273 | } | |
5274 | } else { | |
5275 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5276 | bbio->stripes[i].physical = |
212a17ab LT |
5277 | map->stripes[stripe_index].physical + |
5278 | stripe_offset + | |
5279 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5280 | bbio->stripes[i].dev = |
212a17ab | 5281 | map->stripes[stripe_index].dev; |
fce3bb9a | 5282 | stripe_index++; |
f2d8d74d | 5283 | } |
593060d7 | 5284 | } |
de11cc12 | 5285 | |
d20983b4 MX |
5286 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
5287 | max_errors = btrfs_chunk_max_errors(map); | |
de11cc12 | 5288 | |
472262f3 SB |
5289 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
5290 | dev_replace->tgtdev != NULL) { | |
5291 | int index_where_to_add; | |
5292 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5293 | ||
5294 | /* | |
5295 | * duplicate the write operations while the dev replace | |
5296 | * procedure is running. Since the copying of the old disk | |
5297 | * to the new disk takes place at run time while the | |
5298 | * filesystem is mounted writable, the regular write | |
5299 | * operations to the old disk have to be duplicated to go | |
5300 | * to the new disk as well. | |
5301 | * Note that device->missing is handled by the caller, and | |
5302 | * that the write to the old disk is already set up in the | |
5303 | * stripes array. | |
5304 | */ | |
5305 | index_where_to_add = num_stripes; | |
5306 | for (i = 0; i < num_stripes; i++) { | |
5307 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5308 | /* write to new disk, too */ | |
5309 | struct btrfs_bio_stripe *new = | |
5310 | bbio->stripes + index_where_to_add; | |
5311 | struct btrfs_bio_stripe *old = | |
5312 | bbio->stripes + i; | |
5313 | ||
5314 | new->physical = old->physical; | |
5315 | new->length = old->length; | |
5316 | new->dev = dev_replace->tgtdev; | |
5317 | index_where_to_add++; | |
5318 | max_errors++; | |
5319 | } | |
5320 | } | |
5321 | num_stripes = index_where_to_add; | |
ad6d620e SB |
5322 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
5323 | dev_replace->tgtdev != NULL) { | |
5324 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5325 | int index_srcdev = 0; | |
5326 | int found = 0; | |
5327 | u64 physical_of_found = 0; | |
5328 | ||
5329 | /* | |
5330 | * During the dev-replace procedure, the target drive can | |
5331 | * also be used to read data in case it is needed to repair | |
5332 | * a corrupt block elsewhere. This is possible if the | |
5333 | * requested area is left of the left cursor. In this area, | |
5334 | * the target drive is a full copy of the source drive. | |
5335 | */ | |
5336 | for (i = 0; i < num_stripes; i++) { | |
5337 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5338 | /* | |
5339 | * In case of DUP, in order to keep it | |
5340 | * simple, only add the mirror with the | |
5341 | * lowest physical address | |
5342 | */ | |
5343 | if (found && | |
5344 | physical_of_found <= | |
5345 | bbio->stripes[i].physical) | |
5346 | continue; | |
5347 | index_srcdev = i; | |
5348 | found = 1; | |
5349 | physical_of_found = bbio->stripes[i].physical; | |
5350 | } | |
5351 | } | |
5352 | if (found) { | |
5353 | u64 length = map->stripe_len; | |
5354 | ||
5355 | if (physical_of_found + length <= | |
5356 | dev_replace->cursor_left) { | |
5357 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5358 | bbio->stripes + num_stripes; | |
5359 | ||
5360 | tgtdev_stripe->physical = physical_of_found; | |
5361 | tgtdev_stripe->length = | |
5362 | bbio->stripes[index_srcdev].length; | |
5363 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5364 | ||
5365 | num_stripes++; | |
5366 | } | |
5367 | } | |
472262f3 SB |
5368 | } |
5369 | ||
de11cc12 LZ |
5370 | *bbio_ret = bbio; |
5371 | bbio->num_stripes = num_stripes; | |
5372 | bbio->max_errors = max_errors; | |
5373 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5374 | |
5375 | /* | |
5376 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5377 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5378 | * available as a mirror | |
5379 | */ | |
5380 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5381 | WARN_ON(num_stripes > 1); | |
5382 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5383 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5384 | bbio->mirror_num = map->num_stripes + 1; | |
5385 | } | |
53b381b3 DW |
5386 | if (raid_map) { |
5387 | sort_parity_stripes(bbio, raid_map); | |
5388 | *raid_map_ret = raid_map; | |
5389 | } | |
cea9e445 | 5390 | out: |
472262f3 SB |
5391 | if (dev_replace_is_ongoing) |
5392 | btrfs_dev_replace_unlock(dev_replace); | |
0b86a832 | 5393 | free_extent_map(em); |
de11cc12 | 5394 | return ret; |
0b86a832 CM |
5395 | } |
5396 | ||
3ec706c8 | 5397 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5398 | u64 logical, u64 *length, |
a1d3c478 | 5399 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5400 | { |
3ec706c8 | 5401 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
53b381b3 | 5402 | mirror_num, NULL); |
f2d8d74d CM |
5403 | } |
5404 | ||
a512bbf8 YZ |
5405 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
5406 | u64 chunk_start, u64 physical, u64 devid, | |
5407 | u64 **logical, int *naddrs, int *stripe_len) | |
5408 | { | |
5409 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5410 | struct extent_map *em; | |
5411 | struct map_lookup *map; | |
5412 | u64 *buf; | |
5413 | u64 bytenr; | |
5414 | u64 length; | |
5415 | u64 stripe_nr; | |
53b381b3 | 5416 | u64 rmap_len; |
a512bbf8 YZ |
5417 | int i, j, nr = 0; |
5418 | ||
890871be | 5419 | read_lock(&em_tree->lock); |
a512bbf8 | 5420 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5421 | read_unlock(&em_tree->lock); |
a512bbf8 | 5422 | |
835d974f | 5423 | if (!em) { |
efe120a0 | 5424 | printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n", |
835d974f JB |
5425 | chunk_start); |
5426 | return -EIO; | |
5427 | } | |
5428 | ||
5429 | if (em->start != chunk_start) { | |
efe120a0 | 5430 | printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n", |
835d974f JB |
5431 | em->start, chunk_start); |
5432 | free_extent_map(em); | |
5433 | return -EIO; | |
5434 | } | |
a512bbf8 YZ |
5435 | map = (struct map_lookup *)em->bdev; |
5436 | ||
5437 | length = em->len; | |
53b381b3 DW |
