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> |
4b4e25f2 | 31 | #include "compat.h" |
0b86a832 CM |
32 | #include "ctree.h" |
33 | #include "extent_map.h" | |
34 | #include "disk-io.h" | |
35 | #include "transaction.h" | |
36 | #include "print-tree.h" | |
37 | #include "volumes.h" | |
53b381b3 | 38 | #include "raid56.h" |
8b712842 | 39 | #include "async-thread.h" |
21adbd5c | 40 | #include "check-integrity.h" |
606686ee | 41 | #include "rcu-string.h" |
3fed40cc | 42 | #include "math.h" |
8dabb742 | 43 | #include "dev-replace.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 | |
8a4b83cc CM |
53 | static DEFINE_MUTEX(uuid_mutex); |
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); | |
77 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
78 | INIT_LIST_HEAD(&fs_devs->list); | |
79 | ||
80 | return fs_devs; | |
81 | } | |
82 | ||
83 | /** | |
84 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
85 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
86 | * generated. | |
87 | * | |
88 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
89 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
90 | * can be destroyed with kfree() right away. | |
91 | */ | |
92 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
93 | { | |
94 | struct btrfs_fs_devices *fs_devs; | |
95 | ||
96 | fs_devs = __alloc_fs_devices(); | |
97 | if (IS_ERR(fs_devs)) | |
98 | return fs_devs; | |
99 | ||
100 | if (fsid) | |
101 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
102 | else | |
103 | generate_random_uuid(fs_devs->fsid); | |
104 | ||
105 | return fs_devs; | |
106 | } | |
107 | ||
e4404d6e YZ |
108 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
109 | { | |
110 | struct btrfs_device *device; | |
111 | WARN_ON(fs_devices->opened); | |
112 | while (!list_empty(&fs_devices->devices)) { | |
113 | device = list_entry(fs_devices->devices.next, | |
114 | struct btrfs_device, dev_list); | |
115 | list_del(&device->dev_list); | |
606686ee | 116 | rcu_string_free(device->name); |
e4404d6e YZ |
117 | kfree(device); |
118 | } | |
119 | kfree(fs_devices); | |
120 | } | |
121 | ||
b8b8ff59 LC |
122 | static void btrfs_kobject_uevent(struct block_device *bdev, |
123 | enum kobject_action action) | |
124 | { | |
125 | int ret; | |
126 | ||
127 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
128 | if (ret) | |
129 | pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n", | |
130 | action, | |
131 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
132 | &disk_to_dev(bdev->bd_disk)->kobj); | |
133 | } | |
134 | ||
143bede5 | 135 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
136 | { |
137 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 138 | |
2b82032c YZ |
139 | while (!list_empty(&fs_uuids)) { |
140 | fs_devices = list_entry(fs_uuids.next, | |
141 | struct btrfs_fs_devices, list); | |
142 | list_del(&fs_devices->list); | |
e4404d6e | 143 | free_fs_devices(fs_devices); |
8a4b83cc | 144 | } |
8a4b83cc CM |
145 | } |
146 | ||
12bd2fc0 ID |
147 | static struct btrfs_device *__alloc_device(void) |
148 | { | |
149 | struct btrfs_device *dev; | |
150 | ||
151 | dev = kzalloc(sizeof(*dev), GFP_NOFS); | |
152 | if (!dev) | |
153 | return ERR_PTR(-ENOMEM); | |
154 | ||
155 | INIT_LIST_HEAD(&dev->dev_list); | |
156 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
157 | ||
158 | spin_lock_init(&dev->io_lock); | |
159 | ||
160 | spin_lock_init(&dev->reada_lock); | |
161 | atomic_set(&dev->reada_in_flight, 0); | |
162 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
163 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
164 | ||
165 | return dev; | |
166 | } | |
167 | ||
a1b32a59 CM |
168 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
169 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
170 | { |
171 | struct btrfs_device *dev; | |
8a4b83cc | 172 | |
c6e30871 | 173 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 174 | if (dev->devid == devid && |
8f18cf13 | 175 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 176 | return dev; |
a443755f | 177 | } |
8a4b83cc CM |
178 | } |
179 | return NULL; | |
180 | } | |
181 | ||
a1b32a59 | 182 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 183 | { |
8a4b83cc CM |
184 | struct btrfs_fs_devices *fs_devices; |
185 | ||
c6e30871 | 186 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
187 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
188 | return fs_devices; | |
189 | } | |
190 | return NULL; | |
191 | } | |
192 | ||
beaf8ab3 SB |
193 | static int |
194 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
195 | int flush, struct block_device **bdev, | |
196 | struct buffer_head **bh) | |
197 | { | |
198 | int ret; | |
199 | ||
200 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
201 | ||
202 | if (IS_ERR(*bdev)) { | |
203 | ret = PTR_ERR(*bdev); | |
204 | printk(KERN_INFO "btrfs: open %s failed\n", device_path); | |
205 | goto error; | |
206 | } | |
207 | ||
208 | if (flush) | |
209 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
210 | ret = set_blocksize(*bdev, 4096); | |
211 | if (ret) { | |
212 | blkdev_put(*bdev, flags); | |
213 | goto error; | |
214 | } | |
215 | invalidate_bdev(*bdev); | |
216 | *bh = btrfs_read_dev_super(*bdev); | |
217 | if (!*bh) { | |
218 | ret = -EINVAL; | |
219 | blkdev_put(*bdev, flags); | |
220 | goto error; | |
221 | } | |
222 | ||
223 | return 0; | |
224 | ||
225 | error: | |
226 | *bdev = NULL; | |
227 | *bh = NULL; | |
228 | return ret; | |
229 | } | |
230 | ||
ffbd517d CM |
231 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
232 | struct bio *head, struct bio *tail) | |
233 | { | |
234 | ||
235 | struct bio *old_head; | |
236 | ||
237 | old_head = pending_bios->head; | |
238 | pending_bios->head = head; | |
239 | if (pending_bios->tail) | |
240 | tail->bi_next = old_head; | |
241 | else | |
242 | pending_bios->tail = tail; | |
243 | } | |
244 | ||
8b712842 CM |
245 | /* |
246 | * we try to collect pending bios for a device so we don't get a large | |
247 | * number of procs sending bios down to the same device. This greatly | |
248 | * improves the schedulers ability to collect and merge the bios. | |
249 | * | |
250 | * But, it also turns into a long list of bios to process and that is sure | |
251 | * to eventually make the worker thread block. The solution here is to | |
252 | * make some progress and then put this work struct back at the end of | |
253 | * the list if the block device is congested. This way, multiple devices | |
254 | * can make progress from a single worker thread. | |
255 | */ | |
143bede5 | 256 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
257 | { |
258 | struct bio *pending; | |
259 | struct backing_dev_info *bdi; | |
b64a2851 | 260 | struct btrfs_fs_info *fs_info; |
ffbd517d | 261 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
262 | struct bio *tail; |
263 | struct bio *cur; | |
264 | int again = 0; | |
ffbd517d | 265 | unsigned long num_run; |
d644d8a1 | 266 | unsigned long batch_run = 0; |
b64a2851 | 267 | unsigned long limit; |
b765ead5 | 268 | unsigned long last_waited = 0; |
d84275c9 | 269 | int force_reg = 0; |
0e588859 | 270 | int sync_pending = 0; |
211588ad CM |
271 | struct blk_plug plug; |
272 | ||
273 | /* | |
274 | * this function runs all the bios we've collected for | |
275 | * a particular device. We don't want to wander off to | |
276 | * another device without first sending all of these down. | |
277 | * So, setup a plug here and finish it off before we return | |
278 | */ | |
279 | blk_start_plug(&plug); | |
8b712842 | 280 | |
bedf762b | 281 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
282 | fs_info = device->dev_root->fs_info; |
283 | limit = btrfs_async_submit_limit(fs_info); | |
284 | limit = limit * 2 / 3; | |
285 | ||
8b712842 CM |
286 | loop: |
287 | spin_lock(&device->io_lock); | |
288 | ||
a6837051 | 289 | loop_lock: |
d84275c9 | 290 | num_run = 0; |
ffbd517d | 291 | |
8b712842 CM |
292 | /* take all the bios off the list at once and process them |
293 | * later on (without the lock held). But, remember the | |
294 | * tail and other pointers so the bios can be properly reinserted | |
295 | * into the list if we hit congestion | |
296 | */ | |
d84275c9 | 297 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 298 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
299 | force_reg = 1; |
300 | } else { | |
ffbd517d | 301 | pending_bios = &device->pending_bios; |
d84275c9 CM |
302 | force_reg = 0; |
303 | } | |
ffbd517d CM |
304 | |
305 | pending = pending_bios->head; | |
306 | tail = pending_bios->tail; | |
8b712842 | 307 | WARN_ON(pending && !tail); |
8b712842 CM |
308 | |
309 | /* | |
310 | * if pending was null this time around, no bios need processing | |
311 | * at all and we can stop. Otherwise it'll loop back up again | |
312 | * and do an additional check so no bios are missed. | |
313 | * | |
314 | * device->running_pending is used to synchronize with the | |
315 | * schedule_bio code. | |
316 | */ | |
ffbd517d CM |
317 | if (device->pending_sync_bios.head == NULL && |
318 | device->pending_bios.head == NULL) { | |
8b712842 CM |
319 | again = 0; |
320 | device->running_pending = 0; | |
ffbd517d CM |
321 | } else { |
322 | again = 1; | |
323 | device->running_pending = 1; | |
8b712842 | 324 | } |
ffbd517d CM |
325 | |
326 | pending_bios->head = NULL; | |
327 | pending_bios->tail = NULL; | |
328 | ||
8b712842 CM |
329 | spin_unlock(&device->io_lock); |
330 | ||
d397712b | 331 | while (pending) { |
ffbd517d CM |
332 | |
333 | rmb(); | |
d84275c9 CM |
334 | /* we want to work on both lists, but do more bios on the |
335 | * sync list than the regular list | |
336 | */ | |
337 | if ((num_run > 32 && | |
338 | pending_bios != &device->pending_sync_bios && | |
339 | device->pending_sync_bios.head) || | |
340 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
341 | device->pending_bios.head)) { | |
ffbd517d CM |
342 | spin_lock(&device->io_lock); |
343 | requeue_list(pending_bios, pending, tail); | |
344 | goto loop_lock; | |
345 | } | |
346 | ||
8b712842 CM |
347 | cur = pending; |
348 | pending = pending->bi_next; | |
349 | cur->bi_next = NULL; | |
b64a2851 | 350 | |
66657b31 | 351 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
352 | waitqueue_active(&fs_info->async_submit_wait)) |
353 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
354 | |
355 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 356 | |
2ab1ba68 CM |
357 | /* |
358 | * if we're doing the sync list, record that our | |
359 | * plug has some sync requests on it | |
360 | * | |
361 | * If we're doing the regular list and there are | |
362 | * sync requests sitting around, unplug before | |
363 | * we add more | |
364 | */ | |
365 | if (pending_bios == &device->pending_sync_bios) { | |
366 | sync_pending = 1; | |
367 | } else if (sync_pending) { | |
368 | blk_finish_plug(&plug); | |
369 | blk_start_plug(&plug); | |
370 | sync_pending = 0; | |
371 | } | |
372 | ||
21adbd5c | 373 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
374 | num_run++; |
375 | batch_run++; | |
7eaceacc | 376 | if (need_resched()) |
ffbd517d | 377 | cond_resched(); |
8b712842 CM |
378 | |
379 | /* | |
380 | * we made progress, there is more work to do and the bdi | |
381 | * is now congested. Back off and let other work structs | |
382 | * run instead | |
383 | */ | |
57fd5a5f | 384 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 385 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 386 | struct io_context *ioc; |
8b712842 | 387 | |
b765ead5 CM |
388 | ioc = current->io_context; |
389 | ||
390 | /* | |
391 | * the main goal here is that we don't want to | |
392 | * block if we're going to be able to submit | |
393 | * more requests without blocking. | |
394 | * | |
395 | * This code does two great things, it pokes into | |
396 | * the elevator code from a filesystem _and_ | |
397 | * it makes assumptions about how batching works. | |
398 | */ | |
399 | if (ioc && ioc->nr_batch_requests > 0 && | |
400 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
401 | (last_waited == 0 || | |
402 | ioc->last_waited == last_waited)) { | |
403 | /* | |
404 | * we want to go through our batch of | |
405 | * requests and stop. So, we copy out | |
406 | * the ioc->last_waited time and test | |
407 | * against it before looping | |
408 | */ | |
409 | last_waited = ioc->last_waited; | |
7eaceacc | 410 | if (need_resched()) |
ffbd517d | 411 | cond_resched(); |
b765ead5 CM |
412 | continue; |
413 | } | |
8b712842 | 414 | spin_lock(&device->io_lock); |
ffbd517d | 415 | requeue_list(pending_bios, pending, tail); |
a6837051 | 416 | device->running_pending = 1; |
8b712842 CM |
417 | |
418 | spin_unlock(&device->io_lock); | |
419 | btrfs_requeue_work(&device->work); | |
420 | goto done; | |
421 | } | |
d85c8a6f CM |
422 | /* unplug every 64 requests just for good measure */ |
423 | if (batch_run % 64 == 0) { | |
424 | blk_finish_plug(&plug); | |
425 | blk_start_plug(&plug); | |
426 | sync_pending = 0; | |
427 | } | |
8b712842 | 428 | } |
ffbd517d | 429 | |
51684082 CM |
430 | cond_resched(); |
431 | if (again) | |
432 | goto loop; | |
433 | ||
434 | spin_lock(&device->io_lock); | |
435 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
436 | goto loop_lock; | |
437 | spin_unlock(&device->io_lock); | |
438 | ||
8b712842 | 439 | done: |
211588ad | 440 | blk_finish_plug(&plug); |
8b712842 CM |
441 | } |
442 | ||
b2950863 | 443 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
444 | { |
445 | struct btrfs_device *device; | |
446 | ||
447 | device = container_of(work, struct btrfs_device, work); | |
448 | run_scheduled_bios(device); | |
449 | } | |
450 | ||
a1b32a59 | 451 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
452 | struct btrfs_super_block *disk_super, |
453 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
454 | { | |
455 | struct btrfs_device *device; | |
456 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 457 | struct rcu_string *name; |
8a4b83cc CM |
458 | u64 found_transid = btrfs_super_generation(disk_super); |
459 | ||
460 | fs_devices = find_fsid(disk_super->fsid); | |
461 | if (!fs_devices) { | |
2208a378 ID |
462 | fs_devices = alloc_fs_devices(disk_super->fsid); |
463 | if (IS_ERR(fs_devices)) | |
464 | return PTR_ERR(fs_devices); | |
465 | ||
8a4b83cc | 466 | list_add(&fs_devices->list, &fs_uuids); |
8a4b83cc CM |
467 | fs_devices->latest_devid = devid; |
468 | fs_devices->latest_trans = found_transid; | |
2208a378 | 469 | |
8a4b83cc CM |
470 | device = NULL; |
471 | } else { | |
a443755f CM |
472 | device = __find_device(&fs_devices->devices, devid, |
473 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
474 | } |
475 | if (!device) { | |
2b82032c YZ |
476 | if (fs_devices->opened) |
477 | return -EBUSY; | |
478 | ||
12bd2fc0 ID |
479 | device = btrfs_alloc_device(NULL, &devid, |
480 | disk_super->dev_item.uuid); | |
481 | if (IS_ERR(device)) { | |
8a4b83cc | 482 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 483 | return PTR_ERR(device); |
8a4b83cc | 484 | } |
606686ee JB |
485 | |
486 | name = rcu_string_strdup(path, GFP_NOFS); | |
487 | if (!name) { | |
8a4b83cc CM |
488 | kfree(device); |
489 | return -ENOMEM; | |
490 | } | |
606686ee | 491 | rcu_assign_pointer(device->name, name); |
90519d66 | 492 | |
e5e9a520 | 493 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 494 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
495 | mutex_unlock(&fs_devices->device_list_mutex); |
496 | ||
2b82032c | 497 | device->fs_devices = fs_devices; |
8a4b83cc | 498 | fs_devices->num_devices++; |
606686ee JB |
499 | } else if (!device->name || strcmp(device->name->str, path)) { |
500 | name = rcu_string_strdup(path, GFP_NOFS); | |
3a0524dc TH |
501 | if (!name) |
502 | return -ENOMEM; | |
606686ee JB |
503 | rcu_string_free(device->name); |
504 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
505 | if (device->missing) { |
506 | fs_devices->missing_devices--; | |
507 | device->missing = 0; | |
508 | } | |
8a4b83cc CM |
509 | } |
510 | ||
511 | if (found_transid > fs_devices->latest_trans) { | |
512 | fs_devices->latest_devid = devid; | |
513 | fs_devices->latest_trans = found_transid; | |
514 | } | |
8a4b83cc CM |
515 | *fs_devices_ret = fs_devices; |
516 | return 0; | |
517 | } | |
518 | ||
e4404d6e YZ |
519 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
520 | { | |
521 | struct btrfs_fs_devices *fs_devices; | |
522 | struct btrfs_device *device; | |
523 | struct btrfs_device *orig_dev; | |
524 | ||
2208a378 ID |
525 | fs_devices = alloc_fs_devices(orig->fsid); |
526 | if (IS_ERR(fs_devices)) | |
527 | return fs_devices; | |
e4404d6e | 528 | |
e4404d6e YZ |
529 | fs_devices->latest_devid = orig->latest_devid; |
530 | fs_devices->latest_trans = orig->latest_trans; | |
02db0844 | 531 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 532 | |
46224705 | 533 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 534 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
535 | struct rcu_string *name; |
536 | ||
12bd2fc0 ID |
537 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
538 | orig_dev->uuid); | |
539 | if (IS_ERR(device)) | |
e4404d6e YZ |
540 | goto error; |
541 | ||
606686ee JB |
542 | /* |
543 | * This is ok to do without rcu read locked because we hold the | |
544 | * uuid mutex so nothing we touch in here is going to disappear. | |
545 | */ | |
546 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
547 | if (!name) { | |
fd2696f3 | 548 | kfree(device); |
e4404d6e | 549 | goto error; |
fd2696f3 | 550 | } |
606686ee | 551 | rcu_assign_pointer(device->name, name); |
e4404d6e | 552 | |
e4404d6e YZ |
553 | list_add(&device->dev_list, &fs_devices->devices); |
554 | device->fs_devices = fs_devices; | |
555 | fs_devices->num_devices++; | |
556 | } | |
557 | return fs_devices; | |
558 | error: | |
559 | free_fs_devices(fs_devices); | |
560 | return ERR_PTR(-ENOMEM); | |
561 | } | |
562 | ||
8dabb742 SB |
563 | void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info, |
564 | struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 565 | { |
c6e30871 | 566 | struct btrfs_device *device, *next; |
dfe25020 | 567 | |
a6b0d5c8 CM |
568 | struct block_device *latest_bdev = NULL; |
569 | u64 latest_devid = 0; | |
570 | u64 latest_transid = 0; | |
571 | ||
dfe25020 CM |
572 | mutex_lock(&uuid_mutex); |
573 | again: | |
46224705 | 574 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 575 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 576 | if (device->in_fs_metadata) { |
63a212ab SB |
577 | if (!device->is_tgtdev_for_dev_replace && |
578 | (!latest_transid || | |
579 | device->generation > latest_transid)) { | |
a6b0d5c8 CM |
580 | latest_devid = device->devid; |
581 | latest_transid = device->generation; | |
582 | latest_bdev = device->bdev; | |
583 | } | |
2b82032c | 584 | continue; |
a6b0d5c8 | 585 | } |
2b82032c | 586 | |
8dabb742 SB |
587 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
588 | /* | |
589 | * In the first step, keep the device which has | |
590 | * the correct fsid and the devid that is used | |
591 | * for the dev_replace procedure. | |
592 | * In the second step, the dev_replace state is | |
593 | * read from the device tree and it is known | |
594 | * whether the procedure is really active or | |
595 | * not, which means whether this device is | |
596 | * used or whether it should be removed. | |
597 | */ | |
598 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
599 | continue; | |
600 | } | |
601 | } | |
2b82032c | 602 | if (device->bdev) { |
d4d77629 | 603 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
604 | device->bdev = NULL; |
605 | fs_devices->open_devices--; | |
606 | } | |
607 | if (device->writeable) { | |
608 | list_del_init(&device->dev_alloc_list); | |
609 | device->writeable = 0; | |
8dabb742 SB |
610 | if (!device->is_tgtdev_for_dev_replace) |
611 | fs_devices->rw_devices--; | |
2b82032c | 612 | } |
e4404d6e YZ |
613 | list_del_init(&device->dev_list); |
614 | fs_devices->num_devices--; | |
606686ee | 615 | rcu_string_free(device->name); |
e4404d6e | 616 | kfree(device); |
dfe25020 | 617 | } |
2b82032c YZ |
618 | |
619 | if (fs_devices->seed) { | |
620 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
621 | goto again; |
622 | } | |
623 | ||
a6b0d5c8 CM |
624 | fs_devices->latest_bdev = latest_bdev; |
625 | fs_devices->latest_devid = latest_devid; | |
626 | fs_devices->latest_trans = latest_transid; | |
627 | ||
dfe25020 | 628 | mutex_unlock(&uuid_mutex); |
dfe25020 | 629 | } |
a0af469b | 630 | |
1f78160c XG |
631 | static void __free_device(struct work_struct *work) |
632 | { | |
633 | struct btrfs_device *device; | |
634 | ||
635 | device = container_of(work, struct btrfs_device, rcu_work); | |
636 | ||
637 | if (device->bdev) | |
638 | blkdev_put(device->bdev, device->mode); | |
639 | ||
606686ee | 640 | rcu_string_free(device->name); |
1f78160c XG |
641 | kfree(device); |
642 | } | |
643 | ||
644 | static void free_device(struct rcu_head *head) | |
645 | { | |
646 | struct btrfs_device *device; | |
647 | ||
648 | device = container_of(head, struct btrfs_device, rcu); | |
649 | ||
650 | INIT_WORK(&device->rcu_work, __free_device); | |
651 | schedule_work(&device->rcu_work); | |
652 | } | |
653 | ||
2b82032c | 654 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 655 | { |
8a4b83cc | 656 | struct btrfs_device *device; |
e4404d6e | 657 | |
2b82032c YZ |
658 | if (--fs_devices->opened > 0) |
659 | return 0; | |
8a4b83cc | 660 | |
c9513edb | 661 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 662 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c | 663 | struct btrfs_device *new_device; |
606686ee | 664 | struct rcu_string *name; |
1f78160c XG |
665 | |
666 | if (device->bdev) | |
a0af469b | 667 | fs_devices->open_devices--; |
1f78160c | 668 | |
8dabb742 | 669 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c YZ |
670 | list_del_init(&device->dev_alloc_list); |
671 | fs_devices->rw_devices--; | |
672 | } | |
673 | ||
d5e2003c JB |
674 | if (device->can_discard) |
675 | fs_devices->num_can_discard--; | |
676 | ||
a1e8780a ID |
677 | new_device = btrfs_alloc_device(NULL, &device->devid, |
678 | device->uuid); | |
679 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
606686ee JB |
680 | |
681 | /* Safe because we are under uuid_mutex */ | |
99f5944b JB |
682 | if (device->name) { |
683 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
a1e8780a | 684 | BUG_ON(!name); /* -ENOMEM */ |
99f5944b JB |
685 | rcu_assign_pointer(new_device->name, name); |
686 | } | |
a1e8780a | 687 | |
1f78160c | 688 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
a1e8780a | 689 | new_device->fs_devices = device->fs_devices; |
1f78160c XG |
690 | |
691 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 692 | } |
c9513edb XG |
693 | mutex_unlock(&fs_devices->device_list_mutex); |
694 | ||
e4404d6e YZ |
695 | WARN_ON(fs_devices->open_devices); |
696 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
697 | fs_devices->opened = 0; |
698 | fs_devices->seeding = 0; | |
2b82032c | 699 | |
8a4b83cc CM |
700 | return 0; |
701 | } | |
702 | ||
2b82032c YZ |
703 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
704 | { | |
e4404d6e | 705 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
706 | int ret; |
707 | ||
708 | mutex_lock(&uuid_mutex); | |
709 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
710 | if (!fs_devices->opened) { |
711 | seed_devices = fs_devices->seed; | |
712 | fs_devices->seed = NULL; | |
713 | } | |
2b82032c | 714 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
715 | |
716 | while (seed_devices) { | |
717 | fs_devices = seed_devices; | |
718 | seed_devices = fs_devices->seed; | |
719 | __btrfs_close_devices(fs_devices); | |
720 | free_fs_devices(fs_devices); | |
721 | } | |
bc178622 ES |
722 | /* |
723 | * Wait for rcu kworkers under __btrfs_close_devices | |
724 | * to finish all blkdev_puts so device is really | |
725 | * free when umount is done. | |
726 | */ | |
727 | rcu_barrier(); | |
2b82032c YZ |
728 | return ret; |
729 | } | |
730 | ||
e4404d6e YZ |
731 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
732 | fmode_t flags, void *holder) | |
8a4b83cc | 733 | { |
d5e2003c | 734 | struct request_queue *q; |
8a4b83cc CM |
735 | struct block_device *bdev; |
736 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 737 | struct btrfs_device *device; |
a0af469b CM |
738 | struct block_device *latest_bdev = NULL; |
739 | struct buffer_head *bh; | |
740 | struct btrfs_super_block *disk_super; | |
741 | u64 latest_devid = 0; | |
742 | u64 latest_transid = 0; | |
a0af469b | 743 | u64 devid; |
2b82032c | 744 | int seeding = 1; |
a0af469b | 745 | int ret = 0; |
8a4b83cc | 746 | |
d4d77629 TH |
747 | flags |= FMODE_EXCL; |
748 | ||
c6e30871 | 749 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
750 | if (device->bdev) |
751 | continue; | |
dfe25020 CM |
752 | if (!device->name) |
753 | continue; | |
754 | ||
f63e0cca ES |
755 | /* Just open everything we can; ignore failures here */ |
756 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
757 | &bdev, &bh)) | |
beaf8ab3 | 758 | continue; |
a0af469b CM |
759 | |
760 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 761 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
762 | if (devid != device->devid) |
763 | goto error_brelse; | |
764 | ||
2b82032c YZ |
765 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
766 | BTRFS_UUID_SIZE)) | |
767 | goto error_brelse; | |
768 | ||
769 | device->generation = btrfs_super_generation(disk_super); | |
770 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 771 | latest_devid = devid; |
2b82032c | 772 | latest_transid = device->generation; |
a0af469b CM |
773 | latest_bdev = bdev; |
774 | } | |
775 | ||
2b82032c YZ |
776 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
777 | device->writeable = 0; | |
778 | } else { | |
779 | device->writeable = !bdev_read_only(bdev); | |
780 | seeding = 0; | |
781 | } | |
782 | ||
d5e2003c JB |
783 | q = bdev_get_queue(bdev); |
784 | if (blk_queue_discard(q)) { | |
785 | device->can_discard = 1; | |
786 | fs_devices->num_can_discard++; | |
787 | } | |
788 | ||
8a4b83cc | 789 | device->bdev = bdev; |
dfe25020 | 790 | device->in_fs_metadata = 0; |
15916de8 CM |
791 | device->mode = flags; |
792 | ||
c289811c CM |
793 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
794 | fs_devices->rotating = 1; | |
795 | ||
a0af469b | 796 | fs_devices->open_devices++; |
8dabb742 | 797 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c YZ |
798 | fs_devices->rw_devices++; |
799 | list_add(&device->dev_alloc_list, | |
800 | &fs_devices->alloc_list); | |
801 | } | |
4f6c9328 | 802 | brelse(bh); |
a0af469b | 803 | continue; |
a061fc8d | 804 | |
a0af469b CM |
805 | error_brelse: |
806 | brelse(bh); | |
d4d77629 | 807 | blkdev_put(bdev, flags); |
a0af469b | 808 | continue; |
8a4b83cc | 809 | } |
a0af469b | 810 | if (fs_devices->open_devices == 0) { |
20bcd649 | 811 | ret = -EINVAL; |
a0af469b CM |
812 | goto out; |
813 | } | |
2b82032c YZ |
814 | fs_devices->seeding = seeding; |
815 | fs_devices->opened = 1; | |
a0af469b CM |
816 | fs_devices->latest_bdev = latest_bdev; |
817 | fs_devices->latest_devid = latest_devid; | |
818 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 819 | fs_devices->total_rw_bytes = 0; |
a0af469b | 820 | out: |
2b82032c YZ |
821 | return ret; |
822 | } | |
823 | ||
824 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 825 | fmode_t flags, void *holder) |
2b82032c YZ |
826 | { |
827 | int ret; | |
828 | ||
829 | mutex_lock(&uuid_mutex); | |
830 | if (fs_devices->opened) { | |
e4404d6e YZ |
831 | fs_devices->opened++; |
832 | ret = 0; | |
2b82032c | 833 | } else { |
15916de8 | 834 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 835 | } |
8a4b83cc | 836 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
837 | return ret; |
838 | } | |
839 | ||
6f60cbd3 DS |
840 | /* |
841 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
842 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
843 | * is read via pagecache | |
844 | */ | |
97288f2c | 845 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
846 | struct btrfs_fs_devices **fs_devices_ret) |
847 | { | |
848 | struct btrfs_super_block *disk_super; | |
849 | struct block_device *bdev; | |
6f60cbd3 DS |
850 | struct page *page; |
851 | void *p; | |
