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 DW |
28 | #include <linux/raid/pq.h> |
29 | #include <asm/div64.h> | |
4b4e25f2 | 30 | #include "compat.h" |
0b86a832 CM |
31 | #include "ctree.h" |
32 | #include "extent_map.h" | |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "print-tree.h" | |
36 | #include "volumes.h" | |
53b381b3 | 37 | #include "raid56.h" |
8b712842 | 38 | #include "async-thread.h" |
21adbd5c | 39 | #include "check-integrity.h" |
606686ee | 40 | #include "rcu-string.h" |
3fed40cc | 41 | #include "math.h" |
8dabb742 | 42 | #include "dev-replace.h" |
0b86a832 | 43 | |
2b82032c YZ |
44 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
45 | struct btrfs_root *root, | |
46 | struct btrfs_device *device); | |
47 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb SB |
48 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
49 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); | |
2b82032c | 50 | |
8a4b83cc CM |
51 | static DEFINE_MUTEX(uuid_mutex); |
52 | static LIST_HEAD(fs_uuids); | |
53 | ||
7d9eb12c CM |
54 | static void lock_chunks(struct btrfs_root *root) |
55 | { | |
7d9eb12c CM |
56 | mutex_lock(&root->fs_info->chunk_mutex); |
57 | } | |
58 | ||
59 | static void unlock_chunks(struct btrfs_root *root) | |
60 | { | |
7d9eb12c CM |
61 | mutex_unlock(&root->fs_info->chunk_mutex); |
62 | } | |
63 | ||
e4404d6e YZ |
64 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
65 | { | |
66 | struct btrfs_device *device; | |
67 | WARN_ON(fs_devices->opened); | |
68 | while (!list_empty(&fs_devices->devices)) { | |
69 | device = list_entry(fs_devices->devices.next, | |
70 | struct btrfs_device, dev_list); | |
71 | list_del(&device->dev_list); | |
606686ee | 72 | rcu_string_free(device->name); |
e4404d6e YZ |
73 | kfree(device); |
74 | } | |
75 | kfree(fs_devices); | |
76 | } | |
77 | ||
b8b8ff59 LC |
78 | static void btrfs_kobject_uevent(struct block_device *bdev, |
79 | enum kobject_action action) | |
80 | { | |
81 | int ret; | |
82 | ||
83 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
84 | if (ret) | |
85 | pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n", | |
86 | action, | |
87 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
88 | &disk_to_dev(bdev->bd_disk)->kobj); | |
89 | } | |
90 | ||
143bede5 | 91 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
92 | { |
93 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 94 | |
2b82032c YZ |
95 | while (!list_empty(&fs_uuids)) { |
96 | fs_devices = list_entry(fs_uuids.next, | |
97 | struct btrfs_fs_devices, list); | |
98 | list_del(&fs_devices->list); | |
e4404d6e | 99 | free_fs_devices(fs_devices); |
8a4b83cc | 100 | } |
8a4b83cc CM |
101 | } |
102 | ||
a1b32a59 CM |
103 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
104 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
105 | { |
106 | struct btrfs_device *dev; | |
8a4b83cc | 107 | |
c6e30871 | 108 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 109 | if (dev->devid == devid && |
8f18cf13 | 110 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 111 | return dev; |
a443755f | 112 | } |
8a4b83cc CM |
113 | } |
114 | return NULL; | |
115 | } | |
116 | ||
a1b32a59 | 117 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 118 | { |
8a4b83cc CM |
119 | struct btrfs_fs_devices *fs_devices; |
120 | ||
c6e30871 | 121 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
122 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
123 | return fs_devices; | |
124 | } | |
125 | return NULL; | |
126 | } | |
127 | ||
beaf8ab3 SB |
128 | static int |
129 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
130 | int flush, struct block_device **bdev, | |
131 | struct buffer_head **bh) | |
132 | { | |
133 | int ret; | |
134 | ||
135 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
136 | ||
137 | if (IS_ERR(*bdev)) { | |
138 | ret = PTR_ERR(*bdev); | |
139 | printk(KERN_INFO "btrfs: open %s failed\n", device_path); | |
140 | goto error; | |
141 | } | |
142 | ||
143 | if (flush) | |
144 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
145 | ret = set_blocksize(*bdev, 4096); | |
146 | if (ret) { | |
147 | blkdev_put(*bdev, flags); | |
148 | goto error; | |
149 | } | |
150 | invalidate_bdev(*bdev); | |
151 | *bh = btrfs_read_dev_super(*bdev); | |
152 | if (!*bh) { | |
153 | ret = -EINVAL; | |
154 | blkdev_put(*bdev, flags); | |
155 | goto error; | |
156 | } | |
157 | ||
158 | return 0; | |
159 | ||
160 | error: | |
161 | *bdev = NULL; | |
162 | *bh = NULL; | |
163 | return ret; | |
164 | } | |
165 | ||
ffbd517d CM |
166 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
167 | struct bio *head, struct bio *tail) | |
168 | { | |
169 | ||
170 | struct bio *old_head; | |
171 | ||
172 | old_head = pending_bios->head; | |
173 | pending_bios->head = head; | |
174 | if (pending_bios->tail) | |
175 | tail->bi_next = old_head; | |
176 | else | |
177 | pending_bios->tail = tail; | |
178 | } | |
179 | ||
8b712842 CM |
180 | /* |
181 | * we try to collect pending bios for a device so we don't get a large | |
182 | * number of procs sending bios down to the same device. This greatly | |
183 | * improves the schedulers ability to collect and merge the bios. | |
184 | * | |
185 | * But, it also turns into a long list of bios to process and that is sure | |
186 | * to eventually make the worker thread block. The solution here is to | |
187 | * make some progress and then put this work struct back at the end of | |
188 | * the list if the block device is congested. This way, multiple devices | |
189 | * can make progress from a single worker thread. | |
190 | */ | |
143bede5 | 191 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
192 | { |
193 | struct bio *pending; | |
194 | struct backing_dev_info *bdi; | |
b64a2851 | 195 | struct btrfs_fs_info *fs_info; |
ffbd517d | 196 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
197 | struct bio *tail; |
198 | struct bio *cur; | |
199 | int again = 0; | |
ffbd517d | 200 | unsigned long num_run; |
d644d8a1 | 201 | unsigned long batch_run = 0; |
b64a2851 | 202 | unsigned long limit; |
b765ead5 | 203 | unsigned long last_waited = 0; |
d84275c9 | 204 | int force_reg = 0; |
0e588859 | 205 | int sync_pending = 0; |
211588ad CM |
206 | struct blk_plug plug; |
207 | ||
208 | /* | |
209 | * this function runs all the bios we've collected for | |
210 | * a particular device. We don't want to wander off to | |
211 | * another device without first sending all of these down. | |
212 | * So, setup a plug here and finish it off before we return | |
213 | */ | |
214 | blk_start_plug(&plug); | |
8b712842 | 215 | |
bedf762b | 216 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
217 | fs_info = device->dev_root->fs_info; |
218 | limit = btrfs_async_submit_limit(fs_info); | |
219 | limit = limit * 2 / 3; | |
220 | ||
8b712842 CM |
221 | loop: |
222 | spin_lock(&device->io_lock); | |
223 | ||
a6837051 | 224 | loop_lock: |
d84275c9 | 225 | num_run = 0; |
ffbd517d | 226 | |
8b712842 CM |
227 | /* take all the bios off the list at once and process them |
228 | * later on (without the lock held). But, remember the | |
229 | * tail and other pointers so the bios can be properly reinserted | |
230 | * into the list if we hit congestion | |
231 | */ | |
d84275c9 | 232 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 233 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
234 | force_reg = 1; |
235 | } else { | |
ffbd517d | 236 | pending_bios = &device->pending_bios; |
d84275c9 CM |
237 | force_reg = 0; |
238 | } | |
ffbd517d CM |
239 | |
240 | pending = pending_bios->head; | |
241 | tail = pending_bios->tail; | |
8b712842 | 242 | WARN_ON(pending && !tail); |
8b712842 CM |
243 | |
244 | /* | |
245 | * if pending was null this time around, no bios need processing | |
246 | * at all and we can stop. Otherwise it'll loop back up again | |
247 | * and do an additional check so no bios are missed. | |
248 | * | |
249 | * device->running_pending is used to synchronize with the | |
250 | * schedule_bio code. | |
251 | */ | |
ffbd517d CM |
252 | if (device->pending_sync_bios.head == NULL && |
253 | device->pending_bios.head == NULL) { | |
8b712842 CM |
254 | again = 0; |
255 | device->running_pending = 0; | |
ffbd517d CM |
256 | } else { |
257 | again = 1; | |
258 | device->running_pending = 1; | |
8b712842 | 259 | } |
ffbd517d CM |
260 | |
261 | pending_bios->head = NULL; | |
262 | pending_bios->tail = NULL; | |
263 | ||
8b712842 CM |
264 | spin_unlock(&device->io_lock); |
265 | ||
d397712b | 266 | while (pending) { |
ffbd517d CM |
267 | |
268 | rmb(); | |
d84275c9 CM |
269 | /* we want to work on both lists, but do more bios on the |
270 | * sync list than the regular list | |
271 | */ | |
272 | if ((num_run > 32 && | |
273 | pending_bios != &device->pending_sync_bios && | |
274 | device->pending_sync_bios.head) || | |
275 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
276 | device->pending_bios.head)) { | |
ffbd517d CM |
277 | spin_lock(&device->io_lock); |
278 | requeue_list(pending_bios, pending, tail); | |
279 | goto loop_lock; | |
280 | } | |
281 | ||
8b712842 CM |
282 | cur = pending; |
283 | pending = pending->bi_next; | |
284 | cur->bi_next = NULL; | |
b64a2851 | 285 | |
66657b31 | 286 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
287 | waitqueue_active(&fs_info->async_submit_wait)) |
288 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
289 | |
290 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 291 | |
2ab1ba68 CM |
292 | /* |
293 | * if we're doing the sync list, record that our | |
294 | * plug has some sync requests on it | |
295 | * | |
296 | * If we're doing the regular list and there are | |
297 | * sync requests sitting around, unplug before | |
298 | * we add more | |
299 | */ | |
300 | if (pending_bios == &device->pending_sync_bios) { | |
301 | sync_pending = 1; | |
302 | } else if (sync_pending) { | |
303 | blk_finish_plug(&plug); | |
304 | blk_start_plug(&plug); | |
305 | sync_pending = 0; | |
306 | } | |
307 | ||
21adbd5c | 308 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
309 | num_run++; |
310 | batch_run++; | |
7eaceacc | 311 | if (need_resched()) |
ffbd517d | 312 | cond_resched(); |
8b712842 CM |
313 | |
314 | /* | |
315 | * we made progress, there is more work to do and the bdi | |
316 | * is now congested. Back off and let other work structs | |
317 | * run instead | |
318 | */ | |
57fd5a5f | 319 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 320 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 321 | struct io_context *ioc; |
8b712842 | 322 | |
b765ead5 CM |
323 | ioc = current->io_context; |
324 | ||
325 | /* | |
326 | * the main goal here is that we don't want to | |
327 | * block if we're going to be able to submit | |
328 | * more requests without blocking. | |
329 | * | |
330 | * This code does two great things, it pokes into | |
331 | * the elevator code from a filesystem _and_ | |
332 | * it makes assumptions about how batching works. | |
333 | */ | |
334 | if (ioc && ioc->nr_batch_requests > 0 && | |
335 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
336 | (last_waited == 0 || | |
337 | ioc->last_waited == last_waited)) { | |
338 | /* | |
339 | * we want to go through our batch of | |
340 | * requests and stop. So, we copy out | |
341 | * the ioc->last_waited time and test | |
342 | * against it before looping | |
343 | */ | |
344 | last_waited = ioc->last_waited; | |
7eaceacc | 345 | if (need_resched()) |
ffbd517d | 346 | cond_resched(); |
b765ead5 CM |
347 | continue; |
348 | } | |
8b712842 | 349 | spin_lock(&device->io_lock); |
ffbd517d | 350 | requeue_list(pending_bios, pending, tail); |
a6837051 | 351 | device->running_pending = 1; |
8b712842 CM |
352 | |
353 | spin_unlock(&device->io_lock); | |
354 | btrfs_requeue_work(&device->work); | |
355 | goto done; | |
356 | } | |
d85c8a6f CM |
357 | /* unplug every 64 requests just for good measure */ |
358 | if (batch_run % 64 == 0) { | |
359 | blk_finish_plug(&plug); | |
360 | blk_start_plug(&plug); | |
361 | sync_pending = 0; | |
362 | } | |
8b712842 | 363 | } |
ffbd517d | 364 | |
51684082 CM |
365 | cond_resched(); |
366 | if (again) | |
367 | goto loop; | |
368 | ||
369 | spin_lock(&device->io_lock); | |
370 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
371 | goto loop_lock; | |
372 | spin_unlock(&device->io_lock); | |
373 | ||
8b712842 | 374 | done: |
211588ad | 375 | blk_finish_plug(&plug); |
8b712842 CM |
376 | } |
377 | ||
b2950863 | 378 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
379 | { |
380 | struct btrfs_device *device; | |
381 | ||
382 | device = container_of(work, struct btrfs_device, work); | |
383 | run_scheduled_bios(device); | |
384 | } | |
385 | ||
a1b32a59 | 386 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
387 | struct btrfs_super_block *disk_super, |
388 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
389 | { | |
390 | struct btrfs_device *device; | |
391 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 392 | struct rcu_string *name; |
8a4b83cc CM |
393 | u64 found_transid = btrfs_super_generation(disk_super); |
394 | ||
395 | fs_devices = find_fsid(disk_super->fsid); | |
396 | if (!fs_devices) { | |
515dc322 | 397 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
398 | if (!fs_devices) |
399 | return -ENOMEM; | |
400 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 401 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
402 | list_add(&fs_devices->list, &fs_uuids); |
403 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
404 | fs_devices->latest_devid = devid; | |
405 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 406 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
407 | device = NULL; |
408 | } else { | |
a443755f CM |
409 | device = __find_device(&fs_devices->devices, devid, |
410 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
411 | } |
412 | if (!device) { | |
2b82032c YZ |
413 | if (fs_devices->opened) |
414 | return -EBUSY; | |
415 | ||
8a4b83cc CM |
416 | device = kzalloc(sizeof(*device), GFP_NOFS); |
417 | if (!device) { | |
418 | /* we can safely leave the fs_devices entry around */ | |
419 | return -ENOMEM; | |
420 | } | |
421 | device->devid = devid; | |
733f4fbb | 422 | device->dev_stats_valid = 0; |
8b712842 | 423 | device->work.func = pending_bios_fn; |
a443755f CM |
424 | memcpy(device->uuid, disk_super->dev_item.uuid, |
425 | BTRFS_UUID_SIZE); | |
b248a415 | 426 | spin_lock_init(&device->io_lock); |
606686ee JB |
427 | |
428 | name = rcu_string_strdup(path, GFP_NOFS); | |
429 | if (!name) { | |
8a4b83cc CM |
430 | kfree(device); |
431 | return -ENOMEM; | |
432 | } | |
606686ee | 433 | rcu_assign_pointer(device->name, name); |
2b82032c | 434 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 | 435 | |
90519d66 AJ |
436 | /* init readahead state */ |
437 | spin_lock_init(&device->reada_lock); | |
438 | device->reada_curr_zone = NULL; | |
439 | atomic_set(&device->reada_in_flight, 0); | |
440 | device->reada_next = 0; | |
441 | INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
442 | INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
443 | ||
e5e9a520 | 444 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 445 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
446 | mutex_unlock(&fs_devices->device_list_mutex); |
447 | ||
2b82032c | 448 | device->fs_devices = fs_devices; |
8a4b83cc | 449 | fs_devices->num_devices++; |
606686ee JB |
450 | } else if (!device->name || strcmp(device->name->str, path)) { |
451 | name = rcu_string_strdup(path, GFP_NOFS); | |
3a0524dc TH |
452 | if (!name) |
453 | return -ENOMEM; | |
606686ee JB |
454 | rcu_string_free(device->name); |
455 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
456 | if (device->missing) { |
457 | fs_devices->missing_devices--; | |
458 | device->missing = 0; | |
459 | } | |
8a4b83cc CM |
460 | } |
461 | ||
462 | if (found_transid > fs_devices->latest_trans) { | |
463 | fs_devices->latest_devid = devid; | |
464 | fs_devices->latest_trans = found_transid; | |
465 | } | |
8a4b83cc CM |
466 | *fs_devices_ret = fs_devices; |
467 | return 0; | |
468 | } | |
469 | ||
e4404d6e YZ |
470 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
471 | { | |
472 | struct btrfs_fs_devices *fs_devices; | |
473 | struct btrfs_device *device; | |
474 | struct btrfs_device *orig_dev; | |
475 | ||
476 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
477 | if (!fs_devices) | |
478 | return ERR_PTR(-ENOMEM); | |
479 | ||
480 | INIT_LIST_HEAD(&fs_devices->devices); | |
481 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
482 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 483 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
484 | fs_devices->latest_devid = orig->latest_devid; |
485 | fs_devices->latest_trans = orig->latest_trans; | |
02db0844 | 486 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
487 | memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); |
488 | ||
46224705 | 489 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 490 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
491 | struct rcu_string *name; |
492 | ||
e4404d6e YZ |
493 | device = kzalloc(sizeof(*device), GFP_NOFS); |
494 | if (!device) | |
495 | goto error; | |
496 | ||
606686ee JB |
497 | /* |
498 | * This is ok to do without rcu read locked because we hold the | |
499 | * uuid mutex so nothing we touch in here is going to disappear. | |
500 | */ | |
501 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
502 | if (!name) { | |
fd2696f3 | 503 | kfree(device); |
e4404d6e | 504 | goto error; |
fd2696f3 | 505 | } |
606686ee | 506 | rcu_assign_pointer(device->name, name); |
e4404d6e YZ |
507 | |
508 | device->devid = orig_dev->devid; | |
509 | device->work.func = pending_bios_fn; | |
510 | memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); | |
e4404d6e YZ |
511 | spin_lock_init(&device->io_lock); |
512 | INIT_LIST_HEAD(&device->dev_list); | |
513 | INIT_LIST_HEAD(&device->dev_alloc_list); | |
514 | ||
515 | list_add(&device->dev_list, &fs_devices->devices); | |
516 | device->fs_devices = fs_devices; | |
517 | fs_devices->num_devices++; | |
518 | } | |
519 | return fs_devices; | |
520 | error: | |
521 | free_fs_devices(fs_devices); | |
522 | return ERR_PTR(-ENOMEM); | |
523 | } | |
524 | ||
8dabb742 SB |
525 | void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info, |
526 | struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 527 | { |
c6e30871 | 528 | struct btrfs_device *device, *next; |
dfe25020 | 529 | |
a6b0d5c8 CM |
530 | struct block_device *latest_bdev = NULL; |
531 | u64 latest_devid = 0; | |
532 | u64 latest_transid = 0; | |
533 | ||
dfe25020 CM |
534 | mutex_lock(&uuid_mutex); |
535 | again: | |
46224705 | 536 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 537 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 538 | if (device->in_fs_metadata) { |
63a212ab SB |
539 | if (!device->is_tgtdev_for_dev_replace && |
540 | (!latest_transid || | |
541 | device->generation > latest_transid)) { | |
a6b0d5c8 CM |
542 | latest_devid = device->devid; |
543 | latest_transid = device->generation; | |
544 | latest_bdev = device->bdev; | |
545 | } | |
2b82032c | 546 | continue; |
a6b0d5c8 | 547 | } |
2b82032c | 548 | |
8dabb742 SB |
549 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
550 | /* | |
551 | * In the first step, keep the device which has | |
552 | * the correct fsid and the devid that is used | |
553 | * for the dev_replace procedure. | |
554 | * In the second step, the dev_replace state is | |
555 | * read from the device tree and it is known | |
556 | * whether the procedure is really active or | |
557 | * not, which means whether this device is | |
558 | * used or whether it should be removed. | |
559 | */ | |
560 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
561 | continue; | |
562 | } | |
563 | } | |
2b82032c | 564 | if (device->bdev) { |
d4d77629 | 565 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
566 | device->bdev = NULL; |
567 | fs_devices->open_devices--; | |
568 | } | |
569 | if (device->writeable) { | |
570 | list_del_init(&device->dev_alloc_list); | |
571 | device->writeable = 0; | |
8dabb742 SB |
572 | if (!device->is_tgtdev_for_dev_replace) |
573 | fs_devices->rw_devices--; | |
2b82032c | 574 | } |
e4404d6e YZ |
575 | list_del_init(&device->dev_list); |
576 | fs_devices->num_devices--; | |
606686ee | 577 | rcu_string_free(device->name); |
e4404d6e | 578 | kfree(device); |
dfe25020 | 579 | } |
2b82032c YZ |
580 | |
581 | if (fs_devices->seed) { | |
582 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
583 | goto again; |
584 | } | |
585 | ||
a6b0d5c8 CM |
586 | fs_devices->latest_bdev = latest_bdev; |
587 | fs_devices->latest_devid = latest_devid; | |
588 | fs_devices->latest_trans = latest_transid; | |
589 | ||
dfe25020 | 590 | mutex_unlock(&uuid_mutex); |
dfe25020 | 591 | } |
a0af469b | 592 | |
1f78160c XG |
593 | static void __free_device(struct work_struct *work) |
594 | { | |
595 | struct btrfs_device *device; | |
596 | ||
597 | device = container_of(work, struct btrfs_device, rcu_work); | |
598 | ||
599 | if (device->bdev) | |
600 | blkdev_put(device->bdev, device->mode); | |
601 | ||
606686ee | 602 | rcu_string_free(device->name); |
1f78160c XG |
603 | kfree(device); |
604 | } | |
605 | ||
606 | static void free_device(struct rcu_head *head) | |
607 | { | |
608 | struct btrfs_device *device; | |
609 | ||
610 | device = container_of(head, struct btrfs_device, rcu); | |
611 | ||
612 | INIT_WORK(&device->rcu_work, __free_device); | |
613 | schedule_work(&device->rcu_work); | |
614 | } | |
615 | ||
2b82032c | 616 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 617 | { |
8a4b83cc | 618 | struct btrfs_device *device; |
e4404d6e | 619 | |
2b82032c YZ |
620 | if (--fs_devices->opened > 0) |
621 | return 0; | |
8a4b83cc | 622 | |
c9513edb | 623 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 624 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c | 625 | struct btrfs_device *new_device; |
606686ee | 626 | struct rcu_string *name; |
1f78160c XG |
627 | |
628 | if (device->bdev) | |
a0af469b | 629 | fs_devices->open_devices--; |
1f78160c | 630 | |
8dabb742 | 631 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c YZ |
632 | list_del_init(&device->dev_alloc_list); |
633 | fs_devices->rw_devices--; | |
634 | } | |
635 | ||
d5e2003c JB |
636 | if (device->can_discard) |
637 | fs_devices->num_can_discard--; | |
638 | ||
1f78160c | 639 | new_device = kmalloc(sizeof(*new_device), GFP_NOFS); |
79787eaa | 640 | BUG_ON(!new_device); /* -ENOMEM */ |
1f78160c | 641 | memcpy(new_device, device, sizeof(*new_device)); |
606686ee JB |
642 | |
643 | /* Safe because we are under uuid_mutex */ | |
99f5944b JB |
644 | if (device->name) { |
645 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
646 | BUG_ON(device->name && !name); /* -ENOMEM */ | |
647 | rcu_assign_pointer(new_device->name, name); | |
648 | } | |
1f78160c XG |
649 | new_device->bdev = NULL; |
650 | new_device->writeable = 0; | |
651 | new_device->in_fs_metadata = 0; | |