5438 | rmap_len = map->stripe_len; |
5439 | ||
a512bbf8 YZ |
5440 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5441 | do_div(length, map->num_stripes / map->sub_stripes); | |
5442 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5443 | do_div(length, map->num_stripes); | |
53b381b3 DW |
5444 | else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | |
5445 | BTRFS_BLOCK_GROUP_RAID6)) { | |
5446 | do_div(length, nr_data_stripes(map)); | |
5447 | rmap_len = map->stripe_len * nr_data_stripes(map); | |
5448 | } | |
a512bbf8 YZ |
5449 | |
5450 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
79787eaa | 5451 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5452 | |
5453 | for (i = 0; i < map->num_stripes; i++) { | |
5454 | if (devid && map->stripes[i].dev->devid != devid) | |
5455 | continue; | |
5456 | if (map->stripes[i].physical > physical || | |
5457 | map->stripes[i].physical + length <= physical) | |
5458 | continue; | |
5459 | ||
5460 | stripe_nr = physical - map->stripes[i].physical; | |
5461 | do_div(stripe_nr, map->stripe_len); | |
5462 | ||
5463 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5464 | stripe_nr = stripe_nr * map->num_stripes + i; | |
5465 | do_div(stripe_nr, map->sub_stripes); | |
5466 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
5467 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5468 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5469 | * Alternatively, just use rmap_len below instead of | |
5470 | * map->stripe_len */ | |
5471 | ||
5472 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5473 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5474 | for (j = 0; j < nr; j++) { |
5475 | if (buf[j] == bytenr) | |
5476 | break; | |
5477 | } | |
934d375b CM |
5478 | if (j == nr) { |
5479 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5480 | buf[nr++] = bytenr; |
934d375b | 5481 | } |
a512bbf8 YZ |
5482 | } |
5483 | ||
a512bbf8 YZ |
5484 | *logical = buf; |
5485 | *naddrs = nr; | |
53b381b3 | 5486 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5487 | |
5488 | free_extent_map(em); | |
5489 | return 0; | |
f2d8d74d CM |
5490 | } |
5491 | ||
8408c716 MX |
5492 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err) |
5493 | { | |
5494 | if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED)) | |
5495 | bio_endio_nodec(bio, err); | |
5496 | else | |
5497 | bio_endio(bio, err); | |
5498 | kfree(bbio); | |
5499 | } | |
5500 | ||
a1d3c478 | 5501 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 5502 | { |
9be3395b | 5503 | struct btrfs_bio *bbio = bio->bi_private; |
c404e0dc | 5504 | struct btrfs_device *dev = bbio->stripes[0].dev; |
7d2b4daa | 5505 | int is_orig_bio = 0; |
8790d502 | 5506 | |
442a4f63 | 5507 | if (err) { |
a1d3c478 | 5508 | atomic_inc(&bbio->error); |
442a4f63 SB |
5509 | if (err == -EIO || err == -EREMOTEIO) { |
5510 | unsigned int stripe_index = | |
9be3395b | 5511 | btrfs_io_bio(bio)->stripe_index; |
442a4f63 SB |
5512 | |
5513 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5514 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5515 | if (dev->bdev) { |
5516 | if (bio->bi_rw & WRITE) | |
5517 | btrfs_dev_stat_inc(dev, | |
5518 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5519 | else | |
5520 | btrfs_dev_stat_inc(dev, | |
5521 | BTRFS_DEV_STAT_READ_ERRS); | |
5522 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5523 | btrfs_dev_stat_inc(dev, | |
5524 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5525 | btrfs_dev_stat_print_on_error(dev); | |
5526 | } | |
442a4f63 SB |
5527 | } |
5528 | } | |
8790d502 | 5529 | |
a1d3c478 | 5530 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5531 | is_orig_bio = 1; |
5532 | ||
c404e0dc MX |
5533 | btrfs_bio_counter_dec(bbio->fs_info); |
5534 | ||
a1d3c478 | 5535 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5536 | if (!is_orig_bio) { |
5537 | bio_put(bio); | |
a1d3c478 | 5538 | bio = bbio->orig_bio; |
7d2b4daa | 5539 | } |
c7b22bb1 | 5540 | |
a1d3c478 JS |
5541 | bio->bi_private = bbio->private; |
5542 | bio->bi_end_io = bbio->end_io; | |
9be3395b | 5543 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 5544 | /* only send an error to the higher layers if it is |
53b381b3 | 5545 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5546 | */ |
a1d3c478 | 5547 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 5548 | err = -EIO; |
5dbc8fca | 5549 | } else { |
1259ab75 CM |
5550 | /* |
5551 | * this bio is actually up to date, we didn't | |
5552 | * go over the max number of errors | |
5553 | */ | |
5554 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 5555 | err = 0; |
1259ab75 | 5556 | } |
c55f1396 | 5557 | |
8408c716 | 5558 | btrfs_end_bbio(bbio, bio, err); |
7d2b4daa | 5559 | } else if (!is_orig_bio) { |
8790d502 CM |
5560 | bio_put(bio); |
5561 | } | |
8790d502 CM |
5562 | } |
5563 | ||
8b712842 CM |
5564 | /* |
5565 | * see run_scheduled_bios for a description of why bios are collected for | |
5566 | * async submit. | |
5567 | * | |
5568 | * This will add one bio to the pending list for a device and make sure | |
5569 | * the work struct is scheduled. | |
5570 | */ | |
48a3b636 ES |
5571 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
5572 | struct btrfs_device *device, | |
5573 | int rw, struct bio *bio) | |
8b712842 CM |
5574 | { |
5575 | int should_queue = 1; | |
ffbd517d | 5576 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5577 | |
53b381b3 DW |
5578 | if (device->missing || !device->bdev) { |
5579 | bio_endio(bio, -EIO); | |
5580 | return; | |
5581 | } | |
5582 | ||
8b712842 | 5583 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5584 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5585 | bio_get(bio); |
21adbd5c | 5586 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5587 | bio_put(bio); |
143bede5 | 5588 | return; |
8b712842 CM |
5589 | } |
5590 | ||
5591 | /* | |
0986fe9e | 5592 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5593 | * higher layers. Otherwise, the async bio makes it appear we have |
5594 | * made progress against dirty pages when we've really just put it | |
5595 | * on a queue for later | |
5596 | */ | |
0986fe9e | 5597 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5598 | WARN_ON(bio->bi_next); |
8b712842 CM |
5599 | bio->bi_next = NULL; |