852 | int ret = -EINVAL; | |
8a4b83cc | 853 | u64 devid; |
f2984462 | 854 | u64 transid; |
02db0844 | 855 | u64 total_devices; |
6f60cbd3 DS |
856 | u64 bytenr; |
857 | pgoff_t index; | |
8a4b83cc | 858 | |
6f60cbd3 DS |
859 | /* |
860 | * we would like to check all the supers, but that would make | |
861 | * a btrfs mount succeed after a mkfs from a different FS. | |
862 | * So, we need to add a special mount option to scan for | |
863 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
864 | */ | |
865 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 866 | flags |= FMODE_EXCL; |
10f6327b | 867 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
868 | |
869 | bdev = blkdev_get_by_path(path, flags, holder); | |
870 | ||
871 | if (IS_ERR(bdev)) { | |
872 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 873 | goto error; |
6f60cbd3 DS |
874 | } |
875 | ||
876 | /* make sure our super fits in the device */ | |
877 | if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode)) | |
878 | goto error_bdev_put; | |
879 | ||
880 | /* make sure our super fits in the page */ | |
881 | if (sizeof(*disk_super) > PAGE_CACHE_SIZE) | |
882 | goto error_bdev_put; | |
883 | ||
884 | /* make sure our super doesn't straddle pages on disk */ | |
885 | index = bytenr >> PAGE_CACHE_SHIFT; | |
886 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index) | |
887 | goto error_bdev_put; | |
888 | ||
889 | /* pull in the page with our super */ | |
890 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
891 | index, GFP_NOFS); | |
892 | ||
893 | if (IS_ERR_OR_NULL(page)) | |
894 | goto error_bdev_put; | |
895 | ||
896 | p = kmap(page); | |
897 | ||
898 | /* align our pointer to the offset of the super block */ | |
899 | disk_super = p + (bytenr & ~PAGE_CACHE_MASK); | |
900 | ||
901 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
3cae210f | 902 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) |
6f60cbd3 DS |
903 | goto error_unmap; |
904 | ||
a343832f | 905 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 906 | transid = btrfs_super_generation(disk_super); |
02db0844 | 907 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 908 | |
d03f918a SB |
909 | if (disk_super->label[0]) { |
910 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
911 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
d397712b | 912 | printk(KERN_INFO "device label %s ", disk_super->label); |
d03f918a | 913 | } else { |
22b63a29 | 914 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); |
d03f918a | 915 | } |
6f60cbd3 | 916 | |
c1c9ff7c | 917 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); |
6f60cbd3 | 918 | |
8a4b83cc | 919 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
02db0844 JB |
920 | if (!ret && fs_devices_ret) |
921 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 DS |
922 | |
923 | error_unmap: | |
924 | kunmap(page); | |
925 | page_cache_release(page); | |
926 | ||
927 | error_bdev_put: | |
d4d77629 | 928 | blkdev_put(bdev, flags); |
8a4b83cc | 929 | error: |
beaf8ab3 | 930 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
931 | return ret; |
932 | } | |
0b86a832 | 933 | |
6d07bcec MX |
934 | /* helper to account the used device space in the range */ |
935 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
936 | u64 end, u64 *length) | |
937 | { | |
938 | struct btrfs_key key; | |
939 | struct btrfs_root *root = device->dev_root; | |
940 | struct btrfs_dev_extent *dev_extent; | |
941 | struct btrfs_path *path; | |
942 | u64 extent_end; | |
943 | int ret; | |
944 | int slot; | |
945 | struct extent_buffer *l; | |
946 | ||
947 | *length = 0; | |
948 | ||
63a212ab | 949 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
950 | return 0; |
951 | ||
952 | path = btrfs_alloc_path(); | |
953 | if (!path) | |
954 | return -ENOMEM; | |
955 | path->reada = 2; | |
956 | ||
957 | key.objectid = device->devid; | |
958 | key.offset = start; | |
959 | key.type = BTRFS_DEV_EXTENT_KEY; | |
960 | ||
961 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
962 | if (ret < 0) | |
963 | goto out; | |
964 | if (ret > 0) { | |
965 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
966 | if (ret < 0) | |
967 | goto out; | |
968 | } | |
969 | ||
970 | while (1) { | |
971 | l = path->nodes[0]; | |
972 | slot = path->slots[0]; | |
973 | if (slot >= btrfs_header_nritems(l)) { | |
974 | ret = btrfs_next_leaf(root, path); | |
975 | if (ret == 0) | |
976 | continue; | |
977 | if (ret < 0) | |
978 | goto out; | |
979 | ||
980 | break; | |
981 | } | |
982 | btrfs_item_key_to_cpu(l, &key, slot); | |
983 | ||
984 | if (key.objectid < device->devid) | |
985 | goto next; | |
986 | ||
987 | if (key.objectid > device->devid) | |
988 | break; | |
989 | ||
990 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
991 | goto next; | |
992 | ||
993 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
994 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
995 | dev_extent); | |
996 | if (key.offset <= start && extent_end > end) { | |
997 | *length = end - start + 1; | |
998 | break; | |
999 | } else if (key.offset <= start && extent_end > start) | |
1000 | *length += extent_end - start; | |
1001 | else if (key.offset > start && extent_end <= end) | |
1002 | *length += extent_end - key.offset; | |
1003 | else if (key.offset > start && key.offset <= end) { | |
1004 | *length += end - key.offset + 1; | |
1005 | break; | |
1006 | } else if (key.offset > end) | |
1007 | break; | |
1008 | ||
1009 | next: | |
1010 | path->slots[0]++; | |
1011 | } | |
1012 | ret = 0; | |
1013 | out: | |
1014 | btrfs_free_path(path); | |
1015 | return ret; | |
1016 | } | |
1017 | ||
6df9a95e JB |
1018 | static int contains_pending_extent(struct btrfs_trans_handle *trans, |
1019 | struct btrfs_device *device, | |
1020 | u64 *start, u64 len) | |
1021 | { | |
1022 | struct extent_map *em; | |
1023 | int ret = 0; | |
1024 | ||
1025 | list_for_each_entry(em, &trans->transaction->pending_chunks, list) { | |
1026 | struct map_lookup *map; | |
1027 | int i; | |
1028 | ||
1029 | map = (struct map_lookup *)em->bdev; | |
1030 | for (i = 0; i < map->num_stripes; i++) { | |
1031 | if (map->stripes[i].dev != device) | |
1032 | continue; | |
1033 | if (map->stripes[i].physical >= *start + len || | |
1034 | map->stripes[i].physical + em->orig_block_len <= | |
1035 | *start) | |
1036 | continue; | |
1037 | *start = map->stripes[i].physical + | |
1038 | em->orig_block_len; | |
1039 | ret = 1; | |
1040 | } | |
1041 | } | |
1042 | ||
1043 | return ret; | |
1044 | } | |
1045 | ||
1046 | ||
0b86a832 | 1047 | /* |
7bfc837d | 1048 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
1049 | * @device: the device which we search the free space in |
1050 | * @num_bytes: the size of the free space that we need | |
1051 | * @start: store the start of the free space. | |
1052 | * @len: the size of the free space. that we find, or the size of the max | |
1053 | * free space if we don't find suitable free space | |
1054 | * | |
0b86a832 CM |
1055 | * this uses a pretty simple search, the expectation is that it is |
1056 | * called very infrequently and that a given device has a small number | |
1057 | * of extents | |
7bfc837d MX |
1058 | * |
1059 | * @start is used to store the start of the free space if we find. But if we | |
1060 | * don't find suitable free space, it will be used to store the start position | |
1061 | * of the max free space. | |
1062 | * | |
1063 | * @len is used to store the size of the free space that we find. | |
1064 | * But if we don't find suitable free space, it is used to store the size of | |
1065 | * the max free space. | |
0b86a832 | 1066 | */ |
6df9a95e JB |
1067 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1068 | struct btrfs_device *device, u64 num_bytes, | |
7bfc837d | 1069 | u64 *start, u64 *len) |
0b86a832 CM |
1070 | { |
1071 | struct btrfs_key key; | |
1072 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1073 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1074 | struct btrfs_path *path; |
7bfc837d MX |
1075 | u64 hole_size; |
1076 | u64 max_hole_start; | |
1077 | u64 max_hole_size; | |
1078 | u64 extent_end; | |
1079 | u64 search_start; | |
0b86a832 CM |
1080 | u64 search_end = device->total_bytes; |
1081 | int ret; | |
7bfc837d | 1082 | int slot; |
0b86a832 CM |
1083 | struct extent_buffer *l; |
1084 | ||
0b86a832 CM |
1085 | /* FIXME use last free of some kind */ |
1086 | ||
8a4b83cc CM |
1087 | /* we don't want to overwrite the superblock on the drive, |
1088 | * so we make sure to start at an offset of at least 1MB | |
1089 | */ | |
a9c9bf68 | 1090 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 1091 | |
6df9a95e JB |
1092 | path = btrfs_alloc_path(); |
1093 | if (!path) | |
1094 | return -ENOMEM; | |
1095 | again: | |
7bfc837d MX |
1096 | max_hole_start = search_start; |
1097 | max_hole_size = 0; | |
38c01b96 | 1098 | hole_size = 0; |
7bfc837d | 1099 | |
63a212ab | 1100 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1101 | ret = -ENOSPC; |
6df9a95e | 1102 | goto out; |
7bfc837d MX |
1103 | } |
1104 | ||
7bfc837d | 1105 | path->reada = 2; |
6df9a95e JB |
1106 | path->search_commit_root = 1; |
1107 | path->skip_locking = 1; | |
7bfc837d | 1108 | |
0b86a832 CM |
1109 | key.objectid = device->devid; |
1110 | key.offset = search_start; | |
1111 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1112 | |
125ccb0a | 1113 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1114 | if (ret < 0) |
7bfc837d | 1115 | goto out; |
1fcbac58 YZ |
1116 | if (ret > 0) { |
1117 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1118 | if (ret < 0) | |
7bfc837d | 1119 | goto out; |
1fcbac58 | 1120 | } |
7bfc837d | 1121 | |
0b86a832 CM |
1122 | while (1) { |
1123 | l = path->nodes[0]; | |
1124 | slot = path->slots[0]; | |
1125 | if (slot >= btrfs_header_nritems(l)) { | |
1126 | ret = btrfs_next_leaf(root, path); | |
1127 | if (ret == 0) | |
1128 | continue; | |
1129 | if (ret < 0) | |
7bfc837d MX |
1130 | goto out; |
1131 | ||
1132 | break; | |
0b86a832 CM |
1133 | } |
1134 | btrfs_item_key_to_cpu(l, &key, slot); | |
1135 | ||
1136 | if (key.objectid < device->devid) | |
1137 | goto next; | |
1138 | ||
1139 | if (key.objectid > device->devid) | |
7bfc837d | 1140 | break; |
0b86a832 | 1141 | |
7bfc837d MX |
1142 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
1143 | goto next; | |
9779b72f | 1144 | |
7bfc837d MX |
1145 | if (key.offset > search_start) { |
1146 | hole_size = key.offset - search_start; | |
9779b72f | 1147 | |
6df9a95e JB |
1148 | /* |
1149 | * Have to check before we set max_hole_start, otherwise | |
1150 | * we could end up sending back this offset anyway. | |
1151 | */ | |
1152 | if (contains_pending_extent(trans, device, | |
1153 | &search_start, | |
1154 | hole_size)) | |
1155 | hole_size = 0; | |
1156 | ||
7bfc837d MX |
1157 | if (hole_size > max_hole_size) { |
1158 | max_hole_start = search_start; | |
1159 | max_hole_size = hole_size; | |
1160 | } | |
9779b72f | 1161 | |
7bfc837d MX |
1162 | /* |
1163 | * If this free space is greater than which we need, | |
1164 | * it must be the max free space that we have found | |
1165 | * until now, so max_hole_start must point to the start | |
1166 | * of this free space and the length of this free space | |
1167 | * is stored in max_hole_size. Thus, we return | |
1168 | * max_hole_start and max_hole_size and go back to the | |
1169 | * caller. | |
1170 | */ | |
1171 | if (hole_size >= num_bytes) { | |
1172 | ret = 0; | |
1173 | goto out; | |
0b86a832 CM |
1174 | } |
1175 | } | |
0b86a832 | 1176 | |
0b86a832 | 1177 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1178 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1179 | dev_extent); | |
1180 | if (extent_end > search_start) | |
1181 | search_start = extent_end; | |
0b86a832 CM |
1182 | next: |
1183 | path->slots[0]++; | |
1184 | cond_resched(); | |
1185 | } | |
0b86a832 | 1186 | |
38c01b96 | 1187 | /* |
1188 | * At this point, search_start should be the end of | |
1189 | * allocated dev extents, and when shrinking the device, | |
1190 | * search_end may be smaller than search_start. | |
1191 | */ | |
1192 | if (search_end > search_start) | |
1193 | hole_size = search_end - search_start; | |
1194 | ||
7bfc837d MX |
1195 | if (hole_size > max_hole_size) { |
1196 | max_hole_start = search_start; | |
1197 | max_hole_size = hole_size; | |
0b86a832 | 1198 | } |
0b86a832 | 1199 | |
6df9a95e JB |
1200 | if (contains_pending_extent(trans, device, &search_start, hole_size)) { |
1201 | btrfs_release_path(path); | |
1202 | goto again; | |
1203 | } | |
1204 | ||
7bfc837d MX |
1205 | /* See above. */ |
1206 | if (hole_size < num_bytes) | |
1207 | ret = -ENOSPC; | |
1208 | else | |
1209 | ret = 0; | |
1210 | ||
1211 | out: | |
2b82032c | 1212 | btrfs_free_path(path); |
7bfc837d | 1213 | *start = max_hole_start; |
b2117a39 | 1214 | if (len) |
7bfc837d | 1215 | *len = max_hole_size; |
0b86a832 CM |
1216 | return ret; |
1217 | } | |
1218 | ||
b2950863 | 1219 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1220 | struct btrfs_device *device, |
1221 | u64 start) | |
1222 | { | |
1223 | int ret; | |
1224 | struct btrfs_path *path; | |
1225 | struct btrfs_root *root = device->dev_root; | |
1226 | struct btrfs_key key; | |
a061fc8d CM |
1227 | struct btrfs_key found_key; |
1228 | struct extent_buffer *leaf = NULL; | |
1229 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1230 | |
1231 | path = btrfs_alloc_path(); | |
1232 | if (!path) | |
1233 | return -ENOMEM; | |
1234 | ||
1235 | key.objectid = device->devid; | |
1236 | key.offset = start; | |
1237 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1238 | again: |
8f18cf13 | 1239 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1240 | if (ret > 0) { |
1241 | ret = btrfs_previous_item(root, path, key.objectid, | |
1242 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1243 | if (ret) |
1244 | goto out; | |
a061fc8d CM |
1245 | leaf = path->nodes[0]; |
1246 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1247 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1248 | struct btrfs_dev_extent); | |
1249 | BUG_ON(found_key.offset > start || found_key.offset + | |
1250 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1251 | key = found_key; |
1252 | btrfs_release_path(path); | |
1253 | goto again; | |
a061fc8d CM |
1254 | } else if (ret == 0) { |
1255 | leaf = path->nodes[0]; | |
1256 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1257 | struct btrfs_dev_extent); | |
79787eaa JM |
1258 | } else { |
1259 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1260 | goto out; | |
a061fc8d | 1261 | } |
8f18cf13 | 1262 | |
2bf64758 JB |
1263 | if (device->bytes_used > 0) { |
1264 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1265 | device->bytes_used -= len; | |
1266 | spin_lock(&root->fs_info->free_chunk_lock); | |
1267 | root->fs_info->free_chunk_space += len; | |
1268 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1269 | } | |
8f18cf13 | 1270 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1271 | if (ret) { |
1272 | btrfs_error(root->fs_info, ret, | |
1273 | "Failed to remove dev extent item"); | |
1274 | } | |
b0b802d7 | 1275 | out: |
8f18cf13 CM |
1276 | btrfs_free_path(path); |
1277 | return ret; | |
1278 | } | |
1279 | ||
48a3b636 ES |
1280 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1281 | struct btrfs_device *device, | |
1282 | u64 chunk_tree, u64 chunk_objectid, | |
1283 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1284 | { |
1285 | int ret; | |
1286 | struct btrfs_path *path; | |
1287 | struct btrfs_root *root = device->dev_root; | |
1288 | struct btrfs_dev_extent *extent; | |
1289 | struct extent_buffer *leaf; | |
1290 | struct btrfs_key key; | |
1291 | ||
dfe25020 | 1292 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1293 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1294 | path = btrfs_alloc_path(); |
1295 | if (!path) | |
1296 | return -ENOMEM; | |
1297 | ||
0b86a832 | 1298 | key.objectid = device->devid; |
2b82032c | 1299 | key.offset = start; |
0b86a832 CM |
1300 | key.type = BTRFS_DEV_EXTENT_KEY; |
1301 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1302 | sizeof(*extent)); | |
2cdcecbc MF |
1303 | if (ret) |
1304 | goto out; | |
0b86a832 CM |
1305 | |
1306 | leaf = path->nodes[0]; | |
1307 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1308 | struct btrfs_dev_extent); | |
e17cade2 CM |
1309 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1310 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1311 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1312 | ||
1313 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1314 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1315 | |
0b86a832 CM |
1316 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1317 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1318 | out: |
0b86a832 CM |
1319 | btrfs_free_path(path); |
1320 | return ret; | |
1321 | } | |
1322 | ||
6df9a95e | 1323 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1324 | { |
6df9a95e JB |
1325 | struct extent_map_tree *em_tree; |
1326 | struct extent_map *em; | |
1327 | struct rb_node *n; | |
1328 | u64 ret = 0; | |
0b86a832 | 1329 | |
6df9a95e JB |
1330 | em_tree = &fs_info->mapping_tree.map_tree; |
1331 | read_lock(&em_tree->lock); | |
1332 | n = rb_last(&em_tree->map); | |
1333 | if (n) { | |
1334 | em = rb_entry(n, struct extent_map, rb_node); | |
1335 | ret = em->start + em->len; | |
0b86a832 | 1336 | } |
6df9a95e JB |
1337 | read_unlock(&em_tree->lock); |
1338 | ||
0b86a832 CM |
1339 | return ret; |
1340 | } | |
1341 | ||
53f10659 ID |
1342 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1343 | u64 *devid_ret) | |
0b86a832 CM |
1344 | { |
1345 | int ret; | |
1346 | struct btrfs_key key; | |
1347 | struct btrfs_key found_key; | |
2b82032c YZ |
1348 | struct btrfs_path *path; |
1349 | ||
2b82032c YZ |
1350 | path = btrfs_alloc_path(); |
1351 | if (!path) | |
1352 | return -ENOMEM; | |
0b86a832 CM |
1353 | |
1354 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1355 | key.type = BTRFS_DEV_ITEM_KEY; | |
1356 | key.offset = (u64)-1; | |
1357 | ||
53f10659 | 1358 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1359 | if (ret < 0) |
1360 | goto error; | |
1361 | ||
79787eaa | 1362 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1363 | |
53f10659 ID |
1364 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1365 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1366 | BTRFS_DEV_ITEM_KEY); |
1367 | if (ret) { | |
53f10659 | 1368 | *devid_ret = 1; |
0b86a832 CM |
1369 | } else { |
1370 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1371 | path->slots[0]); | |
53f10659 | 1372 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1373 | } |
1374 | ret = 0; | |
1375 | error: | |
2b82032c | 1376 | btrfs_free_path(path); |
0b86a832 CM |
1377 | return ret; |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * the device information is stored in the chunk root | |
1382 | * the btrfs_device struct should be fully filled in | |
1383 | */ | |
48a3b636 ES |
1384 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1385 | struct btrfs_root *root, | |
1386 | struct btrfs_device *device) | |
0b86a832 CM |
1387 | { |
1388 | int ret; | |
1389 | struct btrfs_path *path; | |
1390 | struct btrfs_dev_item *dev_item; | |
1391 | struct extent_buffer *leaf; | |
1392 | struct btrfs_key key; | |
1393 | unsigned long ptr; | |
0b86a832 CM |
1394 | |
1395 | root = root->fs_info->chunk_root; | |
1396 | ||
1397 | path = btrfs_alloc_path(); | |
1398 | if (!path) | |
1399 | return -ENOMEM; | |
1400 | ||
0b86a832 CM |
1401 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1402 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1403 | key.offset = device->devid; |
0b86a832 CM |
1404 | |
1405 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1406 | sizeof(*dev_item)); |
0b86a832 CM |
1407 | if (ret) |
1408 | goto out; | |
1409 | ||
1410 | leaf = path->nodes[0]; | |
1411 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1412 | ||
1413 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1414 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1415 | btrfs_set_device_type(leaf, dev_item, device->type); |
1416 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1417 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1418 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1419 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1420 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1421 | btrfs_set_device_group(leaf, dev_item, 0); |
1422 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1423 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1424 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1425 | |
410ba3a2 | 1426 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1427 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1428 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1429 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1430 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1431 | |
2b82032c | 1432 | ret = 0; |
0b86a832 CM |
1433 | out: |
1434 | btrfs_free_path(path); | |
1435 | return ret; | |
1436 | } | |
8f18cf13 | 1437 | |
a061fc8d CM |
1438 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1439 | struct btrfs_device *device) | |
1440 | { | |
1441 | int ret; | |
1442 | struct btrfs_path *path; | |
a061fc8d | 1443 | struct btrfs_key key; |
a061fc8d CM |
1444 | struct btrfs_trans_handle *trans; |
1445 | ||
1446 | root = root->fs_info->chunk_root; | |
1447 | ||
1448 | path = btrfs_alloc_path(); | |
1449 | if (!path) | |
1450 | return -ENOMEM; | |
1451 | ||
a22285a6 | 1452 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1453 | if (IS_ERR(trans)) { |
1454 | btrfs_free_path(path); | |
1455 | return PTR_ERR(trans); | |
1456 | } | |
a061fc8d CM |
1457 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1458 | key.type = BTRFS_DEV_ITEM_KEY; | |
1459 | key.offset = device->devid; | |
7d9eb12c | 1460 | lock_chunks(root); |
a061fc8d CM |
1461 | |
1462 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1463 | if (ret < 0) | |
1464 | goto out; | |
1465 | ||
1466 | if (ret > 0) { | |
1467 | ret = -ENOENT; | |
1468 | goto out; | |
1469 | } | |
1470 | ||
1471 | ret = btrfs_del_item(trans, root, path); | |
1472 | if (ret) | |
1473 | goto out; | |
a061fc8d CM |
1474 | out: |
1475 | btrfs_free_path(path); | |
7d9eb12c | 1476 | unlock_chunks(root); |
a061fc8d CM |
1477 | btrfs_commit_transaction(trans, root); |
1478 | return ret; | |
1479 | } | |
1480 | ||
1481 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1482 | { | |
1483 | struct btrfs_device *device; | |
2b82032c | 1484 | struct btrfs_device *next_device; |
a061fc8d | 1485 | struct block_device *bdev; |
dfe25020 | 1486 | struct buffer_head *bh = NULL; |
a061fc8d | 1487 | struct btrfs_super_block *disk_super; |
1f78160c | 1488 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1489 | u64 all_avail; |
1490 | u64 devid; | |
2b82032c YZ |
1491 | u64 num_devices; |
1492 | u8 *dev_uuid; | |
de98ced9 | 1493 | unsigned seq; |
a061fc8d | 1494 | int ret = 0; |
1f78160c | 1495 | bool clear_super = false; |
a061fc8d | 1496 | |
a061fc8d CM |
1497 | mutex_lock(&uuid_mutex); |
1498 | ||
de98ced9 MX |
1499 | do { |
1500 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
1501 | ||
1502 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1503 | root->fs_info->avail_system_alloc_bits | | |
1504 | root->fs_info->avail_metadata_alloc_bits; | |
1505 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
a061fc8d | 1506 | |
8dabb742 SB |
1507 | num_devices = root->fs_info->fs_devices->num_devices; |
1508 | btrfs_dev_replace_lock(&root->fs_info->dev_replace); | |
1509 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1510 | WARN_ON(num_devices < 1); | |
1511 | num_devices--; | |
1512 | } | |
1513 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace); | |
1514 | ||
1515 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) { | |
183860f6 | 1516 | ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET; |
a061fc8d CM |
1517 | goto out; |
1518 | } | |
1519 | ||
8dabb742 | 1520 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) { |
183860f6 | 1521 | ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET; |
a061fc8d CM |
1522 | goto out; |
1523 | } | |
1524 | ||
53b381b3 DW |
1525 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && |
1526 | root->fs_info->fs_devices->rw_devices <= 2) { | |
183860f6 | 1527 | ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET; |
53b381b3 DW |
1528 | goto out; |
1529 | } | |
1530 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1531 | root->fs_info->fs_devices->rw_devices <= 3) { | |
183860f6 | 1532 | ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET; |
53b381b3 DW |
1533 | goto out; |
1534 | } | |
1535 | ||
dfe25020 | 1536 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1537 | struct list_head *devices; |
1538 | struct btrfs_device *tmp; | |
a061fc8d | 1539 | |
dfe25020 CM |
1540 | device = NULL; |
1541 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1542 | /* |
1543 | * It is safe to read the devices since the volume_mutex | |
1544 | * is held. | |
1545 | */ | |
c6e30871 | 1546 | list_for_each_entry(tmp, devices, dev_list) { |
63a212ab SB |
1547 | if (tmp->in_fs_metadata && |
1548 | !tmp->is_tgtdev_for_dev_replace && | |
1549 | !tmp->bdev) { | |
dfe25020 CM |
1550 | device = tmp; |
1551 | break; | |
1552 | } | |
1553 | } | |
1554 | bdev = NULL; | |
1555 | bh = NULL; | |
1556 | disk_super = NULL; | |
1557 | if (!device) { | |
183860f6 | 1558 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
dfe25020 CM |
1559 | goto out; |
1560 | } | |
dfe25020 | 1561 | } else { |
beaf8ab3 | 1562 | ret = btrfs_get_bdev_and_sb(device_path, |
cc975eb4 | 1563 | FMODE_WRITE | FMODE_EXCL, |
beaf8ab3 SB |
1564 | root->fs_info->bdev_holder, 0, |
1565 | &bdev, &bh); | |
1566 | if (ret) | |
dfe25020 | 1567 | goto out; |
dfe25020 | 1568 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1569 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c | 1570 | dev_uuid = disk_super->dev_item.uuid; |
aa1b8cd4 | 1571 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2b82032c | 1572 | disk_super->fsid); |
dfe25020 CM |
1573 | if (!device) { |
1574 | ret = -ENOENT; | |
1575 | goto error_brelse; | |
1576 | } | |
2b82032c | 1577 | } |
dfe25020 | 1578 | |
63a212ab | 1579 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1580 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
63a212ab SB |
1581 | goto error_brelse; |
1582 | } | |
1583 | ||
2b82032c | 1584 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1585 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
2b82032c YZ |
1586 | goto error_brelse; |
1587 | } | |
1588 | ||
1589 | if (device->writeable) { | |
0c1daee0 | 1590 | lock_chunks(root); |
2b82032c | 1591 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1592 | unlock_chunks(root); |
2b82032c | 1593 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1594 | clear_super = true; |
dfe25020 | 1595 | } |
a061fc8d | 1596 | |
d7901554 | 1597 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1598 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1599 | mutex_lock(&uuid_mutex); |
a061fc8d | 1600 | if (ret) |
9b3517e9 | 1601 | goto error_undo; |
a061fc8d | 1602 | |
63a212ab SB |
1603 | /* |
1604 | * TODO: the superblock still includes this device in its num_devices | |
1605 | * counter although write_all_supers() is not locked out. This | |
1606 | * could give a filesystem state which requires a degraded mount. | |
1607 | */ | |
a061fc8d CM |
1608 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1609 | if (ret) | |
9b3517e9 | 1610 | goto error_undo; |
a061fc8d | 1611 | |
2bf64758 JB |
1612 | spin_lock(&root->fs_info->free_chunk_lock); |
1613 | root->fs_info->free_chunk_space = device->total_bytes - | |
1614 | device->bytes_used; | |
1615 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1616 | ||
2b82032c | 1617 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1618 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1619 | |
1620 | /* | |
1621 | * the device list mutex makes sure that we don't change | |
1622 | * the device list while someone else is writing out all | |
1623 | * the device supers. | |
1624 | */ | |
1f78160c XG |
1625 | |
1626 | cur_devices = device->fs_devices; | |