d5e2003c | 652 | new_device->can_discard = 0; |
1cba0cdf | 653 | spin_lock_init(&new_device->io_lock); |
1f78160c XG |
654 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
655 | ||
656 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 657 | } |
c9513edb XG |
658 | mutex_unlock(&fs_devices->device_list_mutex); |
659 | ||
e4404d6e YZ |
660 | WARN_ON(fs_devices->open_devices); |
661 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
662 | fs_devices->opened = 0; |
663 | fs_devices->seeding = 0; | |
2b82032c | 664 | |
8a4b83cc CM |
665 | return 0; |
666 | } | |
667 | ||
2b82032c YZ |
668 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
669 | { | |
e4404d6e | 670 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
671 | int ret; |
672 | ||
673 | mutex_lock(&uuid_mutex); | |
674 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
675 | if (!fs_devices->opened) { |
676 | seed_devices = fs_devices->seed; | |
677 | fs_devices->seed = NULL; | |
678 | } | |
2b82032c | 679 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
680 | |
681 | while (seed_devices) { | |
682 | fs_devices = seed_devices; | |
683 | seed_devices = fs_devices->seed; | |
684 | __btrfs_close_devices(fs_devices); | |
685 | free_fs_devices(fs_devices); | |
686 | } | |
bc178622 ES |
687 | /* |
688 | * Wait for rcu kworkers under __btrfs_close_devices | |
689 | * to finish all blkdev_puts so device is really | |
690 | * free when umount is done. | |
691 | */ | |
692 | rcu_barrier(); | |
2b82032c YZ |
693 | return ret; |
694 | } | |
695 | ||
e4404d6e YZ |
696 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
697 | fmode_t flags, void *holder) | |
8a4b83cc | 698 | { |
d5e2003c | 699 | struct request_queue *q; |
8a4b83cc CM |
700 | struct block_device *bdev; |
701 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 702 | struct btrfs_device *device; |
a0af469b CM |
703 | struct block_device *latest_bdev = NULL; |
704 | struct buffer_head *bh; | |
705 | struct btrfs_super_block *disk_super; | |
706 | u64 latest_devid = 0; | |
707 | u64 latest_transid = 0; | |
a0af469b | 708 | u64 devid; |
2b82032c | 709 | int seeding = 1; |
a0af469b | 710 | int ret = 0; |
8a4b83cc | 711 | |
d4d77629 TH |
712 | flags |= FMODE_EXCL; |
713 | ||
c6e30871 | 714 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
715 | if (device->bdev) |
716 | continue; | |
dfe25020 CM |
717 | if (!device->name) |
718 | continue; | |
719 | ||
beaf8ab3 SB |
720 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, |
721 | &bdev, &bh); | |
722 | if (ret) | |
723 | continue; | |
a0af469b CM |
724 | |
725 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 726 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
727 | if (devid != device->devid) |
728 | goto error_brelse; | |
729 | ||
2b82032c YZ |
730 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
731 | BTRFS_UUID_SIZE)) | |
732 | goto error_brelse; | |
733 | ||
734 | device->generation = btrfs_super_generation(disk_super); | |
735 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 736 | latest_devid = devid; |
2b82032c | 737 | latest_transid = device->generation; |
a0af469b CM |
738 | latest_bdev = bdev; |
739 | } | |
740 | ||
2b82032c YZ |
741 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
742 | device->writeable = 0; | |
743 | } else { | |
744 | device->writeable = !bdev_read_only(bdev); | |
745 | seeding = 0; | |
746 | } | |
747 | ||
d5e2003c JB |
748 | q = bdev_get_queue(bdev); |
749 | if (blk_queue_discard(q)) { | |
750 | device->can_discard = 1; | |
751 | fs_devices->num_can_discard++; | |
752 | } | |
753 | ||
8a4b83cc | 754 | device->bdev = bdev; |
dfe25020 | 755 | device->in_fs_metadata = 0; |
15916de8 CM |
756 | device->mode = flags; |
757 | ||
c289811c CM |
758 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
759 | fs_devices->rotating = 1; | |
760 | ||
a0af469b | 761 | fs_devices->open_devices++; |
8dabb742 | 762 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c YZ |
763 | fs_devices->rw_devices++; |
764 | list_add(&device->dev_alloc_list, | |
765 | &fs_devices->alloc_list); | |
766 | } | |
4f6c9328 | 767 | brelse(bh); |
a0af469b | 768 | continue; |
a061fc8d | 769 | |
a0af469b CM |
770 | error_brelse: |
771 | brelse(bh); | |
d4d77629 | 772 | blkdev_put(bdev, flags); |
a0af469b | 773 | continue; |
8a4b83cc | 774 | } |
a0af469b | 775 | if (fs_devices->open_devices == 0) { |
20bcd649 | 776 | ret = -EINVAL; |
a0af469b CM |
777 | goto out; |
778 | } | |
2b82032c YZ |
779 | fs_devices->seeding = seeding; |
780 | fs_devices->opened = 1; | |
a0af469b CM |
781 | fs_devices->latest_bdev = latest_bdev; |
782 | fs_devices->latest_devid = latest_devid; | |
783 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 784 | fs_devices->total_rw_bytes = 0; |
a0af469b | 785 | out: |
2b82032c YZ |
786 | return ret; |
787 | } | |
788 | ||
789 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 790 | fmode_t flags, void *holder) |
2b82032c YZ |
791 | { |
792 | int ret; | |
793 | ||
794 | mutex_lock(&uuid_mutex); | |
795 | if (fs_devices->opened) { | |
e4404d6e YZ |
796 | fs_devices->opened++; |
797 | ret = 0; | |
2b82032c | 798 | } else { |
15916de8 | 799 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 800 | } |
8a4b83cc | 801 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
802 | return ret; |
803 | } | |
804 | ||
6f60cbd3 DS |
805 | /* |
806 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
807 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
808 | * is read via pagecache | |
809 | */ | |
97288f2c | 810 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
811 | struct btrfs_fs_devices **fs_devices_ret) |
812 | { | |
813 | struct btrfs_super_block *disk_super; | |
814 | struct block_device *bdev; | |
6f60cbd3 DS |
815 | struct page *page; |
816 | void *p; | |
817 | int ret = -EINVAL; | |
8a4b83cc | 818 | u64 devid; |
f2984462 | 819 | u64 transid; |
02db0844 | 820 | u64 total_devices; |
6f60cbd3 DS |
821 | u64 bytenr; |
822 | pgoff_t index; | |
8a4b83cc | 823 | |
6f60cbd3 DS |
824 | /* |
825 | * we would like to check all the supers, but that would make | |
826 | * a btrfs mount succeed after a mkfs from a different FS. | |
827 | * So, we need to add a special mount option to scan for | |
828 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
829 | */ | |
830 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 831 | flags |= FMODE_EXCL; |
10f6327b | 832 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
833 | |
834 | bdev = blkdev_get_by_path(path, flags, holder); | |
835 | ||
836 | if (IS_ERR(bdev)) { | |
837 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 838 | goto error; |
6f60cbd3 DS |
839 | } |
840 | ||
841 | /* make sure our super fits in the device */ | |
842 | if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode)) | |
843 | goto error_bdev_put; | |
844 | ||
845 | /* make sure our super fits in the page */ | |
846 | if (sizeof(*disk_super) > PAGE_CACHE_SIZE) | |
847 | goto error_bdev_put; | |
848 | ||
849 | /* make sure our super doesn't straddle pages on disk */ | |
850 | index = bytenr >> PAGE_CACHE_SHIFT; | |
851 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index) | |
852 | goto error_bdev_put; | |
853 | ||
854 | /* pull in the page with our super */ | |
855 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
856 | index, GFP_NOFS); | |
857 | ||
858 | if (IS_ERR_OR_NULL(page)) | |
859 | goto error_bdev_put; | |
860 | ||
861 | p = kmap(page); | |
862 | ||
863 | /* align our pointer to the offset of the super block */ | |
864 | disk_super = p + (bytenr & ~PAGE_CACHE_MASK); | |
865 | ||
866 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
cdb4c574 | 867 | disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) |
6f60cbd3 DS |
868 | goto error_unmap; |
869 | ||
a343832f | 870 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 871 | transid = btrfs_super_generation(disk_super); |
02db0844 | 872 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 873 | |
d03f918a SB |
874 | if (disk_super->label[0]) { |
875 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
876 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
d397712b | 877 | printk(KERN_INFO "device label %s ", disk_super->label); |
d03f918a | 878 | } else { |
22b63a29 | 879 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); |
d03f918a | 880 | } |
6f60cbd3 | 881 | |
119e10cf | 882 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 883 | (unsigned long long)devid, (unsigned long long)transid, path); |
6f60cbd3 | 884 | |
8a4b83cc | 885 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
02db0844 JB |
886 | if (!ret && fs_devices_ret) |
887 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 DS |
888 | |
889 | error_unmap: | |
890 | kunmap(page); | |
891 | page_cache_release(page); | |
892 | ||
893 | error_bdev_put: | |
d4d77629 | 894 | blkdev_put(bdev, flags); |
8a4b83cc | 895 | error: |
beaf8ab3 | 896 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
897 | return ret; |
898 | } | |
0b86a832 | 899 | |
6d07bcec MX |
900 | /* helper to account the used device space in the range */ |
901 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
902 | u64 end, u64 *length) | |
903 | { | |
904 | struct btrfs_key key; | |
905 | struct btrfs_root *root = device->dev_root; | |
906 | struct btrfs_dev_extent *dev_extent; | |
907 | struct btrfs_path *path; | |
908 | u64 extent_end; | |
909 | int ret; | |
910 | int slot; | |
911 | struct extent_buffer *l; | |
912 | ||
913 | *length = 0; | |
914 | ||
63a212ab | 915 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
916 | return 0; |
917 | ||
918 | path = btrfs_alloc_path(); | |
919 | if (!path) | |
920 | return -ENOMEM; | |
921 | path->reada = 2; | |
922 | ||
923 | key.objectid = device->devid; | |
924 | key.offset = start; | |
925 | key.type = BTRFS_DEV_EXTENT_KEY; | |
926 | ||
927 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
928 | if (ret < 0) | |
929 | goto out; | |
930 | if (ret > 0) { | |
931 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
932 | if (ret < 0) | |
933 | goto out; | |
934 | } | |
935 | ||
936 | while (1) { | |
937 | l = path->nodes[0]; | |
938 | slot = path->slots[0]; | |
939 | if (slot >= btrfs_header_nritems(l)) { | |
940 | ret = btrfs_next_leaf(root, path); | |
941 | if (ret == 0) | |
942 | continue; | |
943 | if (ret < 0) | |
944 | goto out; | |
945 | ||
946 | break; | |
947 | } | |
948 | btrfs_item_key_to_cpu(l, &key, slot); | |
949 | ||
950 | if (key.objectid < device->devid) | |
951 | goto next; | |
952 | ||
953 | if (key.objectid > device->devid) | |
954 | break; | |
955 | ||
956 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
957 | goto next; | |
958 | ||
959 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
960 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
961 | dev_extent); | |
962 | if (key.offset <= start && extent_end > end) { | |
963 | *length = end - start + 1; | |
964 | break; | |
965 | } else if (key.offset <= start && extent_end > start) | |
966 | *length += extent_end - start; | |
967 | else if (key.offset > start && extent_end <= end) | |
968 | *length += extent_end - key.offset; | |
969 | else if (key.offset > start && key.offset <= end) { | |
970 | *length += end - key.offset + 1; | |
971 | break; | |
972 | } else if (key.offset > end) | |
973 | break; | |
974 | ||
975 | next: | |
976 | path->slots[0]++; | |
977 | } | |
978 | ret = 0; | |
979 | out: | |
980 | btrfs_free_path(path); | |
981 | return ret; | |
982 | } | |
983 | ||
0b86a832 | 984 | /* |
7bfc837d | 985 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
986 | * @device: the device which we search the free space in |
987 | * @num_bytes: the size of the free space that we need | |
988 | * @start: store the start of the free space. | |
989 | * @len: the size of the free space. that we find, or the size of the max | |
990 | * free space if we don't find suitable free space | |
991 | * | |
0b86a832 CM |
992 | * this uses a pretty simple search, the expectation is that it is |
993 | * called very infrequently and that a given device has a small number | |
994 | * of extents | |
7bfc837d MX |
995 | * |
996 | * @start is used to store the start of the free space if we find. But if we | |
997 | * don't find suitable free space, it will be used to store the start position | |
998 | * of the max free space. | |
999 | * | |
1000 | * @len is used to store the size of the free space that we find. | |
1001 | * But if we don't find suitable free space, it is used to store the size of | |
1002 | * the max free space. | |
0b86a832 | 1003 | */ |
125ccb0a | 1004 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
7bfc837d | 1005 | u64 *start, u64 *len) |
0b86a832 CM |
1006 | { |
1007 | struct btrfs_key key; | |
1008 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1009 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1010 | struct btrfs_path *path; |
7bfc837d MX |
1011 | u64 hole_size; |
1012 | u64 max_hole_start; | |
1013 | u64 max_hole_size; | |
1014 | u64 extent_end; | |
1015 | u64 search_start; | |
0b86a832 CM |
1016 | u64 search_end = device->total_bytes; |
1017 | int ret; | |
7bfc837d | 1018 | int slot; |
0b86a832 CM |
1019 | struct extent_buffer *l; |
1020 | ||
0b86a832 CM |
1021 | /* FIXME use last free of some kind */ |
1022 | ||
8a4b83cc CM |
1023 | /* we don't want to overwrite the superblock on the drive, |
1024 | * so we make sure to start at an offset of at least 1MB | |
1025 | */ | |
a9c9bf68 | 1026 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 1027 | |
7bfc837d MX |
1028 | max_hole_start = search_start; |
1029 | max_hole_size = 0; | |
38c01b96 | 1030 | hole_size = 0; |
7bfc837d | 1031 | |
63a212ab | 1032 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d MX |
1033 | ret = -ENOSPC; |
1034 | goto error; | |
1035 | } | |
1036 | ||
1037 | path = btrfs_alloc_path(); | |
1038 | if (!path) { | |
1039 | ret = -ENOMEM; | |
1040 | goto error; | |
1041 | } | |
1042 | path->reada = 2; | |
1043 | ||
0b86a832 CM |
1044 | key.objectid = device->devid; |
1045 | key.offset = search_start; | |
1046 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1047 | |
125ccb0a | 1048 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1049 | if (ret < 0) |
7bfc837d | 1050 | goto out; |
1fcbac58 YZ |
1051 | if (ret > 0) { |
1052 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1053 | if (ret < 0) | |
7bfc837d | 1054 | goto out; |
1fcbac58 | 1055 | } |
7bfc837d | 1056 | |
0b86a832 CM |
1057 | while (1) { |
1058 | l = path->nodes[0]; | |
1059 | slot = path->slots[0]; | |
1060 | if (slot >= btrfs_header_nritems(l)) { | |
1061 | ret = btrfs_next_leaf(root, path); | |
1062 | if (ret == 0) | |
1063 | continue; | |
1064 | if (ret < 0) | |
7bfc837d MX |
1065 | goto out; |
1066 | ||
1067 | break; | |
0b86a832 CM |
1068 | } |
1069 | btrfs_item_key_to_cpu(l, &key, slot); | |
1070 | ||
1071 | if (key.objectid < device->devid) | |
1072 | goto next; | |
1073 | ||
1074 | if (key.objectid > device->devid) | |
7bfc837d | 1075 | break; |
0b86a832 | 1076 | |
7bfc837d MX |
1077 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
1078 | goto next; | |
9779b72f | 1079 | |
7bfc837d MX |
1080 | if (key.offset > search_start) { |
1081 | hole_size = key.offset - search_start; | |
9779b72f | 1082 | |
7bfc837d MX |
1083 | if (hole_size > max_hole_size) { |
1084 | max_hole_start = search_start; | |
1085 | max_hole_size = hole_size; | |
1086 | } | |
9779b72f | 1087 | |
7bfc837d MX |
1088 | /* |
1089 | * If this free space is greater than which we need, | |
1090 | * it must be the max free space that we have found | |
1091 | * until now, so max_hole_start must point to the start | |
1092 | * of this free space and the length of this free space | |
1093 | * is stored in max_hole_size. Thus, we return | |
1094 | * max_hole_start and max_hole_size and go back to the | |
1095 | * caller. | |
1096 | */ | |
1097 | if (hole_size >= num_bytes) { | |
1098 | ret = 0; | |
1099 | goto out; | |
0b86a832 CM |
1100 | } |
1101 | } | |
0b86a832 | 1102 | |
0b86a832 | 1103 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1104 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1105 | dev_extent); | |
1106 | if (extent_end > search_start) | |
1107 | search_start = extent_end; | |
0b86a832 CM |
1108 | next: |
1109 | path->slots[0]++; | |
1110 | cond_resched(); | |
1111 | } | |
0b86a832 | 1112 | |
38c01b96 | 1113 | /* |
1114 | * At this point, search_start should be the end of | |
1115 | * allocated dev extents, and when shrinking the device, | |
1116 | * search_end may be smaller than search_start. | |
1117 | */ | |
1118 | if (search_end > search_start) | |
1119 | hole_size = search_end - search_start; | |
1120 | ||
7bfc837d MX |
1121 | if (hole_size > max_hole_size) { |
1122 | max_hole_start = search_start; | |
1123 | max_hole_size = hole_size; | |
0b86a832 | 1124 | } |
0b86a832 | 1125 | |
7bfc837d MX |
1126 | /* See above. */ |
1127 | if (hole_size < num_bytes) | |
1128 | ret = -ENOSPC; | |
1129 | else | |
1130 | ret = 0; | |
1131 | ||
1132 | out: | |
2b82032c | 1133 | btrfs_free_path(path); |
7bfc837d MX |
1134 | error: |
1135 | *start = max_hole_start; | |
b2117a39 | 1136 | if (len) |
7bfc837d | 1137 | *len = max_hole_size; |
0b86a832 CM |
1138 | return ret; |
1139 | } | |
1140 | ||
b2950863 | 1141 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1142 | struct btrfs_device *device, |
1143 | u64 start) | |
1144 | { | |
1145 | int ret; | |
1146 | struct btrfs_path *path; | |
1147 | struct btrfs_root *root = device->dev_root; | |
1148 | struct btrfs_key key; | |
a061fc8d CM |
1149 | struct btrfs_key found_key; |
1150 | struct extent_buffer *leaf = NULL; | |
1151 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1152 | |
1153 | path = btrfs_alloc_path(); | |
1154 | if (!path) | |
1155 | return -ENOMEM; | |
1156 | ||
1157 | key.objectid = device->devid; | |
1158 | key.offset = start; | |
1159 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1160 | again: |
8f18cf13 | 1161 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1162 | if (ret > 0) { |
1163 | ret = btrfs_previous_item(root, path, key.objectid, | |
1164 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1165 | if (ret) |
1166 | goto out; | |
a061fc8d CM |
1167 | leaf = path->nodes[0]; |
1168 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1169 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1170 | struct btrfs_dev_extent); | |
1171 | BUG_ON(found_key.offset > start || found_key.offset + | |
1172 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1173 | key = found_key; |
1174 | btrfs_release_path(path); | |
1175 | goto again; | |
a061fc8d CM |
1176 | } else if (ret == 0) { |
1177 | leaf = path->nodes[0]; | |
1178 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1179 | struct btrfs_dev_extent); | |
79787eaa JM |
1180 | } else { |
1181 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1182 | goto out; | |
a061fc8d | 1183 | } |
8f18cf13 | 1184 | |
2bf64758 JB |
1185 | if (device->bytes_used > 0) { |
1186 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1187 | device->bytes_used -= len; | |
1188 | spin_lock(&root->fs_info->free_chunk_lock); | |
1189 | root->fs_info->free_chunk_space += len; | |
1190 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1191 | } | |
8f18cf13 | 1192 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1193 | if (ret) { |
1194 | btrfs_error(root->fs_info, ret, | |
1195 | "Failed to remove dev extent item"); | |
1196 | } | |
b0b802d7 | 1197 | out: |
8f18cf13 CM |
1198 | btrfs_free_path(path); |
1199 | return ret; | |
1200 | } | |
1201 | ||
2b82032c | 1202 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1203 | struct btrfs_device *device, |
e17cade2 | 1204 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1205 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1206 | { |
1207 | int ret; | |
1208 | struct btrfs_path *path; | |
1209 | struct btrfs_root *root = device->dev_root; | |
1210 | struct btrfs_dev_extent *extent; | |
1211 | struct extent_buffer *leaf; | |
1212 | struct btrfs_key key; | |
1213 | ||
dfe25020 | 1214 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1215 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1216 | path = btrfs_alloc_path(); |
1217 | if (!path) | |
1218 | return -ENOMEM; | |
1219 | ||
0b86a832 | 1220 | key.objectid = device->devid; |
2b82032c | 1221 | key.offset = start; |
0b86a832 CM |
1222 | key.type = BTRFS_DEV_EXTENT_KEY; |
1223 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1224 | sizeof(*extent)); | |
2cdcecbc MF |
1225 | if (ret) |
1226 | goto out; | |
0b86a832 CM |
1227 | |
1228 | leaf = path->nodes[0]; | |
1229 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1230 | struct btrfs_dev_extent); | |
e17cade2 CM |
1231 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1232 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1233 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1234 | ||
1235 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1236 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1237 | BTRFS_UUID_SIZE); | |
1238 | ||
0b86a832 CM |
1239 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1240 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1241 | out: |
0b86a832 CM |
1242 | btrfs_free_path(path); |
1243 | return ret; | |
1244 | } | |
1245 | ||
a1b32a59 CM |
1246 | static noinline int find_next_chunk(struct btrfs_root *root, |
1247 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1248 | { |
1249 | struct btrfs_path *path; | |
1250 | int ret; | |
1251 | struct btrfs_key key; | |
e17cade2 | 1252 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1253 | struct btrfs_key found_key; |
1254 | ||
1255 | path = btrfs_alloc_path(); | |
92b8e897 MF |
1256 | if (!path) |
1257 | return -ENOMEM; | |
0b86a832 | 1258 | |
e17cade2 | 1259 | key.objectid = objectid; |
0b86a832 CM |
1260 | key.offset = (u64)-1; |
1261 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1262 | ||
1263 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1264 | if (ret < 0) | |
1265 | goto error; | |
1266 | ||
79787eaa | 1267 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1268 | |
1269 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1270 | if (ret) { | |
e17cade2 | 1271 | *offset = 0; |
0b86a832 CM |
1272 | } else { |
1273 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1274 | path->slots[0]); | |
e17cade2 CM |
1275 | if (found_key.objectid != objectid) |
1276 | *offset = 0; | |
1277 | else { | |
1278 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1279 | struct btrfs_chunk); | |
1280 | *offset = found_key.offset + | |
1281 | btrfs_chunk_length(path->nodes[0], chunk); | |
1282 | } | |
0b86a832 CM |
1283 | } |
1284 | ret = 0; | |
1285 | error: | |
1286 | btrfs_free_path(path); | |
1287 | return ret; | |
1288 | } | |
1289 | ||
2b82032c | 1290 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1291 | { |
1292 | int ret; | |
1293 | struct btrfs_key key; | |
1294 | struct btrfs_key found_key; | |
2b82032c YZ |
1295 | struct btrfs_path *path; |
1296 | ||
1297 | root = root->fs_info->chunk_root; | |
1298 | ||