5600 | bio->bi_rw |= rw; | |
5601 | ||
5602 | spin_lock(&device->io_lock); | |
7b6d91da | 5603 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5604 | pending_bios = &device->pending_sync_bios; |
5605 | else | |
5606 | pending_bios = &device->pending_bios; | |
8b712842 | 5607 | |
ffbd517d CM |
5608 | if (pending_bios->tail) |
5609 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5610 | |
ffbd517d CM |
5611 | pending_bios->tail = bio; |
5612 | if (!pending_bios->head) | |
5613 | pending_bios->head = bio; | |
8b712842 CM |
5614 | if (device->running_pending) |
5615 | should_queue = 0; | |
5616 | ||
5617 | spin_unlock(&device->io_lock); | |
5618 | ||
5619 | if (should_queue) | |
a8c93d4e QW |
5620 | btrfs_queue_work(root->fs_info->submit_workers, |
5621 | &device->work); | |
8b712842 CM |
5622 | } |
5623 | ||
de1ee92a JB |
5624 | static int bio_size_ok(struct block_device *bdev, struct bio *bio, |
5625 | sector_t sector) | |
5626 | { | |
5627 | struct bio_vec *prev; | |
5628 | struct request_queue *q = bdev_get_queue(bdev); | |
475bf36f | 5629 | unsigned int max_sectors = queue_max_sectors(q); |
de1ee92a JB |
5630 | struct bvec_merge_data bvm = { |
5631 | .bi_bdev = bdev, | |
5632 | .bi_sector = sector, | |
5633 | .bi_rw = bio->bi_rw, | |
5634 | }; | |
5635 | ||
fae7f21c | 5636 | if (WARN_ON(bio->bi_vcnt == 0)) |
de1ee92a | 5637 | return 1; |
de1ee92a JB |
5638 | |
5639 | prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
aa8b57aa | 5640 | if (bio_sectors(bio) > max_sectors) |
de1ee92a JB |
5641 | return 0; |
5642 | ||
5643 | if (!q->merge_bvec_fn) | |
5644 | return 1; | |
5645 | ||
4f024f37 | 5646 | bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len; |
de1ee92a JB |
5647 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) |
5648 | return 0; | |
5649 | return 1; | |
5650 | } | |
5651 | ||
5652 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5653 | struct bio *bio, u64 physical, int dev_nr, | |
5654 | int rw, int async) | |
5655 | { | |
5656 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5657 | ||
5658 | bio->bi_private = bbio; | |
9be3395b | 5659 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 5660 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 5661 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
5662 | #ifdef DEBUG |
5663 | { | |
5664 | struct rcu_string *name; | |
5665 | ||
5666 | rcu_read_lock(); | |
5667 | name = rcu_dereference(dev->name); | |
d1423248 | 5668 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a JB |
5669 | "(%s id %llu), size=%u\n", rw, |
5670 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
5671 | name->str, dev->devid, bio->bi_size); | |
5672 | rcu_read_unlock(); | |
5673 | } | |
5674 | #endif | |
5675 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
5676 | |
5677 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
5678 | ||
de1ee92a | 5679 | if (async) |
53b381b3 | 5680 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
5681 | else |
5682 | btrfsic_submit_bio(rw, bio); | |
5683 | } | |
5684 | ||
5685 | static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5686 | struct bio *first_bio, struct btrfs_device *dev, | |
5687 | int dev_nr, int rw, int async) | |
5688 | { | |
5689 | struct bio_vec *bvec = first_bio->bi_io_vec; | |
5690 | struct bio *bio; | |
5691 | int nr_vecs = bio_get_nr_vecs(dev->bdev); | |
5692 | u64 physical = bbio->stripes[dev_nr].physical; | |
5693 | ||
5694 | again: | |
5695 | bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS); | |
5696 | if (!bio) | |
5697 | return -ENOMEM; | |
5698 | ||
5699 | while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) { | |
5700 | if (bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5701 | bvec->bv_offset) < bvec->bv_len) { | |
4f024f37 | 5702 | u64 len = bio->bi_iter.bi_size; |
de1ee92a JB |
5703 | |
5704 | atomic_inc(&bbio->stripes_pending); | |
5705 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, | |
5706 | rw, async); | |
5707 | physical += len; | |
5708 | goto again; | |
5709 | } | |
5710 | bvec++; | |
5711 | } | |
5712 | ||
5713 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async); | |
5714 | return 0; | |
5715 | } | |
5716 | ||
5717 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
5718 | { | |
5719 | atomic_inc(&bbio->error); | |
5720 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
8408c716 MX |
5721 | /* Shoud be the original bio. */ |
5722 | WARN_ON(bio != bbio->orig_bio); | |
5723 | ||
de1ee92a JB |
5724 | bio->bi_private = bbio->private; |
5725 | bio->bi_end_io = bbio->end_io; | |
9be3395b | 5726 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 5727 | bio->bi_iter.bi_sector = logical >> 9; |
8408c716 MX |
5728 | |
5729 | btrfs_end_bbio(bbio, bio, -EIO); | |
de1ee92a JB |
5730 | } |
5731 | } | |
5732 | ||
f188591e | 5733 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 5734 | int mirror_num, int async_submit) |
0b86a832 | 5735 | { |
0b86a832 | 5736 | struct btrfs_device *dev; |
8790d502 | 5737 | struct bio *first_bio = bio; |
4f024f37 | 5738 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
5739 | u64 length = 0; |
5740 | u64 map_length; | |
53b381b3 | 5741 | u64 *raid_map = NULL; |
0b86a832 | 5742 | int ret; |
8790d502 CM |
5743 | int dev_nr = 0; |
5744 | int total_devs = 1; | |
a1d3c478 | 5745 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 5746 | |
4f024f37 | 5747 | length = bio->bi_iter.bi_size; |
0b86a832 | 5748 | map_length = length; |
cea9e445 | 5749 | |
c404e0dc | 5750 | btrfs_bio_counter_inc_blocked(root->fs_info); |
53b381b3 DW |
5751 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
5752 | mirror_num, &raid_map); | |
c404e0dc MX |
5753 | if (ret) { |
5754 | btrfs_bio_counter_dec(root->fs_info); | |
79787eaa | 5755 | return ret; |
c404e0dc | 5756 | } |
cea9e445 | 5757 | |
a1d3c478 | 5758 | total_devs = bbio->num_stripes; |
53b381b3 DW |
5759 | bbio->orig_bio = first_bio; |
5760 | bbio->private = first_bio->bi_private; | |
5761 | bbio->end_io = first_bio->bi_end_io; | |
c404e0dc | 5762 | bbio->fs_info = root->fs_info; |
53b381b3 DW |
5763 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
5764 | ||
5765 | if (raid_map) { | |
5766 | /* In this case, map_length has been set to the length of | |
5767 | a single stripe; not the whole write */ | |