e5e9a520 | 1627 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1628 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1629 | |
e4404d6e | 1630 | device->fs_devices->num_devices--; |
02db0844 | 1631 | device->fs_devices->total_devices--; |
2b82032c | 1632 | |
cd02dca5 CM |
1633 | if (device->missing) |
1634 | root->fs_info->fs_devices->missing_devices--; | |
1635 | ||
2b82032c YZ |
1636 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1637 | struct btrfs_device, dev_list); | |
1638 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1639 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1640 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1641 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1642 | ||
1f78160c | 1643 | if (device->bdev) |
e4404d6e | 1644 | device->fs_devices->open_devices--; |
1f78160c XG |
1645 | |
1646 | call_rcu(&device->rcu, free_device); | |
1647 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1648 | |
6c41761f DS |
1649 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1650 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1651 | |
1f78160c | 1652 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1653 | struct btrfs_fs_devices *fs_devices; |
1654 | fs_devices = root->fs_info->fs_devices; | |
1655 | while (fs_devices) { | |
1f78160c | 1656 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1657 | break; |
1658 | fs_devices = fs_devices->seed; | |
2b82032c | 1659 | } |
1f78160c XG |
1660 | fs_devices->seed = cur_devices->seed; |
1661 | cur_devices->seed = NULL; | |
0c1daee0 | 1662 | lock_chunks(root); |
1f78160c | 1663 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1664 | unlock_chunks(root); |
1f78160c | 1665 | free_fs_devices(cur_devices); |
2b82032c YZ |
1666 | } |
1667 | ||
5af3e8cc SB |
1668 | root->fs_info->num_tolerated_disk_barrier_failures = |
1669 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1670 | ||
2b82032c YZ |
1671 | /* |
1672 | * at this point, the device is zero sized. We want to | |
1673 | * remove it from the devices list and zero out the old super | |
1674 | */ | |
aa1b8cd4 | 1675 | if (clear_super && disk_super) { |
dfe25020 CM |
1676 | /* make sure this device isn't detected as part of |
1677 | * the FS anymore | |
1678 | */ | |
1679 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1680 | set_buffer_dirty(bh); | |
1681 | sync_dirty_buffer(bh); | |
dfe25020 | 1682 | } |
a061fc8d | 1683 | |
a061fc8d | 1684 | ret = 0; |
a061fc8d | 1685 | |
b8b8ff59 | 1686 | /* Notify udev that device has changed */ |
3c911608 ES |
1687 | if (bdev) |
1688 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
b8b8ff59 | 1689 | |
a061fc8d CM |
1690 | error_brelse: |
1691 | brelse(bh); | |
dfe25020 | 1692 | if (bdev) |
e525fd89 | 1693 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1694 | out: |
1695 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1696 | return ret; |
9b3517e9 ID |
1697 | error_undo: |
1698 | if (device->writeable) { | |
0c1daee0 | 1699 | lock_chunks(root); |
9b3517e9 ID |
1700 | list_add(&device->dev_alloc_list, |
1701 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1702 | unlock_chunks(root); |
9b3517e9 ID |
1703 | root->fs_info->fs_devices->rw_devices++; |
1704 | } | |
1705 | goto error_brelse; | |
a061fc8d CM |
1706 | } |
1707 | ||
e93c89c1 SB |
1708 | void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info, |
1709 | struct btrfs_device *srcdev) | |
1710 | { | |
1711 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); | |
1712 | list_del_rcu(&srcdev->dev_list); | |
1713 | list_del_rcu(&srcdev->dev_alloc_list); | |
1714 | fs_info->fs_devices->num_devices--; | |
1715 | if (srcdev->missing) { | |
1716 | fs_info->fs_devices->missing_devices--; | |
1717 | fs_info->fs_devices->rw_devices++; | |
1718 | } | |
1719 | if (srcdev->can_discard) | |
1720 | fs_info->fs_devices->num_can_discard--; | |
1721 | if (srcdev->bdev) | |
1722 | fs_info->fs_devices->open_devices--; | |
1723 | ||
1724 | call_rcu(&srcdev->rcu, free_device); | |
1725 | } | |
1726 | ||
1727 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
1728 | struct btrfs_device *tgtdev) | |
1729 | { | |
1730 | struct btrfs_device *next_device; | |
1731 | ||
1732 | WARN_ON(!tgtdev); | |
1733 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
1734 | if (tgtdev->bdev) { | |
1735 | btrfs_scratch_superblock(tgtdev); | |
1736 | fs_info->fs_devices->open_devices--; | |
1737 | } | |
1738 | fs_info->fs_devices->num_devices--; | |
1739 | if (tgtdev->can_discard) | |
1740 | fs_info->fs_devices->num_can_discard++; | |
1741 | ||
1742 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
1743 | struct btrfs_device, dev_list); | |
1744 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
1745 | fs_info->sb->s_bdev = next_device->bdev; | |
1746 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
1747 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1748 | list_del_rcu(&tgtdev->dev_list); | |
1749 | ||
1750 | call_rcu(&tgtdev->rcu, free_device); | |
1751 | ||
1752 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
1753 | } | |
1754 | ||
48a3b636 ES |
1755 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
1756 | struct btrfs_device **device) | |
7ba15b7d SB |
1757 | { |
1758 | int ret = 0; | |
1759 | struct btrfs_super_block *disk_super; | |
1760 | u64 devid; | |
1761 | u8 *dev_uuid; | |
1762 | struct block_device *bdev; | |
1763 | struct buffer_head *bh; | |
1764 | ||
1765 | *device = NULL; | |
1766 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
1767 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
1768 | if (ret) | |
1769 | return ret; | |
1770 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1771 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
1772 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 1773 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
1774 | disk_super->fsid); |
1775 | brelse(bh); | |
1776 | if (!*device) | |
1777 | ret = -ENOENT; | |
1778 | blkdev_put(bdev, FMODE_READ); | |
1779 | return ret; | |
1780 | } | |
1781 | ||
1782 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
1783 | char *device_path, | |
1784 | struct btrfs_device **device) | |
1785 | { | |
1786 | *device = NULL; | |
1787 | if (strcmp(device_path, "missing") == 0) { | |
1788 | struct list_head *devices; | |
1789 | struct btrfs_device *tmp; | |
1790 | ||
1791 | devices = &root->fs_info->fs_devices->devices; | |
1792 | /* | |
1793 | * It is safe to read the devices since the volume_mutex | |
1794 | * is held by the caller. | |
1795 | */ | |
1796 | list_for_each_entry(tmp, devices, dev_list) { | |
1797 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
1798 | *device = tmp; | |
1799 | break; | |
1800 | } | |
1801 | } | |
1802 | ||
1803 | if (!*device) { | |
1804 | pr_err("btrfs: no missing device found\n"); | |
1805 | return -ENOENT; | |
1806 | } | |
1807 | ||
1808 | return 0; | |
1809 | } else { | |
1810 | return btrfs_find_device_by_path(root, device_path, device); | |
1811 | } | |
1812 | } | |
1813 | ||
2b82032c YZ |
1814 | /* |
1815 | * does all the dirty work required for changing file system's UUID. | |
1816 | */ | |
125ccb0a | 1817 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1818 | { |
1819 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1820 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1821 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1822 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1823 | struct btrfs_device *device; |
1824 | u64 super_flags; | |
1825 | ||
1826 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1827 | if (!fs_devices->seeding) |
2b82032c YZ |
1828 | return -EINVAL; |
1829 | ||
2208a378 ID |
1830 | seed_devices = __alloc_fs_devices(); |
1831 | if (IS_ERR(seed_devices)) | |
1832 | return PTR_ERR(seed_devices); | |
2b82032c | 1833 | |
e4404d6e YZ |
1834 | old_devices = clone_fs_devices(fs_devices); |
1835 | if (IS_ERR(old_devices)) { | |
1836 | kfree(seed_devices); | |
1837 | return PTR_ERR(old_devices); | |
2b82032c | 1838 | } |
e4404d6e | 1839 | |
2b82032c YZ |
1840 | list_add(&old_devices->list, &fs_uuids); |
1841 | ||
e4404d6e YZ |
1842 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1843 | seed_devices->opened = 1; | |
1844 | INIT_LIST_HEAD(&seed_devices->devices); | |
1845 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1846 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1847 | |
1848 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1849 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1850 | synchronize_rcu); | |
c9513edb XG |
1851 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1852 | ||
e4404d6e YZ |
1853 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1854 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1855 | device->fs_devices = seed_devices; | |
1856 | } | |
1857 | ||
2b82032c YZ |
1858 | fs_devices->seeding = 0; |
1859 | fs_devices->num_devices = 0; | |
1860 | fs_devices->open_devices = 0; | |
02db0844 | 1861 | fs_devices->total_devices = 0; |
e4404d6e | 1862 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1863 | |
1864 | generate_random_uuid(fs_devices->fsid); | |
1865 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1866 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1867 | super_flags = btrfs_super_flags(disk_super) & | |
1868 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1869 | btrfs_set_super_flags(disk_super, super_flags); | |
1870 | ||
1871 | return 0; | |
1872 | } | |
1873 | ||
1874 | /* | |
1875 | * strore the expected generation for seed devices in device items. | |
1876 | */ | |
1877 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1878 | struct btrfs_root *root) | |
1879 | { | |
1880 | struct btrfs_path *path; | |
1881 | struct extent_buffer *leaf; | |
1882 | struct btrfs_dev_item *dev_item; | |
1883 | struct btrfs_device *device; | |
1884 | struct btrfs_key key; | |
1885 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1886 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1887 | u64 devid; | |
1888 | int ret; | |
1889 | ||
1890 | path = btrfs_alloc_path(); | |
1891 | if (!path) | |
1892 | return -ENOMEM; | |
1893 | ||
1894 | root = root->fs_info->chunk_root; | |
1895 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1896 | key.offset = 0; | |
1897 | key.type = BTRFS_DEV_ITEM_KEY; | |
1898 | ||
1899 | while (1) { | |
1900 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1901 | if (ret < 0) | |
1902 | goto error; | |
1903 | ||
1904 | leaf = path->nodes[0]; | |
1905 | next_slot: | |
1906 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1907 | ret = btrfs_next_leaf(root, path); | |
1908 | if (ret > 0) | |
1909 | break; | |
1910 | if (ret < 0) | |
1911 | goto error; | |
1912 | leaf = path->nodes[0]; | |
1913 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1914 | btrfs_release_path(path); |
2b82032c YZ |
1915 | continue; |
1916 | } | |
1917 | ||
1918 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1919 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1920 | key.type != BTRFS_DEV_ITEM_KEY) | |
1921 | break; | |
1922 | ||
1923 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1924 | struct btrfs_dev_item); | |
1925 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 1926 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 1927 | BTRFS_UUID_SIZE); |
1473b24e | 1928 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 1929 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
1930 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
1931 | fs_uuid); | |
79787eaa | 1932 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
1933 | |
1934 | if (device->fs_devices->seeding) { | |
1935 | btrfs_set_device_generation(leaf, dev_item, | |
1936 | device->generation); | |
1937 | btrfs_mark_buffer_dirty(leaf); | |
1938 | } | |
1939 | ||
1940 | path->slots[0]++; | |
1941 | goto next_slot; | |
1942 | } | |
1943 | ret = 0; | |
1944 | error: | |
1945 | btrfs_free_path(path); | |
1946 | return ret; | |
1947 | } | |
1948 | ||
788f20eb CM |
1949 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1950 | { | |
d5e2003c | 1951 | struct request_queue *q; |
788f20eb CM |
1952 | struct btrfs_trans_handle *trans; |
1953 | struct btrfs_device *device; | |
1954 | struct block_device *bdev; | |
788f20eb | 1955 | struct list_head *devices; |
2b82032c | 1956 | struct super_block *sb = root->fs_info->sb; |
606686ee | 1957 | struct rcu_string *name; |
788f20eb | 1958 | u64 total_bytes; |
2b82032c | 1959 | int seeding_dev = 0; |
788f20eb CM |
1960 | int ret = 0; |
1961 | ||
2b82032c | 1962 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 1963 | return -EROFS; |
788f20eb | 1964 | |
a5d16333 | 1965 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1966 | root->fs_info->bdev_holder); |
7f59203a JB |
1967 | if (IS_ERR(bdev)) |
1968 | return PTR_ERR(bdev); | |
a2135011 | 1969 | |
2b82032c YZ |
1970 | if (root->fs_info->fs_devices->seeding) { |
1971 | seeding_dev = 1; | |
1972 | down_write(&sb->s_umount); | |
1973 | mutex_lock(&uuid_mutex); | |
1974 | } | |
1975 | ||
8c8bee1d | 1976 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 1977 | |
788f20eb | 1978 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
1979 | |
1980 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 1981 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1982 | if (device->bdev == bdev) { |
1983 | ret = -EEXIST; | |
d25628bd LB |
1984 | mutex_unlock( |
1985 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 1986 | goto error; |
788f20eb CM |
1987 | } |
1988 | } | |
d25628bd | 1989 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1990 | |
12bd2fc0 ID |
1991 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
1992 | if (IS_ERR(device)) { | |
788f20eb | 1993 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 1994 | ret = PTR_ERR(device); |
2b82032c | 1995 | goto error; |
788f20eb CM |
1996 | } |
1997 | ||
606686ee JB |
1998 | name = rcu_string_strdup(device_path, GFP_NOFS); |
1999 | if (!name) { | |
788f20eb | 2000 | kfree(device); |
2b82032c YZ |
2001 | ret = -ENOMEM; |
2002 | goto error; | |
788f20eb | 2003 | } |
606686ee | 2004 | rcu_assign_pointer(device->name, name); |
2b82032c | 2005 | |
a22285a6 | 2006 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2007 | if (IS_ERR(trans)) { |
606686ee | 2008 | rcu_string_free(device->name); |
98d5dc13 TI |
2009 | kfree(device); |
2010 | ret = PTR_ERR(trans); | |
2011 | goto error; | |
2012 | } | |
2013 | ||
2b82032c YZ |
2014 | lock_chunks(root); |
2015 | ||
d5e2003c JB |
2016 | q = bdev_get_queue(bdev); |
2017 | if (blk_queue_discard(q)) | |
2018 | device->can_discard = 1; | |
2b82032c | 2019 | device->writeable = 1; |
2b82032c | 2020 | device->generation = trans->transid; |
788f20eb CM |
2021 | device->io_width = root->sectorsize; |
2022 | device->io_align = root->sectorsize; | |
2023 | device->sector_size = root->sectorsize; | |
2024 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2025 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
2026 | device->dev_root = root->fs_info->dev_root; |
2027 | device->bdev = bdev; | |
dfe25020 | 2028 | device->in_fs_metadata = 1; |
63a212ab | 2029 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2030 | device->mode = FMODE_EXCL; |
2b82032c | 2031 | set_blocksize(device->bdev, 4096); |
788f20eb | 2032 | |
2b82032c YZ |
2033 | if (seeding_dev) { |
2034 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2035 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2036 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2037 | } |
788f20eb | 2038 | |
2b82032c | 2039 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2040 | |
e5e9a520 | 2041 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 2042 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2043 | list_add(&device->dev_alloc_list, |
2044 | &root->fs_info->fs_devices->alloc_list); | |
2045 | root->fs_info->fs_devices->num_devices++; | |
2046 | root->fs_info->fs_devices->open_devices++; | |
2047 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2048 | root->fs_info->fs_devices->total_devices++; |
d5e2003c JB |
2049 | if (device->can_discard) |
2050 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 2051 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2052 | |
2bf64758 JB |
2053 | spin_lock(&root->fs_info->free_chunk_lock); |
2054 | root->fs_info->free_chunk_space += device->total_bytes; | |
2055 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2056 | ||
c289811c CM |
2057 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2058 | root->fs_info->fs_devices->rotating = 1; | |
2059 | ||
6c41761f DS |
2060 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
2061 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
2062 | total_bytes + device->total_bytes); |
2063 | ||
6c41761f DS |
2064 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
2065 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 2066 | total_bytes + 1); |
e5e9a520 | 2067 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2068 | |
2b82032c YZ |
2069 | if (seeding_dev) { |
2070 | ret = init_first_rw_device(trans, root, device); | |
005d6427 DS |
2071 | if (ret) { |
2072 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2073 | goto error_trans; |
005d6427 | 2074 | } |
2b82032c | 2075 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2076 | if (ret) { |
2077 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2078 | goto error_trans; |
005d6427 | 2079 | } |
2b82032c YZ |
2080 | } else { |
2081 | ret = btrfs_add_device(trans, root, device); | |
005d6427 DS |
2082 | if (ret) { |
2083 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2084 | goto error_trans; |
005d6427 | 2085 | } |
2b82032c YZ |
2086 | } |
2087 | ||
913d952e CM |
2088 | /* |
2089 | * we've got more storage, clear any full flags on the space | |
2090 | * infos | |
2091 | */ | |
2092 | btrfs_clear_space_info_full(root->fs_info); | |
2093 | ||
7d9eb12c | 2094 | unlock_chunks(root); |
5af3e8cc SB |
2095 | root->fs_info->num_tolerated_disk_barrier_failures = |
2096 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2097 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2098 | |
2b82032c YZ |
2099 | if (seeding_dev) { |
2100 | mutex_unlock(&uuid_mutex); | |
2101 | up_write(&sb->s_umount); | |
788f20eb | 2102 | |
79787eaa JM |
2103 | if (ret) /* transaction commit */ |
2104 | return ret; | |
2105 | ||
2b82032c | 2106 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
2107 | if (ret < 0) |
2108 | btrfs_error(root->fs_info, ret, | |
2109 | "Failed to relocate sys chunks after " | |
2110 | "device initialization. This can be fixed " | |
2111 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2112 | trans = btrfs_attach_transaction(root); |
2113 | if (IS_ERR(trans)) { | |
2114 | if (PTR_ERR(trans) == -ENOENT) | |
2115 | return 0; | |
2116 | return PTR_ERR(trans); | |
2117 | } | |
2118 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2119 | } |
c9e9f97b | 2120 | |
2b82032c | 2121 | return ret; |
79787eaa JM |
2122 | |
2123 | error_trans: | |
2124 | unlock_chunks(root); | |
79787eaa | 2125 | btrfs_end_transaction(trans, root); |
606686ee | 2126 | rcu_string_free(device->name); |
79787eaa | 2127 | kfree(device); |
2b82032c | 2128 | error: |
e525fd89 | 2129 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2130 | if (seeding_dev) { |
2131 | mutex_unlock(&uuid_mutex); | |
2132 | up_write(&sb->s_umount); | |
2133 | } | |
c9e9f97b | 2134 | return ret; |
788f20eb CM |
2135 | } |
2136 | ||
e93c89c1 SB |
2137 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
2138 | struct btrfs_device **device_out) | |
2139 | { | |
2140 | struct request_queue *q; | |
2141 | struct btrfs_device *device; | |
2142 | struct block_device *bdev; | |
2143 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2144 | struct list_head *devices; | |
2145 | struct rcu_string *name; | |
12bd2fc0 | 2146 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2147 | int ret = 0; |
2148 | ||
2149 | *device_out = NULL; | |
2150 | if (fs_info->fs_devices->seeding) | |
2151 | return -EINVAL; | |
2152 | ||
2153 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2154 | fs_info->bdev_holder); | |
2155 | if (IS_ERR(bdev)) | |
2156 | return PTR_ERR(bdev); | |
2157 | ||
2158 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2159 | ||
2160 | devices = &fs_info->fs_devices->devices; | |
2161 | list_for_each_entry(device, devices, dev_list) { | |
2162 | if (device->bdev == bdev) { | |
2163 | ret = -EEXIST; | |
2164 | goto error; | |
2165 | } | |
2166 | } | |
2167 | ||
12bd2fc0 ID |
2168 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2169 | if (IS_ERR(device)) { | |
2170 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2171 | goto error; |
2172 | } | |
2173 | ||
2174 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2175 | if (!name) { | |
2176 | kfree(device); | |
2177 | ret = -ENOMEM; | |
2178 | goto error; | |
2179 | } | |
2180 | rcu_assign_pointer(device->name, name); | |
2181 | ||
2182 | q = bdev_get_queue(bdev); | |
2183 | if (blk_queue_discard(q)) | |
2184 | device->can_discard = 1; | |
2185 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2186 | device->writeable = 1; | |
e93c89c1 SB |
2187 | device->generation = 0; |
2188 | device->io_width = root->sectorsize; | |
2189 | device->io_align = root->sectorsize; | |
2190 | device->sector_size = root->sectorsize; | |
2191 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2192 | device->disk_total_bytes = device->total_bytes; | |
2193 | device->dev_root = fs_info->dev_root; | |
2194 | device->bdev = bdev; | |
2195 | device->in_fs_metadata = 1; | |
2196 | device->is_tgtdev_for_dev_replace = 1; | |
2197 | device->mode = FMODE_EXCL; | |
2198 | set_blocksize(device->bdev, 4096); | |
2199 | device->fs_devices = fs_info->fs_devices; | |
2200 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2201 | fs_info->fs_devices->num_devices++; | |
2202 | fs_info->fs_devices->open_devices++; | |
2203 | if (device->can_discard) | |
2204 | fs_info->fs_devices->num_can_discard++; | |
2205 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2206 | ||
2207 | *device_out = device; | |
2208 | return ret; | |
2209 | ||
2210 | error: | |
2211 | blkdev_put(bdev, FMODE_EXCL); | |
2212 | return ret; | |
2213 | } | |
2214 | ||
2215 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2216 | struct btrfs_device *tgtdev) | |
2217 | { | |
2218 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2219 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2220 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2221 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2222 | tgtdev->dev_root = fs_info->dev_root; | |
2223 | tgtdev->in_fs_metadata = 1; | |
2224 | } | |
2225 | ||
d397712b CM |
2226 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2227 | struct btrfs_device *device) | |
0b86a832 CM |
2228 | { |
2229 | int ret; | |
2230 | struct btrfs_path *path; | |
2231 | struct btrfs_root *root; | |
2232 | struct btrfs_dev_item *dev_item; | |
2233 | struct extent_buffer *leaf; | |
2234 | struct btrfs_key key; | |
2235 | ||
2236 | root = device->dev_root->fs_info->chunk_root; | |
2237 | ||
2238 | path = btrfs_alloc_path(); | |
2239 | if (!path) | |
2240 | return -ENOMEM; | |
2241 | ||
2242 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2243 | key.type = BTRFS_DEV_ITEM_KEY; | |
2244 | key.offset = device->devid; | |
2245 | ||
2246 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2247 | if (ret < 0) | |
2248 | goto out; | |
2249 | ||
2250 | if (ret > 0) { | |
2251 | ret = -ENOENT; | |
2252 | goto out; | |
2253 | } | |
2254 | ||
2255 | leaf = path->nodes[0]; | |
2256 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2257 | ||
2258 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2259 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2260 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2261 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2262 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 2263 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
2264 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
2265 | btrfs_mark_buffer_dirty(leaf); | |
2266 | ||
2267 | out: | |
2268 | btrfs_free_path(path); | |
2269 | return ret; | |
2270 | } | |
2271 | ||
7d9eb12c | 2272 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2273 | struct btrfs_device *device, u64 new_size) |
2274 | { | |
2275 | struct btrfs_super_block *super_copy = | |
6c41761f | 2276 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
2277 | u64 old_total = btrfs_super_total_bytes(super_copy); |
2278 | u64 diff = new_size - device->total_bytes; | |
2279 | ||
2b82032c YZ |
2280 | if (!device->writeable) |
2281 | return -EACCES; | |
63a212ab SB |
2282 | if (new_size <= device->total_bytes || |
2283 | device->is_tgtdev_for_dev_replace) | |
2b82032c YZ |
2284 | return -EINVAL; |
2285 | ||
8f18cf13 | 2286 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2287 | device->fs_devices->total_rw_bytes += diff; |
2288 | ||
2289 | device->total_bytes = new_size; | |
9779b72f | 2290 | device->disk_total_bytes = new_size; |
4184ea7f CM |
2291 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
2292 | ||
8f18cf13 CM |
2293 | return btrfs_update_device(trans, device); |
2294 | } | |
2295 | ||
7d9eb12c CM |
2296 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
2297 | struct btrfs_device *device, u64 new_size) | |
2298 | { | |
2299 | int ret; | |
2300 | lock_chunks(device->dev_root); | |
2301 | ret = __btrfs_grow_device(trans, device, new_size); | |
2302 | unlock_chunks(device->dev_root); | |
2303 | return ret; | |
2304 | } | |
2305 | ||
8f18cf13 CM |
2306 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
2307 | struct btrfs_root *root, | |
2308 | u64 chunk_tree, u64 chunk_objectid, | |
2309 | u64 chunk_offset) | |
2310 | { | |
2311 | int ret; | |
2312 | struct btrfs_path *path; | |
2313 | struct btrfs_key key; | |
2314 | ||
2315 | root = root->fs_info->chunk_root; | |
2316 | path = btrfs_alloc_path(); | |
2317 | if (!path) | |
2318 | return -ENOMEM; | |
2319 | ||
2320 | key.objectid = chunk_objectid; | |
2321 | key.offset = chunk_offset; | |
2322 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2323 | ||
2324 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2325 | if (ret < 0) |
2326 | goto out; | |
2327 | else if (ret > 0) { /* Logic error or corruption */ | |
2328 | btrfs_error(root->fs_info, -ENOENT, | |
2329 | "Failed lookup while freeing chunk."); | |
2330 | ret = -ENOENT; | |
2331 | goto out; | |
2332 | } | |
8f18cf13 CM |
2333 | |
2334 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
2335 | if (ret < 0) |
2336 | btrfs_error(root->fs_info, ret, | |
2337 | "Failed to delete chunk item."); | |
2338 | out: | |
8f18cf13 | 2339 | btrfs_free_path(path); |
65a246c5 | 2340 | return ret; |
8f18cf13 CM |
2341 | } |
2342 | ||
b2950863 | 2343 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2344 | chunk_offset) |
2345 | { | |
6c41761f | 2346 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2347 | struct btrfs_disk_key *disk_key; |
2348 | struct btrfs_chunk *chunk; | |
2349 | u8 *ptr; | |
2350 | int ret = 0; | |
2351 | u32 num_stripes; | |
2352 | u32 array_size; | |
2353 | u32 len = 0; | |
2354 | u32 cur; | |
2355 | struct btrfs_key key; | |
2356 | ||
2357 | array_size = btrfs_super_sys_array_size(super_copy); | |
2358 | ||
2359 | ptr = super_copy->sys_chunk_array; | |
2360 | cur = 0; | |
2361 | ||
2362 | while (cur < array_size) { | |
2363 | disk_key = (struct btrfs_disk_key *)ptr; | |
2364 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2365 | ||
2366 | len = sizeof(*disk_key); | |
2367 | ||
2368 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2369 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2370 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2371 | len += btrfs_chunk_item_size(num_stripes); | |
2372 | } else { | |
2373 | ret = -EIO; | |
2374 | break; | |
2375 | } | |
2376 | if (key.objectid == chunk_objectid && | |
2377 | key.offset == chunk_offset) { | |
2378 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2379 | array_size -= len; | |
2380 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2381 | } else { | |
2382 | ptr += len; | |
2383 | cur += len; | |
2384 | } | |
2385 | } | |
2386 | return ret; | |
2387 | } | |
2388 | ||
b2950863 | 2389 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
2390 | u64 chunk_tree, u64 chunk_objectid, |
2391 | u64 chunk_offset) | |
2392 | { | |