1299 | path = btrfs_alloc_path(); | |
1300 | if (!path) | |
1301 | return -ENOMEM; | |
0b86a832 CM |
1302 | |
1303 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1304 | key.type = BTRFS_DEV_ITEM_KEY; | |
1305 | key.offset = (u64)-1; | |
1306 | ||
1307 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1308 | if (ret < 0) | |
1309 | goto error; | |
1310 | ||
79787eaa | 1311 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1312 | |
1313 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1314 | BTRFS_DEV_ITEM_KEY); | |
1315 | if (ret) { | |
1316 | *objectid = 1; | |
1317 | } else { | |
1318 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1319 | path->slots[0]); | |
1320 | *objectid = found_key.offset + 1; | |
1321 | } | |
1322 | ret = 0; | |
1323 | error: | |
2b82032c | 1324 | btrfs_free_path(path); |
0b86a832 CM |
1325 | return ret; |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * the device information is stored in the chunk root | |
1330 | * the btrfs_device struct should be fully filled in | |
1331 | */ | |
1332 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1333 | struct btrfs_root *root, | |
1334 | struct btrfs_device *device) | |
1335 | { | |
1336 | int ret; | |
1337 | struct btrfs_path *path; | |
1338 | struct btrfs_dev_item *dev_item; | |
1339 | struct extent_buffer *leaf; | |
1340 | struct btrfs_key key; | |
1341 | unsigned long ptr; | |
0b86a832 CM |
1342 | |
1343 | root = root->fs_info->chunk_root; | |
1344 | ||
1345 | path = btrfs_alloc_path(); | |
1346 | if (!path) | |
1347 | return -ENOMEM; | |
1348 | ||
0b86a832 CM |
1349 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1350 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1351 | key.offset = device->devid; |
0b86a832 CM |
1352 | |
1353 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1354 | sizeof(*dev_item)); |
0b86a832 CM |
1355 | if (ret) |
1356 | goto out; | |
1357 | ||
1358 | leaf = path->nodes[0]; | |
1359 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1360 | ||
1361 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1362 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1363 | btrfs_set_device_type(leaf, dev_item, device->type); |
1364 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1365 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1366 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1367 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1368 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1369 | btrfs_set_device_group(leaf, dev_item, 0); |
1370 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1371 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1372 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1373 | |
0b86a832 | 1374 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1375 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1376 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1377 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1378 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1379 | |
2b82032c | 1380 | ret = 0; |
0b86a832 CM |
1381 | out: |
1382 | btrfs_free_path(path); | |
1383 | return ret; | |
1384 | } | |
8f18cf13 | 1385 | |
a061fc8d CM |
1386 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1387 | struct btrfs_device *device) | |
1388 | { | |
1389 | int ret; | |
1390 | struct btrfs_path *path; | |
a061fc8d | 1391 | struct btrfs_key key; |
a061fc8d CM |
1392 | struct btrfs_trans_handle *trans; |
1393 | ||
1394 | root = root->fs_info->chunk_root; | |
1395 | ||
1396 | path = btrfs_alloc_path(); | |
1397 | if (!path) | |
1398 | return -ENOMEM; | |
1399 | ||
a22285a6 | 1400 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1401 | if (IS_ERR(trans)) { |
1402 | btrfs_free_path(path); | |
1403 | return PTR_ERR(trans); | |
1404 | } | |
a061fc8d CM |
1405 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1406 | key.type = BTRFS_DEV_ITEM_KEY; | |
1407 | key.offset = device->devid; | |
7d9eb12c | 1408 | lock_chunks(root); |
a061fc8d CM |
1409 | |
1410 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1411 | if (ret < 0) | |
1412 | goto out; | |
1413 | ||
1414 | if (ret > 0) { | |
1415 | ret = -ENOENT; | |
1416 | goto out; | |
1417 | } | |
1418 | ||
1419 | ret = btrfs_del_item(trans, root, path); | |
1420 | if (ret) | |
1421 | goto out; | |
a061fc8d CM |
1422 | out: |
1423 | btrfs_free_path(path); | |
7d9eb12c | 1424 | unlock_chunks(root); |
a061fc8d CM |
1425 | btrfs_commit_transaction(trans, root); |
1426 | return ret; | |
1427 | } | |
1428 | ||
1429 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1430 | { | |
1431 | struct btrfs_device *device; | |
2b82032c | 1432 | struct btrfs_device *next_device; |
a061fc8d | 1433 | struct block_device *bdev; |
dfe25020 | 1434 | struct buffer_head *bh = NULL; |
a061fc8d | 1435 | struct btrfs_super_block *disk_super; |
1f78160c | 1436 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1437 | u64 all_avail; |
1438 | u64 devid; | |
2b82032c YZ |
1439 | u64 num_devices; |
1440 | u8 *dev_uuid; | |
de98ced9 | 1441 | unsigned seq; |
a061fc8d | 1442 | int ret = 0; |
1f78160c | 1443 | bool clear_super = false; |
a061fc8d | 1444 | |
a061fc8d CM |
1445 | mutex_lock(&uuid_mutex); |
1446 | ||
de98ced9 MX |
1447 | do { |
1448 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
1449 | ||
1450 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1451 | root->fs_info->avail_system_alloc_bits | | |
1452 | root->fs_info->avail_metadata_alloc_bits; | |
1453 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
a061fc8d | 1454 | |
8dabb742 SB |
1455 | num_devices = root->fs_info->fs_devices->num_devices; |
1456 | btrfs_dev_replace_lock(&root->fs_info->dev_replace); | |
1457 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1458 | WARN_ON(num_devices < 1); | |
1459 | num_devices--; | |
1460 | } | |
1461 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace); | |
1462 | ||
1463 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) { | |
d397712b CM |
1464 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1465 | "on raid10\n"); | |
a061fc8d CM |
1466 | ret = -EINVAL; |
1467 | goto out; | |
1468 | } | |
1469 | ||
8dabb742 | 1470 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) { |
d397712b CM |
1471 | printk(KERN_ERR "btrfs: unable to go below two " |
1472 | "devices on raid1\n"); | |
a061fc8d CM |
1473 | ret = -EINVAL; |
1474 | goto out; | |
1475 | } | |
1476 | ||
53b381b3 DW |
1477 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && |
1478 | root->fs_info->fs_devices->rw_devices <= 2) { | |
1479 | printk(KERN_ERR "btrfs: unable to go below two " | |
1480 | "devices on raid5\n"); | |
1481 | ret = -EINVAL; | |
1482 | goto out; | |
1483 | } | |
1484 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1485 | root->fs_info->fs_devices->rw_devices <= 3) { | |
1486 | printk(KERN_ERR "btrfs: unable to go below three " | |
1487 | "devices on raid6\n"); | |
1488 | ret = -EINVAL; | |
1489 | goto out; | |
1490 | } | |
1491 | ||
dfe25020 | 1492 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1493 | struct list_head *devices; |
1494 | struct btrfs_device *tmp; | |
a061fc8d | 1495 | |
dfe25020 CM |
1496 | device = NULL; |
1497 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1498 | /* |
1499 | * It is safe to read the devices since the volume_mutex | |
1500 | * is held. | |
1501 | */ | |
c6e30871 | 1502 | list_for_each_entry(tmp, devices, dev_list) { |
63a212ab SB |
1503 | if (tmp->in_fs_metadata && |
1504 | !tmp->is_tgtdev_for_dev_replace && | |
1505 | !tmp->bdev) { | |
dfe25020 CM |
1506 | device = tmp; |
1507 | break; | |
1508 | } | |
1509 | } | |
1510 | bdev = NULL; | |
1511 | bh = NULL; | |
1512 | disk_super = NULL; | |
1513 | if (!device) { | |
d397712b CM |
1514 | printk(KERN_ERR "btrfs: no missing devices found to " |
1515 | "remove\n"); | |
dfe25020 CM |
1516 | goto out; |
1517 | } | |
dfe25020 | 1518 | } else { |
beaf8ab3 | 1519 | ret = btrfs_get_bdev_and_sb(device_path, |
cc975eb4 | 1520 | FMODE_WRITE | FMODE_EXCL, |
beaf8ab3 SB |
1521 | root->fs_info->bdev_holder, 0, |
1522 | &bdev, &bh); | |
1523 | if (ret) | |
dfe25020 | 1524 | goto out; |
dfe25020 | 1525 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1526 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c | 1527 | dev_uuid = disk_super->dev_item.uuid; |
aa1b8cd4 | 1528 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2b82032c | 1529 | disk_super->fsid); |
dfe25020 CM |
1530 | if (!device) { |
1531 | ret = -ENOENT; | |
1532 | goto error_brelse; | |
1533 | } | |
2b82032c | 1534 | } |
dfe25020 | 1535 | |
63a212ab SB |
1536 | if (device->is_tgtdev_for_dev_replace) { |
1537 | pr_err("btrfs: unable to remove the dev_replace target dev\n"); | |
1538 | ret = -EINVAL; | |
1539 | goto error_brelse; | |
1540 | } | |
1541 | ||
2b82032c | 1542 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1543 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1544 | "device\n"); | |
2b82032c YZ |
1545 | ret = -EINVAL; |
1546 | goto error_brelse; | |
1547 | } | |
1548 | ||
1549 | if (device->writeable) { | |
0c1daee0 | 1550 | lock_chunks(root); |
2b82032c | 1551 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1552 | unlock_chunks(root); |
2b82032c | 1553 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1554 | clear_super = true; |
dfe25020 | 1555 | } |
a061fc8d CM |
1556 | |
1557 | ret = btrfs_shrink_device(device, 0); | |
1558 | if (ret) | |
9b3517e9 | 1559 | goto error_undo; |
a061fc8d | 1560 | |
63a212ab SB |
1561 | /* |
1562 | * TODO: the superblock still includes this device in its num_devices | |
1563 | * counter although write_all_supers() is not locked out. This | |
1564 | * could give a filesystem state which requires a degraded mount. | |
1565 | */ | |
a061fc8d CM |
1566 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1567 | if (ret) | |
9b3517e9 | 1568 | goto error_undo; |
a061fc8d | 1569 | |
2bf64758 JB |
1570 | spin_lock(&root->fs_info->free_chunk_lock); |
1571 | root->fs_info->free_chunk_space = device->total_bytes - | |
1572 | device->bytes_used; | |
1573 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1574 | ||
2b82032c | 1575 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1576 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1577 | |
1578 | /* | |
1579 | * the device list mutex makes sure that we don't change | |
1580 | * the device list while someone else is writing out all | |
1581 | * the device supers. | |
1582 | */ | |
1f78160c XG |
1583 | |
1584 | cur_devices = device->fs_devices; | |
e5e9a520 | 1585 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1586 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1587 | |
e4404d6e | 1588 | device->fs_devices->num_devices--; |
02db0844 | 1589 | device->fs_devices->total_devices--; |
2b82032c | 1590 | |
cd02dca5 CM |
1591 | if (device->missing) |
1592 | root->fs_info->fs_devices->missing_devices--; | |
1593 | ||
2b82032c YZ |
1594 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1595 | struct btrfs_device, dev_list); | |
1596 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1597 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1598 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1599 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1600 | ||
1f78160c | 1601 | if (device->bdev) |
e4404d6e | 1602 | device->fs_devices->open_devices--; |
1f78160c XG |
1603 | |
1604 | call_rcu(&device->rcu, free_device); | |
1605 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1606 | |
6c41761f DS |
1607 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1608 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1609 | |
1f78160c | 1610 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1611 | struct btrfs_fs_devices *fs_devices; |
1612 | fs_devices = root->fs_info->fs_devices; | |
1613 | while (fs_devices) { | |
1f78160c | 1614 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1615 | break; |
1616 | fs_devices = fs_devices->seed; | |
2b82032c | 1617 | } |
1f78160c XG |
1618 | fs_devices->seed = cur_devices->seed; |
1619 | cur_devices->seed = NULL; | |
0c1daee0 | 1620 | lock_chunks(root); |
1f78160c | 1621 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1622 | unlock_chunks(root); |
1f78160c | 1623 | free_fs_devices(cur_devices); |
2b82032c YZ |
1624 | } |
1625 | ||
5af3e8cc SB |
1626 | root->fs_info->num_tolerated_disk_barrier_failures = |
1627 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1628 | ||
2b82032c YZ |
1629 | /* |
1630 | * at this point, the device is zero sized. We want to | |
1631 | * remove it from the devices list and zero out the old super | |
1632 | */ | |
aa1b8cd4 | 1633 | if (clear_super && disk_super) { |
dfe25020 CM |
1634 | /* make sure this device isn't detected as part of |
1635 | * the FS anymore | |
1636 | */ | |
1637 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1638 | set_buffer_dirty(bh); | |
1639 | sync_dirty_buffer(bh); | |
dfe25020 | 1640 | } |
a061fc8d | 1641 | |
a061fc8d | 1642 | ret = 0; |
a061fc8d | 1643 | |
b8b8ff59 | 1644 | /* Notify udev that device has changed */ |
3c911608 ES |
1645 | if (bdev) |
1646 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
b8b8ff59 | 1647 | |
a061fc8d CM |
1648 | error_brelse: |
1649 | brelse(bh); | |
dfe25020 | 1650 | if (bdev) |
e525fd89 | 1651 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1652 | out: |
1653 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1654 | return ret; |
9b3517e9 ID |
1655 | error_undo: |
1656 | if (device->writeable) { | |
0c1daee0 | 1657 | lock_chunks(root); |
9b3517e9 ID |
1658 | list_add(&device->dev_alloc_list, |
1659 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1660 | unlock_chunks(root); |
9b3517e9 ID |
1661 | root->fs_info->fs_devices->rw_devices++; |
1662 | } | |
1663 | goto error_brelse; | |
a061fc8d CM |
1664 | } |
1665 | ||
e93c89c1 SB |
1666 | void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info, |
1667 | struct btrfs_device *srcdev) | |
1668 | { | |
1669 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); | |
1670 | list_del_rcu(&srcdev->dev_list); | |
1671 | list_del_rcu(&srcdev->dev_alloc_list); | |
1672 | fs_info->fs_devices->num_devices--; | |
1673 | if (srcdev->missing) { | |
1674 | fs_info->fs_devices->missing_devices--; | |
1675 | fs_info->fs_devices->rw_devices++; | |
1676 | } | |
1677 | if (srcdev->can_discard) | |
1678 | fs_info->fs_devices->num_can_discard--; | |
1679 | if (srcdev->bdev) | |
1680 | fs_info->fs_devices->open_devices--; | |
1681 | ||
1682 | call_rcu(&srcdev->rcu, free_device); | |
1683 | } | |
1684 | ||
1685 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
1686 | struct btrfs_device *tgtdev) | |
1687 | { | |
1688 | struct btrfs_device *next_device; | |
1689 | ||
1690 | WARN_ON(!tgtdev); | |
1691 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
1692 | if (tgtdev->bdev) { | |
1693 | btrfs_scratch_superblock(tgtdev); | |
1694 | fs_info->fs_devices->open_devices--; | |
1695 | } | |
1696 | fs_info->fs_devices->num_devices--; | |
1697 | if (tgtdev->can_discard) | |
1698 | fs_info->fs_devices->num_can_discard++; | |
1699 | ||
1700 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
1701 | struct btrfs_device, dev_list); | |
1702 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
1703 | fs_info->sb->s_bdev = next_device->bdev; | |
1704 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
1705 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1706 | list_del_rcu(&tgtdev->dev_list); | |
1707 | ||
1708 | call_rcu(&tgtdev->rcu, free_device); | |
1709 | ||
1710 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
1711 | } | |
1712 | ||
7ba15b7d SB |
1713 | int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
1714 | struct btrfs_device **device) | |
1715 | { | |
1716 | int ret = 0; | |
1717 | struct btrfs_super_block *disk_super; | |
1718 | u64 devid; | |
1719 | u8 *dev_uuid; | |
1720 | struct block_device *bdev; | |
1721 | struct buffer_head *bh; | |
1722 | ||
1723 | *device = NULL; | |
1724 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
1725 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
1726 | if (ret) | |
1727 | return ret; | |
1728 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1729 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
1730 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 1731 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
1732 | disk_super->fsid); |
1733 | brelse(bh); | |
1734 | if (!*device) | |
1735 | ret = -ENOENT; | |
1736 | blkdev_put(bdev, FMODE_READ); | |
1737 | return ret; | |
1738 | } | |
1739 | ||
1740 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
1741 | char *device_path, | |
1742 | struct btrfs_device **device) | |
1743 | { | |
1744 | *device = NULL; | |
1745 | if (strcmp(device_path, "missing") == 0) { | |
1746 | struct list_head *devices; | |
1747 | struct btrfs_device *tmp; | |
1748 | ||
1749 | devices = &root->fs_info->fs_devices->devices; | |
1750 | /* | |
1751 | * It is safe to read the devices since the volume_mutex | |
1752 | * is held by the caller. | |
1753 | */ | |
1754 | list_for_each_entry(tmp, devices, dev_list) { | |
1755 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
1756 | *device = tmp; | |
1757 | break; | |
1758 | } | |
1759 | } | |
1760 | ||
1761 | if (!*device) { | |
1762 | pr_err("btrfs: no missing device found\n"); | |
1763 | return -ENOENT; | |
1764 | } | |
1765 | ||
1766 | return 0; | |
1767 | } else { | |
1768 | return btrfs_find_device_by_path(root, device_path, device); | |
1769 | } | |
1770 | } | |
1771 | ||
2b82032c YZ |
1772 | /* |
1773 | * does all the dirty work required for changing file system's UUID. | |
1774 | */ | |
125ccb0a | 1775 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1776 | { |
1777 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1778 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1779 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1780 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1781 | struct btrfs_device *device; |
1782 | u64 super_flags; | |
1783 | ||
1784 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1785 | if (!fs_devices->seeding) |
2b82032c YZ |
1786 | return -EINVAL; |
1787 | ||
e4404d6e YZ |
1788 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1789 | if (!seed_devices) | |
2b82032c YZ |
1790 | return -ENOMEM; |
1791 | ||
e4404d6e YZ |
1792 | old_devices = clone_fs_devices(fs_devices); |
1793 | if (IS_ERR(old_devices)) { | |
1794 | kfree(seed_devices); | |
1795 | return PTR_ERR(old_devices); | |
2b82032c | 1796 | } |
e4404d6e | 1797 | |
2b82032c YZ |
1798 | list_add(&old_devices->list, &fs_uuids); |
1799 | ||
e4404d6e YZ |
1800 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1801 | seed_devices->opened = 1; | |
1802 | INIT_LIST_HEAD(&seed_devices->devices); | |
1803 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1804 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1805 | |
1806 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1807 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1808 | synchronize_rcu); | |
c9513edb XG |
1809 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1810 | ||
e4404d6e YZ |
1811 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1812 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1813 | device->fs_devices = seed_devices; | |
1814 | } | |
1815 | ||
2b82032c YZ |
1816 | fs_devices->seeding = 0; |
1817 | fs_devices->num_devices = 0; | |
1818 | fs_devices->open_devices = 0; | |
02db0844 | 1819 | fs_devices->total_devices = 0; |
e4404d6e | 1820 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1821 | |
1822 | generate_random_uuid(fs_devices->fsid); | |
1823 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1824 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1825 | super_flags = btrfs_super_flags(disk_super) & | |
1826 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1827 | btrfs_set_super_flags(disk_super, super_flags); | |
1828 | ||
1829 | return 0; | |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * strore the expected generation for seed devices in device items. | |
1834 | */ | |
1835 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1836 | struct btrfs_root *root) | |
1837 | { | |
1838 | struct btrfs_path *path; | |
1839 | struct extent_buffer *leaf; | |
1840 | struct btrfs_dev_item *dev_item; | |
1841 | struct btrfs_device *device; | |
1842 | struct btrfs_key key; | |
1843 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1844 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1845 | u64 devid; | |
1846 | int ret; | |
1847 | ||
1848 | path = btrfs_alloc_path(); | |
1849 | if (!path) | |
1850 | return -ENOMEM; | |
1851 | ||
1852 | root = root->fs_info->chunk_root; | |
1853 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1854 | key.offset = 0; | |
1855 | key.type = BTRFS_DEV_ITEM_KEY; | |
1856 | ||
1857 | while (1) { | |
1858 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1859 | if (ret < 0) | |
1860 | goto error; | |
1861 | ||
1862 | leaf = path->nodes[0]; | |
1863 | next_slot: | |
1864 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1865 | ret = btrfs_next_leaf(root, path); | |
1866 | if (ret > 0) | |
1867 | break; | |
1868 | if (ret < 0) | |
1869 | goto error; | |
1870 | leaf = path->nodes[0]; | |
1871 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1872 | btrfs_release_path(path); |
2b82032c YZ |
1873 | continue; |
1874 | } | |
1875 | ||
1876 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1877 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1878 | key.type != BTRFS_DEV_ITEM_KEY) | |
1879 | break; | |
1880 | ||
1881 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1882 | struct btrfs_dev_item); | |
1883 | devid = btrfs_device_id(leaf, dev_item); | |
1884 | read_extent_buffer(leaf, dev_uuid, | |
1885 | (unsigned long)btrfs_device_uuid(dev_item), | |
1886 | BTRFS_UUID_SIZE); | |
1887 | read_extent_buffer(leaf, fs_uuid, | |
1888 | (unsigned long)btrfs_device_fsid(dev_item), | |
1889 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
1890 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
1891 | fs_uuid); | |
79787eaa | 1892 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
1893 | |
1894 | if (device->fs_devices->seeding) { | |
1895 | btrfs_set_device_generation(leaf, dev_item, | |
1896 | device->generation); | |
1897 | btrfs_mark_buffer_dirty(leaf); | |
1898 | } | |
1899 | ||
1900 | path->slots[0]++; | |
1901 | goto next_slot; | |
1902 | } | |
1903 | ret = 0; | |
1904 | error: | |
1905 | btrfs_free_path(path); | |
1906 | return ret; | |
1907 | } | |
1908 | ||
788f20eb CM |
1909 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1910 | { | |
d5e2003c | 1911 | struct request_queue *q; |
788f20eb CM |
1912 | struct btrfs_trans_handle *trans; |
1913 | struct btrfs_device *device; | |
1914 | struct block_device *bdev; | |
788f20eb | 1915 | struct list_head *devices; |
2b82032c | 1916 | struct super_block *sb = root->fs_info->sb; |
606686ee | 1917 | struct rcu_string *name; |
788f20eb | 1918 | u64 total_bytes; |
2b82032c | 1919 | int seeding_dev = 0; |
788f20eb CM |
1920 | int ret = 0; |
1921 | ||