5768 | if (rw & WRITE) { | |
c404e0dc MX |
5769 | ret = raid56_parity_write(root, bio, bbio, |
5770 | raid_map, map_length); | |
53b381b3 | 5771 | } else { |
c404e0dc MX |
5772 | ret = raid56_parity_recover(root, bio, bbio, |
5773 | raid_map, map_length, | |
5774 | mirror_num); | |
53b381b3 | 5775 | } |
c404e0dc MX |
5776 | /* |
5777 | * FIXME, replace dosen't support raid56 yet, please fix | |
5778 | * it in the future. | |
5779 | */ | |
5780 | btrfs_bio_counter_dec(root->fs_info); | |
5781 | return ret; | |
53b381b3 DW |
5782 | } |
5783 | ||
cea9e445 | 5784 | if (map_length < length) { |
c2cf52eb | 5785 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", |
c1c9ff7c | 5786 | logical, length, map_length); |
cea9e445 CM |
5787 | BUG(); |
5788 | } | |
a1d3c478 | 5789 | |
d397712b | 5790 | while (dev_nr < total_devs) { |
de1ee92a JB |
5791 | dev = bbio->stripes[dev_nr].dev; |
5792 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
5793 | bbio_error(bbio, first_bio, logical); | |
5794 | dev_nr++; | |
5795 | continue; | |
5796 | } | |
5797 | ||
5798 | /* | |
5799 | * Check and see if we're ok with this bio based on it's size | |
5800 | * and offset with the given device. | |
5801 | */ | |
5802 | if (!bio_size_ok(dev->bdev, first_bio, | |
5803 | bbio->stripes[dev_nr].physical >> 9)) { | |
5804 | ret = breakup_stripe_bio(root, bbio, first_bio, dev, | |
5805 | dev_nr, rw, async_submit); | |
5806 | BUG_ON(ret); | |
5807 | dev_nr++; | |
5808 | continue; | |
5809 | } | |
5810 | ||
a1d3c478 | 5811 | if (dev_nr < total_devs - 1) { |
9be3395b | 5812 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 5813 | BUG_ON(!bio); /* -ENOMEM */ |
a1d3c478 JS |
5814 | } else { |
5815 | bio = first_bio; | |
c55f1396 | 5816 | bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED; |
8790d502 | 5817 | } |
de1ee92a JB |
5818 | |
5819 | submit_stripe_bio(root, bbio, bio, | |
5820 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
5821 | async_submit); | |
8790d502 CM |
5822 | dev_nr++; |
5823 | } | |
c404e0dc | 5824 | btrfs_bio_counter_dec(root->fs_info); |
0b86a832 CM |
5825 | return 0; |
5826 | } | |
5827 | ||
aa1b8cd4 | 5828 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 5829 | u8 *uuid, u8 *fsid) |
0b86a832 | 5830 | { |
2b82032c YZ |
5831 | struct btrfs_device *device; |
5832 | struct btrfs_fs_devices *cur_devices; | |
5833 | ||
aa1b8cd4 | 5834 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
5835 | while (cur_devices) { |
5836 | if (!fsid || | |
5837 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
5838 | device = __find_device(&cur_devices->devices, | |
5839 | devid, uuid); | |
5840 | if (device) | |
5841 | return device; | |
5842 | } | |
5843 | cur_devices = cur_devices->seed; | |
5844 | } | |
5845 | return NULL; | |
0b86a832 CM |
5846 | } |
5847 | ||
dfe25020 | 5848 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 5849 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
5850 | u64 devid, u8 *dev_uuid) |
5851 | { | |
5852 | struct btrfs_device *device; | |
dfe25020 | 5853 | |
12bd2fc0 ID |
5854 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
5855 | if (IS_ERR(device)) | |
7cbd8a83 | 5856 | return NULL; |
12bd2fc0 ID |
5857 | |
5858 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 5859 | device->fs_devices = fs_devices; |
dfe25020 | 5860 | fs_devices->num_devices++; |
12bd2fc0 ID |
5861 | |
5862 | device->missing = 1; | |
cd02dca5 | 5863 | fs_devices->missing_devices++; |
12bd2fc0 | 5864 | |
dfe25020 CM |
5865 | return device; |
5866 | } | |
5867 | ||
12bd2fc0 ID |
5868 | /** |
5869 | * btrfs_alloc_device - allocate struct btrfs_device | |
5870 | * @fs_info: used only for generating a new devid, can be NULL if | |
5871 | * devid is provided (i.e. @devid != NULL). | |
5872 | * @devid: a pointer to devid for this device. If NULL a new devid | |
5873 | * is generated. | |
5874 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
5875 | * is generated. | |
5876 | * | |
5877 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
5878 | * on error. Returned struct is not linked onto any lists and can be | |
5879 | * destroyed with kfree() right away. | |
5880 | */ | |
5881 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
5882 | const u64 *devid, | |
5883 | const u8 *uuid) | |
5884 | { | |
5885 | struct btrfs_device *dev; | |
5886 | u64 tmp; | |
5887 | ||
fae7f21c | 5888 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 5889 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
5890 | |
5891 | dev = __alloc_device(); | |
5892 | if (IS_ERR(dev)) | |
5893 | return dev; | |
5894 | ||
5895 | if (devid) | |
5896 | tmp = *devid; | |
5897 | else { | |
5898 | int ret; | |
5899 | ||
5900 | ret = find_next_devid(fs_info, &tmp); | |
5901 | if (ret) { | |
5902 | kfree(dev); | |
5903 | return ERR_PTR(ret); | |
5904 | } | |
5905 | } | |
5906 | dev->devid = tmp; | |
5907 | ||
5908 | if (uuid) | |
5909 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
5910 | else | |
5911 | generate_random_uuid(dev->uuid); | |
5912 | ||
9e0af237 LB |
5913 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
5914 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
5915 | |
5916 | return dev; | |
5917 | } | |
5918 | ||
0b86a832 CM |
5919 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
5920 | struct extent_buffer *leaf, | |
5921 | struct btrfs_chunk *chunk) | |
5922 | { | |
5923 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5924 | struct map_lookup *map; | |
5925 | struct extent_map *em; | |
5926 | u64 logical; | |
5927 | u64 length; | |
5928 | u64 devid; | |
a443755f | 5929 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 5930 | int num_stripes; |
0b86a832 | 5931 | int ret; |
593060d7 | 5932 | int i; |
0b86a832 | 5933 | |
e17cade2 CM |
5934 | logical = key->offset; |
5935 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 5936 | |
890871be | 5937 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 5938 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 5939 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
5940 | |
5941 | /* already mapped? */ | |
5942 | if (em && em->start <= logical && em->start + em->len > logical) { | |
5943 | free_extent_map(em); | |
0b86a832 CM |
5944 | return 0; |
5945 | } else if (em) { | |
5946 | free_extent_map(em); | |
5947 | } | |