2393 | struct extent_map_tree *em_tree; | |
2394 | struct btrfs_root *extent_root; | |
2395 | struct btrfs_trans_handle *trans; | |
2396 | struct extent_map *em; | |
2397 | struct map_lookup *map; | |
2398 | int ret; | |
2399 | int i; | |
2400 | ||
2401 | root = root->fs_info->chunk_root; | |
2402 | extent_root = root->fs_info->extent_root; | |
2403 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
2404 | ||
ba1bf481 JB |
2405 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2406 | if (ret) | |
2407 | return -ENOSPC; | |
2408 | ||
8f18cf13 | 2409 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 2410 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
2411 | if (ret) |
2412 | return ret; | |
8f18cf13 | 2413 | |
a22285a6 | 2414 | trans = btrfs_start_transaction(root, 0); |
0f788c58 LB |
2415 | if (IS_ERR(trans)) { |
2416 | ret = PTR_ERR(trans); | |
2417 | btrfs_std_error(root->fs_info, ret); | |
2418 | return ret; | |
2419 | } | |
8f18cf13 | 2420 | |
7d9eb12c CM |
2421 | lock_chunks(root); |
2422 | ||
8f18cf13 CM |
2423 | /* |
2424 | * step two, delete the device extents and the | |
2425 | * chunk tree entries | |
2426 | */ | |
890871be | 2427 | read_lock(&em_tree->lock); |
8f18cf13 | 2428 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2429 | read_unlock(&em_tree->lock); |
8f18cf13 | 2430 | |
285190d9 | 2431 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 2432 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
2433 | map = (struct map_lookup *)em->bdev; |
2434 | ||
2435 | for (i = 0; i < map->num_stripes; i++) { | |
2436 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
2437 | map->stripes[i].physical); | |
2438 | BUG_ON(ret); | |
a061fc8d | 2439 | |
dfe25020 CM |
2440 | if (map->stripes[i].dev) { |
2441 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2442 | BUG_ON(ret); | |
2443 | } | |
8f18cf13 CM |
2444 | } |
2445 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
2446 | chunk_offset); | |
2447 | ||
2448 | BUG_ON(ret); | |
2449 | ||
1abe9b8a | 2450 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2451 | ||
8f18cf13 CM |
2452 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2453 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
2454 | BUG_ON(ret); | |
8f18cf13 CM |
2455 | } |
2456 | ||
2b82032c YZ |
2457 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
2458 | BUG_ON(ret); | |
2459 | ||
890871be | 2460 | write_lock(&em_tree->lock); |
2b82032c | 2461 | remove_extent_mapping(em_tree, em); |
890871be | 2462 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2463 | |
2464 | kfree(map); | |
2465 | em->bdev = NULL; | |
2466 | ||
2467 | /* once for the tree */ | |
2468 | free_extent_map(em); | |
2469 | /* once for us */ | |
2470 | free_extent_map(em); | |
2471 | ||
2472 | unlock_chunks(root); | |
2473 | btrfs_end_transaction(trans, root); | |
2474 | return 0; | |
2475 | } | |
2476 | ||
2477 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2478 | { | |
2479 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2480 | struct btrfs_path *path; | |
2481 | struct extent_buffer *leaf; | |
2482 | struct btrfs_chunk *chunk; | |
2483 | struct btrfs_key key; | |
2484 | struct btrfs_key found_key; | |
2485 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2486 | u64 chunk_type; | |
ba1bf481 JB |
2487 | bool retried = false; |
2488 | int failed = 0; | |
2b82032c YZ |
2489 | int ret; |
2490 | ||
2491 | path = btrfs_alloc_path(); | |
2492 | if (!path) | |
2493 | return -ENOMEM; | |
2494 | ||
ba1bf481 | 2495 | again: |
2b82032c YZ |
2496 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2497 | key.offset = (u64)-1; | |
2498 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2499 | ||
2500 | while (1) { | |
2501 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2502 | if (ret < 0) | |
2503 | goto error; | |
79787eaa | 2504 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2505 | |
2506 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2507 | key.type); | |
2508 | if (ret < 0) | |
2509 | goto error; | |
2510 | if (ret > 0) | |
2511 | break; | |
1a40e23b | 2512 | |
2b82032c YZ |
2513 | leaf = path->nodes[0]; |
2514 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2515 | |
2b82032c YZ |
2516 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2517 | struct btrfs_chunk); | |
2518 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2519 | btrfs_release_path(path); |
8f18cf13 | 2520 | |
2b82032c YZ |
2521 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2522 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2523 | found_key.objectid, | |
2524 | found_key.offset); | |
ba1bf481 JB |
2525 | if (ret == -ENOSPC) |
2526 | failed++; | |
2527 | else if (ret) | |
2528 | BUG(); | |
2b82032c | 2529 | } |
8f18cf13 | 2530 | |
2b82032c YZ |
2531 | if (found_key.offset == 0) |
2532 | break; | |
2533 | key.offset = found_key.offset - 1; | |
2534 | } | |
2535 | ret = 0; | |
ba1bf481 JB |
2536 | if (failed && !retried) { |
2537 | failed = 0; | |
2538 | retried = true; | |
2539 | goto again; | |
2540 | } else if (failed && retried) { | |
2541 | WARN_ON(1); | |
2542 | ret = -ENOSPC; | |
2543 | } | |
2b82032c YZ |
2544 | error: |
2545 | btrfs_free_path(path); | |
2546 | return ret; | |
8f18cf13 CM |
2547 | } |
2548 | ||
0940ebf6 ID |
2549 | static int insert_balance_item(struct btrfs_root *root, |
2550 | struct btrfs_balance_control *bctl) | |
2551 | { | |
2552 | struct btrfs_trans_handle *trans; | |
2553 | struct btrfs_balance_item *item; | |
2554 | struct btrfs_disk_balance_args disk_bargs; | |
2555 | struct btrfs_path *path; | |
2556 | struct extent_buffer *leaf; | |
2557 | struct btrfs_key key; | |
2558 | int ret, err; | |
2559 | ||
2560 | path = btrfs_alloc_path(); | |
2561 | if (!path) | |
2562 | return -ENOMEM; | |
2563 | ||
2564 | trans = btrfs_start_transaction(root, 0); | |
2565 | if (IS_ERR(trans)) { | |
2566 | btrfs_free_path(path); | |
2567 | return PTR_ERR(trans); | |
2568 | } | |
2569 | ||
2570 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2571 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2572 | key.offset = 0; | |
2573 | ||
2574 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2575 | sizeof(*item)); | |
2576 | if (ret) | |
2577 | goto out; | |
2578 | ||
2579 | leaf = path->nodes[0]; | |
2580 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2581 | ||
2582 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2583 | ||
2584 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2585 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2586 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2587 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2588 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2589 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2590 | ||
2591 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2592 | ||
2593 | btrfs_mark_buffer_dirty(leaf); | |
2594 | out: | |
2595 | btrfs_free_path(path); | |
2596 | err = btrfs_commit_transaction(trans, root); | |
2597 | if (err && !ret) | |
2598 | ret = err; | |
2599 | return ret; | |
2600 | } | |
2601 | ||
2602 | static int del_balance_item(struct btrfs_root *root) | |
2603 | { | |
2604 | struct btrfs_trans_handle *trans; | |
2605 | struct btrfs_path *path; | |
2606 | struct btrfs_key key; | |
2607 | int ret, err; | |
2608 | ||
2609 | path = btrfs_alloc_path(); | |
2610 | if (!path) | |
2611 | return -ENOMEM; | |
2612 | ||
2613 | trans = btrfs_start_transaction(root, 0); | |
2614 | if (IS_ERR(trans)) { | |
2615 | btrfs_free_path(path); | |
2616 | return PTR_ERR(trans); | |
2617 | } | |
2618 | ||
2619 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2620 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2621 | key.offset = 0; | |
2622 | ||
2623 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2624 | if (ret < 0) | |
2625 | goto out; | |
2626 | if (ret > 0) { | |
2627 | ret = -ENOENT; | |
2628 | goto out; | |
2629 | } | |
2630 | ||
2631 | ret = btrfs_del_item(trans, root, path); | |
2632 | out: | |
2633 | btrfs_free_path(path); | |
2634 | err = btrfs_commit_transaction(trans, root); | |
2635 | if (err && !ret) | |
2636 | ret = err; | |
2637 | return ret; | |
2638 | } | |
2639 | ||
59641015 ID |
2640 | /* |
2641 | * This is a heuristic used to reduce the number of chunks balanced on | |
2642 | * resume after balance was interrupted. | |
2643 | */ | |
2644 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2645 | { | |
2646 | /* | |
2647 | * Turn on soft mode for chunk types that were being converted. | |
2648 | */ | |
2649 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2650 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2651 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2652 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2653 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2654 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2655 | ||
2656 | /* | |
2657 | * Turn on usage filter if is not already used. The idea is | |
2658 | * that chunks that we have already balanced should be | |
2659 | * reasonably full. Don't do it for chunks that are being | |
2660 | * converted - that will keep us from relocating unconverted | |
2661 | * (albeit full) chunks. | |
2662 | */ | |
2663 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2664 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2665 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2666 | bctl->data.usage = 90; | |
2667 | } | |
2668 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2669 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2670 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2671 | bctl->sys.usage = 90; | |
2672 | } | |
2673 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2674 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2675 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2676 | bctl->meta.usage = 90; | |
2677 | } | |
2678 | } | |
2679 | ||
c9e9f97b ID |
2680 | /* |
2681 | * Should be called with both balance and volume mutexes held to | |
2682 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2683 | * restriper. Same goes for unset_balance_control. | |
2684 | */ | |
2685 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2686 | { | |
2687 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2688 | ||
2689 | BUG_ON(fs_info->balance_ctl); | |
2690 | ||
2691 | spin_lock(&fs_info->balance_lock); | |
2692 | fs_info->balance_ctl = bctl; | |
2693 | spin_unlock(&fs_info->balance_lock); | |
2694 | } | |
2695 | ||
2696 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2697 | { | |
2698 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2699 | ||
2700 | BUG_ON(!fs_info->balance_ctl); | |
2701 | ||
2702 | spin_lock(&fs_info->balance_lock); | |
2703 | fs_info->balance_ctl = NULL; | |
2704 | spin_unlock(&fs_info->balance_lock); | |
2705 | ||
2706 | kfree(bctl); | |
2707 | } | |
2708 | ||
ed25e9b2 ID |
2709 | /* |
2710 | * Balance filters. Return 1 if chunk should be filtered out | |
2711 | * (should not be balanced). | |
2712 | */ | |
899c81ea | 2713 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2714 | struct btrfs_balance_args *bargs) |
2715 | { | |
899c81ea ID |
2716 | chunk_type = chunk_to_extended(chunk_type) & |
2717 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2718 | |
899c81ea | 2719 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2720 | return 0; |
2721 | ||
2722 | return 1; | |
2723 | } | |
2724 | ||
5ce5b3c0 ID |
2725 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
2726 | struct btrfs_balance_args *bargs) | |
2727 | { | |
2728 | struct btrfs_block_group_cache *cache; | |
2729 | u64 chunk_used, user_thresh; | |
2730 | int ret = 1; | |
2731 | ||
2732 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2733 | chunk_used = btrfs_block_group_used(&cache->item); | |
2734 | ||
a105bb88 | 2735 | if (bargs->usage == 0) |
3e39cea6 | 2736 | user_thresh = 1; |
a105bb88 ID |
2737 | else if (bargs->usage > 100) |
2738 | user_thresh = cache->key.offset; | |
2739 | else | |
2740 | user_thresh = div_factor_fine(cache->key.offset, | |
2741 | bargs->usage); | |
2742 | ||
5ce5b3c0 ID |
2743 | if (chunk_used < user_thresh) |
2744 | ret = 0; | |
2745 | ||
2746 | btrfs_put_block_group(cache); | |
2747 | return ret; | |
2748 | } | |
2749 | ||
409d404b ID |
2750 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2751 | struct btrfs_chunk *chunk, | |
2752 | struct btrfs_balance_args *bargs) | |
2753 | { | |
2754 | struct btrfs_stripe *stripe; | |
2755 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2756 | int i; | |
2757 | ||
2758 | for (i = 0; i < num_stripes; i++) { | |
2759 | stripe = btrfs_stripe_nr(chunk, i); | |
2760 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2761 | return 0; | |
2762 | } | |
2763 | ||
2764 | return 1; | |
2765 | } | |
2766 | ||
94e60d5a ID |
2767 | /* [pstart, pend) */ |
2768 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2769 | struct btrfs_chunk *chunk, | |
2770 | u64 chunk_offset, | |
2771 | struct btrfs_balance_args *bargs) | |
2772 | { | |
2773 | struct btrfs_stripe *stripe; | |
2774 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2775 | u64 stripe_offset; | |
2776 | u64 stripe_length; | |
2777 | int factor; | |
2778 | int i; | |
2779 | ||
2780 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2781 | return 0; | |
2782 | ||
2783 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
2784 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
2785 | factor = num_stripes / 2; | |
2786 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
2787 | factor = num_stripes - 1; | |
2788 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
2789 | factor = num_stripes - 2; | |
2790 | } else { | |
2791 | factor = num_stripes; | |
2792 | } | |
94e60d5a ID |
2793 | |
2794 | for (i = 0; i < num_stripes; i++) { | |
2795 | stripe = btrfs_stripe_nr(chunk, i); | |
2796 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2797 | continue; | |
2798 | ||
2799 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2800 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2801 | do_div(stripe_length, factor); | |
2802 | ||
2803 | if (stripe_offset < bargs->pend && | |
2804 | stripe_offset + stripe_length > bargs->pstart) | |
2805 | return 0; | |
2806 | } | |
2807 | ||
2808 | return 1; | |
2809 | } | |
2810 | ||
ea67176a ID |
2811 | /* [vstart, vend) */ |
2812 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
2813 | struct btrfs_chunk *chunk, | |
2814 | u64 chunk_offset, | |
2815 | struct btrfs_balance_args *bargs) | |
2816 | { | |
2817 | if (chunk_offset < bargs->vend && | |
2818 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
2819 | /* at least part of the chunk is inside this vrange */ | |
2820 | return 0; | |
2821 | ||
2822 | return 1; | |
2823 | } | |
2824 | ||
899c81ea | 2825 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
2826 | struct btrfs_balance_args *bargs) |
2827 | { | |
2828 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
2829 | return 0; | |
2830 | ||
899c81ea ID |
2831 | chunk_type = chunk_to_extended(chunk_type) & |
2832 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 2833 | |
899c81ea | 2834 | if (bargs->target == chunk_type) |
cfa4c961 ID |
2835 | return 1; |
2836 | ||
2837 | return 0; | |
2838 | } | |
2839 | ||
f43ffb60 ID |
2840 | static int should_balance_chunk(struct btrfs_root *root, |
2841 | struct extent_buffer *leaf, | |
2842 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
2843 | { | |
2844 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
2845 | struct btrfs_balance_args *bargs = NULL; | |
2846 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
2847 | ||
2848 | /* type filter */ | |
2849 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
2850 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
2851 | return 0; | |
2852 | } | |
2853 | ||
2854 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
2855 | bargs = &bctl->data; | |
2856 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
2857 | bargs = &bctl->sys; | |
2858 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
2859 | bargs = &bctl->meta; | |
2860 | ||
ed25e9b2 ID |
2861 | /* profiles filter */ |
2862 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
2863 | chunk_profiles_filter(chunk_type, bargs)) { | |
2864 | return 0; | |
5ce5b3c0 ID |
2865 | } |
2866 | ||
2867 | /* usage filter */ | |
2868 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2869 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
2870 | return 0; | |
409d404b ID |
2871 | } |
2872 | ||
2873 | /* devid filter */ | |
2874 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
2875 | chunk_devid_filter(leaf, chunk, bargs)) { | |
2876 | return 0; | |
94e60d5a ID |
2877 | } |
2878 | ||
2879 | /* drange filter, makes sense only with devid filter */ | |
2880 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
2881 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2882 | return 0; | |
ea67176a ID |
2883 | } |
2884 | ||
2885 | /* vrange filter */ | |
2886 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
2887 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2888 | return 0; | |
ed25e9b2 ID |
2889 | } |
2890 | ||
cfa4c961 ID |
2891 | /* soft profile changing mode */ |
2892 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
2893 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
2894 | return 0; | |
2895 | } | |
2896 | ||
f43ffb60 ID |
2897 | return 1; |
2898 | } | |
2899 | ||
c9e9f97b | 2900 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 2901 | { |
19a39dce | 2902 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
2903 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2904 | struct btrfs_root *dev_root = fs_info->dev_root; | |
2905 | struct list_head *devices; | |
ec44a35c CM |
2906 | struct btrfs_device *device; |
2907 | u64 old_size; | |
2908 | u64 size_to_free; | |
f43ffb60 | 2909 | struct btrfs_chunk *chunk; |
ec44a35c CM |
2910 | struct btrfs_path *path; |
2911 | struct btrfs_key key; | |
ec44a35c | 2912 | struct btrfs_key found_key; |
c9e9f97b | 2913 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
2914 | struct extent_buffer *leaf; |
2915 | int slot; | |
c9e9f97b ID |
2916 | int ret; |
2917 | int enospc_errors = 0; | |
19a39dce | 2918 | bool counting = true; |
ec44a35c | 2919 | |
ec44a35c | 2920 | /* step one make some room on all the devices */ |
c9e9f97b | 2921 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 2922 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2923 | old_size = device->total_bytes; |
2924 | size_to_free = div_factor(old_size, 1); | |
2925 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c | 2926 | if (!device->writeable || |
63a212ab SB |
2927 | device->total_bytes - device->bytes_used > size_to_free || |
2928 | device->is_tgtdev_for_dev_replace) | |
ec44a35c CM |
2929 | continue; |
2930 | ||
2931 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2932 | if (ret == -ENOSPC) |
2933 | break; | |
ec44a35c CM |
2934 | BUG_ON(ret); |
2935 | ||
a22285a6 | 2936 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2937 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2938 | |
2939 | ret = btrfs_grow_device(trans, device, old_size); | |
2940 | BUG_ON(ret); | |
2941 | ||
2942 | btrfs_end_transaction(trans, dev_root); | |
2943 | } | |
2944 | ||
2945 | /* step two, relocate all the chunks */ | |
2946 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2947 | if (!path) { |
2948 | ret = -ENOMEM; | |
2949 | goto error; | |
2950 | } | |
19a39dce ID |
2951 | |
2952 | /* zero out stat counters */ | |
2953 | spin_lock(&fs_info->balance_lock); | |
2954 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
2955 | spin_unlock(&fs_info->balance_lock); | |
2956 | again: | |
ec44a35c CM |
2957 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2958 | key.offset = (u64)-1; | |
2959 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2960 | ||
d397712b | 2961 | while (1) { |
19a39dce | 2962 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 2963 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
2964 | ret = -ECANCELED; |
2965 | goto error; | |
2966 | } | |
2967 | ||
ec44a35c CM |
2968 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2969 | if (ret < 0) | |
2970 | goto error; | |
2971 | ||
2972 | /* | |
2973 | * this shouldn't happen, it means the last relocate | |
2974 | * failed | |
2975 | */ | |
2976 | if (ret == 0) | |
c9e9f97b | 2977 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
2978 | |
2979 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2980 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
2981 | if (ret) { |
2982 | ret = 0; | |
ec44a35c | 2983 | break; |
c9e9f97b | 2984 | } |
7d9eb12c | 2985 | |
f43ffb60 ID |
2986 | leaf = path->nodes[0]; |
2987 | slot = path->slots[0]; | |
2988 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 2989 | |
ec44a35c CM |
2990 | if (found_key.objectid != key.objectid) |
2991 | break; | |
7d9eb12c | 2992 | |
ec44a35c | 2993 | /* chunk zero is special */ |
ba1bf481 | 2994 | if (found_key.offset == 0) |
ec44a35c CM |
2995 | break; |
2996 | ||
f43ffb60 ID |
2997 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
2998 | ||
19a39dce ID |
2999 | if (!counting) { |
3000 | spin_lock(&fs_info->balance_lock); | |
3001 | bctl->stat.considered++; | |
3002 | spin_unlock(&fs_info->balance_lock); | |
3003 | } | |
3004 | ||
f43ffb60 ID |
3005 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3006 | found_key.offset); | |
b3b4aa74 | 3007 | btrfs_release_path(path); |
f43ffb60 ID |
3008 | if (!ret) |
3009 | goto loop; | |
3010 | ||
19a39dce ID |
3011 | if (counting) { |
3012 | spin_lock(&fs_info->balance_lock); | |
3013 | bctl->stat.expected++; | |
3014 | spin_unlock(&fs_info->balance_lock); | |
3015 | goto loop; | |
3016 | } | |
3017 | ||
ec44a35c CM |
3018 | ret = btrfs_relocate_chunk(chunk_root, |
3019 | chunk_root->root_key.objectid, | |
3020 | found_key.objectid, | |
3021 | found_key.offset); | |
508794eb JB |
3022 | if (ret && ret != -ENOSPC) |
3023 | goto error; | |
19a39dce | 3024 | if (ret == -ENOSPC) { |
c9e9f97b | 3025 | enospc_errors++; |
19a39dce ID |
3026 | } else { |
3027 | spin_lock(&fs_info->balance_lock); | |
3028 | bctl->stat.completed++; | |
3029 | spin_unlock(&fs_info->balance_lock); | |
3030 | } | |
f43ffb60 | 3031 | loop: |
ba1bf481 | 3032 | key.offset = found_key.offset - 1; |
ec44a35c | 3033 | } |
c9e9f97b | 3034 | |
19a39dce ID |
3035 | if (counting) { |
3036 | btrfs_release_path(path); | |
3037 | counting = false; | |
3038 | goto again; | |
3039 | } | |
ec44a35c CM |
3040 | error: |
3041 | btrfs_free_path(path); | |
c9e9f97b ID |
3042 | if (enospc_errors) { |
3043 | printk(KERN_INFO "btrfs: %d enospc errors during balance\n", | |
3044 | enospc_errors); | |
3045 | if (!ret) | |
3046 | ret = -ENOSPC; | |
3047 | } | |
3048 | ||
ec44a35c CM |
3049 | return ret; |
3050 | } | |
3051 | ||
0c460c0d ID |
3052 | /** |
3053 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3054 | * @flags: profile to validate | |
3055 | * @extended: if true @flags is treated as an extended profile | |
3056 | */ | |
3057 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3058 | { | |
3059 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3060 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3061 | ||
3062 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3063 | ||
3064 | /* 1) check that all other bits are zeroed */ | |
3065 | if (flags & ~mask) | |
3066 | return 0; | |
3067 | ||
3068 | /* 2) see if profile is reduced */ | |
3069 | if (flags == 0) | |
3070 | return !extended; /* "0" is valid for usual profiles */ | |
3071 | ||
3072 | /* true if exactly one bit set */ | |
3073 | return (flags & (flags - 1)) == 0; | |
3074 | } | |
3075 | ||
837d5b6e ID |
3076 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3077 | { | |
a7e99c69 ID |
3078 | /* cancel requested || normal exit path */ |
3079 | return atomic_read(&fs_info->balance_cancel_req) || | |
3080 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3081 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3082 | } |
3083 | ||
c9e9f97b ID |
3084 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3085 | { | |
0940ebf6 ID |
3086 | int ret; |
3087 | ||
c9e9f97b | 3088 | unset_balance_control(fs_info); |
0940ebf6 | 3089 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 LB |
3090 | if (ret) |
3091 | btrfs_std_error(fs_info, ret); | |
ed0fb78f ID |
3092 | |
3093 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3094 | } |
3095 | ||
c9e9f97b ID |
3096 | /* |
3097 | * Should be called with both balance and volume mutexes held | |
3098 | */ | |
3099 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3100 | struct btrfs_ioctl_balance_args *bargs) | |
3101 | { | |
3102 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3103 | u64 allowed; |
e4837f8f | 3104 | int mixed = 0; |
c9e9f97b | 3105 | int ret; |
8dabb742 | 3106 | u64 num_devices; |
de98ced9 | 3107 | unsigned seq; |
c9e9f97b | 3108 | |
837d5b6e | 3109 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3110 | atomic_read(&fs_info->balance_pause_req) || |
3111 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3112 | ret = -EINVAL; |
3113 | goto out; | |
3114 | } | |
3115 | ||
e4837f8f ID |
3116 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3117 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3118 | mixed = 1; | |
3119 | ||
f43ffb60 ID |
3120 | /* |
3121 | * In case of mixed groups both data and meta should be picked, | |
3122 | * and identical options should be given for both of them. | |
3123 | */ | |
e4837f8f ID |
3124 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3125 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3126 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3127 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3128 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
3129 | printk(KERN_ERR "btrfs: with mixed groups data and " | |
3130 | "metadata balance options must be the same\n"); | |
3131 | ret = -EINVAL; | |
3132 | goto out; | |
3133 | } | |
3134 | } | |
3135 | ||
8dabb742 SB |
3136 | num_devices = fs_info->fs_devices->num_devices; |
3137 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
3138 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3139 | BUG_ON(num_devices < 1); | |
3140 | num_devices--; | |
3141 | } | |
3142 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
e4d8ec0f | 3143 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3144 | if (num_devices == 1) |
e4d8ec0f | 3145 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8250dabe | 3146 | else if (num_devices > 1) |
e4d8ec0f | 3147 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3148 | if (num_devices > 2) |
3149 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3150 | if (num_devices > 3) | |
3151 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3152 | BTRFS_BLOCK_GROUP_RAID6); | |
6728b198 ID |
3153 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3154 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
3155 | (bctl->data.target & ~allowed))) { | |
e4d8ec0f ID |
3156 | printk(KERN_ERR "btrfs: unable to start balance with target " |
3157 | "data profile %llu\n", | |
c1c9ff7c | 3158 | bctl->data.target); |
e4d8ec0f ID |
3159 | ret = -EINVAL; |
3160 | goto out; | |
3161 | } | |
6728b198 ID |
3162 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3163 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
3164 | (bctl->meta.target & ~allowed))) { | |
e4d8ec0f ID |
3165 | printk(KERN_ERR "btrfs: unable to start balance with target " |
3166 | "metadata profile %llu\n", | |
c1c9ff7c | 3167 | bctl->meta.target); |
e4d8ec0f ID |
3168 | ret = -EINVAL; |
3169 | goto out; | |
3170 | } | |
6728b198 ID |
3171 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3172 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
3173 | (bctl->sys.target & ~allowed))) { | |
e4d8ec0f ID |
3174 | printk(KERN_ERR "btrfs: unable to start balance with target " |
3175 | "system profile %llu\n", | |
c1c9ff7c | 3176 | bctl->sys.target); |
e4d8ec0f ID |
3177 | ret = -EINVAL; |
3178 | goto out; | |
3179 | } | |
3180 | ||
e4837f8f ID |
3181 | /* allow dup'ed data chunks only in mixed mode */ |
3182 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 3183 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
e4d8ec0f ID |
3184 | printk(KERN_ERR "btrfs: dup for data is not allowed\n"); |
3185 | ret = -EINVAL; | |
3186 | goto out; | |