2b82032c | 1922 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 1923 | return -EROFS; |
788f20eb | 1924 | |
a5d16333 | 1925 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1926 | root->fs_info->bdev_holder); |
7f59203a JB |
1927 | if (IS_ERR(bdev)) |
1928 | return PTR_ERR(bdev); | |
a2135011 | 1929 | |
2b82032c YZ |
1930 | if (root->fs_info->fs_devices->seeding) { |
1931 | seeding_dev = 1; | |
1932 | down_write(&sb->s_umount); | |
1933 | mutex_lock(&uuid_mutex); | |
1934 | } | |
1935 | ||
8c8bee1d | 1936 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 1937 | |
788f20eb | 1938 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
1939 | |
1940 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 1941 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1942 | if (device->bdev == bdev) { |
1943 | ret = -EEXIST; | |
d25628bd LB |
1944 | mutex_unlock( |
1945 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 1946 | goto error; |
788f20eb CM |
1947 | } |
1948 | } | |
d25628bd | 1949 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb CM |
1950 | |
1951 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1952 | if (!device) { | |
1953 | /* we can safely leave the fs_devices entry around */ | |
1954 | ret = -ENOMEM; | |
2b82032c | 1955 | goto error; |
788f20eb CM |
1956 | } |
1957 | ||
606686ee JB |
1958 | name = rcu_string_strdup(device_path, GFP_NOFS); |
1959 | if (!name) { | |
788f20eb | 1960 | kfree(device); |
2b82032c YZ |
1961 | ret = -ENOMEM; |
1962 | goto error; | |
788f20eb | 1963 | } |
606686ee | 1964 | rcu_assign_pointer(device->name, name); |
2b82032c YZ |
1965 | |
1966 | ret = find_next_devid(root, &device->devid); | |
1967 | if (ret) { | |
606686ee | 1968 | rcu_string_free(device->name); |
2b82032c YZ |
1969 | kfree(device); |
1970 | goto error; | |
1971 | } | |
1972 | ||
a22285a6 | 1973 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1974 | if (IS_ERR(trans)) { |
606686ee | 1975 | rcu_string_free(device->name); |
98d5dc13 TI |
1976 | kfree(device); |
1977 | ret = PTR_ERR(trans); | |
1978 | goto error; | |
1979 | } | |
1980 | ||
2b82032c YZ |
1981 | lock_chunks(root); |
1982 | ||
d5e2003c JB |
1983 | q = bdev_get_queue(bdev); |
1984 | if (blk_queue_discard(q)) | |
1985 | device->can_discard = 1; | |
2b82032c YZ |
1986 | device->writeable = 1; |
1987 | device->work.func = pending_bios_fn; | |
1988 | generate_random_uuid(device->uuid); | |
1989 | spin_lock_init(&device->io_lock); | |
1990 | device->generation = trans->transid; | |
788f20eb CM |
1991 | device->io_width = root->sectorsize; |
1992 | device->io_align = root->sectorsize; | |
1993 | device->sector_size = root->sectorsize; | |
1994 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1995 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1996 | device->dev_root = root->fs_info->dev_root; |
1997 | device->bdev = bdev; | |
dfe25020 | 1998 | device->in_fs_metadata = 1; |
63a212ab | 1999 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2000 | device->mode = FMODE_EXCL; |
2b82032c | 2001 | set_blocksize(device->bdev, 4096); |
788f20eb | 2002 | |
2b82032c YZ |
2003 | if (seeding_dev) { |
2004 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2005 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2006 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2007 | } |
788f20eb | 2008 | |
2b82032c | 2009 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2010 | |
e5e9a520 | 2011 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 2012 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2013 | list_add(&device->dev_alloc_list, |
2014 | &root->fs_info->fs_devices->alloc_list); | |
2015 | root->fs_info->fs_devices->num_devices++; | |
2016 | root->fs_info->fs_devices->open_devices++; | |
2017 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2018 | root->fs_info->fs_devices->total_devices++; |
d5e2003c JB |
2019 | if (device->can_discard) |
2020 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 2021 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2022 | |
2bf64758 JB |
2023 | spin_lock(&root->fs_info->free_chunk_lock); |
2024 | root->fs_info->free_chunk_space += device->total_bytes; | |
2025 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2026 | ||
c289811c CM |
2027 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2028 | root->fs_info->fs_devices->rotating = 1; | |
2029 | ||
6c41761f DS |
2030 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
2031 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
2032 | total_bytes + device->total_bytes); |
2033 | ||
6c41761f DS |
2034 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
2035 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 2036 | total_bytes + 1); |
e5e9a520 | 2037 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2038 | |
2b82032c YZ |
2039 | if (seeding_dev) { |
2040 | ret = init_first_rw_device(trans, root, device); | |
005d6427 DS |
2041 | if (ret) { |
2042 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2043 | goto error_trans; |
005d6427 | 2044 | } |
2b82032c | 2045 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2046 | if (ret) { |
2047 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2048 | goto error_trans; |
005d6427 | 2049 | } |
2b82032c YZ |
2050 | } else { |
2051 | ret = btrfs_add_device(trans, root, device); | |
005d6427 DS |
2052 | if (ret) { |
2053 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2054 | goto error_trans; |
005d6427 | 2055 | } |
2b82032c YZ |
2056 | } |
2057 | ||
913d952e CM |
2058 | /* |
2059 | * we've got more storage, clear any full flags on the space | |
2060 | * infos | |
2061 | */ | |
2062 | btrfs_clear_space_info_full(root->fs_info); | |
2063 | ||
7d9eb12c | 2064 | unlock_chunks(root); |
5af3e8cc SB |
2065 | root->fs_info->num_tolerated_disk_barrier_failures = |
2066 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2067 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2068 | |
2b82032c YZ |
2069 | if (seeding_dev) { |
2070 | mutex_unlock(&uuid_mutex); | |
2071 | up_write(&sb->s_umount); | |
788f20eb | 2072 | |
79787eaa JM |
2073 | if (ret) /* transaction commit */ |
2074 | return ret; | |
2075 | ||
2b82032c | 2076 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
2077 | if (ret < 0) |
2078 | btrfs_error(root->fs_info, ret, | |
2079 | "Failed to relocate sys chunks after " | |
2080 | "device initialization. This can be fixed " | |
2081 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2082 | trans = btrfs_attach_transaction(root); |
2083 | if (IS_ERR(trans)) { | |
2084 | if (PTR_ERR(trans) == -ENOENT) | |
2085 | return 0; | |
2086 | return PTR_ERR(trans); | |
2087 | } | |
2088 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2089 | } |
c9e9f97b | 2090 | |
2b82032c | 2091 | return ret; |
79787eaa JM |
2092 | |
2093 | error_trans: | |
2094 | unlock_chunks(root); | |
79787eaa | 2095 | btrfs_end_transaction(trans, root); |
606686ee | 2096 | rcu_string_free(device->name); |
79787eaa | 2097 | kfree(device); |
2b82032c | 2098 | error: |
e525fd89 | 2099 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2100 | if (seeding_dev) { |
2101 | mutex_unlock(&uuid_mutex); | |
2102 | up_write(&sb->s_umount); | |
2103 | } | |
c9e9f97b | 2104 | return ret; |
788f20eb CM |
2105 | } |
2106 | ||
e93c89c1 SB |
2107 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
2108 | struct btrfs_device **device_out) | |
2109 | { | |
2110 | struct request_queue *q; | |
2111 | struct btrfs_device *device; | |
2112 | struct block_device *bdev; | |
2113 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2114 | struct list_head *devices; | |
2115 | struct rcu_string *name; | |
2116 | int ret = 0; | |
2117 | ||
2118 | *device_out = NULL; | |
2119 | if (fs_info->fs_devices->seeding) | |
2120 | return -EINVAL; | |
2121 | ||
2122 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2123 | fs_info->bdev_holder); | |
2124 | if (IS_ERR(bdev)) | |
2125 | return PTR_ERR(bdev); | |
2126 | ||
2127 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2128 | ||
2129 | devices = &fs_info->fs_devices->devices; | |
2130 | list_for_each_entry(device, devices, dev_list) { | |
2131 | if (device->bdev == bdev) { | |
2132 | ret = -EEXIST; | |
2133 | goto error; | |
2134 | } | |
2135 | } | |
2136 | ||
2137 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
2138 | if (!device) { | |
2139 | ret = -ENOMEM; | |
2140 | goto error; | |
2141 | } | |
2142 | ||
2143 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2144 | if (!name) { | |
2145 | kfree(device); | |
2146 | ret = -ENOMEM; | |
2147 | goto error; | |
2148 | } | |
2149 | rcu_assign_pointer(device->name, name); | |
2150 | ||
2151 | q = bdev_get_queue(bdev); | |
2152 | if (blk_queue_discard(q)) | |
2153 | device->can_discard = 1; | |
2154 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2155 | device->writeable = 1; | |
2156 | device->work.func = pending_bios_fn; | |
2157 | generate_random_uuid(device->uuid); | |
2158 | device->devid = BTRFS_DEV_REPLACE_DEVID; | |
2159 | spin_lock_init(&device->io_lock); | |
2160 | device->generation = 0; | |
2161 | device->io_width = root->sectorsize; | |
2162 | device->io_align = root->sectorsize; | |
2163 | device->sector_size = root->sectorsize; | |
2164 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2165 | device->disk_total_bytes = device->total_bytes; | |
2166 | device->dev_root = fs_info->dev_root; | |
2167 | device->bdev = bdev; | |
2168 | device->in_fs_metadata = 1; | |
2169 | device->is_tgtdev_for_dev_replace = 1; | |
2170 | device->mode = FMODE_EXCL; | |
2171 | set_blocksize(device->bdev, 4096); | |
2172 | device->fs_devices = fs_info->fs_devices; | |
2173 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2174 | fs_info->fs_devices->num_devices++; | |
2175 | fs_info->fs_devices->open_devices++; | |
2176 | if (device->can_discard) | |
2177 | fs_info->fs_devices->num_can_discard++; | |
2178 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2179 | ||
2180 | *device_out = device; | |
2181 | return ret; | |
2182 | ||
2183 | error: | |
2184 | blkdev_put(bdev, FMODE_EXCL); | |
2185 | return ret; | |
2186 | } | |
2187 | ||
2188 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2189 | struct btrfs_device *tgtdev) | |
2190 | { | |
2191 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2192 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2193 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2194 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2195 | tgtdev->dev_root = fs_info->dev_root; | |
2196 | tgtdev->in_fs_metadata = 1; | |
2197 | } | |
2198 | ||
d397712b CM |
2199 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2200 | struct btrfs_device *device) | |
0b86a832 CM |
2201 | { |
2202 | int ret; | |
2203 | struct btrfs_path *path; | |
2204 | struct btrfs_root *root; | |
2205 | struct btrfs_dev_item *dev_item; | |
2206 | struct extent_buffer *leaf; | |
2207 | struct btrfs_key key; | |
2208 | ||
2209 | root = device->dev_root->fs_info->chunk_root; | |
2210 | ||
2211 | path = btrfs_alloc_path(); | |
2212 | if (!path) | |
2213 | return -ENOMEM; | |
2214 | ||
2215 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2216 | key.type = BTRFS_DEV_ITEM_KEY; | |
2217 | key.offset = device->devid; | |
2218 | ||
2219 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2220 | if (ret < 0) | |
2221 | goto out; | |
2222 | ||
2223 | if (ret > 0) { | |
2224 | ret = -ENOENT; | |
2225 | goto out; | |
2226 | } | |
2227 | ||
2228 | leaf = path->nodes[0]; | |
2229 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2230 | ||
2231 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2232 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2233 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2234 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2235 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 2236 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
2237 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
2238 | btrfs_mark_buffer_dirty(leaf); | |
2239 | ||
2240 | out: | |
2241 | btrfs_free_path(path); | |
2242 | return ret; | |
2243 | } | |
2244 | ||
7d9eb12c | 2245 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2246 | struct btrfs_device *device, u64 new_size) |
2247 | { | |
2248 | struct btrfs_super_block *super_copy = | |
6c41761f | 2249 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
2250 | u64 old_total = btrfs_super_total_bytes(super_copy); |
2251 | u64 diff = new_size - device->total_bytes; | |
2252 | ||
2b82032c YZ |
2253 | if (!device->writeable) |
2254 | return -EACCES; | |
63a212ab SB |
2255 | if (new_size <= device->total_bytes || |
2256 | device->is_tgtdev_for_dev_replace) | |
2b82032c YZ |
2257 | return -EINVAL; |
2258 | ||
8f18cf13 | 2259 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2260 | device->fs_devices->total_rw_bytes += diff; |
2261 | ||
2262 | device->total_bytes = new_size; | |
9779b72f | 2263 | device->disk_total_bytes = new_size; |
4184ea7f CM |
2264 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
2265 | ||
8f18cf13 CM |
2266 | return btrfs_update_device(trans, device); |
2267 | } | |
2268 | ||
7d9eb12c CM |
2269 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
2270 | struct btrfs_device *device, u64 new_size) | |
2271 | { | |
2272 | int ret; | |
2273 | lock_chunks(device->dev_root); | |
2274 | ret = __btrfs_grow_device(trans, device, new_size); | |
2275 | unlock_chunks(device->dev_root); | |
2276 | return ret; | |
2277 | } | |
2278 | ||
8f18cf13 CM |
2279 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
2280 | struct btrfs_root *root, | |
2281 | u64 chunk_tree, u64 chunk_objectid, | |
2282 | u64 chunk_offset) | |
2283 | { | |
2284 | int ret; | |
2285 | struct btrfs_path *path; | |
2286 | struct btrfs_key key; | |
2287 | ||
2288 | root = root->fs_info->chunk_root; | |
2289 | path = btrfs_alloc_path(); | |
2290 | if (!path) | |
2291 | return -ENOMEM; | |
2292 | ||
2293 | key.objectid = chunk_objectid; | |
2294 | key.offset = chunk_offset; | |
2295 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2296 | ||
2297 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2298 | if (ret < 0) |
2299 | goto out; | |
2300 | else if (ret > 0) { /* Logic error or corruption */ | |
2301 | btrfs_error(root->fs_info, -ENOENT, | |
2302 | "Failed lookup while freeing chunk."); | |
2303 | ret = -ENOENT; | |
2304 | goto out; | |
2305 | } | |
8f18cf13 CM |
2306 | |
2307 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
2308 | if (ret < 0) |
2309 | btrfs_error(root->fs_info, ret, | |
2310 | "Failed to delete chunk item."); | |
2311 | out: | |
8f18cf13 | 2312 | btrfs_free_path(path); |
65a246c5 | 2313 | return ret; |
8f18cf13 CM |
2314 | } |
2315 | ||
b2950863 | 2316 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2317 | chunk_offset) |
2318 | { | |
6c41761f | 2319 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2320 | struct btrfs_disk_key *disk_key; |
2321 | struct btrfs_chunk *chunk; | |
2322 | u8 *ptr; | |
2323 | int ret = 0; | |
2324 | u32 num_stripes; | |
2325 | u32 array_size; | |
2326 | u32 len = 0; | |
2327 | u32 cur; | |
2328 | struct btrfs_key key; | |
2329 | ||
2330 | array_size = btrfs_super_sys_array_size(super_copy); | |
2331 | ||
2332 | ptr = super_copy->sys_chunk_array; | |
2333 | cur = 0; | |
2334 | ||
2335 | while (cur < array_size) { | |
2336 | disk_key = (struct btrfs_disk_key *)ptr; | |
2337 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2338 | ||
2339 | len = sizeof(*disk_key); | |
2340 | ||
2341 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2342 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2343 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2344 | len += btrfs_chunk_item_size(num_stripes); | |
2345 | } else { | |
2346 | ret = -EIO; | |
2347 | break; | |
2348 | } | |
2349 | if (key.objectid == chunk_objectid && | |
2350 | key.offset == chunk_offset) { | |
2351 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2352 | array_size -= len; | |
2353 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2354 | } else { | |
2355 | ptr += len; | |
2356 | cur += len; | |
2357 | } | |
2358 | } | |
2359 | return ret; | |
2360 | } | |
2361 | ||
b2950863 | 2362 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
2363 | u64 chunk_tree, u64 chunk_objectid, |
2364 | u64 chunk_offset) | |
2365 | { | |
2366 | struct extent_map_tree *em_tree; | |
2367 | struct btrfs_root *extent_root; | |
2368 | struct btrfs_trans_handle *trans; | |
2369 | struct extent_map *em; | |
2370 | struct map_lookup *map; | |
2371 | int ret; | |
2372 | int i; | |
2373 | ||
2374 | root = root->fs_info->chunk_root; | |
2375 | extent_root = root->fs_info->extent_root; | |
2376 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
2377 | ||
ba1bf481 JB |
2378 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2379 | if (ret) | |
2380 | return -ENOSPC; | |
2381 | ||
8f18cf13 | 2382 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 2383 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
2384 | if (ret) |
2385 | return ret; | |
8f18cf13 | 2386 | |
a22285a6 | 2387 | trans = btrfs_start_transaction(root, 0); |
0f788c58 LB |
2388 | if (IS_ERR(trans)) { |
2389 | ret = PTR_ERR(trans); | |
2390 | btrfs_std_error(root->fs_info, ret); | |
2391 | return ret; | |
2392 | } | |
8f18cf13 | 2393 | |
7d9eb12c CM |
2394 | lock_chunks(root); |
2395 | ||
8f18cf13 CM |
2396 | /* |
2397 | * step two, delete the device extents and the | |
2398 | * chunk tree entries | |
2399 | */ | |
890871be | 2400 | read_lock(&em_tree->lock); |
8f18cf13 | 2401 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2402 | read_unlock(&em_tree->lock); |
8f18cf13 | 2403 | |
285190d9 | 2404 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 2405 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
2406 | map = (struct map_lookup *)em->bdev; |
2407 | ||
2408 | for (i = 0; i < map->num_stripes; i++) { | |
2409 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
2410 | map->stripes[i].physical); | |
2411 | BUG_ON(ret); | |
a061fc8d | 2412 | |
dfe25020 CM |
2413 | if (map->stripes[i].dev) { |
2414 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2415 | BUG_ON(ret); | |
2416 | } | |
8f18cf13 CM |
2417 | } |
2418 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
2419 | chunk_offset); | |
2420 | ||
2421 | BUG_ON(ret); | |
2422 | ||
1abe9b8a | 2423 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2424 | ||
8f18cf13 CM |
2425 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2426 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
2427 | BUG_ON(ret); | |
8f18cf13 CM |
2428 | } |
2429 | ||
2b82032c YZ |
2430 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
2431 | BUG_ON(ret); | |
2432 | ||
890871be | 2433 | write_lock(&em_tree->lock); |
2b82032c | 2434 | remove_extent_mapping(em_tree, em); |
890871be | 2435 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2436 | |
2437 | kfree(map); | |
2438 | em->bdev = NULL; | |
2439 | ||
2440 | /* once for the tree */ | |
2441 | free_extent_map(em); | |
2442 | /* once for us */ | |
2443 | free_extent_map(em); | |
2444 | ||
2445 | unlock_chunks(root); | |
2446 | btrfs_end_transaction(trans, root); | |
2447 | return 0; | |
2448 | } | |
2449 | ||
2450 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2451 | { | |
2452 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2453 | struct btrfs_path *path; | |
2454 | struct extent_buffer *leaf; | |
2455 | struct btrfs_chunk *chunk; | |
2456 | struct btrfs_key key; | |
2457 | struct btrfs_key found_key; | |
2458 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2459 | u64 chunk_type; | |
ba1bf481 JB |
2460 | bool retried = false; |
2461 | int failed = 0; | |
2b82032c YZ |
2462 | int ret; |
2463 | ||
2464 | path = btrfs_alloc_path(); | |
2465 | if (!path) | |
2466 | return -ENOMEM; | |
2467 | ||
ba1bf481 | 2468 | again: |
2b82032c YZ |
2469 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2470 | key.offset = (u64)-1; | |
2471 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2472 | ||
2473 | while (1) { | |
2474 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2475 | if (ret < 0) | |
2476 | goto error; | |
79787eaa | 2477 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2478 | |
2479 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2480 | key.type); | |
2481 | if (ret < 0) | |
2482 | goto error; | |
2483 | if (ret > 0) | |
2484 | break; | |
1a40e23b | 2485 | |
2b82032c YZ |
2486 | leaf = path->nodes[0]; |
2487 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2488 | |
2b82032c YZ |
2489 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2490 | struct btrfs_chunk); | |
2491 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2492 | btrfs_release_path(path); |
8f18cf13 | 2493 | |
2b82032c YZ |
2494 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2495 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2496 | found_key.objectid, | |
2497 | found_key.offset); | |
ba1bf481 JB |
2498 | if (ret == -ENOSPC) |
2499 | failed++; | |
2500 | else if (ret) | |
2501 | BUG(); | |
2b82032c | 2502 | } |
8f18cf13 | 2503 | |
2b82032c YZ |
2504 | if (found_key.offset == 0) |
2505 | break; | |
2506 | key.offset = found_key.offset - 1; | |
2507 | } | |
2508 | ret = 0; | |
ba1bf481 JB |
2509 | if (failed && !retried) { |
2510 | failed = 0; | |
2511 | retried = true; | |
2512 | goto again; | |
2513 | } else if (failed && retried) { | |
2514 | WARN_ON(1); | |
2515 | ret = -ENOSPC; | |
2516 | } | |
2b82032c YZ |
2517 | error: |
2518 | btrfs_free_path(path); | |
2519 | return ret; | |
8f18cf13 CM |
2520 | } |
2521 | ||
0940ebf6 ID |
2522 | static int insert_balance_item(struct btrfs_root *root, |
2523 | struct btrfs_balance_control *bctl) | |
2524 | { | |
2525 | struct btrfs_trans_handle *trans; | |
2526 | struct btrfs_balance_item *item; | |
2527 | struct btrfs_disk_balance_args disk_bargs; | |
2528 | struct btrfs_path *path; | |
2529 | struct extent_buffer *leaf; | |
2530 | struct btrfs_key key; | |
2531 | int ret, err; | |
2532 | ||
2533 | path = btrfs_alloc_path(); | |
2534 | if (!path) | |
2535 | return -ENOMEM; | |
2536 | ||
2537 | trans = btrfs_start_transaction(root, 0); | |
2538 | if (IS_ERR(trans)) { | |
2539 | btrfs_free_path(path); | |
2540 | return PTR_ERR(trans); | |
2541 | } | |
2542 | ||
2543 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2544 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2545 | key.offset = 0; | |
2546 | ||
2547 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2548 | sizeof(*item)); | |
2549 | if (ret) | |
2550 | goto out; | |
2551 | ||
2552 | leaf = path->nodes[0]; | |
2553 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2554 | ||
2555 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2556 | ||
2557 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2558 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2559 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2560 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2561 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2562 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2563 | ||
2564 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2565 | ||