0b86a832 | 5948 | |
172ddd60 | 5949 | em = alloc_extent_map(); |
0b86a832 CM |
5950 | if (!em) |
5951 | return -ENOMEM; | |
593060d7 CM |
5952 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
5953 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
5954 | if (!map) { |
5955 | free_extent_map(em); | |
5956 | return -ENOMEM; | |
5957 | } | |
5958 | ||
298a8f9c | 5959 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
0b86a832 CM |
5960 | em->bdev = (struct block_device *)map; |
5961 | em->start = logical; | |
5962 | em->len = length; | |
70c8a91c | 5963 | em->orig_start = 0; |
0b86a832 | 5964 | em->block_start = 0; |
c8b97818 | 5965 | em->block_len = em->len; |
0b86a832 | 5966 | |
593060d7 CM |
5967 | map->num_stripes = num_stripes; |
5968 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
5969 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
5970 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
5971 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
5972 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 5973 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
5974 | for (i = 0; i < num_stripes; i++) { |
5975 | map->stripes[i].physical = | |
5976 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
5977 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
5978 | read_extent_buffer(leaf, uuid, (unsigned long) |
5979 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
5980 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
5981 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
5982 | uuid, NULL); | |
dfe25020 | 5983 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
5984 | free_extent_map(em); |
5985 | return -EIO; | |
5986 | } | |
dfe25020 CM |
5987 | if (!map->stripes[i].dev) { |
5988 | map->stripes[i].dev = | |
5f375835 MX |
5989 | add_missing_dev(root, root->fs_info->fs_devices, |
5990 | devid, uuid); | |
dfe25020 | 5991 | if (!map->stripes[i].dev) { |
dfe25020 CM |
5992 | free_extent_map(em); |
5993 | return -EIO; | |
5994 | } | |
5995 | } | |
5996 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
5997 | } |
5998 | ||
890871be | 5999 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6000 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6001 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6002 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6003 | free_extent_map(em); |
6004 | ||
6005 | return 0; | |
6006 | } | |
6007 | ||
143bede5 | 6008 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6009 | struct btrfs_dev_item *dev_item, |
6010 | struct btrfs_device *device) | |
6011 | { | |
6012 | unsigned long ptr; | |
0b86a832 CM |
6013 | |
6014 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6015 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6016 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6017 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6018 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6019 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6020 | device->type = btrfs_device_type(leaf, dev_item); |
6021 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6022 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6023 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6024 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6025 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6026 | |
410ba3a2 | 6027 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6028 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6029 | } |
6030 | ||
5f375835 MX |
6031 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
6032 | u8 *fsid) | |
2b82032c YZ |
6033 | { |
6034 | struct btrfs_fs_devices *fs_devices; | |
6035 | int ret; | |
6036 | ||
b367e47f | 6037 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
6038 | |
6039 | fs_devices = root->fs_info->fs_devices->seed; | |
6040 | while (fs_devices) { | |
5f375835 MX |
6041 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6042 | return fs_devices; | |
6043 | ||
2b82032c YZ |
6044 | fs_devices = fs_devices->seed; |
6045 | } | |
6046 | ||
6047 | fs_devices = find_fsid(fsid); | |
6048 | if (!fs_devices) { | |
5f375835 MX |
6049 | if (!btrfs_test_opt(root, DEGRADED)) |
6050 | return ERR_PTR(-ENOENT); | |
6051 | ||
6052 | fs_devices = alloc_fs_devices(fsid); | |
6053 | if (IS_ERR(fs_devices)) | |
6054 | return fs_devices; | |
6055 | ||
6056 | fs_devices->seeding = 1; | |
6057 | fs_devices->opened = 1; | |
6058 | return fs_devices; | |
2b82032c | 6059 | } |
e4404d6e YZ |
6060 | |
6061 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6062 | if (IS_ERR(fs_devices)) |
6063 | return fs_devices; | |
2b82032c | 6064 | |
97288f2c | 6065 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 6066 | root->fs_info->bdev_holder); |
48d28232 JL |
6067 | if (ret) { |
6068 | free_fs_devices(fs_devices); | |
5f375835 | 6069 | fs_devices = ERR_PTR(ret); |
2b82032c | 6070 | goto out; |
48d28232 | 6071 | } |
2b82032c YZ |
6072 | |
6073 | if (!fs_devices->seeding) { | |
6074 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6075 | free_fs_devices(fs_devices); |
5f375835 | 6076 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6077 | goto out; |
6078 | } | |
6079 | ||
6080 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6081 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6082 | out: |
5f375835 | 6083 | return fs_devices; |
2b82032c YZ |
6084 | } |
6085 | ||
0d81ba5d | 6086 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6087 | struct extent_buffer *leaf, |
6088 | struct btrfs_dev_item *dev_item) | |
6089 | { | |
5f375835 | 6090 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
0b86a832 CM |
6091 | struct btrfs_device *device; |
6092 | u64 devid; | |
6093 | int ret; | |
2b82032c | 6094 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6095 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6096 | ||
0b86a832 | 6097 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6098 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6099 | BTRFS_UUID_SIZE); |
1473b24e | 6100 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6101 | BTRFS_UUID_SIZE); |
6102 | ||
6103 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5f375835 MX |
6104 | fs_devices = open_seed_devices(root, fs_uuid); |
6105 | if (IS_ERR(fs_devices)) | |