3187 | } | |
3188 | ||
3189 | /* allow to reduce meta or sys integrity only if force set */ | |
3190 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3191 | BTRFS_BLOCK_GROUP_RAID10 | |
3192 | BTRFS_BLOCK_GROUP_RAID5 | | |
3193 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3194 | do { |
3195 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3196 | ||
3197 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3198 | (fs_info->avail_system_alloc_bits & allowed) && | |
3199 | !(bctl->sys.target & allowed)) || | |
3200 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3201 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3202 | !(bctl->meta.target & allowed))) { | |
3203 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
3204 | printk(KERN_INFO "btrfs: force reducing metadata " | |
3205 | "integrity\n"); | |
3206 | } else { | |
3207 | printk(KERN_ERR "btrfs: balance will reduce metadata " | |
3208 | "integrity, use force if you want this\n"); | |
3209 | ret = -EINVAL; | |
3210 | goto out; | |
3211 | } | |
e4d8ec0f | 3212 | } |
de98ced9 | 3213 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3214 | |
5af3e8cc SB |
3215 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3216 | int num_tolerated_disk_barrier_failures; | |
3217 | u64 target = bctl->sys.target; | |
3218 | ||
3219 | num_tolerated_disk_barrier_failures = | |
3220 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3221 | if (num_tolerated_disk_barrier_failures > 0 && | |
3222 | (target & | |
3223 | (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | | |
3224 | BTRFS_AVAIL_ALLOC_BIT_SINGLE))) | |
3225 | num_tolerated_disk_barrier_failures = 0; | |
3226 | else if (num_tolerated_disk_barrier_failures > 1 && | |
3227 | (target & | |
3228 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))) | |
3229 | num_tolerated_disk_barrier_failures = 1; | |
3230 | ||
3231 | fs_info->num_tolerated_disk_barrier_failures = | |
3232 | num_tolerated_disk_barrier_failures; | |
3233 | } | |
3234 | ||
0940ebf6 | 3235 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3236 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3237 | goto out; |
3238 | ||
59641015 ID |
3239 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3240 | BUG_ON(ret == -EEXIST); | |
3241 | set_balance_control(bctl); | |
3242 | } else { | |
3243 | BUG_ON(ret != -EEXIST); | |
3244 | spin_lock(&fs_info->balance_lock); | |
3245 | update_balance_args(bctl); | |
3246 | spin_unlock(&fs_info->balance_lock); | |
3247 | } | |
c9e9f97b | 3248 | |
837d5b6e | 3249 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3250 | mutex_unlock(&fs_info->balance_mutex); |
3251 | ||
3252 | ret = __btrfs_balance(fs_info); | |
3253 | ||
3254 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3255 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3256 | |
bf023ecf ID |
3257 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3258 | fs_info->num_tolerated_disk_barrier_failures = | |
3259 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3260 | } | |
3261 | ||
c9e9f97b ID |
3262 | if (bargs) { |
3263 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3264 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3265 | } |
3266 | ||
3a01aa7a ID |
3267 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3268 | balance_need_close(fs_info)) { | |
3269 | __cancel_balance(fs_info); | |
3270 | } | |
3271 | ||
837d5b6e | 3272 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3273 | |
3274 | return ret; | |
3275 | out: | |
59641015 ID |
3276 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3277 | __cancel_balance(fs_info); | |
ed0fb78f | 3278 | else { |
59641015 | 3279 | kfree(bctl); |
ed0fb78f ID |
3280 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3281 | } | |
59641015 ID |
3282 | return ret; |
3283 | } | |
3284 | ||
3285 | static int balance_kthread(void *data) | |
3286 | { | |
2b6ba629 | 3287 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3288 | int ret = 0; |
59641015 ID |
3289 | |
3290 | mutex_lock(&fs_info->volume_mutex); | |
3291 | mutex_lock(&fs_info->balance_mutex); | |
3292 | ||
2b6ba629 | 3293 | if (fs_info->balance_ctl) { |
9555c6c1 | 3294 | printk(KERN_INFO "btrfs: continuing balance\n"); |
2b6ba629 | 3295 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3296 | } |
59641015 ID |
3297 | |
3298 | mutex_unlock(&fs_info->balance_mutex); | |
3299 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3300 | |
59641015 ID |
3301 | return ret; |
3302 | } | |
3303 | ||
2b6ba629 ID |
3304 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3305 | { | |
3306 | struct task_struct *tsk; | |
3307 | ||
3308 | spin_lock(&fs_info->balance_lock); | |
3309 | if (!fs_info->balance_ctl) { | |
3310 | spin_unlock(&fs_info->balance_lock); | |
3311 | return 0; | |
3312 | } | |
3313 | spin_unlock(&fs_info->balance_lock); | |
3314 | ||
3315 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
3316 | printk(KERN_INFO "btrfs: force skipping balance\n"); | |
3317 | return 0; | |
3318 | } | |
3319 | ||
3320 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
97a184fe | 3321 | return PTR_RET(tsk); |
2b6ba629 ID |
3322 | } |
3323 | ||
68310a5e | 3324 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3325 | { |
59641015 ID |
3326 | struct btrfs_balance_control *bctl; |
3327 | struct btrfs_balance_item *item; | |
3328 | struct btrfs_disk_balance_args disk_bargs; | |
3329 | struct btrfs_path *path; | |
3330 | struct extent_buffer *leaf; | |
3331 | struct btrfs_key key; | |
3332 | int ret; | |
3333 | ||
3334 | path = btrfs_alloc_path(); | |
3335 | if (!path) | |
3336 | return -ENOMEM; | |
3337 | ||
59641015 ID |
3338 | key.objectid = BTRFS_BALANCE_OBJECTID; |
3339 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3340 | key.offset = 0; | |
3341 | ||
68310a5e | 3342 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3343 | if (ret < 0) |
68310a5e | 3344 | goto out; |
59641015 ID |
3345 | if (ret > 0) { /* ret = -ENOENT; */ |
3346 | ret = 0; | |
68310a5e ID |
3347 | goto out; |
3348 | } | |
3349 | ||
3350 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3351 | if (!bctl) { | |
3352 | ret = -ENOMEM; | |
3353 | goto out; | |
59641015 ID |
3354 | } |
3355 | ||
3356 | leaf = path->nodes[0]; | |
3357 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3358 | ||
68310a5e ID |
3359 | bctl->fs_info = fs_info; |
3360 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3361 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3362 | |
3363 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3364 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3365 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3366 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3367 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3368 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3369 | ||
ed0fb78f ID |
3370 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3371 | ||
68310a5e ID |
3372 | mutex_lock(&fs_info->volume_mutex); |
3373 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3374 | |
68310a5e ID |
3375 | set_balance_control(bctl); |
3376 | ||
3377 | mutex_unlock(&fs_info->balance_mutex); | |
3378 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3379 | out: |
3380 | btrfs_free_path(path); | |
ec44a35c CM |
3381 | return ret; |
3382 | } | |
3383 | ||
837d5b6e ID |
3384 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3385 | { | |
3386 | int ret = 0; | |
3387 | ||
3388 | mutex_lock(&fs_info->balance_mutex); | |
3389 | if (!fs_info->balance_ctl) { | |
3390 | mutex_unlock(&fs_info->balance_mutex); | |
3391 | return -ENOTCONN; | |
3392 | } | |
3393 | ||
3394 | if (atomic_read(&fs_info->balance_running)) { | |
3395 | atomic_inc(&fs_info->balance_pause_req); | |
3396 | mutex_unlock(&fs_info->balance_mutex); | |
3397 | ||
3398 | wait_event(fs_info->balance_wait_q, | |
3399 | atomic_read(&fs_info->balance_running) == 0); | |
3400 | ||
3401 | mutex_lock(&fs_info->balance_mutex); | |
3402 | /* we are good with balance_ctl ripped off from under us */ | |
3403 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3404 | atomic_dec(&fs_info->balance_pause_req); | |
3405 | } else { | |
3406 | ret = -ENOTCONN; | |
3407 | } | |
3408 | ||
3409 | mutex_unlock(&fs_info->balance_mutex); | |
3410 | return ret; | |
3411 | } | |
3412 | ||
a7e99c69 ID |
3413 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3414 | { | |
3415 | mutex_lock(&fs_info->balance_mutex); | |
3416 | if (!fs_info->balance_ctl) { | |
3417 | mutex_unlock(&fs_info->balance_mutex); | |
3418 | return -ENOTCONN; | |
3419 | } | |
3420 | ||
3421 | atomic_inc(&fs_info->balance_cancel_req); | |
3422 | /* | |
3423 | * if we are running just wait and return, balance item is | |
3424 | * deleted in btrfs_balance in this case | |
3425 | */ | |
3426 | if (atomic_read(&fs_info->balance_running)) { | |
3427 | mutex_unlock(&fs_info->balance_mutex); | |
3428 | wait_event(fs_info->balance_wait_q, | |
3429 | atomic_read(&fs_info->balance_running) == 0); | |
3430 | mutex_lock(&fs_info->balance_mutex); | |
3431 | } else { | |
3432 | /* __cancel_balance needs volume_mutex */ | |
3433 | mutex_unlock(&fs_info->balance_mutex); | |
3434 | mutex_lock(&fs_info->volume_mutex); | |
3435 | mutex_lock(&fs_info->balance_mutex); | |
3436 | ||
3437 | if (fs_info->balance_ctl) | |
3438 | __cancel_balance(fs_info); | |
3439 | ||
3440 | mutex_unlock(&fs_info->volume_mutex); | |
3441 | } | |
3442 | ||
3443 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3444 | atomic_dec(&fs_info->balance_cancel_req); | |
3445 | mutex_unlock(&fs_info->balance_mutex); | |
3446 | return 0; | |
3447 | } | |
3448 | ||
803b2f54 SB |
3449 | static int btrfs_uuid_scan_kthread(void *data) |
3450 | { | |
3451 | struct btrfs_fs_info *fs_info = data; | |
3452 | struct btrfs_root *root = fs_info->tree_root; | |
3453 | struct btrfs_key key; | |
3454 | struct btrfs_key max_key; | |
3455 | struct btrfs_path *path = NULL; | |
3456 | int ret = 0; | |
3457 | struct extent_buffer *eb; | |
3458 | int slot; | |
3459 | struct btrfs_root_item root_item; | |
3460 | u32 item_size; | |
3461 | struct btrfs_trans_handle *trans; | |
3462 | ||
3463 | path = btrfs_alloc_path(); | |
3464 | if (!path) { | |
3465 | ret = -ENOMEM; | |
3466 | goto out; | |
3467 | } | |
3468 | ||
3469 | key.objectid = 0; | |
3470 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3471 | key.offset = 0; | |
3472 | ||
3473 | max_key.objectid = (u64)-1; | |
3474 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
3475 | max_key.offset = (u64)-1; | |
3476 | ||
3477 | path->keep_locks = 1; | |
3478 | ||
3479 | while (1) { | |
3480 | ret = btrfs_search_forward(root, &key, &max_key, path, 0); | |
3481 | if (ret) { | |
3482 | if (ret > 0) | |
3483 | ret = 0; | |
3484 | break; | |
3485 | } | |
3486 | ||
3487 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
3488 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
3489 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
3490 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
3491 | goto skip; | |
3492 | ||
3493 | eb = path->nodes[0]; | |
3494 | slot = path->slots[0]; | |
3495 | item_size = btrfs_item_size_nr(eb, slot); | |
3496 | if (item_size < sizeof(root_item)) | |
3497 | goto skip; | |
3498 | ||
3499 | trans = NULL; | |
3500 | read_extent_buffer(eb, &root_item, | |
3501 | btrfs_item_ptr_offset(eb, slot), | |
3502 | (int)sizeof(root_item)); | |
3503 | if (btrfs_root_refs(&root_item) == 0) | |
3504 | goto skip; | |
3505 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
3506 | /* | |
3507 | * 1 - subvol uuid item | |
3508 | * 1 - received_subvol uuid item | |
3509 | */ | |
3510 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
3511 | if (IS_ERR(trans)) { | |
3512 | ret = PTR_ERR(trans); | |
3513 | break; | |
3514 | } | |
3515 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
3516 | root_item.uuid, | |
3517 | BTRFS_UUID_KEY_SUBVOL, | |
3518 | key.objectid); | |
3519 | if (ret < 0) { | |
3520 | pr_warn("btrfs: uuid_tree_add failed %d\n", | |
3521 | ret); | |
3522 | btrfs_end_transaction(trans, | |
3523 | fs_info->uuid_root); | |
3524 | break; | |
3525 | } | |
3526 | } | |
3527 | ||
3528 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
3529 | if (!trans) { | |
3530 | /* 1 - received_subvol uuid item */ | |
3531 | trans = btrfs_start_transaction( | |
3532 | fs_info->uuid_root, 1); | |
3533 | if (IS_ERR(trans)) { | |
3534 | ret = PTR_ERR(trans); | |
3535 | break; | |
3536 | } | |
3537 | } | |
3538 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
3539 | root_item.received_uuid, | |
3540 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
3541 | key.objectid); | |
3542 | if (ret < 0) { | |
3543 | pr_warn("btrfs: uuid_tree_add failed %d\n", | |
3544 | ret); | |
3545 | btrfs_end_transaction(trans, | |
3546 | fs_info->uuid_root); | |
3547 | break; | |
3548 | } | |
3549 | } | |
3550 | ||
3551 | if (trans) { | |
3552 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
3553 | if (ret) | |
3554 | break; | |
3555 | } | |
3556 | ||
3557 | skip: | |
3558 | btrfs_release_path(path); | |
3559 | if (key.offset < (u64)-1) { | |
3560 | key.offset++; | |
3561 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
3562 | key.offset = 0; | |
3563 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3564 | } else if (key.objectid < (u64)-1) { | |
3565 | key.offset = 0; | |
3566 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3567 | key.objectid++; | |
3568 | } else { | |
3569 | break; | |
3570 | } | |
3571 | cond_resched(); | |
3572 | } | |
3573 | ||
3574 | out: | |
3575 | btrfs_free_path(path); | |
3576 | if (ret) | |
3577 | pr_warn("btrfs: btrfs_uuid_scan_kthread failed %d\n", ret); | |
70f80175 SB |
3578 | else |
3579 | fs_info->update_uuid_tree_gen = 1; | |
803b2f54 SB |
3580 | up(&fs_info->uuid_tree_rescan_sem); |
3581 | return 0; | |
3582 | } | |
3583 | ||
70f80175 SB |
3584 | /* |
3585 | * Callback for btrfs_uuid_tree_iterate(). | |
3586 | * returns: | |
3587 | * 0 check succeeded, the entry is not outdated. | |
3588 | * < 0 if an error occured. | |
3589 | * > 0 if the check failed, which means the caller shall remove the entry. | |
3590 | */ | |
3591 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
3592 | u8 *uuid, u8 type, u64 subid) | |
3593 | { | |
3594 | struct btrfs_key key; | |
3595 | int ret = 0; | |
3596 | struct btrfs_root *subvol_root; | |
3597 | ||
3598 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
3599 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
3600 | goto out; | |
3601 | ||
3602 | key.objectid = subid; | |
3603 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3604 | key.offset = (u64)-1; | |
3605 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
3606 | if (IS_ERR(subvol_root)) { | |
3607 | ret = PTR_ERR(subvol_root); | |
3608 | if (ret == -ENOENT) | |
3609 | ret = 1; | |
3610 | goto out; | |
3611 | } | |
3612 | ||
3613 | switch (type) { | |
3614 | case BTRFS_UUID_KEY_SUBVOL: | |
3615 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
3616 | ret = 1; | |
3617 | break; | |
3618 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
3619 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
3620 | BTRFS_UUID_SIZE)) | |
3621 | ret = 1; | |
3622 | break; | |
3623 | } | |
3624 | ||
3625 | out: | |
3626 | return ret; | |
3627 | } | |
3628 | ||
3629 | static int btrfs_uuid_rescan_kthread(void *data) | |
3630 | { | |
3631 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
3632 | int ret; | |
3633 | ||
3634 | /* | |
3635 | * 1st step is to iterate through the existing UUID tree and | |
3636 | * to delete all entries that contain outdated data. | |
3637 | * 2nd step is to add all missing entries to the UUID tree. | |
3638 | */ | |
3639 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
3640 | if (ret < 0) { | |
3641 | pr_warn("btrfs: iterating uuid_tree failed %d\n", ret); | |
3642 | up(&fs_info->uuid_tree_rescan_sem); | |
3643 | return ret; | |
3644 | } | |
3645 | return btrfs_uuid_scan_kthread(data); | |
3646 | } | |
3647 | ||
f7a81ea4 SB |
3648 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
3649 | { | |
3650 | struct btrfs_trans_handle *trans; | |
3651 | struct btrfs_root *tree_root = fs_info->tree_root; | |
3652 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
3653 | struct task_struct *task; |
3654 | int ret; | |
f7a81ea4 SB |
3655 | |
3656 | /* | |
3657 | * 1 - root node | |
3658 | * 1 - root item | |
3659 | */ | |
3660 | trans = btrfs_start_transaction(tree_root, 2); | |
3661 | if (IS_ERR(trans)) | |
3662 | return PTR_ERR(trans); | |
3663 | ||
3664 | uuid_root = btrfs_create_tree(trans, fs_info, | |
3665 | BTRFS_UUID_TREE_OBJECTID); | |
3666 | if (IS_ERR(uuid_root)) { | |
3667 | btrfs_abort_transaction(trans, tree_root, | |
3668 | PTR_ERR(uuid_root)); | |
3669 | return PTR_ERR(uuid_root); | |
3670 | } | |
3671 | ||
3672 | fs_info->uuid_root = uuid_root; | |
3673 | ||
803b2f54 SB |
3674 | ret = btrfs_commit_transaction(trans, tree_root); |
3675 | if (ret) | |
3676 | return ret; | |
3677 | ||
3678 | down(&fs_info->uuid_tree_rescan_sem); | |
3679 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
3680 | if (IS_ERR(task)) { | |
70f80175 | 3681 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
803b2f54 SB |
3682 | pr_warn("btrfs: failed to start uuid_scan task\n"); |
3683 | up(&fs_info->uuid_tree_rescan_sem); | |
3684 | return PTR_ERR(task); | |
3685 | } | |
3686 | ||
3687 | return 0; | |
f7a81ea4 | 3688 | } |
803b2f54 | 3689 | |
70f80175 SB |
3690 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
3691 | { | |
3692 | struct task_struct *task; | |
3693 | ||
3694 | down(&fs_info->uuid_tree_rescan_sem); | |
3695 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
3696 | if (IS_ERR(task)) { | |
3697 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
3698 | pr_warn("btrfs: failed to start uuid_rescan task\n"); | |
3699 | up(&fs_info->uuid_tree_rescan_sem); | |
3700 | return PTR_ERR(task); | |
3701 | } | |
3702 | ||
3703 | return 0; | |
3704 | } | |
3705 | ||
8f18cf13 CM |
3706 | /* |
3707 | * shrinking a device means finding all of the device extents past | |
3708 | * the new size, and then following the back refs to the chunks. | |
3709 | * The chunk relocation code actually frees the device extent | |
3710 | */ | |
3711 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
3712 | { | |
3713 | struct btrfs_trans_handle *trans; | |
3714 | struct btrfs_root *root = device->dev_root; | |
3715 | struct btrfs_dev_extent *dev_extent = NULL; | |
3716 | struct btrfs_path *path; | |
3717 | u64 length; | |
3718 | u64 chunk_tree; | |
3719 | u64 chunk_objectid; | |
3720 | u64 chunk_offset; | |
3721 | int ret; | |
3722 | int slot; | |
ba1bf481 JB |
3723 | int failed = 0; |
3724 | bool retried = false; | |
8f18cf13 CM |
3725 | struct extent_buffer *l; |
3726 | struct btrfs_key key; | |
6c41761f | 3727 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 3728 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 3729 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
3730 | u64 diff = device->total_bytes - new_size; |
3731 | ||
63a212ab SB |
3732 | if (device->is_tgtdev_for_dev_replace) |
3733 | return -EINVAL; | |
3734 | ||
8f18cf13 CM |
3735 | path = btrfs_alloc_path(); |
3736 | if (!path) | |
3737 | return -ENOMEM; | |
3738 | ||
8f18cf13 CM |
3739 | path->reada = 2; |
3740 | ||
7d9eb12c CM |
3741 | lock_chunks(root); |
3742 | ||
8f18cf13 | 3743 | device->total_bytes = new_size; |
2bf64758 | 3744 | if (device->writeable) { |
2b82032c | 3745 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
3746 | spin_lock(&root->fs_info->free_chunk_lock); |
3747 | root->fs_info->free_chunk_space -= diff; | |
3748 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3749 | } | |
7d9eb12c | 3750 | unlock_chunks(root); |
8f18cf13 | 3751 | |
ba1bf481 | 3752 | again: |
8f18cf13 CM |
3753 | key.objectid = device->devid; |
3754 | key.offset = (u64)-1; | |
3755 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3756 | ||
213e64da | 3757 | do { |
8f18cf13 CM |
3758 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3759 | if (ret < 0) | |
3760 | goto done; | |
3761 | ||
3762 | ret = btrfs_previous_item(root, path, 0, key.type); | |
3763 | if (ret < 0) | |
3764 | goto done; | |
3765 | if (ret) { | |
3766 | ret = 0; | |
b3b4aa74 | 3767 | btrfs_release_path(path); |
bf1fb512 | 3768 | break; |
8f18cf13 CM |
3769 | } |
3770 | ||
3771 | l = path->nodes[0]; | |
3772 | slot = path->slots[0]; | |
3773 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
3774 | ||
ba1bf481 | 3775 | if (key.objectid != device->devid) { |
b3b4aa74 | 3776 | btrfs_release_path(path); |
bf1fb512 | 3777 | break; |
ba1bf481 | 3778 | } |
8f18cf13 CM |
3779 | |
3780 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3781 | length = btrfs_dev_extent_length(l, dev_extent); | |
3782 | ||
ba1bf481 | 3783 | if (key.offset + length <= new_size) { |
b3b4aa74 | 3784 | btrfs_release_path(path); |
d6397bae | 3785 | break; |
ba1bf481 | 3786 | } |
8f18cf13 CM |
3787 | |
3788 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3789 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3790 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3791 | btrfs_release_path(path); |
8f18cf13 CM |
3792 | |
3793 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
3794 | chunk_offset); | |
ba1bf481 | 3795 | if (ret && ret != -ENOSPC) |
8f18cf13 | 3796 | goto done; |
ba1bf481 JB |
3797 | if (ret == -ENOSPC) |
3798 | failed++; | |
213e64da | 3799 | } while (key.offset-- > 0); |
ba1bf481 JB |
3800 | |
3801 | if (failed && !retried) { | |
3802 | failed = 0; | |
3803 | retried = true; | |
3804 | goto again; | |
3805 | } else if (failed && retried) { | |
3806 | ret = -ENOSPC; | |
3807 | lock_chunks(root); | |
3808 | ||
3809 | device->total_bytes = old_size; | |
3810 | if (device->writeable) | |
3811 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
3812 | spin_lock(&root->fs_info->free_chunk_lock); |
3813 | root->fs_info->free_chunk_space += diff; | |
3814 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
3815 | unlock_chunks(root); |
3816 | goto done; | |
8f18cf13 CM |
3817 | } |
3818 | ||
d6397bae | 3819 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 3820 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
3821 | if (IS_ERR(trans)) { |
3822 | ret = PTR_ERR(trans); | |
3823 | goto done; | |
3824 | } | |
3825 | ||
d6397bae CB |
3826 | lock_chunks(root); |
3827 | ||
3828 | device->disk_total_bytes = new_size; | |
3829 | /* Now btrfs_update_device() will change the on-disk size. */ | |
3830 | ret = btrfs_update_device(trans, device); | |
3831 | if (ret) { | |
3832 | unlock_chunks(root); | |
3833 | btrfs_end_transaction(trans, root); | |
3834 | goto done; | |
3835 | } | |
3836 | WARN_ON(diff > old_total); | |
3837 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
3838 | unlock_chunks(root); | |
3839 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
3840 | done: |
3841 | btrfs_free_path(path); | |
3842 | return ret; | |
3843 | } | |
3844 | ||
125ccb0a | 3845 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
3846 | struct btrfs_key *key, |
3847 | struct btrfs_chunk *chunk, int item_size) | |
3848 | { | |
6c41761f | 3849 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
3850 | struct btrfs_disk_key disk_key; |
3851 | u32 array_size; | |
3852 | u8 *ptr; | |
3853 | ||
3854 | array_size = btrfs_super_sys_array_size(super_copy); | |
3855 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
3856 | return -EFBIG; | |
3857 | ||
3858 | ptr = super_copy->sys_chunk_array + array_size; | |
3859 | btrfs_cpu_key_to_disk(&disk_key, key); | |
3860 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
3861 | ptr += sizeof(disk_key); | |
3862 | memcpy(ptr, chunk, item_size); | |
3863 | item_size += sizeof(disk_key); | |
3864 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
3865 | return 0; | |
3866 | } | |
3867 | ||
73c5de00 AJ |
3868 | /* |
3869 | * sort the devices in descending order by max_avail, total_avail | |
3870 | */ | |
3871 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 3872 | { |
73c5de00 AJ |
3873 | const struct btrfs_device_info *di_a = a; |
3874 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 3875 | |
73c5de00 | 3876 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 3877 | return -1; |
73c5de00 | 3878 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 3879 | return 1; |
73c5de00 AJ |
3880 | if (di_a->total_avail > di_b->total_avail) |
3881 | return -1; | |
3882 | if (di_a->total_avail < di_b->total_avail) | |
3883 | return 1; | |
3884 | return 0; | |
b2117a39 | 3885 | } |
0b86a832 | 3886 | |
48a3b636 | 3887 | static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
e6ec716f MX |
3888 | [BTRFS_RAID_RAID10] = { |
3889 | .sub_stripes = 2, | |
3890 | .dev_stripes = 1, | |
3891 | .devs_max = 0, /* 0 == as many as possible */ | |
3892 | .devs_min = 4, | |
3893 | .devs_increment = 2, | |
3894 | .ncopies = 2, | |
3895 | }, | |
3896 | [BTRFS_RAID_RAID1] = { | |
3897 | .sub_stripes = 1, | |
3898 | .dev_stripes = 1, | |
3899 | .devs_max = 2, | |
3900 | .devs_min = 2, | |
3901 | .devs_increment = 2, | |
3902 | .ncopies = 2, | |
3903 | }, | |
3904 | [BTRFS_RAID_DUP] = { | |
3905 | .sub_stripes = 1, | |
3906 | .dev_stripes = 2, | |
3907 | .devs_max = 1, | |
3908 | .devs_min = 1, | |
3909 | .devs_increment = 1, | |
3910 | .ncopies = 2, | |
3911 | }, | |
3912 | [BTRFS_RAID_RAID0] = { | |
3913 | .sub_stripes = 1, | |
3914 | .dev_stripes = 1, | |
3915 | .devs_max = 0, | |
3916 | .devs_min = 2, | |
3917 | .devs_increment = 1, | |
3918 | .ncopies = 1, | |
3919 | }, | |
3920 | [BTRFS_RAID_SINGLE] = { | |
3921 | .sub_stripes = 1, | |
3922 | .dev_stripes = 1, | |
3923 | .devs_max = 1, | |
3924 | .devs_min = 1, | |
3925 | .devs_increment = 1, | |
3926 | .ncopies = 1, | |
3927 | }, | |
e942f883 CM |
3928 | [BTRFS_RAID_RAID5] = { |
3929 | .sub_stripes = 1, | |
3930 | .dev_stripes = 1, | |
3931 | .devs_max = 0, | |
3932 | .devs_min = 2, | |
3933 | .devs_increment = 1, | |
3934 | .ncopies = 2, | |
3935 | }, | |
3936 | [BTRFS_RAID_RAID6] = { | |
3937 | .sub_stripes = 1, | |
3938 | .dev_stripes = 1, | |
3939 | .devs_max = 0, | |
3940 | .devs_min = 3, | |
3941 | .devs_increment = 1, | |
3942 | .ncopies = 3, | |
3943 | }, | |
31e50229 LB |
3944 | }; |
3945 | ||
53b381b3 DW |
3946 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
3947 | { | |
3948 | /* TODO allow them to set a preferred stripe size */ | |
3949 | return 64 * 1024; | |
3950 | } | |
3951 | ||
3952 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
3953 | { | |
53b381b3 DW |
3954 | if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))) |
3955 | return; | |
3956 | ||
ceda0864 | 3957 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
3958 | } |
3959 | ||
73c5de00 | 3960 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
3961 | struct btrfs_root *extent_root, u64 start, |
3962 | u64 type) | |
b2117a39 | 3963 | { |
73c5de00 AJ |
3964 | struct btrfs_fs_info *info = extent_root->fs_info; |
3965 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
3966 | struct list_head *cur; | |
3967 | struct map_lookup *map = NULL; | |
3968 | struct extent_map_tree *em_tree; | |
3969 | struct extent_map *em; | |
3970 | struct btrfs_device_info *devices_info = NULL; | |
3971 | u64 total_avail; | |
3972 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
3973 | int data_stripes; /* number of stripes that count for |
3974 | block group size */ | |
73c5de00 AJ |
3975 | int sub_stripes; /* sub_stripes info for map */ |
3976 | int dev_stripes; /* stripes per dev */ | |
3977 | int devs_max; /* max devs to use */ | |
3978 | int devs_min; /* min devs needed */ | |
3979 | int devs_increment; /* ndevs has to be a multiple of this */ | |
3980 | int ncopies; /* how many copies to data has */ | |
3981 | int ret; | |
3982 | u64 max_stripe_size; | |
3983 | u64 max_chunk_size; | |
3984 | u64 stripe_size; | |
3985 | u64 num_bytes; | |
53b381b3 | 3986 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
3987 | int ndevs; |
3988 | int i; | |
3989 | int j; | |
31e50229 | 3990 | int index; |
593060d7 | 3991 | |
0c460c0d | 3992 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 3993 | |
73c5de00 AJ |
3994 | if (list_empty(&fs_devices->alloc_list)) |
3995 | return -ENOSPC; | |
b2117a39 | 3996 | |
31e50229 | 3997 | index = __get_raid_index(type); |
73c5de00 | 3998 | |
31e50229 LB |