2566 | btrfs_mark_buffer_dirty(leaf); | |
2567 | out: | |
2568 | btrfs_free_path(path); | |
2569 | err = btrfs_commit_transaction(trans, root); | |
2570 | if (err && !ret) | |
2571 | ret = err; | |
2572 | return ret; | |
2573 | } | |
2574 | ||
2575 | static int del_balance_item(struct btrfs_root *root) | |
2576 | { | |
2577 | struct btrfs_trans_handle *trans; | |
2578 | struct btrfs_path *path; | |
2579 | struct btrfs_key key; | |
2580 | int ret, err; | |
2581 | ||
2582 | path = btrfs_alloc_path(); | |
2583 | if (!path) | |
2584 | return -ENOMEM; | |
2585 | ||
2586 | trans = btrfs_start_transaction(root, 0); | |
2587 | if (IS_ERR(trans)) { | |
2588 | btrfs_free_path(path); | |
2589 | return PTR_ERR(trans); | |
2590 | } | |
2591 | ||
2592 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2593 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2594 | key.offset = 0; | |
2595 | ||
2596 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2597 | if (ret < 0) | |
2598 | goto out; | |
2599 | if (ret > 0) { | |
2600 | ret = -ENOENT; | |
2601 | goto out; | |
2602 | } | |
2603 | ||
2604 | ret = btrfs_del_item(trans, root, path); | |
2605 | out: | |
2606 | btrfs_free_path(path); | |
2607 | err = btrfs_commit_transaction(trans, root); | |
2608 | if (err && !ret) | |
2609 | ret = err; | |
2610 | return ret; | |
2611 | } | |
2612 | ||
59641015 ID |
2613 | /* |
2614 | * This is a heuristic used to reduce the number of chunks balanced on | |
2615 | * resume after balance was interrupted. | |
2616 | */ | |
2617 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2618 | { | |
2619 | /* | |
2620 | * Turn on soft mode for chunk types that were being converted. | |
2621 | */ | |
2622 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2623 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2624 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2625 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2626 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2627 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2628 | ||
2629 | /* | |
2630 | * Turn on usage filter if is not already used. The idea is | |
2631 | * that chunks that we have already balanced should be | |
2632 | * reasonably full. Don't do it for chunks that are being | |
2633 | * converted - that will keep us from relocating unconverted | |
2634 | * (albeit full) chunks. | |
2635 | */ | |
2636 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2637 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2638 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2639 | bctl->data.usage = 90; | |
2640 | } | |
2641 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2642 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2643 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2644 | bctl->sys.usage = 90; | |
2645 | } | |
2646 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2647 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2648 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2649 | bctl->meta.usage = 90; | |
2650 | } | |
2651 | } | |
2652 | ||
c9e9f97b ID |
2653 | /* |
2654 | * Should be called with both balance and volume mutexes held to | |
2655 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2656 | * restriper. Same goes for unset_balance_control. | |
2657 | */ | |
2658 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2659 | { | |
2660 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2661 | ||
2662 | BUG_ON(fs_info->balance_ctl); | |
2663 | ||
2664 | spin_lock(&fs_info->balance_lock); | |
2665 | fs_info->balance_ctl = bctl; | |
2666 | spin_unlock(&fs_info->balance_lock); | |
2667 | } | |
2668 | ||
2669 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2670 | { | |
2671 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2672 | ||
2673 | BUG_ON(!fs_info->balance_ctl); | |
2674 | ||
2675 | spin_lock(&fs_info->balance_lock); | |
2676 | fs_info->balance_ctl = NULL; | |
2677 | spin_unlock(&fs_info->balance_lock); | |
2678 | ||
2679 | kfree(bctl); | |
2680 | } | |
2681 | ||
ed25e9b2 ID |
2682 | /* |
2683 | * Balance filters. Return 1 if chunk should be filtered out | |
2684 | * (should not be balanced). | |
2685 | */ | |
899c81ea | 2686 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2687 | struct btrfs_balance_args *bargs) |
2688 | { | |
899c81ea ID |
2689 | chunk_type = chunk_to_extended(chunk_type) & |
2690 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2691 | |
899c81ea | 2692 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2693 | return 0; |
2694 | ||
2695 | return 1; | |
2696 | } | |
2697 | ||
5ce5b3c0 ID |
2698 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
2699 | struct btrfs_balance_args *bargs) | |
2700 | { | |
2701 | struct btrfs_block_group_cache *cache; | |
2702 | u64 chunk_used, user_thresh; | |
2703 | int ret = 1; | |
2704 | ||
2705 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2706 | chunk_used = btrfs_block_group_used(&cache->item); | |
2707 | ||
a105bb88 | 2708 | if (bargs->usage == 0) |
3e39cea6 | 2709 | user_thresh = 1; |
a105bb88 ID |
2710 | else if (bargs->usage > 100) |
2711 | user_thresh = cache->key.offset; | |
2712 | else | |
2713 | user_thresh = div_factor_fine(cache->key.offset, | |
2714 | bargs->usage); | |
2715 | ||
5ce5b3c0 ID |
2716 | if (chunk_used < user_thresh) |
2717 | ret = 0; | |
2718 | ||
2719 | btrfs_put_block_group(cache); | |
2720 | return ret; | |
2721 | } | |
2722 | ||
409d404b ID |
2723 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2724 | struct btrfs_chunk *chunk, | |
2725 | struct btrfs_balance_args *bargs) | |
2726 | { | |
2727 | struct btrfs_stripe *stripe; | |
2728 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2729 | int i; | |
2730 | ||
2731 | for (i = 0; i < num_stripes; i++) { | |
2732 | stripe = btrfs_stripe_nr(chunk, i); | |
2733 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2734 | return 0; | |
2735 | } | |
2736 | ||
2737 | return 1; | |
2738 | } | |
2739 | ||
94e60d5a ID |
2740 | /* [pstart, pend) */ |
2741 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2742 | struct btrfs_chunk *chunk, | |
2743 | u64 chunk_offset, | |
2744 | struct btrfs_balance_args *bargs) | |
2745 | { | |
2746 | struct btrfs_stripe *stripe; | |
2747 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2748 | u64 stripe_offset; | |
2749 | u64 stripe_length; | |
2750 | int factor; | |
2751 | int i; | |
2752 | ||
2753 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2754 | return 0; | |
2755 | ||
2756 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
2757 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
2758 | factor = num_stripes / 2; | |
2759 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
2760 | factor = num_stripes - 1; | |
2761 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
2762 | factor = num_stripes - 2; | |
2763 | } else { | |
2764 | factor = num_stripes; | |
2765 | } | |
94e60d5a ID |
2766 | |
2767 | for (i = 0; i < num_stripes; i++) { | |
2768 | stripe = btrfs_stripe_nr(chunk, i); | |
2769 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2770 | continue; | |
2771 | ||
2772 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2773 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2774 | do_div(stripe_length, factor); | |
2775 | ||
2776 | if (stripe_offset < bargs->pend && | |
2777 | stripe_offset + stripe_length > bargs->pstart) | |
2778 | return 0; | |
2779 | } | |
2780 | ||
2781 | return 1; | |
2782 | } | |
2783 | ||
ea67176a ID |
2784 | /* [vstart, vend) */ |
2785 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
2786 | struct btrfs_chunk *chunk, | |
2787 | u64 chunk_offset, | |
2788 | struct btrfs_balance_args *bargs) | |
2789 | { | |
2790 | if (chunk_offset < bargs->vend && | |
2791 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
2792 | /* at least part of the chunk is inside this vrange */ | |
2793 | return 0; | |
2794 | ||
2795 | return 1; | |
2796 | } | |
2797 | ||
899c81ea | 2798 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
2799 | struct btrfs_balance_args *bargs) |
2800 | { | |
2801 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
2802 | return 0; | |
2803 | ||
899c81ea ID |
2804 | chunk_type = chunk_to_extended(chunk_type) & |
2805 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 2806 | |
899c81ea | 2807 | if (bargs->target == chunk_type) |
cfa4c961 ID |
2808 | return 1; |
2809 | ||
2810 | return 0; | |
2811 | } | |
2812 | ||
f43ffb60 ID |
2813 | static int should_balance_chunk(struct btrfs_root *root, |
2814 | struct extent_buffer *leaf, | |
2815 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
2816 | { | |
2817 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
2818 | struct btrfs_balance_args *bargs = NULL; | |
2819 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
2820 | ||
2821 | /* type filter */ | |
2822 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
2823 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
2824 | return 0; | |
2825 | } | |
2826 | ||
2827 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
2828 | bargs = &bctl->data; | |
2829 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
2830 | bargs = &bctl->sys; | |
2831 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
2832 | bargs = &bctl->meta; | |
2833 | ||
ed25e9b2 ID |
2834 | /* profiles filter */ |
2835 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
2836 | chunk_profiles_filter(chunk_type, bargs)) { | |
2837 | return 0; | |
5ce5b3c0 ID |
2838 | } |
2839 | ||
2840 | /* usage filter */ | |
2841 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2842 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
2843 | return 0; | |
409d404b ID |
2844 | } |
2845 | ||
2846 | /* devid filter */ | |
2847 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
2848 | chunk_devid_filter(leaf, chunk, bargs)) { | |
2849 | return 0; | |
94e60d5a ID |
2850 | } |
2851 | ||
2852 | /* drange filter, makes sense only with devid filter */ | |
2853 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
2854 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2855 | return 0; | |
ea67176a ID |
2856 | } |
2857 | ||
2858 | /* vrange filter */ | |
2859 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
2860 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2861 | return 0; | |
ed25e9b2 ID |
2862 | } |
2863 | ||
cfa4c961 ID |
2864 | /* soft profile changing mode */ |
2865 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
2866 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
2867 | return 0; | |
2868 | } | |
2869 | ||
f43ffb60 ID |
2870 | return 1; |
2871 | } | |
2872 | ||
c9e9f97b | 2873 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 2874 | { |
19a39dce | 2875 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
2876 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2877 | struct btrfs_root *dev_root = fs_info->dev_root; | |
2878 | struct list_head *devices; | |
ec44a35c CM |
2879 | struct btrfs_device *device; |
2880 | u64 old_size; | |
2881 | u64 size_to_free; | |
f43ffb60 | 2882 | struct btrfs_chunk *chunk; |
ec44a35c CM |
2883 | struct btrfs_path *path; |
2884 | struct btrfs_key key; | |
ec44a35c | 2885 | struct btrfs_key found_key; |
c9e9f97b | 2886 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
2887 | struct extent_buffer *leaf; |
2888 | int slot; | |
c9e9f97b ID |
2889 | int ret; |
2890 | int enospc_errors = 0; | |
19a39dce | 2891 | bool counting = true; |
ec44a35c | 2892 | |
ec44a35c | 2893 | /* step one make some room on all the devices */ |
c9e9f97b | 2894 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 2895 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2896 | old_size = device->total_bytes; |
2897 | size_to_free = div_factor(old_size, 1); | |
2898 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c | 2899 | if (!device->writeable || |
63a212ab SB |
2900 | device->total_bytes - device->bytes_used > size_to_free || |
2901 | device->is_tgtdev_for_dev_replace) | |
ec44a35c CM |
2902 | continue; |
2903 | ||
2904 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2905 | if (ret == -ENOSPC) |
2906 | break; | |
ec44a35c CM |
2907 | BUG_ON(ret); |
2908 | ||
a22285a6 | 2909 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2910 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2911 | |
2912 | ret = btrfs_grow_device(trans, device, old_size); | |
2913 | BUG_ON(ret); | |
2914 | ||
2915 | btrfs_end_transaction(trans, dev_root); | |
2916 | } | |
2917 | ||
2918 | /* step two, relocate all the chunks */ | |
2919 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2920 | if (!path) { |
2921 | ret = -ENOMEM; | |
2922 | goto error; | |
2923 | } | |
19a39dce ID |
2924 | |
2925 | /* zero out stat counters */ | |
2926 | spin_lock(&fs_info->balance_lock); | |
2927 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
2928 | spin_unlock(&fs_info->balance_lock); | |
2929 | again: | |
ec44a35c CM |
2930 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2931 | key.offset = (u64)-1; | |
2932 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2933 | ||
d397712b | 2934 | while (1) { |
19a39dce | 2935 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 2936 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
2937 | ret = -ECANCELED; |
2938 | goto error; | |
2939 | } | |
2940 | ||
ec44a35c CM |
2941 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2942 | if (ret < 0) | |
2943 | goto error; | |
2944 | ||
2945 | /* | |
2946 | * this shouldn't happen, it means the last relocate | |
2947 | * failed | |
2948 | */ | |
2949 | if (ret == 0) | |
c9e9f97b | 2950 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
2951 | |
2952 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2953 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
2954 | if (ret) { |
2955 | ret = 0; | |
ec44a35c | 2956 | break; |
c9e9f97b | 2957 | } |
7d9eb12c | 2958 | |
f43ffb60 ID |
2959 | leaf = path->nodes[0]; |
2960 | slot = path->slots[0]; | |
2961 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 2962 | |
ec44a35c CM |
2963 | if (found_key.objectid != key.objectid) |
2964 | break; | |
7d9eb12c | 2965 | |
ec44a35c | 2966 | /* chunk zero is special */ |
ba1bf481 | 2967 | if (found_key.offset == 0) |
ec44a35c CM |
2968 | break; |
2969 | ||
f43ffb60 ID |
2970 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
2971 | ||
19a39dce ID |
2972 | if (!counting) { |
2973 | spin_lock(&fs_info->balance_lock); | |
2974 | bctl->stat.considered++; | |
2975 | spin_unlock(&fs_info->balance_lock); | |
2976 | } | |
2977 | ||
f43ffb60 ID |
2978 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
2979 | found_key.offset); | |
b3b4aa74 | 2980 | btrfs_release_path(path); |
f43ffb60 ID |
2981 | if (!ret) |
2982 | goto loop; | |
2983 | ||
19a39dce ID |
2984 | if (counting) { |
2985 | spin_lock(&fs_info->balance_lock); | |
2986 | bctl->stat.expected++; | |
2987 | spin_unlock(&fs_info->balance_lock); | |
2988 | goto loop; | |
2989 | } | |
2990 | ||
ec44a35c CM |
2991 | ret = btrfs_relocate_chunk(chunk_root, |
2992 | chunk_root->root_key.objectid, | |
2993 | found_key.objectid, | |
2994 | found_key.offset); | |
508794eb JB |
2995 | if (ret && ret != -ENOSPC) |
2996 | goto error; | |
19a39dce | 2997 | if (ret == -ENOSPC) { |
c9e9f97b | 2998 | enospc_errors++; |
19a39dce ID |
2999 | } else { |
3000 | spin_lock(&fs_info->balance_lock); | |
3001 | bctl->stat.completed++; | |
3002 | spin_unlock(&fs_info->balance_lock); | |
3003 | } | |
f43ffb60 | 3004 | loop: |
ba1bf481 | 3005 | key.offset = found_key.offset - 1; |
ec44a35c | 3006 | } |
c9e9f97b | 3007 | |
19a39dce ID |
3008 | if (counting) { |
3009 | btrfs_release_path(path); | |
3010 | counting = false; | |
3011 | goto again; | |
3012 | } | |
ec44a35c CM |
3013 | error: |
3014 | btrfs_free_path(path); | |
c9e9f97b ID |
3015 | if (enospc_errors) { |
3016 | printk(KERN_INFO "btrfs: %d enospc errors during balance\n", | |
3017 | enospc_errors); | |
3018 | if (!ret) | |
3019 | ret = -ENOSPC; | |
3020 | } | |
3021 | ||
ec44a35c CM |
3022 | return ret; |
3023 | } | |
3024 | ||
0c460c0d ID |
3025 | /** |
3026 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3027 | * @flags: profile to validate | |
3028 | * @extended: if true @flags is treated as an extended profile | |
3029 | */ | |
3030 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3031 | { | |
3032 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3033 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3034 | ||
3035 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3036 | ||
3037 | /* 1) check that all other bits are zeroed */ | |
3038 | if (flags & ~mask) | |
3039 | return 0; | |
3040 | ||
3041 | /* 2) see if profile is reduced */ | |
3042 | if (flags == 0) | |
3043 | return !extended; /* "0" is valid for usual profiles */ | |
3044 | ||
3045 | /* true if exactly one bit set */ | |
3046 | return (flags & (flags - 1)) == 0; | |
3047 | } | |
3048 | ||
837d5b6e ID |
3049 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3050 | { | |
a7e99c69 ID |
3051 | /* cancel requested || normal exit path */ |
3052 | return atomic_read(&fs_info->balance_cancel_req) || | |
3053 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3054 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3055 | } |
3056 | ||
c9e9f97b ID |
3057 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3058 | { | |
0940ebf6 ID |
3059 | int ret; |
3060 | ||
c9e9f97b | 3061 | unset_balance_control(fs_info); |
0940ebf6 | 3062 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 LB |
3063 | if (ret) |
3064 | btrfs_std_error(fs_info, ret); | |
ed0fb78f ID |
3065 | |
3066 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3067 | } |
3068 | ||
19a39dce | 3069 | void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, |
c9e9f97b ID |
3070 | struct btrfs_ioctl_balance_args *bargs); |
3071 | ||
3072 | /* | |
3073 | * Should be called with both balance and volume mutexes held | |
3074 | */ | |
3075 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3076 | struct btrfs_ioctl_balance_args *bargs) | |
3077 | { | |
3078 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3079 | u64 allowed; |
e4837f8f | 3080 | int mixed = 0; |
c9e9f97b | 3081 | int ret; |
8dabb742 | 3082 | u64 num_devices; |
de98ced9 | 3083 | unsigned seq; |
c9e9f97b | 3084 | |
837d5b6e | 3085 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3086 | atomic_read(&fs_info->balance_pause_req) || |
3087 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3088 | ret = -EINVAL; |
3089 | goto out; | |
3090 | } | |
3091 | ||
e4837f8f ID |
3092 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3093 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3094 | mixed = 1; | |
3095 | ||
f43ffb60 ID |
3096 | /* |
3097 | * In case of mixed groups both data and meta should be picked, | |
3098 | * and identical options should be given for both of them. | |
3099 | */ | |
e4837f8f ID |
3100 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3101 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3102 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3103 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3104 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
3105 | printk(KERN_ERR "btrfs: with mixed groups data and " | |
3106 | "metadata balance options must be the same\n"); | |
3107 | ret = -EINVAL; | |
3108 | goto out; | |
3109 | } | |
3110 | } | |
3111 | ||
8dabb742 SB |
3112 | num_devices = fs_info->fs_devices->num_devices; |
3113 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
3114 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3115 | BUG_ON(num_devices < 1); | |
3116 | num_devices--; | |
3117 | } | |
3118 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
e4d8ec0f | 3119 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3120 | if (num_devices == 1) |
e4d8ec0f | 3121 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8dabb742 | 3122 | else if (num_devices < 4) |
e4d8ec0f ID |
3123 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
3124 | else | |
3125 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3126 | BTRFS_BLOCK_GROUP_RAID10 | |
3127 | BTRFS_BLOCK_GROUP_RAID5 | | |
3128 | BTRFS_BLOCK_GROUP_RAID6); | |
e4d8ec0f | 3129 | |
6728b198 ID |
3130 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3131 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
3132 | (bctl->data.target & ~allowed))) { | |
e4d8ec0f ID |
3133 | printk(KERN_ERR "btrfs: unable to start balance with target " |
3134 | "data profile %llu\n", | |
3135 | (unsigned long long)bctl->data.target); | |
3136 | ret = -EINVAL; | |
3137 | goto out; | |
3138 | } | |
6728b198 ID |
3139 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3140 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
3141 | (bctl->meta.target & ~allowed))) { | |
e4d8ec0f ID |
3142 | printk(KERN_ERR "btrfs: unable to start balance with target " |
3143 | "metadata profile %llu\n", | |
3144 | (unsigned long long)bctl->meta.target); | |
3145 | ret = -EINVAL; | |
3146 | goto out; | |
3147 | } | |
6728b198 ID |
3148 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3149 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
3150 | (bctl->sys.target & ~allowed))) { | |
e4d8ec0f ID |
3151 | printk(KERN_ERR "btrfs: unable to start balance with target " |
3152 | "system profile %llu\n", | |
3153 | (unsigned long long)bctl->sys.target); | |
3154 | ret = -EINVAL; | |
3155 | goto out; | |
3156 | } | |
3157 | ||
e4837f8f ID |
3158 | /* allow dup'ed data chunks only in mixed mode */ |
3159 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 3160 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
e4d8ec0f ID |
3161 | printk(KERN_ERR "btrfs: dup for data is not allowed\n"); |
3162 | ret = -EINVAL; | |
3163 | goto out; | |
3164 | } | |
3165 | ||
3166 | /* allow to reduce meta or sys integrity only if force set */ | |
3167 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3168 | BTRFS_BLOCK_GROUP_RAID10 | |
3169 | BTRFS_BLOCK_GROUP_RAID5 | | |
3170 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3171 | do { |
3172 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3173 | ||
3174 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3175 | (fs_info->avail_system_alloc_bits & allowed) && | |
3176 | !(bctl->sys.target & allowed)) || | |
3177 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3178 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3179 | !(bctl->meta.target & allowed))) { | |
3180 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
3181 | printk(KERN_INFO "btrfs: force reducing metadata " | |
3182 | "integrity\n"); | |
3183 | } else { | |
3184 | printk(KERN_ERR "btrfs: balance will reduce metadata " | |
3185 | "integrity, use force if you want this\n"); | |
3186 | ret = -EINVAL; | |
3187 | goto out; | |
3188 | } | |
e4d8ec0f | 3189 | } |
de98ced9 | 3190 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3191 | |
5af3e8cc SB |
3192 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3193 | int num_tolerated_disk_barrier_failures; | |
3194 | u64 target = bctl->sys.target; | |
3195 | ||
3196 | num_tolerated_disk_barrier_failures = | |