6106 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6107 | } |
6108 | ||
aa1b8cd4 | 6109 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6110 | if (!device) { |
e4404d6e | 6111 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
6112 | return -EIO; |
6113 | ||
5f375835 MX |
6114 | btrfs_warn(root->fs_info, "devid %llu missing", devid); |
6115 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); | |
6116 | if (!device) | |
6117 | return -ENOMEM; | |
6118 | } else { | |
6119 | if (!device->bdev && !btrfs_test_opt(root, DEGRADED)) | |
6120 | return -EIO; | |
6121 | ||
6122 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6123 | /* |
6124 | * this happens when a device that was properly setup | |
6125 | * in the device info lists suddenly goes bad. | |
6126 | * device->bdev is NULL, and so we have to set | |
6127 | * device->missing to one here | |
6128 | */ | |
5f375835 | 6129 | device->fs_devices->missing_devices++; |
cd02dca5 | 6130 | device->missing = 1; |
2b82032c | 6131 | } |
5f375835 MX |
6132 | |
6133 | /* Move the device to its own fs_devices */ | |
6134 | if (device->fs_devices != fs_devices) { | |
6135 | ASSERT(device->missing); | |
6136 | ||
6137 | list_move(&device->dev_list, &fs_devices->devices); | |
6138 | device->fs_devices->num_devices--; | |
6139 | fs_devices->num_devices++; | |
6140 | ||
6141 | device->fs_devices->missing_devices--; | |
6142 | fs_devices->missing_devices++; | |
6143 | ||
6144 | device->fs_devices = fs_devices; | |
6145 | } | |
2b82032c YZ |
6146 | } |
6147 | ||
6148 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6149 | BUG_ON(device->writeable); | |
6150 | if (device->generation != | |
6151 | btrfs_device_generation(leaf, dev_item)) | |
6152 | return -EINVAL; | |
6324fbf3 | 6153 | } |
0b86a832 CM |
6154 | |
6155 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6156 | device->in_fs_metadata = 1; |
63a212ab | 6157 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6158 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
6159 | spin_lock(&root->fs_info->free_chunk_lock); |
6160 | root->fs_info->free_chunk_space += device->total_bytes - | |
6161 | device->bytes_used; | |
6162 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6163 | } | |
0b86a832 | 6164 | ret = 0; |
0b86a832 CM |
6165 | return ret; |
6166 | } | |
6167 | ||
e4404d6e | 6168 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 6169 | { |
6c41761f | 6170 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 6171 | struct extent_buffer *sb; |
0b86a832 | 6172 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6173 | struct btrfs_chunk *chunk; |
84eed90f CM |
6174 | u8 *ptr; |
6175 | unsigned long sb_ptr; | |
6176 | int ret = 0; | |
0b86a832 CM |
6177 | u32 num_stripes; |
6178 | u32 array_size; | |
6179 | u32 len = 0; | |
0b86a832 | 6180 | u32 cur; |
84eed90f | 6181 | struct btrfs_key key; |
0b86a832 | 6182 | |
e4404d6e | 6183 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
6184 | BTRFS_SUPER_INFO_SIZE); |
6185 | if (!sb) | |
6186 | return -ENOMEM; | |
6187 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 6188 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
6189 | /* |
6190 | * The sb extent buffer is artifical and just used to read the system array. | |
6191 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
6192 | * pages up-to-date when the page is larger: extent does not cover the | |
6193 | * whole page and consequently check_page_uptodate does not find all | |
6194 | * the page's extents up-to-date (the hole beyond sb), | |
6195 | * write_extent_buffer then triggers a WARN_ON. | |
6196 | * | |
6197 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6198 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6199 | * to silence the warning eg. on PowerPC 64. | |
6200 | */ | |
6201 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 6202 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6203 | |
a061fc8d | 6204 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6205 | array_size = btrfs_super_sys_array_size(super_copy); |
6206 | ||
0b86a832 CM |
6207 | ptr = super_copy->sys_chunk_array; |
6208 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6209 | cur = 0; | |
6210 | ||
6211 | while (cur < array_size) { | |
6212 | disk_key = (struct btrfs_disk_key *)ptr; | |
6213 | btrfs_disk_key_to_cpu(&key, disk_key); | |
6214 | ||
a061fc8d | 6215 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
6216 | sb_ptr += len; |
6217 | cur += len; | |
6218 | ||
0d81ba5d | 6219 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 6220 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 6221 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6222 | if (ret) |
6223 | break; | |
0b86a832 CM |
6224 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
6225 | len = btrfs_chunk_item_size(num_stripes); | |
6226 | } else { | |
84eed90f CM |
6227 | ret = -EIO; |
6228 | break; | |
0b86a832 CM |
6229 | } |
6230 | ptr += len; | |
6231 | sb_ptr += len; | |
6232 | cur += len; | |
6233 | } | |
a061fc8d | 6234 | free_extent_buffer(sb); |
84eed90f | 6235 | return ret; |
0b86a832 CM |
6236 | } |
6237 | ||
6238 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6239 | { | |
6240 | struct btrfs_path *path; | |
6241 | struct extent_buffer *leaf; | |
6242 | struct btrfs_key key; | |
6243 | struct btrfs_key found_key; | |
6244 | int ret; | |
6245 | int slot; | |
6246 | ||
6247 | root = root->fs_info->chunk_root; | |
6248 | ||
6249 | path = btrfs_alloc_path(); | |
6250 | if (!path) | |
6251 | return -ENOMEM; | |
6252 | ||
b367e47f LZ |
6253 | mutex_lock(&uuid_mutex); |
6254 | lock_chunks(root); | |
6255 | ||
395927a9 FDBM |
6256 | /* |
6257 | * Read all device items, and then all the chunk items. All | |
6258 | * device items are found before any chunk item (their object id | |
6259 | * is smaller than the lowest possible object id for a chunk | |
6260 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6261 | */ |
6262 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6263 | key.offset = 0; | |
6264 | key.type = 0; | |
0b86a832 | 6265 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6266 | if (ret < 0) |
6267 | goto error; | |
d397712b | 6268 | while (1) { |
0b86a832 CM |
6269 | leaf = path->nodes[0]; |
6270 | slot = path->slots[0]; | |
6271 | if (slot >= btrfs_header_nritems(leaf)) { | |