3999 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4000 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4001 | devs_max = btrfs_raid_array[index].devs_max; | |
4002 | devs_min = btrfs_raid_array[index].devs_min; | |
4003 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4004 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4005 | |
9b3f68b9 | 4006 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
4007 | max_stripe_size = 1024 * 1024 * 1024; |
4008 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 4009 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
4010 | /* for larger filesystems, use larger metadata chunks */ |
4011 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
4012 | max_stripe_size = 1024 * 1024 * 1024; | |
4013 | else | |
4014 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 4015 | max_chunk_size = max_stripe_size; |
a40a90a0 | 4016 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 4017 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 AJ |
4018 | max_chunk_size = 2 * max_stripe_size; |
4019 | } else { | |
4020 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
4021 | type); | |
4022 | BUG_ON(1); | |
9b3f68b9 CM |
4023 | } |
4024 | ||
2b82032c YZ |
4025 | /* we don't want a chunk larger than 10% of writeable space */ |
4026 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4027 | max_chunk_size); | |
9b3f68b9 | 4028 | |
73c5de00 AJ |
4029 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
4030 | GFP_NOFS); | |
4031 | if (!devices_info) | |
4032 | return -ENOMEM; | |
0cad8a11 | 4033 | |
73c5de00 | 4034 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4035 | |
9f680ce0 | 4036 | /* |
73c5de00 AJ |
4037 | * in the first pass through the devices list, we gather information |
4038 | * about the available holes on each device. | |
9f680ce0 | 4039 | */ |
73c5de00 AJ |
4040 | ndevs = 0; |
4041 | while (cur != &fs_devices->alloc_list) { | |
4042 | struct btrfs_device *device; | |
4043 | u64 max_avail; | |
4044 | u64 dev_offset; | |
b2117a39 | 4045 | |
73c5de00 | 4046 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4047 | |
73c5de00 | 4048 | cur = cur->next; |
b2117a39 | 4049 | |
73c5de00 | 4050 | if (!device->writeable) { |
31b1a2bd | 4051 | WARN(1, KERN_ERR |
73c5de00 | 4052 | "btrfs: read-only device in alloc_list\n"); |
73c5de00 AJ |
4053 | continue; |
4054 | } | |
b2117a39 | 4055 | |
63a212ab SB |
4056 | if (!device->in_fs_metadata || |
4057 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4058 | continue; |
b2117a39 | 4059 | |
73c5de00 AJ |
4060 | if (device->total_bytes > device->bytes_used) |
4061 | total_avail = device->total_bytes - device->bytes_used; | |
4062 | else | |
4063 | total_avail = 0; | |
38c01b96 | 4064 | |
4065 | /* If there is no space on this device, skip it. */ | |
4066 | if (total_avail == 0) | |
4067 | continue; | |
b2117a39 | 4068 | |
6df9a95e | 4069 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4070 | max_stripe_size * dev_stripes, |
4071 | &dev_offset, &max_avail); | |
4072 | if (ret && ret != -ENOSPC) | |
4073 | goto error; | |
b2117a39 | 4074 | |
73c5de00 AJ |
4075 | if (ret == 0) |
4076 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4077 | |
73c5de00 AJ |
4078 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4079 | continue; | |
b2117a39 | 4080 | |
063d006f ES |
4081 | if (ndevs == fs_devices->rw_devices) { |
4082 | WARN(1, "%s: found more than %llu devices\n", | |
4083 | __func__, fs_devices->rw_devices); | |
4084 | break; | |
4085 | } | |
73c5de00 AJ |
4086 | devices_info[ndevs].dev_offset = dev_offset; |
4087 | devices_info[ndevs].max_avail = max_avail; | |
4088 | devices_info[ndevs].total_avail = total_avail; | |
4089 | devices_info[ndevs].dev = device; | |
4090 | ++ndevs; | |
4091 | } | |
b2117a39 | 4092 | |
73c5de00 AJ |
4093 | /* |
4094 | * now sort the devices by hole size / available space | |
4095 | */ | |
4096 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4097 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4098 | |
73c5de00 AJ |
4099 | /* round down to number of usable stripes */ |
4100 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4101 | |
73c5de00 AJ |
4102 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4103 | ret = -ENOSPC; | |
4104 | goto error; | |
b2117a39 | 4105 | } |
9f680ce0 | 4106 | |
73c5de00 AJ |
4107 | if (devs_max && ndevs > devs_max) |
4108 | ndevs = devs_max; | |
4109 | /* | |
4110 | * the primary goal is to maximize the number of stripes, so use as many | |
4111 | * devices as possible, even if the stripes are not maximum sized. | |
4112 | */ | |
4113 | stripe_size = devices_info[ndevs-1].max_avail; | |
4114 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4115 | |
53b381b3 DW |
4116 | /* |
4117 | * this will have to be fixed for RAID1 and RAID10 over | |
4118 | * more drives | |
4119 | */ | |
4120 | data_stripes = num_stripes / ncopies; | |
4121 | ||
53b381b3 DW |
4122 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4123 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4124 | btrfs_super_stripesize(info->super_copy)); | |
4125 | data_stripes = num_stripes - 1; | |
4126 | } | |
4127 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4128 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4129 | btrfs_super_stripesize(info->super_copy)); | |
4130 | data_stripes = num_stripes - 2; | |
4131 | } | |
86db2578 CM |
4132 | |
4133 | /* | |
4134 | * Use the number of data stripes to figure out how big this chunk | |
4135 | * is really going to be in terms of logical address space, | |
4136 | * and compare that answer with the max chunk size | |
4137 | */ | |
4138 | if (stripe_size * data_stripes > max_chunk_size) { | |
4139 | u64 mask = (1ULL << 24) - 1; | |
4140 | stripe_size = max_chunk_size; | |
4141 | do_div(stripe_size, data_stripes); | |
4142 | ||
4143 | /* bump the answer up to a 16MB boundary */ | |
4144 | stripe_size = (stripe_size + mask) & ~mask; | |
4145 | ||
4146 | /* but don't go higher than the limits we found | |
4147 | * while searching for free extents | |
4148 | */ | |
4149 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4150 | stripe_size = devices_info[ndevs-1].max_avail; | |
4151 | } | |
4152 | ||
73c5de00 | 4153 | do_div(stripe_size, dev_stripes); |
37db63a4 ID |
4154 | |
4155 | /* align to BTRFS_STRIPE_LEN */ | |
53b381b3 DW |
4156 | do_div(stripe_size, raid_stripe_len); |
4157 | stripe_size *= raid_stripe_len; | |
b2117a39 MX |
4158 | |
4159 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4160 | if (!map) { | |
4161 | ret = -ENOMEM; | |
4162 | goto error; | |
4163 | } | |
4164 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4165 | |
73c5de00 AJ |
4166 | for (i = 0; i < ndevs; ++i) { |
4167 | for (j = 0; j < dev_stripes; ++j) { | |
4168 | int s = i * dev_stripes + j; | |
4169 | map->stripes[s].dev = devices_info[i].dev; | |
4170 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4171 | j * stripe_size; | |
6324fbf3 | 4172 | } |
6324fbf3 | 4173 | } |
2b82032c | 4174 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4175 | map->stripe_len = raid_stripe_len; |
4176 | map->io_align = raid_stripe_len; | |
4177 | map->io_width = raid_stripe_len; | |
2b82032c | 4178 | map->type = type; |
2b82032c | 4179 | map->sub_stripes = sub_stripes; |
0b86a832 | 4180 | |
53b381b3 | 4181 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4182 | |
73c5de00 | 4183 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4184 | |
172ddd60 | 4185 | em = alloc_extent_map(); |
2b82032c | 4186 | if (!em) { |
b2117a39 MX |
4187 | ret = -ENOMEM; |
4188 | goto error; | |
593060d7 | 4189 | } |
2b82032c YZ |
4190 | em->bdev = (struct block_device *)map; |
4191 | em->start = start; | |
73c5de00 | 4192 | em->len = num_bytes; |
2b82032c YZ |
4193 | em->block_start = 0; |
4194 | em->block_len = em->len; | |
6df9a95e | 4195 | em->orig_block_len = stripe_size; |
593060d7 | 4196 | |
2b82032c | 4197 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4198 | write_lock(&em_tree->lock); |
09a2a8f9 | 4199 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4200 | if (!ret) { |
4201 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4202 | atomic_inc(&em->refs); | |
4203 | } | |
890871be | 4204 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4205 | if (ret) { |
4206 | free_extent_map(em); | |
1dd4602f | 4207 | goto error; |
0f5d42b2 | 4208 | } |
0b86a832 | 4209 | |
04487488 JB |
4210 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4211 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4212 | start, num_bytes); | |
6df9a95e JB |
4213 | if (ret) |
4214 | goto error_del_extent; | |
2b82032c | 4215 | |
0f5d42b2 | 4216 | free_extent_map(em); |
53b381b3 DW |
4217 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4218 | ||
b2117a39 | 4219 | kfree(devices_info); |
2b82032c | 4220 | return 0; |
b2117a39 | 4221 | |
6df9a95e | 4222 | error_del_extent: |
0f5d42b2 JB |
4223 | write_lock(&em_tree->lock); |
4224 | remove_extent_mapping(em_tree, em); | |
4225 | write_unlock(&em_tree->lock); | |
4226 | ||
4227 | /* One for our allocation */ | |
4228 | free_extent_map(em); | |
4229 | /* One for the tree reference */ | |
4230 | free_extent_map(em); | |
b2117a39 MX |
4231 | error: |
4232 | kfree(map); | |
4233 | kfree(devices_info); | |
4234 | return ret; | |
2b82032c YZ |
4235 | } |
4236 | ||
6df9a95e | 4237 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4238 | struct btrfs_root *extent_root, |
6df9a95e | 4239 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4240 | { |
2b82032c YZ |
4241 | struct btrfs_key key; |
4242 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4243 | struct btrfs_device *device; | |
4244 | struct btrfs_chunk *chunk; | |
4245 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4246 | struct extent_map_tree *em_tree; |
4247 | struct extent_map *em; | |
4248 | struct map_lookup *map; | |
4249 | size_t item_size; | |
4250 | u64 dev_offset; | |
4251 | u64 stripe_size; | |
4252 | int i = 0; | |
2b82032c YZ |
4253 | int ret; |
4254 | ||
6df9a95e JB |
4255 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4256 | read_lock(&em_tree->lock); | |
4257 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4258 | read_unlock(&em_tree->lock); | |
4259 | ||
4260 | if (!em) { | |
4261 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4262 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4263 | return -EINVAL; | |
4264 | } | |
4265 | ||
4266 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4267 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
4268 | " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset, | |
4269 | chunk_size, em->start, em->len); | |
4270 | free_extent_map(em); | |
4271 | return -EINVAL; | |
4272 | } | |
4273 | ||
4274 | map = (struct map_lookup *)em->bdev; | |
4275 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
4276 | stripe_size = em->orig_block_len; | |
4277 | ||
2b82032c | 4278 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4279 | if (!chunk) { |
4280 | ret = -ENOMEM; | |
4281 | goto out; | |
4282 | } | |
4283 | ||
4284 | for (i = 0; i < map->num_stripes; i++) { | |
4285 | device = map->stripes[i].dev; | |
4286 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4287 | |
2b82032c | 4288 | device->bytes_used += stripe_size; |
0b86a832 | 4289 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4290 | if (ret) |
6df9a95e JB |
4291 | goto out; |
4292 | ret = btrfs_alloc_dev_extent(trans, device, | |
4293 | chunk_root->root_key.objectid, | |
4294 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4295 | chunk_offset, dev_offset, | |
4296 | stripe_size); | |
4297 | if (ret) | |
4298 | goto out; | |
2b82032c YZ |
4299 | } |
4300 | ||
2bf64758 JB |
4301 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4302 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4303 | map->num_stripes); | |
4304 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4305 | ||
2b82032c | 4306 | stripe = &chunk->stripe; |
6df9a95e JB |
4307 | for (i = 0; i < map->num_stripes; i++) { |
4308 | device = map->stripes[i].dev; | |
4309 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4310 | |
e17cade2 CM |
4311 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4312 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4313 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4314 | stripe++; |
0b86a832 CM |
4315 | } |
4316 | ||
2b82032c | 4317 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4318 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4319 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4320 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4321 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4322 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4323 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4324 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4325 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4326 | |
2b82032c YZ |
4327 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4328 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4329 | key.offset = chunk_offset; | |
0b86a832 | 4330 | |
2b82032c | 4331 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4332 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4333 | /* | |
4334 | * TODO: Cleanup of inserted chunk root in case of | |
4335 | * failure. | |
4336 | */ | |
125ccb0a | 4337 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4338 | item_size); |
8f18cf13 | 4339 | } |
1abe9b8a | 4340 | |
6df9a95e | 4341 | out: |
0b86a832 | 4342 | kfree(chunk); |
6df9a95e | 4343 | free_extent_map(em); |
4ed1d16e | 4344 | return ret; |
2b82032c | 4345 | } |
0b86a832 | 4346 | |
2b82032c YZ |
4347 | /* |
4348 | * Chunk allocation falls into two parts. The first part does works | |
4349 | * that make the new allocated chunk useable, but not do any operation | |
4350 | * that modifies the chunk tree. The second part does the works that | |
4351 | * require modifying the chunk tree. This division is important for the | |
4352 | * bootstrap process of adding storage to a seed btrfs. | |
4353 | */ | |
4354 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4355 | struct btrfs_root *extent_root, u64 type) | |
4356 | { | |
4357 | u64 chunk_offset; | |
2b82032c | 4358 | |
6df9a95e JB |
4359 | chunk_offset = find_next_chunk(extent_root->fs_info); |
4360 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
4361 | } |
4362 | ||
d397712b | 4363 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4364 | struct btrfs_root *root, |
4365 | struct btrfs_device *device) | |
4366 | { | |
4367 | u64 chunk_offset; | |
4368 | u64 sys_chunk_offset; | |
2b82032c | 4369 | u64 alloc_profile; |
2b82032c YZ |
4370 | struct btrfs_fs_info *fs_info = root->fs_info; |
4371 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4372 | int ret; | |
4373 | ||
6df9a95e | 4374 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 4375 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
4376 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
4377 | alloc_profile); | |
79787eaa JM |
4378 | if (ret) |
4379 | return ret; | |
2b82032c | 4380 | |
6df9a95e | 4381 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 4382 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
4383 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
4384 | alloc_profile); | |
005d6427 DS |
4385 | if (ret) { |
4386 | btrfs_abort_transaction(trans, root, ret); | |
4387 | goto out; | |
4388 | } | |
2b82032c YZ |
4389 | |
4390 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
79787eaa | 4391 | if (ret) |
005d6427 | 4392 | btrfs_abort_transaction(trans, root, ret); |
005d6427 | 4393 | out: |
79787eaa | 4394 | return ret; |
2b82032c YZ |
4395 | } |
4396 | ||
4397 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4398 | { | |
4399 | struct extent_map *em; | |
4400 | struct map_lookup *map; | |
4401 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4402 | int readonly = 0; | |
4403 | int i; | |
4404 | ||
890871be | 4405 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4406 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4407 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4408 | if (!em) |
4409 | return 1; | |
4410 | ||
f48b9075 JB |
4411 | if (btrfs_test_opt(root, DEGRADED)) { |
4412 | free_extent_map(em); | |
4413 | return 0; | |
4414 | } | |
4415 | ||
2b82032c YZ |
4416 | map = (struct map_lookup *)em->bdev; |
4417 | for (i = 0; i < map->num_stripes; i++) { | |
4418 | if (!map->stripes[i].dev->writeable) { | |
4419 | readonly = 1; | |
4420 | break; | |
4421 | } | |
4422 | } | |
0b86a832 | 4423 | free_extent_map(em); |
2b82032c | 4424 | return readonly; |
0b86a832 CM |
4425 | } |
4426 | ||
4427 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4428 | { | |
a8067e02 | 4429 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
4430 | } |
4431 | ||
4432 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4433 | { | |
4434 | struct extent_map *em; | |
4435 | ||
d397712b | 4436 | while (1) { |
890871be | 4437 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
4438 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
4439 | if (em) | |
4440 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 4441 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
4442 | if (!em) |
4443 | break; | |
4444 | kfree(em->bdev); | |
4445 | /* once for us */ | |
4446 | free_extent_map(em); | |
4447 | /* once for the tree */ | |
4448 | free_extent_map(em); | |
4449 | } | |
4450 | } | |
4451 | ||
5d964051 | 4452 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 4453 | { |
5d964051 | 4454 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
4455 | struct extent_map *em; |
4456 | struct map_lookup *map; | |
4457 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4458 | int ret; | |
4459 | ||
890871be | 4460 | read_lock(&em_tree->lock); |
f188591e | 4461 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 4462 | read_unlock(&em_tree->lock); |
f188591e | 4463 | |
fb7669b5 JB |
4464 | /* |
4465 | * We could return errors for these cases, but that could get ugly and | |
4466 | * we'd probably do the same thing which is just not do anything else | |
4467 | * and exit, so return 1 so the callers don't try to use other copies. | |
4468 | */ | |
4469 | if (!em) { | |
ccf39f92 | 4470 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu\n", logical, |
fb7669b5 JB |
4471 | logical+len); |
4472 | return 1; | |
4473 | } | |
4474 | ||
4475 | if (em->start > logical || em->start + em->len < logical) { | |
ccf39f92 | 4476 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " |
fb7669b5 JB |
4477 | "%Lu-%Lu\n", logical, logical+len, em->start, |
4478 | em->start + em->len); | |
4479 | return 1; | |
4480 | } | |
4481 | ||
f188591e CM |
4482 | map = (struct map_lookup *)em->bdev; |
4483 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
4484 | ret = map->num_stripes; | |
321aecc6 CM |
4485 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
4486 | ret = map->sub_stripes; | |
53b381b3 DW |
4487 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
4488 | ret = 2; | |
4489 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4490 | ret = 3; | |
f188591e CM |
4491 | else |
4492 | ret = 1; | |
4493 | free_extent_map(em); | |
ad6d620e SB |
4494 | |
4495 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
4496 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
4497 | ret++; | |
4498 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
4499 | ||
f188591e CM |
4500 | return ret; |
4501 | } | |
4502 | ||
53b381b3 DW |
4503 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
4504 | struct btrfs_mapping_tree *map_tree, | |
4505 | u64 logical) | |
4506 | { | |
4507 | struct extent_map *em; | |
4508 | struct map_lookup *map; | |
4509 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4510 | unsigned long len = root->sectorsize; | |
4511 | ||
4512 | read_lock(&em_tree->lock); | |
4513 | em = lookup_extent_mapping(em_tree, logical, len); | |
4514 | read_unlock(&em_tree->lock); | |
4515 | BUG_ON(!em); | |
4516 | ||
4517 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4518 | map = (struct map_lookup *)em->bdev; | |
4519 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4520 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4521 | len = map->stripe_len * nr_data_stripes(map); | |
4522 | } | |
4523 | free_extent_map(em); | |
4524 | return len; | |
4525 | } | |
4526 | ||
4527 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
4528 | u64 logical, u64 len, int mirror_num) | |
4529 | { | |
4530 | struct extent_map *em; | |
4531 | struct map_lookup *map; | |
4532 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4533 | int ret = 0; | |
4534 | ||
4535 | read_lock(&em_tree->lock); | |
4536 | em = lookup_extent_mapping(em_tree, logical, len); | |
4537 | read_unlock(&em_tree->lock); | |
4538 | BUG_ON(!em); | |
4539 | ||
4540 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4541 | map = (struct map_lookup *)em->bdev; | |
4542 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4543 | BTRFS_BLOCK_GROUP_RAID6)) | |
4544 | ret = 1; | |
4545 | free_extent_map(em); | |
4546 | return ret; | |
4547 | } | |
4548 | ||
30d9861f SB |
4549 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
4550 | struct map_lookup *map, int first, int num, | |
4551 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
4552 | { |
4553 | int i; | |
30d9861f SB |
4554 | int tolerance; |
4555 | struct btrfs_device *srcdev; | |
4556 | ||
4557 | if (dev_replace_is_ongoing && | |
4558 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
4559 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
4560 | srcdev = fs_info->dev_replace.srcdev; | |
4561 | else | |
4562 | srcdev = NULL; | |
4563 | ||
4564 | /* | |
4565 | * try to avoid the drive that is the source drive for a | |
4566 | * dev-replace procedure, only choose it if no other non-missing | |
4567 | * mirror is available | |
4568 | */ | |
4569 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
4570 | if (map->stripes[optimal].dev->bdev && | |
4571 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
4572 | return optimal; | |
4573 | for (i = first; i < first + num; i++) { | |
4574 | if (map->stripes[i].dev->bdev && | |
4575 | (tolerance || map->stripes[i].dev != srcdev)) | |
4576 | return i; | |
4577 | } | |
dfe25020 | 4578 | } |
30d9861f | 4579 | |
dfe25020 CM |
4580 | /* we couldn't find one that doesn't fail. Just return something |
4581 | * and the io error handling code will clean up eventually | |
4582 | */ | |
4583 | return optimal; | |
4584 | } | |
4585 | ||
53b381b3 DW |
4586 | static inline int parity_smaller(u64 a, u64 b) |
4587 | { | |
4588 | return a > b; | |
4589 | } | |
4590 | ||
4591 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
4592 | static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map) | |
4593 | { | |
4594 | struct btrfs_bio_stripe s; | |
4595 | int i; | |
4596 | u64 l; | |
4597 | int again = 1; | |
4598 | ||
4599 | while (again) { | |
4600 | again = 0; | |
4601 | for (i = 0; i < bbio->num_stripes - 1; i++) { | |
4602 | if (parity_smaller(raid_map[i], raid_map[i+1])) { | |
4603 | s = bbio->stripes[i]; | |
4604 | l = raid_map[i]; | |
4605 | bbio->stripes[i] = bbio->stripes[i+1]; | |
4606 | raid_map[i] = raid_map[i+1]; | |
4607 | bbio->stripes[i+1] = s; | |
4608 | raid_map[i+1] = l; | |
4609 | again = 1; | |
4610 | } | |
4611 | } | |
4612 | } | |
4613 | } | |
4614 | ||
3ec706c8 | 4615 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 4616 | u64 logical, u64 *length, |
a1d3c478 | 4617 | struct btrfs_bio **bbio_ret, |
53b381b3 | 4618 | int mirror_num, u64 **raid_map_ret) |
0b86a832 CM |
4619 | { |
4620 | struct extent_map *em; | |
4621 | struct map_lookup *map; | |
3ec706c8 | 4622 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
4623 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
4624 | u64 offset; | |
593060d7 | 4625 | u64 stripe_offset; |
fce3bb9a | 4626 | u64 stripe_end_offset; |
593060d7 | 4627 | u64 stripe_nr; |
fce3bb9a LD |
4628 | u64 stripe_nr_orig; |
4629 | u64 stripe_nr_end; | |
53b381b3 DW |
4630 | u64 stripe_len; |
4631 | u64 *raid_map = NULL; | |
593060d7 | 4632 | int stripe_index; |
cea9e445 | 4633 | int i; |
de11cc12 | 4634 | int ret = 0; |
f2d8d74d | 4635 | int num_stripes; |
a236aed1 | 4636 | int max_errors = 0; |
a1d3c478 | 4637 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
4638 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
4639 | int dev_replace_is_ongoing = 0; | |
4640 | int num_alloc_stripes; | |
ad6d620e SB |
4641 | int patch_the_first_stripe_for_dev_replace = 0; |
4642 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 4643 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 4644 | |
890871be | 4645 | read_lock(&em_tree->lock); |
0b86a832 | 4646 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 4647 | read_unlock(&em_tree->lock); |
f2d8d74d | 4648 | |
3b951516 | 4649 | if (!em) { |
c2cf52eb | 4650 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 4651 | logical, *length); |
9bb91873 JB |
4652 | return -EINVAL; |
4653 | } | |
4654 | ||
4655 | if (em->start > logical || em->start + em->len < logical) { | |
4656 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
4657 | "found %Lu-%Lu\n", logical, em->start, | |
4658 | em->start + em->len); | |
4659 | return -EINVAL; | |
3b951516 | 4660 | } |
0b86a832 | 4661 | |
0b86a832 CM |
4662 | map = (struct map_lookup *)em->bdev; |
4663 | offset = logical - em->start; | |
593060d7 | 4664 | |
53b381b3 | 4665 | stripe_len = map->stripe_len; |
593060d7 CM |
4666 | stripe_nr = offset; |
4667 | /* | |
4668 | * stripe_nr counts the total number of stripes we have to stride | |
4669 | * to get to this block | |
4670 | */ | |
53b381b3 | 4671 | do_div(stripe_nr, stripe_len); |
593060d7 | 4672 | |
53b381b3 | 4673 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
4674 | BUG_ON(offset < stripe_offset); |
4675 | ||
4676 | /* stripe_offset is the offset of this block in its stripe*/ | |
4677 | stripe_offset = offset - stripe_offset; | |
4678 | ||
53b381b3 DW |
4679 | /* if we're here for raid56, we need to know the stripe aligned start */ |
4680 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4681 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); | |
4682 | raid56_full_stripe_start = offset; | |
4683 | ||
4684 | /* allow a write of a full stripe, but make sure we don't | |
4685 | * allow straddling of stripes | |
4686 | */ | |
4687 | do_div(raid56_full_stripe_start, full_stripe_len); | |
4688 | raid56_full_stripe_start *= full_stripe_len; | |
4689 | } | |
4690 | ||
4691 | if (rw & REQ_DISCARD) { | |
4692 | /* we don't discard raid56 yet */ | |
4693 | if (map->type & | |
4694 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4695 | ret = -EOPNOTSUPP; | |
4696 | goto out; | |
4697 | } | |
fce3bb9a | 4698 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
4699 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
4700 | u64 max_len; | |
4701 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
4702 | For other RAID types and for RAID[56] reads, just allow a single | |
4703 | stripe (on a single disk). */ | |
4704 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) && | |
4705 | (rw & REQ_WRITE)) { | |
4706 | max_len = stripe_len * nr_data_stripes(map) - | |
4707 | (offset - raid56_full_stripe_start); | |
4708 | } else { | |
4709 | /* we limit the length of each bio to what fits in a stripe */ | |
4710 | max_len = stripe_len - stripe_offset; | |
4711 | } | |
4712 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
4713 | } else { |
4714 | *length = em->len - offset; | |
4715 | } | |
f2d8d74d | 4716 | |
53b381b3 DW |
4717 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
4718 | it cares about is the length */ | |
a1d3c478 | 4719 | if (!bbio_ret) |
cea9e445 CM |
4720 | goto out; |
4721 | ||
472262f3 SB |
4722 | btrfs_dev_replace_lock(dev_replace); |
4723 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
4724 | if (!dev_replace_is_ongoing) | |
4725 | btrfs_dev_replace_unlock(dev_replace); | |
4726 | ||
ad6d620e SB |
4727 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
4728 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
4729 | dev_replace->tgtdev != NULL) { | |
4730 | /* | |
4731 | * in dev-replace case, for repair case (that's the only | |
4732 | * case where the mirror is selected explicitly when | |
4733 | * calling btrfs_map_block), blocks left of the left cursor | |
4734 | * can also be read from the target drive. | |