3197 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3198 | if (num_tolerated_disk_barrier_failures > 0 && | |
3199 | (target & | |
3200 | (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | | |
3201 | BTRFS_AVAIL_ALLOC_BIT_SINGLE))) | |
3202 | num_tolerated_disk_barrier_failures = 0; | |
3203 | else if (num_tolerated_disk_barrier_failures > 1 && | |
3204 | (target & | |
3205 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))) | |
3206 | num_tolerated_disk_barrier_failures = 1; | |
3207 | ||
3208 | fs_info->num_tolerated_disk_barrier_failures = | |
3209 | num_tolerated_disk_barrier_failures; | |
3210 | } | |
3211 | ||
0940ebf6 | 3212 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3213 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3214 | goto out; |
3215 | ||
59641015 ID |
3216 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3217 | BUG_ON(ret == -EEXIST); | |
3218 | set_balance_control(bctl); | |
3219 | } else { | |
3220 | BUG_ON(ret != -EEXIST); | |
3221 | spin_lock(&fs_info->balance_lock); | |
3222 | update_balance_args(bctl); | |
3223 | spin_unlock(&fs_info->balance_lock); | |
3224 | } | |
c9e9f97b | 3225 | |
837d5b6e | 3226 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3227 | mutex_unlock(&fs_info->balance_mutex); |
3228 | ||
3229 | ret = __btrfs_balance(fs_info); | |
3230 | ||
3231 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3232 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3233 | |
bf023ecf ID |
3234 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3235 | fs_info->num_tolerated_disk_barrier_failures = | |
3236 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3237 | } | |
3238 | ||
c9e9f97b ID |
3239 | if (bargs) { |
3240 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3241 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3242 | } |
3243 | ||
3a01aa7a ID |
3244 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3245 | balance_need_close(fs_info)) { | |
3246 | __cancel_balance(fs_info); | |
3247 | } | |
3248 | ||
837d5b6e | 3249 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3250 | |
3251 | return ret; | |
3252 | out: | |
59641015 ID |
3253 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3254 | __cancel_balance(fs_info); | |
ed0fb78f | 3255 | else { |
59641015 | 3256 | kfree(bctl); |
ed0fb78f ID |
3257 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3258 | } | |
59641015 ID |
3259 | return ret; |
3260 | } | |
3261 | ||
3262 | static int balance_kthread(void *data) | |
3263 | { | |
2b6ba629 | 3264 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3265 | int ret = 0; |
59641015 ID |
3266 | |
3267 | mutex_lock(&fs_info->volume_mutex); | |
3268 | mutex_lock(&fs_info->balance_mutex); | |
3269 | ||
2b6ba629 | 3270 | if (fs_info->balance_ctl) { |
9555c6c1 | 3271 | printk(KERN_INFO "btrfs: continuing balance\n"); |
2b6ba629 | 3272 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3273 | } |
59641015 ID |
3274 | |
3275 | mutex_unlock(&fs_info->balance_mutex); | |
3276 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3277 | |
59641015 ID |
3278 | return ret; |
3279 | } | |
3280 | ||
2b6ba629 ID |
3281 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3282 | { | |
3283 | struct task_struct *tsk; | |
3284 | ||
3285 | spin_lock(&fs_info->balance_lock); | |
3286 | if (!fs_info->balance_ctl) { | |
3287 | spin_unlock(&fs_info->balance_lock); | |
3288 | return 0; | |
3289 | } | |
3290 | spin_unlock(&fs_info->balance_lock); | |
3291 | ||
3292 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
3293 | printk(KERN_INFO "btrfs: force skipping balance\n"); | |
3294 | return 0; | |
3295 | } | |
3296 | ||
3297 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
3298 | if (IS_ERR(tsk)) | |
3299 | return PTR_ERR(tsk); | |
3300 | ||
3301 | return 0; | |
3302 | } | |
3303 | ||
68310a5e | 3304 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3305 | { |
59641015 ID |
3306 | struct btrfs_balance_control *bctl; |
3307 | struct btrfs_balance_item *item; | |
3308 | struct btrfs_disk_balance_args disk_bargs; | |
3309 | struct btrfs_path *path; | |
3310 | struct extent_buffer *leaf; | |
3311 | struct btrfs_key key; | |
3312 | int ret; | |
3313 | ||
3314 | path = btrfs_alloc_path(); | |
3315 | if (!path) | |
3316 | return -ENOMEM; | |
3317 | ||
59641015 ID |
3318 | key.objectid = BTRFS_BALANCE_OBJECTID; |
3319 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3320 | key.offset = 0; | |
3321 | ||
68310a5e | 3322 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3323 | if (ret < 0) |
68310a5e | 3324 | goto out; |
59641015 ID |
3325 | if (ret > 0) { /* ret = -ENOENT; */ |
3326 | ret = 0; | |
68310a5e ID |
3327 | goto out; |
3328 | } | |
3329 | ||
3330 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3331 | if (!bctl) { | |
3332 | ret = -ENOMEM; | |
3333 | goto out; | |
59641015 ID |
3334 | } |
3335 | ||
3336 | leaf = path->nodes[0]; | |
3337 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3338 | ||
68310a5e ID |
3339 | bctl->fs_info = fs_info; |
3340 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3341 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3342 | |
3343 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3344 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3345 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3346 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3347 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3348 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3349 | ||
ed0fb78f ID |
3350 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3351 | ||
68310a5e ID |
3352 | mutex_lock(&fs_info->volume_mutex); |
3353 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3354 | |
68310a5e ID |
3355 | set_balance_control(bctl); |
3356 | ||
3357 | mutex_unlock(&fs_info->balance_mutex); | |
3358 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3359 | out: |
3360 | btrfs_free_path(path); | |
ec44a35c CM |
3361 | return ret; |
3362 | } | |
3363 | ||
837d5b6e ID |
3364 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3365 | { | |
3366 | int ret = 0; | |
3367 | ||
3368 | mutex_lock(&fs_info->balance_mutex); | |
3369 | if (!fs_info->balance_ctl) { | |
3370 | mutex_unlock(&fs_info->balance_mutex); | |
3371 | return -ENOTCONN; | |
3372 | } | |
3373 | ||
3374 | if (atomic_read(&fs_info->balance_running)) { | |
3375 | atomic_inc(&fs_info->balance_pause_req); | |
3376 | mutex_unlock(&fs_info->balance_mutex); | |
3377 | ||
3378 | wait_event(fs_info->balance_wait_q, | |
3379 | atomic_read(&fs_info->balance_running) == 0); | |
3380 | ||
3381 | mutex_lock(&fs_info->balance_mutex); | |
3382 | /* we are good with balance_ctl ripped off from under us */ | |
3383 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3384 | atomic_dec(&fs_info->balance_pause_req); | |
3385 | } else { | |
3386 | ret = -ENOTCONN; | |
3387 | } | |
3388 | ||
3389 | mutex_unlock(&fs_info->balance_mutex); | |
3390 | return ret; | |
3391 | } | |
3392 | ||
a7e99c69 ID |
3393 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3394 | { | |
3395 | mutex_lock(&fs_info->balance_mutex); | |
3396 | if (!fs_info->balance_ctl) { | |
3397 | mutex_unlock(&fs_info->balance_mutex); | |
3398 | return -ENOTCONN; | |
3399 | } | |
3400 | ||
3401 | atomic_inc(&fs_info->balance_cancel_req); | |
3402 | /* | |
3403 | * if we are running just wait and return, balance item is | |
3404 | * deleted in btrfs_balance in this case | |
3405 | */ | |
3406 | if (atomic_read(&fs_info->balance_running)) { | |
3407 | mutex_unlock(&fs_info->balance_mutex); | |
3408 | wait_event(fs_info->balance_wait_q, | |
3409 | atomic_read(&fs_info->balance_running) == 0); | |
3410 | mutex_lock(&fs_info->balance_mutex); | |
3411 | } else { | |
3412 | /* __cancel_balance needs volume_mutex */ | |
3413 | mutex_unlock(&fs_info->balance_mutex); | |
3414 | mutex_lock(&fs_info->volume_mutex); | |
3415 | mutex_lock(&fs_info->balance_mutex); | |
3416 | ||
3417 | if (fs_info->balance_ctl) | |
3418 | __cancel_balance(fs_info); | |
3419 | ||
3420 | mutex_unlock(&fs_info->volume_mutex); | |
3421 | } | |
3422 | ||
3423 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3424 | atomic_dec(&fs_info->balance_cancel_req); | |
3425 | mutex_unlock(&fs_info->balance_mutex); | |
3426 | return 0; | |
3427 | } | |
3428 | ||
8f18cf13 CM |
3429 | /* |
3430 | * shrinking a device means finding all of the device extents past | |
3431 | * the new size, and then following the back refs to the chunks. | |
3432 | * The chunk relocation code actually frees the device extent | |
3433 | */ | |
3434 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
3435 | { | |
3436 | struct btrfs_trans_handle *trans; | |
3437 | struct btrfs_root *root = device->dev_root; | |
3438 | struct btrfs_dev_extent *dev_extent = NULL; | |
3439 | struct btrfs_path *path; | |
3440 | u64 length; | |
3441 | u64 chunk_tree; | |
3442 | u64 chunk_objectid; | |
3443 | u64 chunk_offset; | |
3444 | int ret; | |
3445 | int slot; | |
ba1bf481 JB |
3446 | int failed = 0; |
3447 | bool retried = false; | |
8f18cf13 CM |
3448 | struct extent_buffer *l; |
3449 | struct btrfs_key key; | |
6c41761f | 3450 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 3451 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 3452 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
3453 | u64 diff = device->total_bytes - new_size; |
3454 | ||
63a212ab SB |
3455 | if (device->is_tgtdev_for_dev_replace) |
3456 | return -EINVAL; | |
3457 | ||
8f18cf13 CM |
3458 | path = btrfs_alloc_path(); |
3459 | if (!path) | |
3460 | return -ENOMEM; | |
3461 | ||
8f18cf13 CM |
3462 | path->reada = 2; |
3463 | ||
7d9eb12c CM |
3464 | lock_chunks(root); |
3465 | ||
8f18cf13 | 3466 | device->total_bytes = new_size; |
2bf64758 | 3467 | if (device->writeable) { |
2b82032c | 3468 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
3469 | spin_lock(&root->fs_info->free_chunk_lock); |
3470 | root->fs_info->free_chunk_space -= diff; | |
3471 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3472 | } | |
7d9eb12c | 3473 | unlock_chunks(root); |
8f18cf13 | 3474 | |
ba1bf481 | 3475 | again: |
8f18cf13 CM |
3476 | key.objectid = device->devid; |
3477 | key.offset = (u64)-1; | |
3478 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3479 | ||
213e64da | 3480 | do { |
8f18cf13 CM |
3481 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3482 | if (ret < 0) | |
3483 | goto done; | |
3484 | ||
3485 | ret = btrfs_previous_item(root, path, 0, key.type); | |
3486 | if (ret < 0) | |
3487 | goto done; | |
3488 | if (ret) { | |
3489 | ret = 0; | |
b3b4aa74 | 3490 | btrfs_release_path(path); |
bf1fb512 | 3491 | break; |
8f18cf13 CM |
3492 | } |
3493 | ||
3494 | l = path->nodes[0]; | |
3495 | slot = path->slots[0]; | |
3496 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
3497 | ||
ba1bf481 | 3498 | if (key.objectid != device->devid) { |
b3b4aa74 | 3499 | btrfs_release_path(path); |
bf1fb512 | 3500 | break; |
ba1bf481 | 3501 | } |
8f18cf13 CM |
3502 | |
3503 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3504 | length = btrfs_dev_extent_length(l, dev_extent); | |
3505 | ||
ba1bf481 | 3506 | if (key.offset + length <= new_size) { |
b3b4aa74 | 3507 | btrfs_release_path(path); |
d6397bae | 3508 | break; |
ba1bf481 | 3509 | } |
8f18cf13 CM |
3510 | |
3511 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3512 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3513 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3514 | btrfs_release_path(path); |
8f18cf13 CM |
3515 | |
3516 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
3517 | chunk_offset); | |
ba1bf481 | 3518 | if (ret && ret != -ENOSPC) |
8f18cf13 | 3519 | goto done; |
ba1bf481 JB |
3520 | if (ret == -ENOSPC) |
3521 | failed++; | |
213e64da | 3522 | } while (key.offset-- > 0); |
ba1bf481 JB |
3523 | |
3524 | if (failed && !retried) { | |
3525 | failed = 0; | |
3526 | retried = true; | |
3527 | goto again; | |
3528 | } else if (failed && retried) { | |
3529 | ret = -ENOSPC; | |
3530 | lock_chunks(root); | |
3531 | ||
3532 | device->total_bytes = old_size; | |
3533 | if (device->writeable) | |
3534 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
3535 | spin_lock(&root->fs_info->free_chunk_lock); |
3536 | root->fs_info->free_chunk_space += diff; | |
3537 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
3538 | unlock_chunks(root); |
3539 | goto done; | |
8f18cf13 CM |
3540 | } |
3541 | ||
d6397bae | 3542 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 3543 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
3544 | if (IS_ERR(trans)) { |
3545 | ret = PTR_ERR(trans); | |
3546 | goto done; | |
3547 | } | |
3548 | ||
d6397bae CB |
3549 | lock_chunks(root); |
3550 | ||
3551 | device->disk_total_bytes = new_size; | |
3552 | /* Now btrfs_update_device() will change the on-disk size. */ | |
3553 | ret = btrfs_update_device(trans, device); | |
3554 | if (ret) { | |
3555 | unlock_chunks(root); | |
3556 | btrfs_end_transaction(trans, root); | |
3557 | goto done; | |
3558 | } | |
3559 | WARN_ON(diff > old_total); | |
3560 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
3561 | unlock_chunks(root); | |
3562 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
3563 | done: |
3564 | btrfs_free_path(path); | |
3565 | return ret; | |
3566 | } | |
3567 | ||
125ccb0a | 3568 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
3569 | struct btrfs_key *key, |
3570 | struct btrfs_chunk *chunk, int item_size) | |
3571 | { | |
6c41761f | 3572 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
3573 | struct btrfs_disk_key disk_key; |
3574 | u32 array_size; | |
3575 | u8 *ptr; | |
3576 | ||
3577 | array_size = btrfs_super_sys_array_size(super_copy); | |
3578 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
3579 | return -EFBIG; | |
3580 | ||
3581 | ptr = super_copy->sys_chunk_array + array_size; | |
3582 | btrfs_cpu_key_to_disk(&disk_key, key); | |
3583 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
3584 | ptr += sizeof(disk_key); | |
3585 | memcpy(ptr, chunk, item_size); | |
3586 | item_size += sizeof(disk_key); | |
3587 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
3588 | return 0; | |
3589 | } | |
3590 | ||
73c5de00 AJ |
3591 | /* |
3592 | * sort the devices in descending order by max_avail, total_avail | |
3593 | */ | |
3594 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 3595 | { |
73c5de00 AJ |
3596 | const struct btrfs_device_info *di_a = a; |
3597 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 3598 | |
73c5de00 | 3599 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 3600 | return -1; |
73c5de00 | 3601 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 3602 | return 1; |
73c5de00 AJ |
3603 | if (di_a->total_avail > di_b->total_avail) |
3604 | return -1; | |
3605 | if (di_a->total_avail < di_b->total_avail) | |
3606 | return 1; | |
3607 | return 0; | |
b2117a39 | 3608 | } |
0b86a832 | 3609 | |
31e50229 | 3610 | struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
e6ec716f MX |
3611 | [BTRFS_RAID_RAID10] = { |
3612 | .sub_stripes = 2, | |
3613 | .dev_stripes = 1, | |
3614 | .devs_max = 0, /* 0 == as many as possible */ | |
3615 | .devs_min = 4, | |
3616 | .devs_increment = 2, | |
3617 | .ncopies = 2, | |
3618 | }, | |
3619 | [BTRFS_RAID_RAID1] = { | |
3620 | .sub_stripes = 1, | |
3621 | .dev_stripes = 1, | |
3622 | .devs_max = 2, | |
3623 | .devs_min = 2, | |
3624 | .devs_increment = 2, | |
3625 | .ncopies = 2, | |
3626 | }, | |
3627 | [BTRFS_RAID_DUP] = { | |
3628 | .sub_stripes = 1, | |
3629 | .dev_stripes = 2, | |
3630 | .devs_max = 1, | |
3631 | .devs_min = 1, | |
3632 | .devs_increment = 1, | |
3633 | .ncopies = 2, | |
3634 | }, | |
3635 | [BTRFS_RAID_RAID0] = { | |
3636 | .sub_stripes = 1, | |
3637 | .dev_stripes = 1, | |
3638 | .devs_max = 0, | |
3639 | .devs_min = 2, | |
3640 | .devs_increment = 1, | |
3641 | .ncopies = 1, | |
3642 | }, | |
3643 | [BTRFS_RAID_SINGLE] = { | |
3644 | .sub_stripes = 1, | |
3645 | .dev_stripes = 1, | |
3646 | .devs_max = 1, | |
3647 | .devs_min = 1, | |
3648 | .devs_increment = 1, | |
3649 | .ncopies = 1, | |
3650 | }, | |
e942f883 CM |
3651 | [BTRFS_RAID_RAID5] = { |
3652 | .sub_stripes = 1, | |
3653 | .dev_stripes = 1, | |
3654 | .devs_max = 0, | |
3655 | .devs_min = 2, | |
3656 | .devs_increment = 1, | |
3657 | .ncopies = 2, | |
3658 | }, | |
3659 | [BTRFS_RAID_RAID6] = { | |
3660 | .sub_stripes = 1, | |
3661 | .dev_stripes = 1, | |
3662 | .devs_max = 0, | |
3663 | .devs_min = 3, | |
3664 | .devs_increment = 1, | |
3665 | .ncopies = 3, | |
3666 | }, | |
31e50229 LB |
3667 | }; |
3668 | ||
53b381b3 DW |
3669 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
3670 | { | |
3671 | /* TODO allow them to set a preferred stripe size */ | |
3672 | return 64 * 1024; | |
3673 | } | |
3674 | ||
3675 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
3676 | { | |
3677 | u64 features; | |
3678 | ||
3679 | if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))) | |
3680 | return; | |
3681 | ||
3682 | features = btrfs_super_incompat_flags(info->super_copy); | |
3683 | if (features & BTRFS_FEATURE_INCOMPAT_RAID56) | |
3684 | return; | |
3685 | ||
3686 | features |= BTRFS_FEATURE_INCOMPAT_RAID56; | |
3687 | btrfs_set_super_incompat_flags(info->super_copy, features); | |
3688 | printk(KERN_INFO "btrfs: setting RAID5/6 feature flag\n"); | |
3689 | } | |
3690 | ||
73c5de00 AJ |
3691 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
3692 | struct btrfs_root *extent_root, | |
3693 | struct map_lookup **map_ret, | |
3694 | u64 *num_bytes_out, u64 *stripe_size_out, | |
3695 | u64 start, u64 type) | |
b2117a39 | 3696 | { |
73c5de00 AJ |
3697 | struct btrfs_fs_info *info = extent_root->fs_info; |
3698 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
3699 | struct list_head *cur; | |
3700 | struct map_lookup *map = NULL; | |
3701 | struct extent_map_tree *em_tree; | |
3702 | struct extent_map *em; | |
3703 | struct btrfs_device_info *devices_info = NULL; | |
3704 | u64 total_avail; | |
3705 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
3706 | int data_stripes; /* number of stripes that count for |
3707 | block group size */ | |
73c5de00 AJ |
3708 | int sub_stripes; /* sub_stripes info for map */ |
3709 | int dev_stripes; /* stripes per dev */ | |
3710 | int devs_max; /* max devs to use */ | |
3711 | int devs_min; /* min devs needed */ | |
3712 | int devs_increment; /* ndevs has to be a multiple of this */ | |
3713 | int ncopies; /* how many copies to data has */ | |
3714 | int ret; | |
3715 | u64 max_stripe_size; | |
3716 | u64 max_chunk_size; | |
3717 | u64 stripe_size; | |
3718 | u64 num_bytes; | |
53b381b3 | 3719 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
3720 | int ndevs; |
3721 | int i; | |
3722 | int j; | |
31e50229 | 3723 | int index; |
593060d7 | 3724 | |
0c460c0d | 3725 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 3726 | |
73c5de00 AJ |
3727 | if (list_empty(&fs_devices->alloc_list)) |
3728 | return -ENOSPC; | |
b2117a39 | 3729 | |
31e50229 | 3730 | index = __get_raid_index(type); |
73c5de00 | 3731 | |
31e50229 LB |
3732 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
3733 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
3734 | devs_max = btrfs_raid_array[index].devs_max; | |
3735 | devs_min = btrfs_raid_array[index].devs_min; | |
3736 | devs_increment = btrfs_raid_array[index].devs_increment; | |
3737 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 3738 | |
9b3f68b9 | 3739 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
3740 | max_stripe_size = 1024 * 1024 * 1024; |
3741 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 3742 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
3743 | /* for larger filesystems, use larger metadata chunks */ |
3744 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
3745 | max_stripe_size = 1024 * 1024 * 1024; | |
3746 | else | |
3747 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 3748 | max_chunk_size = max_stripe_size; |
a40a90a0 | 3749 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 3750 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 AJ |
3751 | max_chunk_size = 2 * max_stripe_size; |
3752 | } else { | |
3753 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
3754 | type); | |
3755 | BUG_ON(1); | |
9b3f68b9 CM |
3756 | } |
3757 | ||
2b82032c YZ |
3758 | /* we don't want a chunk larger than 10% of writeable space */ |
3759 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
3760 | max_chunk_size); | |
9b3f68b9 | 3761 | |
73c5de00 AJ |
3762 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
3763 | GFP_NOFS); | |
3764 | if (!devices_info) | |
3765 | return -ENOMEM; | |
0cad8a11 | 3766 | |
73c5de00 | 3767 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 3768 | |
9f680ce0 | 3769 | /* |
73c5de00 AJ |
3770 | * in the first pass through the devices list, we gather information |
3771 | * about the available holes on each device. | |
9f680ce0 | 3772 | */ |
73c5de00 AJ |
3773 | ndevs = 0; |
3774 | while (cur != &fs_devices->alloc_list) { | |
3775 | struct btrfs_device *device; | |
3776 | u64 max_avail; | |
3777 | u64 dev_offset; | |
b2117a39 | 3778 | |
73c5de00 | 3779 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 3780 | |
73c5de00 | 3781 | cur = cur->next; |
b2117a39 | 3782 | |
73c5de00 | 3783 | if (!device->writeable) { |
31b1a2bd | 3784 | WARN(1, KERN_ERR |
73c5de00 | 3785 | "btrfs: read-only device in alloc_list\n"); |
73c5de00 AJ |
3786 | continue; |
3787 | } | |
b2117a39 | 3788 | |
63a212ab SB |
3789 | if (!device->in_fs_metadata || |
3790 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 3791 | continue; |
b2117a39 | 3792 | |
73c5de00 AJ |
3793 | if (device->total_bytes > device->bytes_used) |
3794 | total_avail = device->total_bytes - device->bytes_used; | |
3795 | else | |
3796 | total_avail = 0; | |
38c01b96 | 3797 | |
3798 | /* If there is no space on this device, skip it. */ | |
3799 | if (total_avail == 0) | |
3800 | continue; | |
b2117a39 | 3801 | |
125ccb0a | 3802 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
3803 | max_stripe_size * dev_stripes, |
3804 | &dev_offset, &max_avail); | |
3805 | if (ret && ret != -ENOSPC) | |
3806 | goto error; | |
b2117a39 | 3807 | |
73c5de00 AJ |
3808 | if (ret == 0) |
3809 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 3810 | |
73c5de00 AJ |
3811 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
3812 | continue; | |
b2117a39 | 3813 | |
063d006f ES |
3814 | if (ndevs == fs_devices->rw_devices) { |
3815 | WARN(1, "%s: found more than %llu devices\n", | |
3816 | __func__, fs_devices->rw_devices); | |
3817 | break; | |
3818 | } | |
73c5de00 AJ |
3819 | devices_info[ndevs].dev_offset = dev_offset; |
3820 | devices_info[ndevs].max_avail = max_avail; | |
3821 | devices_info[ndevs].total_avail = total_avail; | |
3822 | devices_info[ndevs].dev = device; | |
3823 | ++ndevs; | |
3824 | } | |
b2117a39 | 3825 | |
73c5de00 AJ |
3826 | /* |
3827 | * now sort the devices by hole size / available space | |
3828 | */ | |
3829 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
3830 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 3831 | |