6272 | ret = btrfs_next_leaf(root, path); | |
6273 | if (ret == 0) | |
6274 | continue; | |
6275 | if (ret < 0) | |
6276 | goto error; | |
6277 | break; | |
6278 | } | |
6279 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6280 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6281 | struct btrfs_dev_item *dev_item; | |
6282 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6283 | struct btrfs_dev_item); |
395927a9 FDBM |
6284 | ret = read_one_dev(root, leaf, dev_item); |
6285 | if (ret) | |
6286 | goto error; | |
0b86a832 CM |
6287 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6288 | struct btrfs_chunk *chunk; | |
6289 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6290 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6291 | if (ret) |
6292 | goto error; | |
0b86a832 CM |
6293 | } |
6294 | path->slots[0]++; | |
6295 | } | |
0b86a832 CM |
6296 | ret = 0; |
6297 | error: | |
b367e47f LZ |
6298 | unlock_chunks(root); |
6299 | mutex_unlock(&uuid_mutex); | |
6300 | ||
2b82032c | 6301 | btrfs_free_path(path); |
0b86a832 CM |
6302 | return ret; |
6303 | } | |
442a4f63 | 6304 | |
cb517eab MX |
6305 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6306 | { | |
6307 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6308 | struct btrfs_device *device; | |
6309 | ||
29cc83f6 LB |
6310 | while (fs_devices) { |
6311 | mutex_lock(&fs_devices->device_list_mutex); | |
6312 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6313 | device->dev_root = fs_info->dev_root; | |
6314 | mutex_unlock(&fs_devices->device_list_mutex); | |
6315 | ||
6316 | fs_devices = fs_devices->seed; | |
6317 | } | |
cb517eab MX |
6318 | } |
6319 | ||
733f4fbb SB |
6320 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6321 | { | |
6322 | int i; | |
6323 | ||
6324 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6325 | btrfs_dev_stat_reset(dev, i); | |
6326 | } | |
6327 | ||
6328 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6329 | { | |
6330 | struct btrfs_key key; | |
6331 | struct btrfs_key found_key; | |
6332 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6333 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6334 | struct extent_buffer *eb; | |
6335 | int slot; | |
6336 | int ret = 0; | |
6337 | struct btrfs_device *device; | |
6338 | struct btrfs_path *path = NULL; | |
6339 | int i; | |
6340 | ||
6341 | path = btrfs_alloc_path(); | |
6342 | if (!path) { | |
6343 | ret = -ENOMEM; | |
6344 | goto out; | |
6345 | } | |
6346 | ||
6347 | mutex_lock(&fs_devices->device_list_mutex); | |
6348 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6349 | int item_size; | |
6350 | struct btrfs_dev_stats_item *ptr; | |
6351 | ||
6352 | key.objectid = 0; | |
6353 | key.type = BTRFS_DEV_STATS_KEY; | |
6354 | key.offset = device->devid; | |
6355 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6356 | if (ret) { | |
733f4fbb SB |
6357 | __btrfs_reset_dev_stats(device); |
6358 | device->dev_stats_valid = 1; | |
6359 | btrfs_release_path(path); | |
6360 | continue; | |
6361 | } | |
6362 | slot = path->slots[0]; | |
6363 | eb = path->nodes[0]; | |
6364 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6365 | item_size = btrfs_item_size_nr(eb, slot); | |
6366 | ||
6367 | ptr = btrfs_item_ptr(eb, slot, | |
6368 | struct btrfs_dev_stats_item); | |
6369 | ||
6370 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6371 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6372 | btrfs_dev_stat_set(device, i, | |
6373 | btrfs_dev_stats_value(eb, ptr, i)); | |
6374 | else | |
6375 | btrfs_dev_stat_reset(device, i); | |
6376 | } | |
6377 | ||
6378 | device->dev_stats_valid = 1; | |
6379 | btrfs_dev_stat_print_on_load(device); | |
6380 | btrfs_release_path(path); | |
6381 | } | |
6382 | mutex_unlock(&fs_devices->device_list_mutex); | |
6383 | ||
6384 | out: | |
6385 | btrfs_free_path(path); | |
6386 | return ret < 0 ? ret : 0; | |
6387 | } | |
6388 | ||
6389 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6390 | struct btrfs_root *dev_root, | |
6391 | struct btrfs_device *device) | |
6392 | { | |
6393 | struct btrfs_path *path; | |
6394 | struct btrfs_key key; | |
6395 | struct extent_buffer *eb; | |
6396 | struct btrfs_dev_stats_item *ptr; | |
6397 | int ret; | |
6398 | int i; | |
6399 | ||
6400 | key.objectid = 0; | |
6401 | key.type = BTRFS_DEV_STATS_KEY; | |
6402 | key.offset = device->devid; | |
6403 | ||
6404 | path = btrfs_alloc_path(); | |
6405 | BUG_ON(!path); | |
6406 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6407 | if (ret < 0) { | |
efe120a0 FH |
6408 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6409 | "error %d while searching for dev_stats item for device %s!\n", | |
606686ee | 6410 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6411 | goto out; |
6412 | } | |
6413 | ||
6414 | if (ret == 0 && | |
6415 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6416 | /* need to delete old one and insert a new one */ | |
6417 | ret = btrfs_del_item(trans, dev_root, path); | |
6418 | if (ret != 0) { | |
efe120a0 FH |
6419 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6420 | "delete too small dev_stats item for device %s failed %d!\n", | |
606686ee | 6421 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6422 | goto out; |
6423 | } | |
6424 | ret = 1; | |
6425 | } | |
6426 | ||
6427 | if (ret == 1) { | |
6428 | /* need to insert a new item */ | |
6429 | btrfs_release_path(path); | |
6430 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6431 | &key, sizeof(*ptr)); | |
6432 | if (ret < 0) { | |
efe120a0 FH |
6433 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6434 | "insert dev_stats item for device %s failed %d!\n", | |
606686ee | 6435 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6436 | goto out; |
6437 | } | |
6438 | } | |
6439 | ||
6440 | eb = path->nodes[0]; | |
6441 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6442 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6443 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6444 | btrfs_dev_stat_read(device, i)); | |
6445 | btrfs_mark_buffer_dirty(eb); | |
6446 | ||
6447 | out: | |
6448 | btrfs_free_path(path); | |
6449 | return ret; | |
6450 | } | |
6451 | ||
6452 | /* | |
6453 | * called from commit_transaction. Writes all changed device stats to disk. | |