4735 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
4736 | * the last one to the array of stripes. For READ, it also | |
4737 | * needs to be supported using the same mirror number. | |
4738 | * If the requested block is not left of the left cursor, | |
4739 | * EIO is returned. This can happen because btrfs_num_copies() | |
4740 | * returns one more in the dev-replace case. | |
4741 | */ | |
4742 | u64 tmp_length = *length; | |
4743 | struct btrfs_bio *tmp_bbio = NULL; | |
4744 | int tmp_num_stripes; | |
4745 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
4746 | int index_srcdev = 0; | |
4747 | int found = 0; | |
4748 | u64 physical_of_found = 0; | |
4749 | ||
4750 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
53b381b3 | 4751 | logical, &tmp_length, &tmp_bbio, 0, NULL); |
ad6d620e SB |
4752 | if (ret) { |
4753 | WARN_ON(tmp_bbio != NULL); | |
4754 | goto out; | |
4755 | } | |
4756 | ||
4757 | tmp_num_stripes = tmp_bbio->num_stripes; | |
4758 | if (mirror_num > tmp_num_stripes) { | |
4759 | /* | |
4760 | * REQ_GET_READ_MIRRORS does not contain this | |
4761 | * mirror, that means that the requested area | |
4762 | * is not left of the left cursor | |
4763 | */ | |
4764 | ret = -EIO; | |
4765 | kfree(tmp_bbio); | |
4766 | goto out; | |
4767 | } | |
4768 | ||
4769 | /* | |
4770 | * process the rest of the function using the mirror_num | |
4771 | * of the source drive. Therefore look it up first. | |
4772 | * At the end, patch the device pointer to the one of the | |
4773 | * target drive. | |
4774 | */ | |
4775 | for (i = 0; i < tmp_num_stripes; i++) { | |
4776 | if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) { | |
4777 | /* | |
4778 | * In case of DUP, in order to keep it | |
4779 | * simple, only add the mirror with the | |
4780 | * lowest physical address | |
4781 | */ | |
4782 | if (found && | |
4783 | physical_of_found <= | |
4784 | tmp_bbio->stripes[i].physical) | |
4785 | continue; | |
4786 | index_srcdev = i; | |
4787 | found = 1; | |
4788 | physical_of_found = | |
4789 | tmp_bbio->stripes[i].physical; | |
4790 | } | |
4791 | } | |
4792 | ||
4793 | if (found) { | |
4794 | mirror_num = index_srcdev + 1; | |
4795 | patch_the_first_stripe_for_dev_replace = 1; | |
4796 | physical_to_patch_in_first_stripe = physical_of_found; | |
4797 | } else { | |
4798 | WARN_ON(1); | |
4799 | ret = -EIO; | |
4800 | kfree(tmp_bbio); | |
4801 | goto out; | |
4802 | } | |
4803 | ||
4804 | kfree(tmp_bbio); | |
4805 | } else if (mirror_num > map->num_stripes) { | |
4806 | mirror_num = 0; | |
4807 | } | |
4808 | ||
f2d8d74d | 4809 | num_stripes = 1; |
cea9e445 | 4810 | stripe_index = 0; |
fce3bb9a | 4811 | stripe_nr_orig = stripe_nr; |
fda2832f | 4812 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
fce3bb9a LD |
4813 | do_div(stripe_nr_end, map->stripe_len); |
4814 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
4815 | (offset + *length); | |
53b381b3 | 4816 | |
fce3bb9a LD |
4817 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
4818 | if (rw & REQ_DISCARD) | |
4819 | num_stripes = min_t(u64, map->num_stripes, | |
4820 | stripe_nr_end - stripe_nr_orig); | |
4821 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
4822 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
29a8d9a0 | 4823 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 4824 | num_stripes = map->num_stripes; |
2fff734f | 4825 | else if (mirror_num) |
f188591e | 4826 | stripe_index = mirror_num - 1; |
dfe25020 | 4827 | else { |
30d9861f | 4828 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 4829 | map->num_stripes, |
30d9861f SB |
4830 | current->pid % map->num_stripes, |
4831 | dev_replace_is_ongoing); | |
a1d3c478 | 4832 | mirror_num = stripe_index + 1; |
dfe25020 | 4833 | } |
2fff734f | 4834 | |
611f0e00 | 4835 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 4836 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 4837 | num_stripes = map->num_stripes; |
a1d3c478 | 4838 | } else if (mirror_num) { |
f188591e | 4839 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
4840 | } else { |
4841 | mirror_num = 1; | |
4842 | } | |
2fff734f | 4843 | |
321aecc6 CM |
4844 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
4845 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
4846 | |
4847 | stripe_index = do_div(stripe_nr, factor); | |
4848 | stripe_index *= map->sub_stripes; | |
4849 | ||
29a8d9a0 | 4850 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 4851 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
4852 | else if (rw & REQ_DISCARD) |
4853 | num_stripes = min_t(u64, map->sub_stripes * | |
4854 | (stripe_nr_end - stripe_nr_orig), | |
4855 | map->num_stripes); | |
321aecc6 CM |
4856 | else if (mirror_num) |
4857 | stripe_index += mirror_num - 1; | |
dfe25020 | 4858 | else { |
3e74317a | 4859 | int old_stripe_index = stripe_index; |
30d9861f SB |
4860 | stripe_index = find_live_mirror(fs_info, map, |
4861 | stripe_index, | |
dfe25020 | 4862 | map->sub_stripes, stripe_index + |
30d9861f SB |
4863 | current->pid % map->sub_stripes, |
4864 | dev_replace_is_ongoing); | |
3e74317a | 4865 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 4866 | } |
53b381b3 DW |
4867 | |
4868 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4869 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4870 | u64 tmp; | |
4871 | ||
4872 | if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1) | |
4873 | && raid_map_ret) { | |
4874 | int i, rot; | |
4875 | ||
4876 | /* push stripe_nr back to the start of the full stripe */ | |
4877 | stripe_nr = raid56_full_stripe_start; | |
4878 | do_div(stripe_nr, stripe_len); | |
4879 | ||
4880 | stripe_index = do_div(stripe_nr, nr_data_stripes(map)); | |
4881 | ||
4882 | /* RAID[56] write or recovery. Return all stripes */ | |
4883 | num_stripes = map->num_stripes; | |
4884 | max_errors = nr_parity_stripes(map); | |
4885 | ||
4886 | raid_map = kmalloc(sizeof(u64) * num_stripes, | |
4887 | GFP_NOFS); | |
4888 | if (!raid_map) { | |
4889 | ret = -ENOMEM; | |
4890 | goto out; | |
4891 | } | |
4892 | ||
4893 | /* Work out the disk rotation on this stripe-set */ | |
4894 | tmp = stripe_nr; | |
4895 | rot = do_div(tmp, num_stripes); | |
4896 | ||
4897 | /* Fill in the logical address of each stripe */ | |
4898 | tmp = stripe_nr * nr_data_stripes(map); | |
4899 | for (i = 0; i < nr_data_stripes(map); i++) | |
4900 | raid_map[(i+rot) % num_stripes] = | |
4901 | em->start + (tmp + i) * map->stripe_len; | |
4902 | ||
4903 | raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
4904 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4905 | raid_map[(i+rot+1) % num_stripes] = | |
4906 | RAID6_Q_STRIPE; | |
4907 | ||
4908 | *length = map->stripe_len; | |
4909 | stripe_index = 0; | |
4910 | stripe_offset = 0; | |
4911 | } else { | |
4912 | /* | |
4913 | * Mirror #0 or #1 means the original data block. | |
4914 | * Mirror #2 is RAID5 parity block. | |
4915 | * Mirror #3 is RAID6 Q block. | |
4916 | */ | |
4917 | stripe_index = do_div(stripe_nr, nr_data_stripes(map)); | |
4918 | if (mirror_num > 1) | |
4919 | stripe_index = nr_data_stripes(map) + | |
4920 | mirror_num - 2; | |
4921 | ||
4922 | /* We distribute the parity blocks across stripes */ | |
4923 | tmp = stripe_nr + stripe_index; | |
4924 | stripe_index = do_div(tmp, map->num_stripes); | |
4925 | } | |
8790d502 CM |
4926 | } else { |
4927 | /* | |
4928 | * after this do_div call, stripe_nr is the number of stripes | |
4929 | * on this device we have to walk to find the data, and | |
4930 | * stripe_index is the number of our device in the stripe array | |
4931 | */ | |
4932 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 4933 | mirror_num = stripe_index + 1; |
8790d502 | 4934 | } |
593060d7 | 4935 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 4936 | |
472262f3 | 4937 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
4938 | if (dev_replace_is_ongoing) { |
4939 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
4940 | num_alloc_stripes <<= 1; | |
4941 | if (rw & REQ_GET_READ_MIRRORS) | |
4942 | num_alloc_stripes++; | |
4943 | } | |
472262f3 | 4944 | bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS); |
de11cc12 | 4945 | if (!bbio) { |
eb2067f7 | 4946 | kfree(raid_map); |
de11cc12 LZ |
4947 | ret = -ENOMEM; |
4948 | goto out; | |
4949 | } | |
4950 | atomic_set(&bbio->error, 0); | |
4951 | ||
fce3bb9a | 4952 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
4953 | int factor = 0; |
4954 | int sub_stripes = 0; | |
4955 | u64 stripes_per_dev = 0; | |
4956 | u32 remaining_stripes = 0; | |
b89203f7 | 4957 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
4958 | |
4959 | if (map->type & | |
4960 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
4961 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
4962 | sub_stripes = 1; | |
4963 | else | |
4964 | sub_stripes = map->sub_stripes; | |
4965 | ||
4966 | factor = map->num_stripes / sub_stripes; | |
4967 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
4968 | stripe_nr_orig, | |
4969 | factor, | |
4970 | &remaining_stripes); | |
b89203f7 LB |
4971 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
4972 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
4973 | } |
4974 | ||
fce3bb9a | 4975 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 4976 | bbio->stripes[i].physical = |
f2d8d74d CM |
4977 | map->stripes[stripe_index].physical + |
4978 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 4979 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 4980 | |
ec9ef7a1 LZ |
4981 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
4982 | BTRFS_BLOCK_GROUP_RAID10)) { | |
4983 | bbio->stripes[i].length = stripes_per_dev * | |
4984 | map->stripe_len; | |
b89203f7 | 4985 | |
ec9ef7a1 LZ |
4986 | if (i / sub_stripes < remaining_stripes) |
4987 | bbio->stripes[i].length += | |
4988 | map->stripe_len; | |
b89203f7 LB |
4989 | |
4990 | /* | |
4991 | * Special for the first stripe and | |
4992 | * the last stripe: | |
4993 | * | |
4994 | * |-------|...|-------| | |
4995 | * |----------| | |
4996 | * off end_off | |
4997 | */ | |
ec9ef7a1 | 4998 | if (i < sub_stripes) |
a1d3c478 | 4999 | bbio->stripes[i].length -= |
fce3bb9a | 5000 | stripe_offset; |
b89203f7 LB |
5001 | |
5002 | if (stripe_index >= last_stripe && | |
5003 | stripe_index <= (last_stripe + | |
5004 | sub_stripes - 1)) | |
a1d3c478 | 5005 | bbio->stripes[i].length -= |
fce3bb9a | 5006 | stripe_end_offset; |
b89203f7 | 5007 | |
ec9ef7a1 LZ |
5008 | if (i == sub_stripes - 1) |
5009 | stripe_offset = 0; | |
fce3bb9a | 5010 | } else |
a1d3c478 | 5011 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5012 | |
5013 | stripe_index++; | |
5014 | if (stripe_index == map->num_stripes) { | |
5015 | /* This could only happen for RAID0/10 */ | |
5016 | stripe_index = 0; | |
5017 | stripe_nr++; | |
5018 | } | |
5019 | } | |
5020 | } else { | |
5021 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5022 | bbio->stripes[i].physical = |
212a17ab LT |
5023 | map->stripes[stripe_index].physical + |
5024 | stripe_offset + | |
5025 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5026 | bbio->stripes[i].dev = |
212a17ab | 5027 | map->stripes[stripe_index].dev; |
fce3bb9a | 5028 | stripe_index++; |
f2d8d74d | 5029 | } |
593060d7 | 5030 | } |
de11cc12 | 5031 | |
29a8d9a0 | 5032 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) { |
de11cc12 LZ |
5033 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | |
5034 | BTRFS_BLOCK_GROUP_RAID10 | | |
53b381b3 | 5035 | BTRFS_BLOCK_GROUP_RAID5 | |
de11cc12 LZ |
5036 | BTRFS_BLOCK_GROUP_DUP)) { |
5037 | max_errors = 1; | |
53b381b3 DW |
5038 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { |
5039 | max_errors = 2; | |
de11cc12 | 5040 | } |
f2d8d74d | 5041 | } |
de11cc12 | 5042 | |
472262f3 SB |
5043 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
5044 | dev_replace->tgtdev != NULL) { | |
5045 | int index_where_to_add; | |
5046 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5047 | ||
5048 | /* | |
5049 | * duplicate the write operations while the dev replace | |
5050 | * procedure is running. Since the copying of the old disk | |
5051 | * to the new disk takes place at run time while the | |
5052 | * filesystem is mounted writable, the regular write | |
5053 | * operations to the old disk have to be duplicated to go | |
5054 | * to the new disk as well. | |
5055 | * Note that device->missing is handled by the caller, and | |
5056 | * that the write to the old disk is already set up in the | |
5057 | * stripes array. | |
5058 | */ | |
5059 | index_where_to_add = num_stripes; | |
5060 | for (i = 0; i < num_stripes; i++) { | |
5061 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5062 | /* write to new disk, too */ | |
5063 | struct btrfs_bio_stripe *new = | |
5064 | bbio->stripes + index_where_to_add; | |
5065 | struct btrfs_bio_stripe *old = | |
5066 | bbio->stripes + i; | |
5067 | ||
5068 | new->physical = old->physical; | |
5069 | new->length = old->length; | |
5070 | new->dev = dev_replace->tgtdev; | |
5071 | index_where_to_add++; | |
5072 | max_errors++; | |
5073 | } | |
5074 | } | |
5075 | num_stripes = index_where_to_add; | |
ad6d620e SB |
5076 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
5077 | dev_replace->tgtdev != NULL) { | |
5078 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5079 | int index_srcdev = 0; | |
5080 | int found = 0; | |
5081 | u64 physical_of_found = 0; | |
5082 | ||
5083 | /* | |
5084 | * During the dev-replace procedure, the target drive can | |
5085 | * also be used to read data in case it is needed to repair | |
5086 | * a corrupt block elsewhere. This is possible if the | |
5087 | * requested area is left of the left cursor. In this area, | |
5088 | * the target drive is a full copy of the source drive. | |
5089 | */ | |
5090 | for (i = 0; i < num_stripes; i++) { | |
5091 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5092 | /* | |
5093 | * In case of DUP, in order to keep it | |
5094 | * simple, only add the mirror with the | |
5095 | * lowest physical address | |
5096 | */ | |
5097 | if (found && | |
5098 | physical_of_found <= | |
5099 | bbio->stripes[i].physical) | |
5100 | continue; | |
5101 | index_srcdev = i; | |
5102 | found = 1; | |
5103 | physical_of_found = bbio->stripes[i].physical; | |
5104 | } | |
5105 | } | |
5106 | if (found) { | |
5107 | u64 length = map->stripe_len; | |
5108 | ||
5109 | if (physical_of_found + length <= | |
5110 | dev_replace->cursor_left) { | |
5111 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5112 | bbio->stripes + num_stripes; | |
5113 | ||
5114 | tgtdev_stripe->physical = physical_of_found; | |
5115 | tgtdev_stripe->length = | |
5116 | bbio->stripes[index_srcdev].length; | |
5117 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5118 | ||
5119 | num_stripes++; | |
5120 | } | |
5121 | } | |
472262f3 SB |
5122 | } |
5123 | ||
de11cc12 LZ |
5124 | *bbio_ret = bbio; |
5125 | bbio->num_stripes = num_stripes; | |
5126 | bbio->max_errors = max_errors; | |
5127 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5128 | |
5129 | /* | |
5130 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5131 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5132 | * available as a mirror | |
5133 | */ | |
5134 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5135 | WARN_ON(num_stripes > 1); | |
5136 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5137 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5138 | bbio->mirror_num = map->num_stripes + 1; | |
5139 | } | |
53b381b3 DW |
5140 | if (raid_map) { |
5141 | sort_parity_stripes(bbio, raid_map); | |
5142 | *raid_map_ret = raid_map; | |
5143 | } | |
cea9e445 | 5144 | out: |
472262f3 SB |
5145 | if (dev_replace_is_ongoing) |
5146 | btrfs_dev_replace_unlock(dev_replace); | |
0b86a832 | 5147 | free_extent_map(em); |
de11cc12 | 5148 | return ret; |
0b86a832 CM |
5149 | } |
5150 | ||
3ec706c8 | 5151 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5152 | u64 logical, u64 *length, |
a1d3c478 | 5153 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5154 | { |
3ec706c8 | 5155 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
53b381b3 | 5156 | mirror_num, NULL); |
f2d8d74d CM |
5157 | } |
5158 | ||
a512bbf8 YZ |
5159 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
5160 | u64 chunk_start, u64 physical, u64 devid, | |
5161 | u64 **logical, int *naddrs, int *stripe_len) | |
5162 | { | |
5163 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5164 | struct extent_map *em; | |
5165 | struct map_lookup *map; | |
5166 | u64 *buf; | |
5167 | u64 bytenr; | |
5168 | u64 length; | |
5169 | u64 stripe_nr; | |
53b381b3 | 5170 | u64 rmap_len; |
a512bbf8 YZ |
5171 | int i, j, nr = 0; |
5172 | ||
890871be | 5173 | read_lock(&em_tree->lock); |
a512bbf8 | 5174 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5175 | read_unlock(&em_tree->lock); |
a512bbf8 | 5176 | |
835d974f JB |
5177 | if (!em) { |
5178 | printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n", | |
5179 | chunk_start); | |
5180 | return -EIO; | |
5181 | } | |
5182 | ||
5183 | if (em->start != chunk_start) { | |
5184 | printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n", | |
5185 | em->start, chunk_start); | |
5186 | free_extent_map(em); | |
5187 | return -EIO; | |
5188 | } | |
a512bbf8 YZ |
5189 | map = (struct map_lookup *)em->bdev; |
5190 | ||
5191 | length = em->len; | |
53b381b3 DW |
5192 | rmap_len = map->stripe_len; |
5193 | ||
a512bbf8 YZ |
5194 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5195 | do_div(length, map->num_stripes / map->sub_stripes); | |
5196 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5197 | do_div(length, map->num_stripes); | |
53b381b3 DW |
5198 | else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | |
5199 | BTRFS_BLOCK_GROUP_RAID6)) { | |
5200 | do_div(length, nr_data_stripes(map)); | |
5201 | rmap_len = map->stripe_len * nr_data_stripes(map); | |
5202 | } | |
a512bbf8 YZ |
5203 | |
5204 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
79787eaa | 5205 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5206 | |
5207 | for (i = 0; i < map->num_stripes; i++) { | |
5208 | if (devid && map->stripes[i].dev->devid != devid) | |
5209 | continue; | |
5210 | if (map->stripes[i].physical > physical || | |
5211 | map->stripes[i].physical + length <= physical) | |
5212 | continue; | |
5213 | ||
5214 | stripe_nr = physical - map->stripes[i].physical; | |
5215 | do_div(stripe_nr, map->stripe_len); | |
5216 | ||
5217 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5218 | stripe_nr = stripe_nr * map->num_stripes + i; | |
5219 | do_div(stripe_nr, map->sub_stripes); | |
5220 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
5221 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5222 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5223 | * Alternatively, just use rmap_len below instead of | |
5224 | * map->stripe_len */ | |
5225 | ||
5226 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5227 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5228 | for (j = 0; j < nr; j++) { |
5229 | if (buf[j] == bytenr) | |
5230 | break; | |
5231 | } | |
934d375b CM |
5232 | if (j == nr) { |
5233 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5234 | buf[nr++] = bytenr; |
934d375b | 5235 | } |
a512bbf8 YZ |
5236 | } |
5237 | ||
a512bbf8 YZ |
5238 | *logical = buf; |
5239 | *naddrs = nr; | |
53b381b3 | 5240 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5241 | |
5242 | free_extent_map(em); | |
5243 | return 0; | |
f2d8d74d CM |
5244 | } |
5245 | ||
a1d3c478 | 5246 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 5247 | { |
9be3395b | 5248 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 5249 | int is_orig_bio = 0; |
8790d502 | 5250 | |
442a4f63 | 5251 | if (err) { |
a1d3c478 | 5252 | atomic_inc(&bbio->error); |
442a4f63 SB |
5253 | if (err == -EIO || err == -EREMOTEIO) { |
5254 | unsigned int stripe_index = | |
9be3395b | 5255 | btrfs_io_bio(bio)->stripe_index; |
442a4f63 SB |
5256 | struct btrfs_device *dev; |
5257 | ||
5258 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5259 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5260 | if (dev->bdev) { |
5261 | if (bio->bi_rw & WRITE) | |
5262 | btrfs_dev_stat_inc(dev, | |
5263 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5264 | else | |
5265 | btrfs_dev_stat_inc(dev, | |
5266 | BTRFS_DEV_STAT_READ_ERRS); | |
5267 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5268 | btrfs_dev_stat_inc(dev, | |
5269 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5270 | btrfs_dev_stat_print_on_error(dev); | |
5271 | } | |
442a4f63 SB |
5272 | } |
5273 | } | |
8790d502 | 5274 | |
a1d3c478 | 5275 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5276 | is_orig_bio = 1; |
5277 | ||
a1d3c478 | 5278 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5279 | if (!is_orig_bio) { |
5280 | bio_put(bio); | |
a1d3c478 | 5281 | bio = bbio->orig_bio; |
7d2b4daa | 5282 | } |
a1d3c478 JS |
5283 | bio->bi_private = bbio->private; |
5284 | bio->bi_end_io = bbio->end_io; | |
9be3395b | 5285 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 5286 | /* only send an error to the higher layers if it is |
53b381b3 | 5287 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5288 | */ |
a1d3c478 | 5289 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 5290 | err = -EIO; |
5dbc8fca | 5291 | } else { |
1259ab75 CM |
5292 | /* |
5293 | * this bio is actually up to date, we didn't | |
5294 | * go over the max number of errors | |
5295 | */ | |
5296 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 5297 | err = 0; |
1259ab75 | 5298 | } |
a1d3c478 | 5299 | kfree(bbio); |
8790d502 CM |
5300 | |
5301 | bio_endio(bio, err); | |
7d2b4daa | 5302 | } else if (!is_orig_bio) { |
8790d502 CM |
5303 | bio_put(bio); |
5304 | } | |
8790d502 CM |
5305 | } |
5306 | ||
8b712842 CM |
5307 | struct async_sched { |
5308 | struct bio *bio; | |
5309 | int rw; | |
5310 | struct btrfs_fs_info *info; | |
5311 | struct btrfs_work work; | |
5312 | }; | |
5313 | ||
5314 | /* | |
5315 | * see run_scheduled_bios for a description of why bios are collected for | |
5316 | * async submit. | |
5317 | * | |
5318 | * This will add one bio to the pending list for a device and make sure | |
5319 | * the work struct is scheduled. | |
5320 | */ | |
48a3b636 ES |
5321 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
5322 | struct btrfs_device *device, | |
5323 | int rw, struct bio *bio) | |
8b712842 CM |
5324 | { |
5325 | int should_queue = 1; | |
ffbd517d | 5326 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5327 | |
53b381b3 DW |
5328 | if (device->missing || !device->bdev) { |
5329 | bio_endio(bio, -EIO); | |
5330 | return; | |
5331 | } | |
5332 | ||
8b712842 | 5333 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5334 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5335 | bio_get(bio); |
21adbd5c | 5336 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5337 | bio_put(bio); |
143bede5 | 5338 | return; |
8b712842 CM |
5339 | } |
5340 | ||
5341 | /* | |
0986fe9e | 5342 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5343 | * higher layers. Otherwise, the async bio makes it appear we have |
5344 | * made progress against dirty pages when we've really just put it | |
5345 | * on a queue for later | |
5346 | */ | |
0986fe9e | 5347 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5348 | WARN_ON(bio->bi_next); |
8b712842 CM |
5349 | bio->bi_next = NULL; |
5350 | bio->bi_rw |= rw; | |
5351 | ||
5352 | spin_lock(&device->io_lock); | |
7b6d91da | 5353 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5354 | pending_bios = &device->pending_sync_bios; |
5355 | else | |
5356 | pending_bios = &device->pending_bios; | |
8b712842 | 5357 | |
ffbd517d CM |
5358 | if (pending_bios->tail) |
5359 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5360 | |
ffbd517d CM |
5361 | pending_bios->tail = bio; |
5362 | if (!pending_bios->head) | |
5363 | pending_bios->head = bio; | |
8b712842 CM |
5364 | if (device->running_pending) |
5365 | should_queue = 0; | |
5366 | ||
5367 | spin_unlock(&device->io_lock); | |
5368 | ||
5369 | if (should_queue) | |
1cc127b5 CM |
5370 | btrfs_queue_worker(&root->fs_info->submit_workers, |
5371 | &device->work); | |
8b712842 CM |
5372 | } |
5373 | ||
de1ee92a JB |
5374 | static int bio_size_ok(struct block_device *bdev, struct bio *bio, |
5375 | sector_t sector) | |
5376 | { | |
5377 | struct bio_vec *prev; | |
5378 | struct request_queue *q = bdev_get_queue(bdev); | |
5379 | unsigned short max_sectors = queue_max_sectors(q); | |
5380 | struct bvec_merge_data bvm = { | |
5381 | .bi_bdev = bdev, | |
5382 | .bi_sector = sector, | |
5383 | .bi_rw = bio->bi_rw, | |
5384 | }; | |
5385 | ||
5386 | if (bio->bi_vcnt == 0) { | |
5387 | WARN_ON(1); | |
5388 | return 1; | |
5389 | } | |
5390 | ||
5391 | prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
aa8b57aa | 5392 | if (bio_sectors(bio) > max_sectors) |
de1ee92a JB |
5393 | return 0; |
5394 | ||
5395 | if (!q->merge_bvec_fn) | |
5396 | return 1; | |
5397 | ||
5398 | bvm.bi_size = bio->bi_size - prev->bv_len; | |
5399 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) | |
5400 | return 0; | |
5401 | return 1; | |
5402 | } | |
5403 | ||
5404 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5405 | struct bio *bio, u64 physical, int dev_nr, | |
5406 | int rw, int async) | |
5407 | { | |
5408 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5409 | ||
5410 | bio->bi_private = bbio; | |
9be3395b | 5411 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a JB |
5412 | bio->bi_end_io = btrfs_end_bio; |
5413 | bio->bi_sector = physical >> 9; | |
5414 | #ifdef DEBUG | |
5415 | { | |
5416 | struct rcu_string *name; | |
5417 | ||
5418 | rcu_read_lock(); | |
5419 | name = rcu_dereference(dev->name); | |
d1423248 | 5420 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a JB |
5421 | "(%s id %llu), size=%u\n", rw, |
5422 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
5423 | name->str, dev->devid, bio->bi_size); | |
5424 | rcu_read_unlock(); | |
5425 | } | |
5426 | #endif | |
5427 | bio->bi_bdev = dev->bdev; | |
5428 | if (async) | |
53b381b3 | 5429 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
5430 | else |
5431 | btrfsic_submit_bio(rw, bio); | |
5432 | } | |
5433 | ||
5434 | static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5435 | struct bio *first_bio, struct btrfs_device *dev, | |
5436 | int dev_nr, int rw, int async) | |
5437 | { | |
5438 | struct bio_vec *bvec = first_bio->bi_io_vec; | |
5439 | struct bio *bio; | |
5440 | int nr_vecs = bio_get_nr_vecs(dev->bdev); | |
5441 | u64 physical = bbio->stripes[dev_nr].physical; | |
5442 | ||
5443 | again: | |
5444 | bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS); | |
5445 | if (!bio) | |
5446 | return -ENOMEM; | |
5447 | ||
5448 | while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) { | |
5449 | if (bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5450 | bvec->bv_offset) < bvec->bv_len) { | |
5451 | u64 len = bio->bi_size; | |
5452 | ||
5453 | atomic_inc(&bbio->stripes_pending); | |
5454 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, | |
5455 | rw, async); | |
5456 | physical += len; | |
5457 | goto again; | |
5458 | } | |
5459 | bvec++; | |
5460 | } | |
5461 | ||
5462 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async); | |
5463 | return 0; | |
5464 | } | |
5465 | ||
5466 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
5467 | { | |
5468 | atomic_inc(&bbio->error); | |
5469 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
5470 | bio->bi_private = bbio->private; | |
5471 | bio->bi_end_io = bbio->end_io; | |
9be3395b | 5472 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
de1ee92a JB |
5473 | bio->bi_sector = logical >> 9; |
5474 | kfree(bbio); | |
5475 | bio_endio(bio, -EIO); | |
5476 | } | |
5477 | } | |
5478 | ||
f188591e | 5479 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 5480 | int mirror_num, int async_submit) |
0b86a832 | 5481 | { |
0b86a832 | 5482 | struct btrfs_device *dev; |
8790d502 | 5483 | struct bio *first_bio = bio; |
a62b9401 | 5484 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
5485 | u64 length = 0; |
5486 | u64 map_length; | |
53b381b3 | 5487 | u64 *raid_map = NULL; |
0b86a832 | 5488 | int ret; |
8790d502 CM |
5489 | int dev_nr = 0; |
5490 | int total_devs = 1; | |