73c5de00 AJ |
3832 | /* round down to number of usable stripes */ |
3833 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 3834 | |
73c5de00 AJ |
3835 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
3836 | ret = -ENOSPC; | |
3837 | goto error; | |
b2117a39 | 3838 | } |
9f680ce0 | 3839 | |
73c5de00 AJ |
3840 | if (devs_max && ndevs > devs_max) |
3841 | ndevs = devs_max; | |
3842 | /* | |
3843 | * the primary goal is to maximize the number of stripes, so use as many | |
3844 | * devices as possible, even if the stripes are not maximum sized. | |
3845 | */ | |
3846 | stripe_size = devices_info[ndevs-1].max_avail; | |
3847 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 3848 | |
53b381b3 DW |
3849 | /* |
3850 | * this will have to be fixed for RAID1 and RAID10 over | |
3851 | * more drives | |
3852 | */ | |
3853 | data_stripes = num_stripes / ncopies; | |
3854 | ||
53b381b3 DW |
3855 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
3856 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
3857 | btrfs_super_stripesize(info->super_copy)); | |
3858 | data_stripes = num_stripes - 1; | |
3859 | } | |
3860 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
3861 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
3862 | btrfs_super_stripesize(info->super_copy)); | |
3863 | data_stripes = num_stripes - 2; | |
3864 | } | |
86db2578 CM |
3865 | |
3866 | /* | |
3867 | * Use the number of data stripes to figure out how big this chunk | |
3868 | * is really going to be in terms of logical address space, | |
3869 | * and compare that answer with the max chunk size | |
3870 | */ | |
3871 | if (stripe_size * data_stripes > max_chunk_size) { | |
3872 | u64 mask = (1ULL << 24) - 1; | |
3873 | stripe_size = max_chunk_size; | |
3874 | do_div(stripe_size, data_stripes); | |
3875 | ||
3876 | /* bump the answer up to a 16MB boundary */ | |
3877 | stripe_size = (stripe_size + mask) & ~mask; | |
3878 | ||
3879 | /* but don't go higher than the limits we found | |
3880 | * while searching for free extents | |
3881 | */ | |
3882 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
3883 | stripe_size = devices_info[ndevs-1].max_avail; | |
3884 | } | |
3885 | ||
73c5de00 | 3886 | do_div(stripe_size, dev_stripes); |
37db63a4 ID |
3887 | |
3888 | /* align to BTRFS_STRIPE_LEN */ | |
53b381b3 DW |
3889 | do_div(stripe_size, raid_stripe_len); |
3890 | stripe_size *= raid_stripe_len; | |
b2117a39 MX |
3891 | |
3892 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
3893 | if (!map) { | |
3894 | ret = -ENOMEM; | |
3895 | goto error; | |
3896 | } | |
3897 | map->num_stripes = num_stripes; | |
9b3f68b9 | 3898 | |
73c5de00 AJ |
3899 | for (i = 0; i < ndevs; ++i) { |
3900 | for (j = 0; j < dev_stripes; ++j) { | |
3901 | int s = i * dev_stripes + j; | |
3902 | map->stripes[s].dev = devices_info[i].dev; | |
3903 | map->stripes[s].physical = devices_info[i].dev_offset + | |
3904 | j * stripe_size; | |
6324fbf3 | 3905 | } |
6324fbf3 | 3906 | } |
2b82032c | 3907 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
3908 | map->stripe_len = raid_stripe_len; |
3909 | map->io_align = raid_stripe_len; | |
3910 | map->io_width = raid_stripe_len; | |
2b82032c | 3911 | map->type = type; |
2b82032c | 3912 | map->sub_stripes = sub_stripes; |
0b86a832 | 3913 | |
2b82032c | 3914 | *map_ret = map; |
53b381b3 | 3915 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 3916 | |
73c5de00 AJ |
3917 | *stripe_size_out = stripe_size; |
3918 | *num_bytes_out = num_bytes; | |
0b86a832 | 3919 | |
73c5de00 | 3920 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 3921 | |
172ddd60 | 3922 | em = alloc_extent_map(); |
2b82032c | 3923 | if (!em) { |
b2117a39 MX |
3924 | ret = -ENOMEM; |
3925 | goto error; | |
593060d7 | 3926 | } |
2b82032c YZ |
3927 | em->bdev = (struct block_device *)map; |
3928 | em->start = start; | |
73c5de00 | 3929 | em->len = num_bytes; |
2b82032c YZ |
3930 | em->block_start = 0; |
3931 | em->block_len = em->len; | |
593060d7 | 3932 | |
2b82032c | 3933 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 3934 | write_lock(&em_tree->lock); |
2b82032c | 3935 | ret = add_extent_mapping(em_tree, em); |
890871be | 3936 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
3937 | if (ret) { |
3938 | free_extent_map(em); | |
1dd4602f | 3939 | goto error; |
0f5d42b2 | 3940 | } |
0b86a832 | 3941 | |
73c5de00 AJ |
3942 | for (i = 0; i < map->num_stripes; ++i) { |
3943 | struct btrfs_device *device; | |
3944 | u64 dev_offset; | |
3945 | ||
3946 | device = map->stripes[i].dev; | |
3947 | dev_offset = map->stripes[i].physical; | |
0b86a832 CM |
3948 | |
3949 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
3950 | info->chunk_root->root_key.objectid, |
3951 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3952 | start, dev_offset, stripe_size); |
04487488 JB |
3953 | if (ret) |
3954 | goto error_dev_extent; | |
3955 | } | |
3956 | ||
3957 | ret = btrfs_make_block_group(trans, extent_root, 0, type, | |
3958 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
3959 | start, num_bytes); | |
3960 | if (ret) { | |
3961 | i = map->num_stripes - 1; | |
3962 | goto error_dev_extent; | |
2b82032c YZ |
3963 | } |
3964 | ||
0f5d42b2 | 3965 | free_extent_map(em); |
53b381b3 DW |
3966 | check_raid56_incompat_flag(extent_root->fs_info, type); |
3967 | ||
b2117a39 | 3968 | kfree(devices_info); |
2b82032c | 3969 | return 0; |
b2117a39 | 3970 | |
04487488 JB |
3971 | error_dev_extent: |
3972 | for (; i >= 0; i--) { | |
3973 | struct btrfs_device *device; | |
3974 | int err; | |
3975 | ||
3976 | device = map->stripes[i].dev; | |
3977 | err = btrfs_free_dev_extent(trans, device, start); | |
3978 | if (err) { | |
3979 | btrfs_abort_transaction(trans, extent_root, err); | |
3980 | break; | |
3981 | } | |
3982 | } | |
0f5d42b2 JB |
3983 | write_lock(&em_tree->lock); |
3984 | remove_extent_mapping(em_tree, em); | |
3985 | write_unlock(&em_tree->lock); | |
3986 | ||
3987 | /* One for our allocation */ | |
3988 | free_extent_map(em); | |
3989 | /* One for the tree reference */ | |
3990 | free_extent_map(em); | |
b2117a39 MX |
3991 | error: |
3992 | kfree(map); | |
3993 | kfree(devices_info); | |
3994 | return ret; | |
2b82032c YZ |
3995 | } |
3996 | ||
3997 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
3998 | struct btrfs_root *extent_root, | |
3999 | struct map_lookup *map, u64 chunk_offset, | |
4000 | u64 chunk_size, u64 stripe_size) | |
4001 | { | |
4002 | u64 dev_offset; | |
4003 | struct btrfs_key key; | |
4004 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4005 | struct btrfs_device *device; | |
4006 | struct btrfs_chunk *chunk; | |
4007 | struct btrfs_stripe *stripe; | |
4008 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
4009 | int index = 0; | |
4010 | int ret; | |
4011 | ||
4012 | chunk = kzalloc(item_size, GFP_NOFS); | |
4013 | if (!chunk) | |
4014 | return -ENOMEM; | |
4015 | ||
4016 | index = 0; | |
4017 | while (index < map->num_stripes) { | |
4018 | device = map->stripes[index].dev; | |
4019 | device->bytes_used += stripe_size; | |
0b86a832 | 4020 | ret = btrfs_update_device(trans, device); |
3acd3953 MF |
4021 | if (ret) |
4022 | goto out_free; | |
2b82032c YZ |
4023 | index++; |
4024 | } | |
4025 | ||
2bf64758 JB |
4026 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4027 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4028 | map->num_stripes); | |
4029 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4030 | ||
2b82032c YZ |
4031 | index = 0; |
4032 | stripe = &chunk->stripe; | |
4033 | while (index < map->num_stripes) { | |
4034 | device = map->stripes[index].dev; | |
4035 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 4036 | |
e17cade2 CM |
4037 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4038 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4039 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4040 | stripe++; |
0b86a832 CM |
4041 | index++; |
4042 | } | |
4043 | ||
2b82032c | 4044 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4045 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4046 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4047 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4048 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4049 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4050 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4051 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4052 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4053 | |
2b82032c YZ |
4054 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4055 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4056 | key.offset = chunk_offset; | |
0b86a832 | 4057 | |
2b82032c | 4058 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
0b86a832 | 4059 | |
4ed1d16e MF |
4060 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4061 | /* | |
4062 | * TODO: Cleanup of inserted chunk root in case of | |
4063 | * failure. | |
4064 | */ | |
125ccb0a | 4065 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4066 | item_size); |
8f18cf13 | 4067 | } |
1abe9b8a | 4068 | |
3acd3953 | 4069 | out_free: |
0b86a832 | 4070 | kfree(chunk); |
4ed1d16e | 4071 | return ret; |
2b82032c | 4072 | } |
0b86a832 | 4073 | |
2b82032c YZ |
4074 | /* |
4075 | * Chunk allocation falls into two parts. The first part does works | |
4076 | * that make the new allocated chunk useable, but not do any operation | |
4077 | * that modifies the chunk tree. The second part does the works that | |
4078 | * require modifying the chunk tree. This division is important for the | |
4079 | * bootstrap process of adding storage to a seed btrfs. | |
4080 | */ | |
4081 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4082 | struct btrfs_root *extent_root, u64 type) | |
4083 | { | |
4084 | u64 chunk_offset; | |
4085 | u64 chunk_size; | |
4086 | u64 stripe_size; | |
4087 | struct map_lookup *map; | |
4088 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4089 | int ret; | |
4090 | ||
4091 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4092 | &chunk_offset); | |
4093 | if (ret) | |
4094 | return ret; | |
4095 | ||
4096 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
4097 | &stripe_size, chunk_offset, type); | |
4098 | if (ret) | |
4099 | return ret; | |
4100 | ||
4101 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
4102 | chunk_size, stripe_size); | |
79787eaa JM |
4103 | if (ret) |
4104 | return ret; | |
2b82032c YZ |
4105 | return 0; |
4106 | } | |
4107 | ||
d397712b | 4108 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4109 | struct btrfs_root *root, |
4110 | struct btrfs_device *device) | |
4111 | { | |
4112 | u64 chunk_offset; | |
4113 | u64 sys_chunk_offset; | |
4114 | u64 chunk_size; | |
4115 | u64 sys_chunk_size; | |
4116 | u64 stripe_size; | |
4117 | u64 sys_stripe_size; | |
4118 | u64 alloc_profile; | |
4119 | struct map_lookup *map; | |
4120 | struct map_lookup *sys_map; | |
4121 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4122 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4123 | int ret; | |
4124 | ||
4125 | ret = find_next_chunk(fs_info->chunk_root, | |
4126 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
92b8e897 MF |
4127 | if (ret) |
4128 | return ret; | |
2b82032c | 4129 | |
de98ced9 | 4130 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
2b82032c YZ |
4131 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, |
4132 | &stripe_size, chunk_offset, alloc_profile); | |
79787eaa JM |
4133 | if (ret) |
4134 | return ret; | |
2b82032c YZ |
4135 | |
4136 | sys_chunk_offset = chunk_offset + chunk_size; | |
4137 | ||
de98ced9 | 4138 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
2b82032c YZ |
4139 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, |
4140 | &sys_chunk_size, &sys_stripe_size, | |
4141 | sys_chunk_offset, alloc_profile); | |
005d6427 DS |
4142 | if (ret) { |
4143 | btrfs_abort_transaction(trans, root, ret); | |
4144 | goto out; | |
4145 | } | |
2b82032c YZ |
4146 | |
4147 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
005d6427 DS |
4148 | if (ret) { |
4149 | btrfs_abort_transaction(trans, root, ret); | |
4150 | goto out; | |
4151 | } | |
2b82032c YZ |
4152 | |
4153 | /* | |
4154 | * Modifying chunk tree needs allocating new blocks from both | |
4155 | * system block group and metadata block group. So we only can | |
4156 | * do operations require modifying the chunk tree after both | |
4157 | * block groups were created. | |
4158 | */ | |
4159 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
4160 | chunk_size, stripe_size); | |
005d6427 DS |
4161 | if (ret) { |
4162 | btrfs_abort_transaction(trans, root, ret); | |
4163 | goto out; | |
4164 | } | |
2b82032c YZ |
4165 | |
4166 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
4167 | sys_chunk_offset, sys_chunk_size, | |
4168 | sys_stripe_size); | |
79787eaa | 4169 | if (ret) |
005d6427 | 4170 | btrfs_abort_transaction(trans, root, ret); |
79787eaa | 4171 | |
005d6427 | 4172 | out: |
79787eaa | 4173 | |
79787eaa | 4174 | return ret; |
2b82032c YZ |
4175 | } |
4176 | ||
4177 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4178 | { | |
4179 | struct extent_map *em; | |
4180 | struct map_lookup *map; | |
4181 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4182 | int readonly = 0; | |
4183 | int i; | |
4184 | ||
890871be | 4185 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4186 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4187 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4188 | if (!em) |
4189 | return 1; | |
4190 | ||
f48b9075 JB |
4191 | if (btrfs_test_opt(root, DEGRADED)) { |
4192 | free_extent_map(em); | |
4193 | return 0; | |
4194 | } | |
4195 | ||
2b82032c YZ |
4196 | map = (struct map_lookup *)em->bdev; |
4197 | for (i = 0; i < map->num_stripes; i++) { | |
4198 | if (!map->stripes[i].dev->writeable) { | |
4199 | readonly = 1; | |
4200 | break; | |
4201 | } | |
4202 | } | |
0b86a832 | 4203 | free_extent_map(em); |
2b82032c | 4204 | return readonly; |
0b86a832 CM |
4205 | } |
4206 | ||
4207 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4208 | { | |
a8067e02 | 4209 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
4210 | } |
4211 | ||
4212 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4213 | { | |
4214 | struct extent_map *em; | |
4215 | ||
d397712b | 4216 | while (1) { |
890871be | 4217 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
4218 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
4219 | if (em) | |
4220 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 4221 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
4222 | if (!em) |
4223 | break; | |
4224 | kfree(em->bdev); | |
4225 | /* once for us */ | |
4226 | free_extent_map(em); | |
4227 | /* once for the tree */ | |
4228 | free_extent_map(em); | |
4229 | } | |
4230 | } | |
4231 | ||
5d964051 | 4232 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 4233 | { |
5d964051 | 4234 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
4235 | struct extent_map *em; |
4236 | struct map_lookup *map; | |
4237 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4238 | int ret; | |
4239 | ||
890871be | 4240 | read_lock(&em_tree->lock); |
f188591e | 4241 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 4242 | read_unlock(&em_tree->lock); |
f188591e CM |
4243 | BUG_ON(!em); |
4244 | ||
4245 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4246 | map = (struct map_lookup *)em->bdev; | |
4247 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
4248 | ret = map->num_stripes; | |
321aecc6 CM |
4249 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
4250 | ret = map->sub_stripes; | |
53b381b3 DW |
4251 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
4252 | ret = 2; | |
4253 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4254 | ret = 3; | |
f188591e CM |
4255 | else |
4256 | ret = 1; | |
4257 | free_extent_map(em); | |
ad6d620e SB |
4258 | |
4259 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
4260 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
4261 | ret++; | |
4262 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
4263 | ||
f188591e CM |
4264 | return ret; |
4265 | } | |
4266 | ||
53b381b3 DW |
4267 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
4268 | struct btrfs_mapping_tree *map_tree, | |
4269 | u64 logical) | |
4270 | { | |
4271 | struct extent_map *em; | |
4272 | struct map_lookup *map; | |
4273 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4274 | unsigned long len = root->sectorsize; | |
4275 | ||
4276 | read_lock(&em_tree->lock); | |
4277 | em = lookup_extent_mapping(em_tree, logical, len); | |
4278 | read_unlock(&em_tree->lock); | |
4279 | BUG_ON(!em); | |
4280 | ||
4281 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4282 | map = (struct map_lookup *)em->bdev; | |
4283 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4284 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4285 | len = map->stripe_len * nr_data_stripes(map); | |
4286 | } | |
4287 | free_extent_map(em); | |
4288 | return len; | |
4289 | } | |
4290 | ||
4291 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
4292 | u64 logical, u64 len, int mirror_num) | |
4293 | { | |
4294 | struct extent_map *em; | |
4295 | struct map_lookup *map; | |
4296 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4297 | int ret = 0; | |
4298 | ||
4299 | read_lock(&em_tree->lock); | |
4300 | em = lookup_extent_mapping(em_tree, logical, len); | |
4301 | read_unlock(&em_tree->lock); | |
4302 | BUG_ON(!em); | |
4303 | ||
4304 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4305 | map = (struct map_lookup *)em->bdev; | |
4306 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4307 | BTRFS_BLOCK_GROUP_RAID6)) | |
4308 | ret = 1; | |
4309 | free_extent_map(em); | |
4310 | return ret; | |
4311 | } | |
4312 | ||
30d9861f SB |
4313 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
4314 | struct map_lookup *map, int first, int num, | |
4315 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
4316 | { |
4317 | int i; | |
30d9861f SB |
4318 | int tolerance; |
4319 | struct btrfs_device *srcdev; | |
4320 | ||
4321 | if (dev_replace_is_ongoing && | |
4322 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
4323 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
4324 | srcdev = fs_info->dev_replace.srcdev; | |
4325 | else | |
4326 | srcdev = NULL; | |
4327 | ||
4328 | /* | |
4329 | * try to avoid the drive that is the source drive for a | |
4330 | * dev-replace procedure, only choose it if no other non-missing | |
4331 | * mirror is available | |
4332 | */ | |
4333 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
4334 | if (map->stripes[optimal].dev->bdev && | |
4335 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
4336 | return optimal; | |
4337 | for (i = first; i < first + num; i++) { | |
4338 | if (map->stripes[i].dev->bdev && | |
4339 | (tolerance || map->stripes[i].dev != srcdev)) | |
4340 | return i; | |
4341 | } | |
dfe25020 | 4342 | } |
30d9861f | 4343 | |
dfe25020 CM |
4344 | /* we couldn't find one that doesn't fail. Just return something |
4345 | * and the io error handling code will clean up eventually | |
4346 | */ | |
4347 | return optimal; | |
4348 | } | |
4349 | ||
53b381b3 DW |
4350 | static inline int parity_smaller(u64 a, u64 b) |
4351 | { | |
4352 | return a > b; | |
4353 | } | |
4354 | ||
4355 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
4356 | static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map) | |
4357 | { | |
4358 | struct btrfs_bio_stripe s; | |
4359 | int i; | |
4360 | u64 l; | |
4361 | int again = 1; | |
4362 | ||
4363 | while (again) { | |
4364 | again = 0; | |
4365 | for (i = 0; i < bbio->num_stripes - 1; i++) { | |
4366 | if (parity_smaller(raid_map[i], raid_map[i+1])) { | |
4367 | s = bbio->stripes[i]; | |
4368 | l = raid_map[i]; | |
4369 | bbio->stripes[i] = bbio->stripes[i+1]; | |
4370 | raid_map[i] = raid_map[i+1]; | |
4371 | bbio->stripes[i+1] = s; | |
4372 | raid_map[i+1] = l; | |
4373 | again = 1; | |
4374 | } | |
4375 | } | |
4376 | } | |
4377 | } | |
4378 | ||
3ec706c8 | 4379 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 4380 | u64 logical, u64 *length, |
a1d3c478 | 4381 | struct btrfs_bio **bbio_ret, |
53b381b3 | 4382 | int mirror_num, u64 **raid_map_ret) |
0b86a832 CM |
4383 | { |
4384 | struct extent_map *em; | |
4385 | struct map_lookup *map; | |
3ec706c8 | 4386 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
4387 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
4388 | u64 offset; | |
593060d7 | 4389 | u64 stripe_offset; |
fce3bb9a | 4390 | u64 stripe_end_offset; |
593060d7 | 4391 | u64 stripe_nr; |
fce3bb9a LD |
4392 | u64 stripe_nr_orig; |
4393 | u64 stripe_nr_end; | |
53b381b3 DW |
4394 | u64 stripe_len; |
4395 | u64 *raid_map = NULL; | |
593060d7 | 4396 | int stripe_index; |
cea9e445 | 4397 | int i; |
de11cc12 | 4398 | int ret = 0; |
f2d8d74d | 4399 | int num_stripes; |
a236aed1 | 4400 | int max_errors = 0; |
a1d3c478 | 4401 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
4402 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
4403 | int dev_replace_is_ongoing = 0; | |
4404 | int num_alloc_stripes; | |
ad6d620e SB |
4405 | int patch_the_first_stripe_for_dev_replace = 0; |
4406 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 4407 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 4408 | |
890871be | 4409 | read_lock(&em_tree->lock); |
0b86a832 | 4410 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 4411 | read_unlock(&em_tree->lock); |
f2d8d74d | 4412 | |
3b951516 | 4413 | if (!em) { |
48940662 | 4414 | printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n", |
d397712b CM |
4415 | (unsigned long long)logical, |
4416 | (unsigned long long)*length); | |
f2d8d74d | 4417 | BUG(); |
3b951516 | 4418 | } |
0b86a832 CM |
4419 | |
4420 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4421 | map = (struct map_lookup *)em->bdev; | |
4422 | offset = logical - em->start; | |
593060d7 | 4423 | |
53b381b3 DW |
4424 | if (mirror_num > map->num_stripes) |
4425 | mirror_num = 0; | |
4426 | ||
4427 | stripe_len = map->stripe_len; | |
593060d7 CM |
4428 | stripe_nr = offset; |
4429 | /* | |
4430 | * stripe_nr counts the total number of stripes we have to stride | |
4431 | * to get to this block | |
4432 | */ | |
53b381b3 | 4433 | do_div(stripe_nr, stripe_len); |
593060d7 | 4434 | |
53b381b3 | 4435 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
4436 | BUG_ON(offset < stripe_offset); |
4437 | ||
4438 | /* stripe_offset is the offset of this block in its stripe*/ | |
4439 | stripe_offset = offset - stripe_offset; | |
4440 | ||
53b381b3 DW |
4441 | /* if we're here for raid56, we need to know the stripe aligned start */ |
4442 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4443 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); | |
4444 | raid56_full_stripe_start = offset; | |
4445 | ||
4446 | /* allow a write of a full stripe, but make sure we don't | |
4447 | * allow straddling of stripes | |
4448 | */ | |
4449 | do_div(raid56_full_stripe_start, full_stripe_len); | |