6454 | */ | |
6455 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6456 | struct btrfs_fs_info *fs_info) | |
6457 | { | |
6458 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6459 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6460 | struct btrfs_device *device; | |
addc3fa7 | 6461 | int stats_cnt; |
733f4fbb SB |
6462 | int ret = 0; |
6463 | ||
6464 | mutex_lock(&fs_devices->device_list_mutex); | |
6465 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 6466 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
6467 | continue; |
6468 | ||
addc3fa7 | 6469 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
733f4fbb SB |
6470 | ret = update_dev_stat_item(trans, dev_root, device); |
6471 | if (!ret) | |
addc3fa7 | 6472 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
6473 | } |
6474 | mutex_unlock(&fs_devices->device_list_mutex); | |
6475 | ||
6476 | return ret; | |
6477 | } | |
6478 | ||
442a4f63 SB |
6479 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
6480 | { | |
6481 | btrfs_dev_stat_inc(dev, index); | |
6482 | btrfs_dev_stat_print_on_error(dev); | |
6483 | } | |
6484 | ||
48a3b636 | 6485 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 6486 | { |
733f4fbb SB |
6487 | if (!dev->dev_stats_valid) |
6488 | return; | |
efe120a0 FH |
6489 | printk_ratelimited_in_rcu(KERN_ERR "BTRFS: " |
6490 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6491 | rcu_str_deref(dev->name), |
442a4f63 SB |
6492 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6493 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6494 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
6495 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
6496 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 6497 | } |
c11d2c23 | 6498 | |
733f4fbb SB |
6499 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
6500 | { | |
a98cdb85 SB |
6501 | int i; |
6502 | ||
6503 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6504 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6505 | break; | |
6506 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6507 | return; /* all values == 0, suppress message */ | |
6508 | ||
efe120a0 FH |
6509 | printk_in_rcu(KERN_INFO "BTRFS: " |
6510 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6511 | rcu_str_deref(dev->name), |
733f4fbb SB |
6512 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6513 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6514 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6515 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6516 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6517 | } | |
6518 | ||
c11d2c23 | 6519 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 6520 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
6521 | { |
6522 | struct btrfs_device *dev; | |
6523 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6524 | int i; | |
6525 | ||
6526 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 6527 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
6528 | mutex_unlock(&fs_devices->device_list_mutex); |
6529 | ||
6530 | if (!dev) { | |
efe120a0 | 6531 | btrfs_warn(root->fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 6532 | return -ENODEV; |
733f4fbb | 6533 | } else if (!dev->dev_stats_valid) { |
efe120a0 | 6534 | btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 6535 | return -ENODEV; |
b27f7c0c | 6536 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
6537 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
6538 | if (stats->nr_items > i) | |
6539 | stats->values[i] = | |
6540 | btrfs_dev_stat_read_and_reset(dev, i); | |
6541 | else | |
6542 | btrfs_dev_stat_reset(dev, i); | |
6543 | } | |
6544 | } else { | |
6545 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6546 | if (stats->nr_items > i) | |
6547 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6548 | } | |
6549 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6550 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6551 | return 0; | |
6552 | } | |
a8a6dab7 SB |
6553 | |
6554 | int btrfs_scratch_superblock(struct btrfs_device *device) | |
6555 | { | |
6556 | struct buffer_head *bh; | |
6557 | struct btrfs_super_block *disk_super; | |
6558 | ||
6559 | bh = btrfs_read_dev_super(device->bdev); | |
6560 | if (!bh) | |
6561 | return -EINVAL; | |
6562 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6563 | ||
6564 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6565 | set_buffer_dirty(bh); | |
6566 | sync_dirty_buffer(bh); | |
6567 | brelse(bh); | |
6568 | ||
6569 | return 0; | |
6570 | } | |
935e5cc9 MX |
6571 | |
6572 | /* | |
6573 | * Update the size of all devices, which is used for writing out the | |
6574 | * super blocks. | |
6575 | */ | |
6576 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
6577 | { | |
6578 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6579 | struct btrfs_device *curr, *next; | |
6580 | ||
6581 | if (list_empty(&fs_devices->resized_devices)) | |
6582 | return; | |
6583 | ||
6584 | mutex_lock(&fs_devices->device_list_mutex); | |
6585 | lock_chunks(fs_info->dev_root); | |
6586 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
6587 | resized_list) { | |
6588 | list_del_init(&curr->resized_list); | |
6589 | curr->commit_total_bytes = curr->disk_total_bytes; | |
6590 | } | |
6591 | unlock_chunks(fs_info->dev_root); | |
6592 | mutex_unlock(&fs_devices->device_list_mutex); | |
6593 | } | |
ce7213c7 MX |
6594 | |
6595 | /* Must be invoked during the transaction commit */ | |
6596 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
6597 | struct btrfs_transaction *transaction) | |
6598 | { | |
6599 | struct extent_map *em; | |
6600 | struct map_lookup *map; | |
6601 | struct btrfs_device *dev; | |
6602 | int i; | |
6603 | ||
6604 | if (list_empty(&transaction->pending_chunks)) | |
6605 | return; | |
6606 | ||
6607 | /* In order to kick the device replace finish process */ | |
6608 | lock_chunks(root); | |
6609 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
6610 | map = (struct map_lookup *)em->bdev; | |
6611 | ||
6612 | for (i = 0; i < map->num_stripes; i++) { | |
6613 | dev = map->stripes[i].dev; | |
6614 | dev->commit_bytes_used = dev->bytes_used; | |
6615 | } | |
6616 | } | |
6617 | unlock_chunks(root); | |
6618 | } |