a1d3c478 | 5491 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 5492 | |
f2d8d74d | 5493 | length = bio->bi_size; |
0b86a832 | 5494 | map_length = length; |
cea9e445 | 5495 | |
53b381b3 DW |
5496 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
5497 | mirror_num, &raid_map); | |
5498 | if (ret) /* -ENOMEM */ | |
79787eaa | 5499 | return ret; |
cea9e445 | 5500 | |
a1d3c478 | 5501 | total_devs = bbio->num_stripes; |
53b381b3 DW |
5502 | bbio->orig_bio = first_bio; |
5503 | bbio->private = first_bio->bi_private; | |
5504 | bbio->end_io = first_bio->bi_end_io; | |
5505 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
5506 | ||
5507 | if (raid_map) { | |
5508 | /* In this case, map_length has been set to the length of | |
5509 | a single stripe; not the whole write */ | |
5510 | if (rw & WRITE) { | |
5511 | return raid56_parity_write(root, bio, bbio, | |
5512 | raid_map, map_length); | |
5513 | } else { | |
5514 | return raid56_parity_recover(root, bio, bbio, | |
5515 | raid_map, map_length, | |
5516 | mirror_num); | |
5517 | } | |
5518 | } | |
5519 | ||
cea9e445 | 5520 | if (map_length < length) { |
c2cf52eb | 5521 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", |
c1c9ff7c | 5522 | logical, length, map_length); |
cea9e445 CM |
5523 | BUG(); |
5524 | } | |
a1d3c478 | 5525 | |
d397712b | 5526 | while (dev_nr < total_devs) { |
de1ee92a JB |
5527 | dev = bbio->stripes[dev_nr].dev; |
5528 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
5529 | bbio_error(bbio, first_bio, logical); | |
5530 | dev_nr++; | |
5531 | continue; | |
5532 | } | |
5533 | ||
5534 | /* | |
5535 | * Check and see if we're ok with this bio based on it's size | |
5536 | * and offset with the given device. | |
5537 | */ | |
5538 | if (!bio_size_ok(dev->bdev, first_bio, | |
5539 | bbio->stripes[dev_nr].physical >> 9)) { | |
5540 | ret = breakup_stripe_bio(root, bbio, first_bio, dev, | |
5541 | dev_nr, rw, async_submit); | |
5542 | BUG_ON(ret); | |
5543 | dev_nr++; | |
5544 | continue; | |
5545 | } | |
5546 | ||
a1d3c478 | 5547 | if (dev_nr < total_devs - 1) { |
9be3395b | 5548 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 5549 | BUG_ON(!bio); /* -ENOMEM */ |
a1d3c478 JS |
5550 | } else { |
5551 | bio = first_bio; | |
8790d502 | 5552 | } |
de1ee92a JB |
5553 | |
5554 | submit_stripe_bio(root, bbio, bio, | |
5555 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
5556 | async_submit); | |
8790d502 CM |
5557 | dev_nr++; |
5558 | } | |
0b86a832 CM |
5559 | return 0; |
5560 | } | |
5561 | ||
aa1b8cd4 | 5562 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 5563 | u8 *uuid, u8 *fsid) |
0b86a832 | 5564 | { |
2b82032c YZ |
5565 | struct btrfs_device *device; |
5566 | struct btrfs_fs_devices *cur_devices; | |
5567 | ||
aa1b8cd4 | 5568 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
5569 | while (cur_devices) { |
5570 | if (!fsid || | |
5571 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
5572 | device = __find_device(&cur_devices->devices, | |
5573 | devid, uuid); | |
5574 | if (device) | |
5575 | return device; | |
5576 | } | |
5577 | cur_devices = cur_devices->seed; | |
5578 | } | |
5579 | return NULL; | |
0b86a832 CM |
5580 | } |
5581 | ||
dfe25020 CM |
5582 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5583 | u64 devid, u8 *dev_uuid) | |
5584 | { | |
5585 | struct btrfs_device *device; | |
5586 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
5587 | ||
12bd2fc0 ID |
5588 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
5589 | if (IS_ERR(device)) | |
7cbd8a83 | 5590 | return NULL; |
12bd2fc0 ID |
5591 | |
5592 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 5593 | device->fs_devices = fs_devices; |
dfe25020 | 5594 | fs_devices->num_devices++; |
12bd2fc0 ID |
5595 | |
5596 | device->missing = 1; | |
cd02dca5 | 5597 | fs_devices->missing_devices++; |
12bd2fc0 | 5598 | |
dfe25020 CM |
5599 | return device; |
5600 | } | |
5601 | ||
12bd2fc0 ID |
5602 | /** |
5603 | * btrfs_alloc_device - allocate struct btrfs_device | |
5604 | * @fs_info: used only for generating a new devid, can be NULL if | |
5605 | * devid is provided (i.e. @devid != NULL). | |
5606 | * @devid: a pointer to devid for this device. If NULL a new devid | |
5607 | * is generated. | |
5608 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
5609 | * is generated. | |
5610 | * | |
5611 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
5612 | * on error. Returned struct is not linked onto any lists and can be | |
5613 | * destroyed with kfree() right away. | |
5614 | */ | |
5615 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
5616 | const u64 *devid, | |
5617 | const u8 *uuid) | |
5618 | { | |
5619 | struct btrfs_device *dev; | |
5620 | u64 tmp; | |
5621 | ||
5622 | if (!devid && !fs_info) { | |
5623 | WARN_ON(1); | |
5624 | return ERR_PTR(-EINVAL); | |
5625 | } | |
5626 | ||
5627 | dev = __alloc_device(); | |
5628 | if (IS_ERR(dev)) | |
5629 | return dev; | |
5630 | ||
5631 | if (devid) | |
5632 | tmp = *devid; | |
5633 | else { | |
5634 | int ret; | |
5635 | ||
5636 | ret = find_next_devid(fs_info, &tmp); | |
5637 | if (ret) { | |
5638 | kfree(dev); | |
5639 | return ERR_PTR(ret); | |
5640 | } | |
5641 | } | |
5642 | dev->devid = tmp; | |
5643 | ||
5644 | if (uuid) | |
5645 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
5646 | else | |
5647 | generate_random_uuid(dev->uuid); | |
5648 | ||
5649 | dev->work.func = pending_bios_fn; | |
5650 | ||
5651 | return dev; | |
5652 | } | |
5653 | ||
0b86a832 CM |
5654 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
5655 | struct extent_buffer *leaf, | |
5656 | struct btrfs_chunk *chunk) | |
5657 | { | |
5658 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5659 | struct map_lookup *map; | |
5660 | struct extent_map *em; | |
5661 | u64 logical; | |
5662 | u64 length; | |
5663 | u64 devid; | |
a443755f | 5664 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 5665 | int num_stripes; |
0b86a832 | 5666 | int ret; |
593060d7 | 5667 | int i; |
0b86a832 | 5668 | |
e17cade2 CM |
5669 | logical = key->offset; |
5670 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 5671 | |
890871be | 5672 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 5673 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 5674 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
5675 | |
5676 | /* already mapped? */ | |
5677 | if (em && em->start <= logical && em->start + em->len > logical) { | |
5678 | free_extent_map(em); | |
0b86a832 CM |
5679 | return 0; |
5680 | } else if (em) { | |
5681 | free_extent_map(em); | |
5682 | } | |
0b86a832 | 5683 | |
172ddd60 | 5684 | em = alloc_extent_map(); |
0b86a832 CM |
5685 | if (!em) |
5686 | return -ENOMEM; | |
593060d7 CM |
5687 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
5688 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
5689 | if (!map) { |
5690 | free_extent_map(em); | |
5691 | return -ENOMEM; | |
5692 | } | |
5693 | ||
5694 | em->bdev = (struct block_device *)map; | |
5695 | em->start = logical; | |
5696 | em->len = length; | |
70c8a91c | 5697 | em->orig_start = 0; |
0b86a832 | 5698 | em->block_start = 0; |
c8b97818 | 5699 | em->block_len = em->len; |
0b86a832 | 5700 | |
593060d7 CM |
5701 | map->num_stripes = num_stripes; |
5702 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
5703 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
5704 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
5705 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
5706 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 5707 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
5708 | for (i = 0; i < num_stripes; i++) { |
5709 | map->stripes[i].physical = | |
5710 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
5711 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
5712 | read_extent_buffer(leaf, uuid, (unsigned long) |
5713 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
5714 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
5715 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
5716 | uuid, NULL); | |
dfe25020 | 5717 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
5718 | kfree(map); |
5719 | free_extent_map(em); | |
5720 | return -EIO; | |
5721 | } | |
dfe25020 CM |
5722 | if (!map->stripes[i].dev) { |
5723 | map->stripes[i].dev = | |
5724 | add_missing_dev(root, devid, uuid); | |
5725 | if (!map->stripes[i].dev) { | |
5726 | kfree(map); | |
5727 | free_extent_map(em); | |
5728 | return -EIO; | |
5729 | } | |
5730 | } | |
5731 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
5732 | } |
5733 | ||
890871be | 5734 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 5735 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 5736 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 5737 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
5738 | free_extent_map(em); |
5739 | ||
5740 | return 0; | |
5741 | } | |
5742 | ||
143bede5 | 5743 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
5744 | struct btrfs_dev_item *dev_item, |
5745 | struct btrfs_device *device) | |
5746 | { | |
5747 | unsigned long ptr; | |
0b86a832 CM |
5748 | |
5749 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
5750 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
5751 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
5752 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
5753 | device->type = btrfs_device_type(leaf, dev_item); | |
5754 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
5755 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
5756 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 5757 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 5758 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 5759 | |
410ba3a2 | 5760 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 5761 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
5762 | } |
5763 | ||
2b82032c YZ |
5764 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
5765 | { | |
5766 | struct btrfs_fs_devices *fs_devices; | |
5767 | int ret; | |
5768 | ||
b367e47f | 5769 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
5770 | |
5771 | fs_devices = root->fs_info->fs_devices->seed; | |
5772 | while (fs_devices) { | |
5773 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
5774 | ret = 0; | |
5775 | goto out; | |
5776 | } | |
5777 | fs_devices = fs_devices->seed; | |
5778 | } | |
5779 | ||
5780 | fs_devices = find_fsid(fsid); | |
5781 | if (!fs_devices) { | |
5782 | ret = -ENOENT; | |
5783 | goto out; | |
5784 | } | |
e4404d6e YZ |
5785 | |
5786 | fs_devices = clone_fs_devices(fs_devices); | |
5787 | if (IS_ERR(fs_devices)) { | |
5788 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
5789 | goto out; |
5790 | } | |
5791 | ||
97288f2c | 5792 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 5793 | root->fs_info->bdev_holder); |
48d28232 JL |
5794 | if (ret) { |
5795 | free_fs_devices(fs_devices); | |
2b82032c | 5796 | goto out; |
48d28232 | 5797 | } |
2b82032c YZ |
5798 | |
5799 | if (!fs_devices->seeding) { | |
5800 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 5801 | free_fs_devices(fs_devices); |
2b82032c YZ |
5802 | ret = -EINVAL; |
5803 | goto out; | |
5804 | } | |
5805 | ||
5806 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
5807 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 5808 | out: |
2b82032c YZ |
5809 | return ret; |
5810 | } | |
5811 | ||
0d81ba5d | 5812 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
5813 | struct extent_buffer *leaf, |
5814 | struct btrfs_dev_item *dev_item) | |
5815 | { | |
5816 | struct btrfs_device *device; | |
5817 | u64 devid; | |
5818 | int ret; | |
2b82032c | 5819 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
5820 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
5821 | ||
0b86a832 | 5822 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 5823 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 5824 | BTRFS_UUID_SIZE); |
1473b24e | 5825 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
5826 | BTRFS_UUID_SIZE); |
5827 | ||
5828 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5829 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 5830 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 5831 | return ret; |
2b82032c YZ |
5832 | } |
5833 | ||
aa1b8cd4 | 5834 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
2b82032c | 5835 | if (!device || !device->bdev) { |
e4404d6e | 5836 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
5837 | return -EIO; |
5838 | ||
5839 | if (!device) { | |
c1c9ff7c | 5840 | btrfs_warn(root->fs_info, "devid %llu missing", devid); |
2b82032c YZ |
5841 | device = add_missing_dev(root, devid, dev_uuid); |
5842 | if (!device) | |
5843 | return -ENOMEM; | |
cd02dca5 CM |
5844 | } else if (!device->missing) { |
5845 | /* | |
5846 | * this happens when a device that was properly setup | |
5847 | * in the device info lists suddenly goes bad. | |
5848 | * device->bdev is NULL, and so we have to set | |
5849 | * device->missing to one here | |
5850 | */ | |
5851 | root->fs_info->fs_devices->missing_devices++; | |
5852 | device->missing = 1; | |
2b82032c YZ |
5853 | } |
5854 | } | |
5855 | ||
5856 | if (device->fs_devices != root->fs_info->fs_devices) { | |
5857 | BUG_ON(device->writeable); | |
5858 | if (device->generation != | |
5859 | btrfs_device_generation(leaf, dev_item)) | |
5860 | return -EINVAL; | |
6324fbf3 | 5861 | } |
0b86a832 CM |
5862 | |
5863 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 5864 | device->in_fs_metadata = 1; |
63a212ab | 5865 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 5866 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
5867 | spin_lock(&root->fs_info->free_chunk_lock); |
5868 | root->fs_info->free_chunk_space += device->total_bytes - | |
5869 | device->bytes_used; | |
5870 | spin_unlock(&root->fs_info->free_chunk_lock); | |
5871 | } | |
0b86a832 | 5872 | ret = 0; |
0b86a832 CM |
5873 | return ret; |
5874 | } | |
5875 | ||
e4404d6e | 5876 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 5877 | { |
6c41761f | 5878 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 5879 | struct extent_buffer *sb; |
0b86a832 | 5880 | struct btrfs_disk_key *disk_key; |
0b86a832 | 5881 | struct btrfs_chunk *chunk; |
84eed90f CM |
5882 | u8 *ptr; |
5883 | unsigned long sb_ptr; | |
5884 | int ret = 0; | |
0b86a832 CM |
5885 | u32 num_stripes; |
5886 | u32 array_size; | |
5887 | u32 len = 0; | |
0b86a832 | 5888 | u32 cur; |
84eed90f | 5889 | struct btrfs_key key; |
0b86a832 | 5890 | |
e4404d6e | 5891 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
5892 | BTRFS_SUPER_INFO_SIZE); |
5893 | if (!sb) | |
5894 | return -ENOMEM; | |
5895 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 5896 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
5897 | /* |
5898 | * The sb extent buffer is artifical and just used to read the system array. | |
5899 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
5900 | * pages up-to-date when the page is larger: extent does not cover the | |
5901 | * whole page and consequently check_page_uptodate does not find all | |
5902 | * the page's extents up-to-date (the hole beyond sb), | |
5903 | * write_extent_buffer then triggers a WARN_ON. | |
5904 | * | |
5905 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
5906 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
5907 | * to silence the warning eg. on PowerPC 64. | |
5908 | */ | |
5909 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 5910 | SetPageUptodate(sb->pages[0]); |
4008c04a | 5911 | |
a061fc8d | 5912 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
5913 | array_size = btrfs_super_sys_array_size(super_copy); |
5914 | ||
0b86a832 CM |
5915 | ptr = super_copy->sys_chunk_array; |
5916 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
5917 | cur = 0; | |
5918 | ||
5919 | while (cur < array_size) { | |
5920 | disk_key = (struct btrfs_disk_key *)ptr; | |
5921 | btrfs_disk_key_to_cpu(&key, disk_key); | |
5922 | ||
a061fc8d | 5923 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
5924 | sb_ptr += len; |
5925 | cur += len; | |
5926 | ||
0d81ba5d | 5927 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 5928 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 5929 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
5930 | if (ret) |
5931 | break; | |
0b86a832 CM |
5932 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
5933 | len = btrfs_chunk_item_size(num_stripes); | |
5934 | } else { | |
84eed90f CM |
5935 | ret = -EIO; |
5936 | break; | |
0b86a832 CM |
5937 | } |
5938 | ptr += len; | |
5939 | sb_ptr += len; | |
5940 | cur += len; | |
5941 | } | |
a061fc8d | 5942 | free_extent_buffer(sb); |
84eed90f | 5943 | return ret; |
0b86a832 CM |
5944 | } |
5945 | ||
5946 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
5947 | { | |
5948 | struct btrfs_path *path; | |
5949 | struct extent_buffer *leaf; | |
5950 | struct btrfs_key key; | |
5951 | struct btrfs_key found_key; | |
5952 | int ret; | |
5953 | int slot; | |
5954 | ||
5955 | root = root->fs_info->chunk_root; | |
5956 | ||
5957 | path = btrfs_alloc_path(); | |
5958 | if (!path) | |
5959 | return -ENOMEM; | |
5960 | ||
b367e47f LZ |
5961 | mutex_lock(&uuid_mutex); |
5962 | lock_chunks(root); | |
5963 | ||
395927a9 FDBM |
5964 | /* |
5965 | * Read all device items, and then all the chunk items. All | |
5966 | * device items are found before any chunk item (their object id | |
5967 | * is smaller than the lowest possible object id for a chunk | |
5968 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
5969 | */ |
5970 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
5971 | key.offset = 0; | |
5972 | key.type = 0; | |
0b86a832 | 5973 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
5974 | if (ret < 0) |
5975 | goto error; | |
d397712b | 5976 | while (1) { |
0b86a832 CM |
5977 | leaf = path->nodes[0]; |
5978 | slot = path->slots[0]; | |
5979 | if (slot >= btrfs_header_nritems(leaf)) { | |
5980 | ret = btrfs_next_leaf(root, path); | |
5981 | if (ret == 0) | |
5982 | continue; | |
5983 | if (ret < 0) | |
5984 | goto error; | |
5985 | break; | |
5986 | } | |
5987 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
5988 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
5989 | struct btrfs_dev_item *dev_item; | |
5990 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 5991 | struct btrfs_dev_item); |
395927a9 FDBM |
5992 | ret = read_one_dev(root, leaf, dev_item); |
5993 | if (ret) | |
5994 | goto error; | |
0b86a832 CM |
5995 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
5996 | struct btrfs_chunk *chunk; | |
5997 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
5998 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
5999 | if (ret) |
6000 | goto error; | |
0b86a832 CM |
6001 | } |
6002 | path->slots[0]++; | |
6003 | } | |
0b86a832 CM |
6004 | ret = 0; |
6005 | error: | |
b367e47f LZ |
6006 | unlock_chunks(root); |
6007 | mutex_unlock(&uuid_mutex); | |
6008 | ||
2b82032c | 6009 | btrfs_free_path(path); |
0b86a832 CM |
6010 | return ret; |
6011 | } | |
442a4f63 | 6012 | |
cb517eab MX |
6013 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6014 | { | |
6015 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6016 | struct btrfs_device *device; | |
6017 | ||
6018 | mutex_lock(&fs_devices->device_list_mutex); | |
6019 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6020 | device->dev_root = fs_info->dev_root; | |
6021 | mutex_unlock(&fs_devices->device_list_mutex); | |
6022 | } | |
6023 | ||
733f4fbb SB |
6024 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6025 | { | |
6026 | int i; | |
6027 | ||
6028 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6029 | btrfs_dev_stat_reset(dev, i); | |
6030 | } | |
6031 | ||
6032 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6033 | { | |
6034 | struct btrfs_key key; | |
6035 | struct btrfs_key found_key; | |
6036 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6037 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6038 | struct extent_buffer *eb; | |
6039 | int slot; | |
6040 | int ret = 0; | |
6041 | struct btrfs_device *device; | |
6042 | struct btrfs_path *path = NULL; | |
6043 | int i; | |
6044 | ||
6045 | path = btrfs_alloc_path(); | |
6046 | if (!path) { | |
6047 | ret = -ENOMEM; | |
6048 | goto out; | |
6049 | } | |
6050 | ||
6051 | mutex_lock(&fs_devices->device_list_mutex); | |
6052 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6053 | int item_size; | |
6054 | struct btrfs_dev_stats_item *ptr; | |
6055 | ||
6056 | key.objectid = 0; | |
6057 | key.type = BTRFS_DEV_STATS_KEY; | |
6058 | key.offset = device->devid; | |
6059 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6060 | if (ret) { | |
733f4fbb SB |
6061 | __btrfs_reset_dev_stats(device); |
6062 | device->dev_stats_valid = 1; | |
6063 | btrfs_release_path(path); | |
6064 | continue; | |
6065 | } | |
6066 | slot = path->slots[0]; | |
6067 | eb = path->nodes[0]; | |
6068 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6069 | item_size = btrfs_item_size_nr(eb, slot); | |
6070 | ||
6071 | ptr = btrfs_item_ptr(eb, slot, | |
6072 | struct btrfs_dev_stats_item); | |
6073 | ||
6074 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6075 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6076 | btrfs_dev_stat_set(device, i, | |
6077 | btrfs_dev_stats_value(eb, ptr, i)); | |
6078 | else | |
6079 | btrfs_dev_stat_reset(device, i); | |
6080 | } | |
6081 | ||
6082 | device->dev_stats_valid = 1; | |
6083 | btrfs_dev_stat_print_on_load(device); | |
6084 | btrfs_release_path(path); | |
6085 | } | |
6086 | mutex_unlock(&fs_devices->device_list_mutex); | |
6087 | ||
6088 | out: | |
6089 | btrfs_free_path(path); | |
6090 | return ret < 0 ? ret : 0; | |
6091 | } | |
6092 | ||
6093 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6094 | struct btrfs_root *dev_root, | |
6095 | struct btrfs_device *device) | |
6096 | { | |
6097 | struct btrfs_path *path; | |
6098 | struct btrfs_key key; | |
6099 | struct extent_buffer *eb; | |
6100 | struct btrfs_dev_stats_item *ptr; | |
6101 | int ret; | |
6102 | int i; | |
6103 | ||
6104 | key.objectid = 0; | |
6105 | key.type = BTRFS_DEV_STATS_KEY; | |
6106 | key.offset = device->devid; | |
6107 | ||
6108 | path = btrfs_alloc_path(); | |
6109 | BUG_ON(!path); | |
6110 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6111 | if (ret < 0) { | |
606686ee JB |
6112 | printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n", |
6113 | ret, rcu_str_deref(device->name)); | |
733f4fbb SB |
6114 | goto out; |
6115 | } | |
6116 | ||
6117 | if (ret == 0 && | |
6118 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6119 | /* need to delete old one and insert a new one */ | |
6120 | ret = btrfs_del_item(trans, dev_root, path); | |
6121 | if (ret != 0) { | |
606686ee JB |
6122 | printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n", |
6123 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
6124 | goto out; |
6125 | } | |
6126 | ret = 1; | |
6127 | } | |
6128 | ||
6129 | if (ret == 1) { | |
6130 | /* need to insert a new item */ | |
6131 | btrfs_release_path(path); | |
6132 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6133 | &key, sizeof(*ptr)); | |
6134 | if (ret < 0) { | |
606686ee JB |
6135 | printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n", |
6136 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
6137 | goto out; |
6138 | } | |
6139 | } | |
6140 | ||
6141 | eb = path->nodes[0]; | |
6142 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6143 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6144 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6145 | btrfs_dev_stat_read(device, i)); | |
6146 | btrfs_mark_buffer_dirty(eb); | |
6147 | ||
6148 | out: | |
6149 | btrfs_free_path(path); | |
6150 | return ret; | |
6151 | } | |
6152 | ||
6153 | /* | |
6154 | * called from commit_transaction. Writes all changed device stats to disk. | |
6155 | */ | |
6156 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6157 | struct btrfs_fs_info *fs_info) | |
6158 | { | |
6159 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6160 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6161 | struct btrfs_device *device; | |
6162 | int ret = 0; | |
6163 | ||
6164 | mutex_lock(&fs_devices->device_list_mutex); | |
6165 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6166 | if (!device->dev_stats_valid || !device->dev_stats_dirty) | |
6167 | continue; | |
6168 | ||
6169 | ret = update_dev_stat_item(trans, dev_root, device); | |
6170 | if (!ret) | |
6171 | device->dev_stats_dirty = 0; | |
6172 | } | |
6173 | mutex_unlock(&fs_devices->device_list_mutex); | |
6174 | ||
6175 | return ret; | |
6176 | } | |
6177 | ||
442a4f63 SB |
6178 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
6179 | { | |
6180 | btrfs_dev_stat_inc(dev, index); | |
6181 | btrfs_dev_stat_print_on_error(dev); | |
6182 | } | |
6183 | ||
48a3b636 | 6184 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 6185 | { |
733f4fbb SB |
6186 | if (!dev->dev_stats_valid) |
6187 | return; | |
606686ee | 6188 | printk_ratelimited_in_rcu(KERN_ERR |
442a4f63 | 6189 | "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
606686ee | 6190 | rcu_str_deref(dev->name), |
442a4f63 SB |
6191 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6192 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6193 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6194 | btrfs_dev_stat_read(dev, | |
6195 | BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6196 | btrfs_dev_stat_read(dev, | |
6197 | BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6198 | } | |
c11d2c23 | 6199 | |
733f4fbb SB |
6200 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
6201 | { | |
a98cdb85 SB |
6202 | int i; |
6203 | ||
6204 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6205 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6206 | break; | |
6207 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6208 | return; /* all values == 0, suppress message */ | |
6209 | ||
606686ee JB |
6210 | printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
6211 | rcu_str_deref(dev->name), | |
733f4fbb SB |
6212 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6213 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6214 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6215 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6216 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6217 | } | |
6218 | ||
c11d2c23 | 6219 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 6220 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
6221 | { |
6222 | struct btrfs_device *dev; | |
6223 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6224 | int i; | |
6225 | ||
6226 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 6227 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
6228 | mutex_unlock(&fs_devices->device_list_mutex); |
6229 | ||
6230 | if (!dev) { | |
6231 | printk(KERN_WARNING | |
6232 | "btrfs: get dev_stats failed, device not found\n"); | |
6233 | return -ENODEV; | |
733f4fbb SB |
6234 | } else if (!dev->dev_stats_valid) { |
6235 | printk(KERN_WARNING | |
6236 | "btrfs: get dev_stats failed, not yet valid\n"); | |
6237 | return -ENODEV; | |
b27f7c0c | 6238 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
6239 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
6240 | if (stats->nr_items > i) | |
6241 | stats->values[i] = | |
6242 | btrfs_dev_stat_read_and_reset(dev, i); | |
6243 | else | |
6244 | btrfs_dev_stat_reset(dev, i); | |
6245 | } | |
6246 | } else { | |
6247 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6248 | if (stats->nr_items > i) | |
6249 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6250 | } | |
6251 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6252 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6253 | return 0; | |
6254 | } | |
a8a6dab7 SB |
6255 | |
6256 | int btrfs_scratch_superblock(struct btrfs_device *device) | |
6257 | { | |
6258 | struct buffer_head *bh; | |
6259 | struct btrfs_super_block *disk_super; | |
6260 | ||
6261 | bh = btrfs_read_dev_super(device->bdev); | |
6262 | if (!bh) | |
6263 | return -EINVAL; | |
6264 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6265 | ||
6266 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6267 | set_buffer_dirty(bh); | |
6268 | sync_dirty_buffer(bh); | |
6269 | brelse(bh); | |
6270 | ||
6271 | return 0; | |
6272 | } |