4450 | raid56_full_stripe_start *= full_stripe_len; | |
4451 | } | |
4452 | ||
4453 | if (rw & REQ_DISCARD) { | |
4454 | /* we don't discard raid56 yet */ | |
4455 | if (map->type & | |
4456 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4457 | ret = -EOPNOTSUPP; | |
4458 | goto out; | |
4459 | } | |
fce3bb9a | 4460 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
4461 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
4462 | u64 max_len; | |
4463 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
4464 | For other RAID types and for RAID[56] reads, just allow a single | |
4465 | stripe (on a single disk). */ | |
4466 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) && | |
4467 | (rw & REQ_WRITE)) { | |
4468 | max_len = stripe_len * nr_data_stripes(map) - | |
4469 | (offset - raid56_full_stripe_start); | |
4470 | } else { | |
4471 | /* we limit the length of each bio to what fits in a stripe */ | |
4472 | max_len = stripe_len - stripe_offset; | |
4473 | } | |
4474 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
4475 | } else { |
4476 | *length = em->len - offset; | |
4477 | } | |
f2d8d74d | 4478 | |
53b381b3 DW |
4479 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
4480 | it cares about is the length */ | |
a1d3c478 | 4481 | if (!bbio_ret) |
cea9e445 CM |
4482 | goto out; |
4483 | ||
472262f3 SB |
4484 | btrfs_dev_replace_lock(dev_replace); |
4485 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
4486 | if (!dev_replace_is_ongoing) | |
4487 | btrfs_dev_replace_unlock(dev_replace); | |
4488 | ||
ad6d620e SB |
4489 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
4490 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
4491 | dev_replace->tgtdev != NULL) { | |
4492 | /* | |
4493 | * in dev-replace case, for repair case (that's the only | |
4494 | * case where the mirror is selected explicitly when | |
4495 | * calling btrfs_map_block), blocks left of the left cursor | |
4496 | * can also be read from the target drive. | |
4497 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
4498 | * the last one to the array of stripes. For READ, it also | |
4499 | * needs to be supported using the same mirror number. | |
4500 | * If the requested block is not left of the left cursor, | |
4501 | * EIO is returned. This can happen because btrfs_num_copies() | |
4502 | * returns one more in the dev-replace case. | |
4503 | */ | |
4504 | u64 tmp_length = *length; | |
4505 | struct btrfs_bio *tmp_bbio = NULL; | |
4506 | int tmp_num_stripes; | |
4507 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
4508 | int index_srcdev = 0; | |
4509 | int found = 0; | |
4510 | u64 physical_of_found = 0; | |
4511 | ||
4512 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
53b381b3 | 4513 | logical, &tmp_length, &tmp_bbio, 0, NULL); |
ad6d620e SB |
4514 | if (ret) { |
4515 | WARN_ON(tmp_bbio != NULL); | |
4516 | goto out; | |
4517 | } | |
4518 | ||
4519 | tmp_num_stripes = tmp_bbio->num_stripes; | |
4520 | if (mirror_num > tmp_num_stripes) { | |
4521 | /* | |
4522 | * REQ_GET_READ_MIRRORS does not contain this | |
4523 | * mirror, that means that the requested area | |
4524 | * is not left of the left cursor | |
4525 | */ | |
4526 | ret = -EIO; | |
4527 | kfree(tmp_bbio); | |
4528 | goto out; | |
4529 | } | |
4530 | ||
4531 | /* | |
4532 | * process the rest of the function using the mirror_num | |
4533 | * of the source drive. Therefore look it up first. | |
4534 | * At the end, patch the device pointer to the one of the | |
4535 | * target drive. | |
4536 | */ | |
4537 | for (i = 0; i < tmp_num_stripes; i++) { | |
4538 | if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) { | |
4539 | /* | |
4540 | * In case of DUP, in order to keep it | |
4541 | * simple, only add the mirror with the | |
4542 | * lowest physical address | |
4543 | */ | |
4544 | if (found && | |
4545 | physical_of_found <= | |
4546 | tmp_bbio->stripes[i].physical) | |
4547 | continue; | |
4548 | index_srcdev = i; | |
4549 | found = 1; | |
4550 | physical_of_found = | |
4551 | tmp_bbio->stripes[i].physical; | |
4552 | } | |
4553 | } | |
4554 | ||
4555 | if (found) { | |
4556 | mirror_num = index_srcdev + 1; | |
4557 | patch_the_first_stripe_for_dev_replace = 1; | |
4558 | physical_to_patch_in_first_stripe = physical_of_found; | |
4559 | } else { | |
4560 | WARN_ON(1); | |
4561 | ret = -EIO; | |
4562 | kfree(tmp_bbio); | |
4563 | goto out; | |
4564 | } | |
4565 | ||
4566 | kfree(tmp_bbio); | |
4567 | } else if (mirror_num > map->num_stripes) { | |
4568 | mirror_num = 0; | |
4569 | } | |
4570 | ||
f2d8d74d | 4571 | num_stripes = 1; |
cea9e445 | 4572 | stripe_index = 0; |
fce3bb9a | 4573 | stripe_nr_orig = stripe_nr; |
fda2832f | 4574 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
fce3bb9a LD |
4575 | do_div(stripe_nr_end, map->stripe_len); |
4576 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
4577 | (offset + *length); | |
53b381b3 | 4578 | |
fce3bb9a LD |
4579 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
4580 | if (rw & REQ_DISCARD) | |
4581 | num_stripes = min_t(u64, map->num_stripes, | |
4582 | stripe_nr_end - stripe_nr_orig); | |
4583 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
4584 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
29a8d9a0 | 4585 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 4586 | num_stripes = map->num_stripes; |
2fff734f | 4587 | else if (mirror_num) |
f188591e | 4588 | stripe_index = mirror_num - 1; |
dfe25020 | 4589 | else { |
30d9861f | 4590 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 4591 | map->num_stripes, |
30d9861f SB |
4592 | current->pid % map->num_stripes, |
4593 | dev_replace_is_ongoing); | |
a1d3c478 | 4594 | mirror_num = stripe_index + 1; |
dfe25020 | 4595 | } |
2fff734f | 4596 | |
611f0e00 | 4597 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 4598 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 4599 | num_stripes = map->num_stripes; |
a1d3c478 | 4600 | } else if (mirror_num) { |
f188591e | 4601 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
4602 | } else { |
4603 | mirror_num = 1; | |
4604 | } | |
2fff734f | 4605 | |
321aecc6 CM |
4606 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
4607 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
4608 | |
4609 | stripe_index = do_div(stripe_nr, factor); | |
4610 | stripe_index *= map->sub_stripes; | |
4611 | ||
29a8d9a0 | 4612 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 4613 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
4614 | else if (rw & REQ_DISCARD) |
4615 | num_stripes = min_t(u64, map->sub_stripes * | |
4616 | (stripe_nr_end - stripe_nr_orig), | |
4617 | map->num_stripes); | |
321aecc6 CM |
4618 | else if (mirror_num) |
4619 | stripe_index += mirror_num - 1; | |
dfe25020 | 4620 | else { |
3e74317a | 4621 | int old_stripe_index = stripe_index; |
30d9861f SB |
4622 | stripe_index = find_live_mirror(fs_info, map, |
4623 | stripe_index, | |
dfe25020 | 4624 | map->sub_stripes, stripe_index + |
30d9861f SB |
4625 | current->pid % map->sub_stripes, |
4626 | dev_replace_is_ongoing); | |
3e74317a | 4627 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 4628 | } |
53b381b3 DW |
4629 | |
4630 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
4631 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4632 | u64 tmp; | |
4633 | ||
4634 | if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1) | |
4635 | && raid_map_ret) { | |
4636 | int i, rot; | |
4637 | ||
4638 | /* push stripe_nr back to the start of the full stripe */ | |
4639 | stripe_nr = raid56_full_stripe_start; | |
4640 | do_div(stripe_nr, stripe_len); | |
4641 | ||
4642 | stripe_index = do_div(stripe_nr, nr_data_stripes(map)); | |
4643 | ||
4644 | /* RAID[56] write or recovery. Return all stripes */ | |
4645 | num_stripes = map->num_stripes; | |
4646 | max_errors = nr_parity_stripes(map); | |
4647 | ||
4648 | raid_map = kmalloc(sizeof(u64) * num_stripes, | |
4649 | GFP_NOFS); | |
4650 | if (!raid_map) { | |
4651 | ret = -ENOMEM; | |
4652 | goto out; | |
4653 | } | |
4654 | ||
4655 | /* Work out the disk rotation on this stripe-set */ | |
4656 | tmp = stripe_nr; | |
4657 | rot = do_div(tmp, num_stripes); | |
4658 | ||
4659 | /* Fill in the logical address of each stripe */ | |
4660 | tmp = stripe_nr * nr_data_stripes(map); | |
4661 | for (i = 0; i < nr_data_stripes(map); i++) | |
4662 | raid_map[(i+rot) % num_stripes] = | |
4663 | em->start + (tmp + i) * map->stripe_len; | |
4664 | ||
4665 | raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
4666 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4667 | raid_map[(i+rot+1) % num_stripes] = | |
4668 | RAID6_Q_STRIPE; | |
4669 | ||
4670 | *length = map->stripe_len; | |
4671 | stripe_index = 0; | |
4672 | stripe_offset = 0; | |
4673 | } else { | |
4674 | /* | |
4675 | * Mirror #0 or #1 means the original data block. | |
4676 | * Mirror #2 is RAID5 parity block. | |
4677 | * Mirror #3 is RAID6 Q block. | |
4678 | */ | |
4679 | stripe_index = do_div(stripe_nr, nr_data_stripes(map)); | |
4680 | if (mirror_num > 1) | |
4681 | stripe_index = nr_data_stripes(map) + | |
4682 | mirror_num - 2; | |
4683 | ||
4684 | /* We distribute the parity blocks across stripes */ | |
4685 | tmp = stripe_nr + stripe_index; | |
4686 | stripe_index = do_div(tmp, map->num_stripes); | |
4687 | } | |
8790d502 CM |
4688 | } else { |
4689 | /* | |
4690 | * after this do_div call, stripe_nr is the number of stripes | |
4691 | * on this device we have to walk to find the data, and | |
4692 | * stripe_index is the number of our device in the stripe array | |
4693 | */ | |
4694 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 4695 | mirror_num = stripe_index + 1; |
8790d502 | 4696 | } |
593060d7 | 4697 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 4698 | |
472262f3 | 4699 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
4700 | if (dev_replace_is_ongoing) { |
4701 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
4702 | num_alloc_stripes <<= 1; | |
4703 | if (rw & REQ_GET_READ_MIRRORS) | |
4704 | num_alloc_stripes++; | |
4705 | } | |
472262f3 | 4706 | bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS); |
de11cc12 LZ |
4707 | if (!bbio) { |
4708 | ret = -ENOMEM; | |
4709 | goto out; | |
4710 | } | |
4711 | atomic_set(&bbio->error, 0); | |
4712 | ||
fce3bb9a | 4713 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
4714 | int factor = 0; |
4715 | int sub_stripes = 0; | |
4716 | u64 stripes_per_dev = 0; | |
4717 | u32 remaining_stripes = 0; | |
b89203f7 | 4718 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
4719 | |
4720 | if (map->type & | |
4721 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
4722 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
4723 | sub_stripes = 1; | |
4724 | else | |
4725 | sub_stripes = map->sub_stripes; | |
4726 | ||
4727 | factor = map->num_stripes / sub_stripes; | |
4728 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
4729 | stripe_nr_orig, | |
4730 | factor, | |
4731 | &remaining_stripes); | |
b89203f7 LB |
4732 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
4733 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
4734 | } |
4735 | ||
fce3bb9a | 4736 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 4737 | bbio->stripes[i].physical = |
f2d8d74d CM |
4738 | map->stripes[stripe_index].physical + |
4739 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 4740 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 4741 | |
ec9ef7a1 LZ |
4742 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
4743 | BTRFS_BLOCK_GROUP_RAID10)) { | |
4744 | bbio->stripes[i].length = stripes_per_dev * | |
4745 | map->stripe_len; | |
b89203f7 | 4746 | |
ec9ef7a1 LZ |
4747 | if (i / sub_stripes < remaining_stripes) |
4748 | bbio->stripes[i].length += | |
4749 | map->stripe_len; | |
b89203f7 LB |
4750 | |
4751 | /* | |
4752 | * Special for the first stripe and | |
4753 | * the last stripe: | |
4754 | * | |
4755 | * |-------|...|-------| | |
4756 | * |----------| | |
4757 | * off end_off | |
4758 | */ | |
ec9ef7a1 | 4759 | if (i < sub_stripes) |
a1d3c478 | 4760 | bbio->stripes[i].length -= |
fce3bb9a | 4761 | stripe_offset; |
b89203f7 LB |
4762 | |
4763 | if (stripe_index >= last_stripe && | |
4764 | stripe_index <= (last_stripe + | |
4765 | sub_stripes - 1)) | |
a1d3c478 | 4766 | bbio->stripes[i].length -= |
fce3bb9a | 4767 | stripe_end_offset; |
b89203f7 | 4768 | |
ec9ef7a1 LZ |
4769 | if (i == sub_stripes - 1) |
4770 | stripe_offset = 0; | |
fce3bb9a | 4771 | } else |
a1d3c478 | 4772 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
4773 | |
4774 | stripe_index++; | |
4775 | if (stripe_index == map->num_stripes) { | |
4776 | /* This could only happen for RAID0/10 */ | |
4777 | stripe_index = 0; | |
4778 | stripe_nr++; | |
4779 | } | |
4780 | } | |
4781 | } else { | |
4782 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 4783 | bbio->stripes[i].physical = |
212a17ab LT |
4784 | map->stripes[stripe_index].physical + |
4785 | stripe_offset + | |
4786 | stripe_nr * map->stripe_len; | |
a1d3c478 | 4787 | bbio->stripes[i].dev = |
212a17ab | 4788 | map->stripes[stripe_index].dev; |
fce3bb9a | 4789 | stripe_index++; |
f2d8d74d | 4790 | } |
593060d7 | 4791 | } |
de11cc12 | 4792 | |
29a8d9a0 | 4793 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) { |
de11cc12 LZ |
4794 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | |
4795 | BTRFS_BLOCK_GROUP_RAID10 | | |
53b381b3 | 4796 | BTRFS_BLOCK_GROUP_RAID5 | |
de11cc12 LZ |
4797 | BTRFS_BLOCK_GROUP_DUP)) { |
4798 | max_errors = 1; | |
53b381b3 DW |
4799 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { |
4800 | max_errors = 2; | |
de11cc12 | 4801 | } |
f2d8d74d | 4802 | } |
de11cc12 | 4803 | |
472262f3 SB |
4804 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
4805 | dev_replace->tgtdev != NULL) { | |
4806 | int index_where_to_add; | |
4807 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
4808 | ||
4809 | /* | |
4810 | * duplicate the write operations while the dev replace | |
4811 | * procedure is running. Since the copying of the old disk | |
4812 | * to the new disk takes place at run time while the | |
4813 | * filesystem is mounted writable, the regular write | |
4814 | * operations to the old disk have to be duplicated to go | |
4815 | * to the new disk as well. | |
4816 | * Note that device->missing is handled by the caller, and | |
4817 | * that the write to the old disk is already set up in the | |
4818 | * stripes array. | |
4819 | */ | |
4820 | index_where_to_add = num_stripes; | |
4821 | for (i = 0; i < num_stripes; i++) { | |
4822 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
4823 | /* write to new disk, too */ | |
4824 | struct btrfs_bio_stripe *new = | |
4825 | bbio->stripes + index_where_to_add; | |
4826 | struct btrfs_bio_stripe *old = | |
4827 | bbio->stripes + i; | |
4828 | ||
4829 | new->physical = old->physical; | |
4830 | new->length = old->length; | |
4831 | new->dev = dev_replace->tgtdev; | |
4832 | index_where_to_add++; | |
4833 | max_errors++; | |
4834 | } | |
4835 | } | |
4836 | num_stripes = index_where_to_add; | |
ad6d620e SB |
4837 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
4838 | dev_replace->tgtdev != NULL) { | |
4839 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
4840 | int index_srcdev = 0; | |
4841 | int found = 0; | |
4842 | u64 physical_of_found = 0; | |
4843 | ||
4844 | /* | |
4845 | * During the dev-replace procedure, the target drive can | |
4846 | * also be used to read data in case it is needed to repair | |
4847 | * a corrupt block elsewhere. This is possible if the | |
4848 | * requested area is left of the left cursor. In this area, | |
4849 | * the target drive is a full copy of the source drive. | |
4850 | */ | |
4851 | for (i = 0; i < num_stripes; i++) { | |
4852 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
4853 | /* | |
4854 | * In case of DUP, in order to keep it | |
4855 | * simple, only add the mirror with the | |
4856 | * lowest physical address | |
4857 | */ | |
4858 | if (found && | |
4859 | physical_of_found <= | |
4860 | bbio->stripes[i].physical) | |
4861 | continue; | |
4862 | index_srcdev = i; | |
4863 | found = 1; | |
4864 | physical_of_found = bbio->stripes[i].physical; | |
4865 | } | |
4866 | } | |
4867 | if (found) { | |
4868 | u64 length = map->stripe_len; | |
4869 | ||
4870 | if (physical_of_found + length <= | |
4871 | dev_replace->cursor_left) { | |
4872 | struct btrfs_bio_stripe *tgtdev_stripe = | |
4873 | bbio->stripes + num_stripes; | |
4874 | ||
4875 | tgtdev_stripe->physical = physical_of_found; | |
4876 | tgtdev_stripe->length = | |
4877 | bbio->stripes[index_srcdev].length; | |
4878 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
4879 | ||
4880 | num_stripes++; | |
4881 | } | |
4882 | } | |
472262f3 SB |
4883 | } |
4884 | ||
de11cc12 LZ |
4885 | *bbio_ret = bbio; |
4886 | bbio->num_stripes = num_stripes; | |
4887 | bbio->max_errors = max_errors; | |
4888 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
4889 | |
4890 | /* | |
4891 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
4892 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
4893 | * available as a mirror | |
4894 | */ | |
4895 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
4896 | WARN_ON(num_stripes > 1); | |
4897 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
4898 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
4899 | bbio->mirror_num = map->num_stripes + 1; | |
4900 | } | |
53b381b3 DW |
4901 | if (raid_map) { |
4902 | sort_parity_stripes(bbio, raid_map); | |
4903 | *raid_map_ret = raid_map; | |
4904 | } | |
cea9e445 | 4905 | out: |
472262f3 SB |
4906 | if (dev_replace_is_ongoing) |
4907 | btrfs_dev_replace_unlock(dev_replace); | |
0b86a832 | 4908 | free_extent_map(em); |
de11cc12 | 4909 | return ret; |
0b86a832 CM |
4910 | } |
4911 | ||
3ec706c8 | 4912 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 4913 | u64 logical, u64 *length, |
a1d3c478 | 4914 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 4915 | { |
3ec706c8 | 4916 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
53b381b3 | 4917 | mirror_num, NULL); |
f2d8d74d CM |
4918 | } |
4919 | ||
a512bbf8 YZ |
4920 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
4921 | u64 chunk_start, u64 physical, u64 devid, | |
4922 | u64 **logical, int *naddrs, int *stripe_len) | |
4923 | { | |
4924 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4925 | struct extent_map *em; | |
4926 | struct map_lookup *map; | |
4927 | u64 *buf; | |
4928 | u64 bytenr; | |
4929 | u64 length; | |
4930 | u64 stripe_nr; | |
53b381b3 | 4931 | u64 rmap_len; |
a512bbf8 YZ |
4932 | int i, j, nr = 0; |
4933 | ||
890871be | 4934 | read_lock(&em_tree->lock); |
a512bbf8 | 4935 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 4936 | read_unlock(&em_tree->lock); |
a512bbf8 | 4937 | |
835d974f JB |
4938 | if (!em) { |
4939 | printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n", | |
4940 | chunk_start); | |
4941 | return -EIO; | |
4942 | } | |
4943 | ||
4944 | if (em->start != chunk_start) { | |
4945 | printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n", | |
4946 | em->start, chunk_start); | |
4947 | free_extent_map(em); | |
4948 | return -EIO; | |
4949 | } | |
a512bbf8 YZ |
4950 | map = (struct map_lookup *)em->bdev; |
4951 | ||
4952 | length = em->len; | |
53b381b3 DW |
4953 | rmap_len = map->stripe_len; |
4954 | ||
a512bbf8 YZ |
4955 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
4956 | do_div(length, map->num_stripes / map->sub_stripes); | |
4957 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
4958 | do_div(length, map->num_stripes); | |
53b381b3 DW |
4959 | else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | |
4960 | BTRFS_BLOCK_GROUP_RAID6)) { | |
4961 | do_div(length, nr_data_stripes(map)); | |
4962 | rmap_len = map->stripe_len * nr_data_stripes(map); | |
4963 | } | |
a512bbf8 YZ |
4964 | |
4965 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
79787eaa | 4966 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
4967 | |
4968 | for (i = 0; i < map->num_stripes; i++) { | |
4969 | if (devid && map->stripes[i].dev->devid != devid) | |
4970 | continue; | |
4971 | if (map->stripes[i].physical > physical || | |
4972 | map->stripes[i].physical + length <= physical) | |
4973 | continue; | |
4974 | ||
4975 | stripe_nr = physical - map->stripes[i].physical; | |
4976 | do_div(stripe_nr, map->stripe_len); | |
4977 | ||
4978 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
4979 | stripe_nr = stripe_nr * map->num_stripes + i; | |
4980 | do_div(stripe_nr, map->sub_stripes); | |
4981 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
4982 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
4983 | } /* else if RAID[56], multiply by nr_data_stripes(). |
4984 | * Alternatively, just use rmap_len below instead of | |
4985 | * map->stripe_len */ | |
4986 | ||
4987 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 4988 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
4989 | for (j = 0; j < nr; j++) { |
4990 | if (buf[j] == bytenr) | |
4991 | break; | |
4992 | } | |
934d375b CM |
4993 | if (j == nr) { |
4994 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 4995 | buf[nr++] = bytenr; |
934d375b | 4996 | } |
a512bbf8 YZ |
4997 | } |
4998 | ||
a512bbf8 YZ |
4999 | *logical = buf; |
5000 | *naddrs = nr; | |
53b381b3 | 5001 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5002 | |
5003 | free_extent_map(em); | |
5004 | return 0; | |
f2d8d74d CM |
5005 | } |
5006 | ||
442a4f63 SB |
5007 | static void *merge_stripe_index_into_bio_private(void *bi_private, |
5008 | unsigned int stripe_index) | |
5009 | { | |
5010 | /* | |
5011 | * with single, dup, RAID0, RAID1 and RAID10, stripe_index is | |
5012 | * at most 1. | |
5013 | * The alternative solution (instead of stealing bits from the | |
5014 | * pointer) would be to allocate an intermediate structure | |
5015 | * that contains the old private pointer plus the stripe_index. | |
5016 | */ | |
5017 | BUG_ON((((uintptr_t)bi_private) & 3) != 0); | |
5018 | BUG_ON(stripe_index > 3); | |
5019 | return (void *)(((uintptr_t)bi_private) | stripe_index); | |
5020 | } | |
5021 | ||
5022 | static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private) | |
5023 | { | |
5024 | return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3)); | |
5025 | } | |
5026 | ||
5027 | static unsigned int extract_stripe_index_from_bio_private(void *bi_private) | |
5028 | { | |
5029 | return (unsigned int)((uintptr_t)bi_private) & 3; | |
5030 | } | |
5031 | ||
a1d3c478 | 5032 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 5033 | { |
442a4f63 | 5034 | struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private); |
7d2b4daa | 5035 | int is_orig_bio = 0; |
8790d502 | 5036 | |
442a4f63 | 5037 | if (err) { |
a1d3c478 | 5038 | atomic_inc(&bbio->error); |
442a4f63 SB |
5039 | if (err == -EIO || err == -EREMOTEIO) { |
5040 | unsigned int stripe_index = | |
5041 | extract_stripe_index_from_bio_private( | |
5042 | bio->bi_private); | |
5043 | struct btrfs_device *dev; | |
5044 | ||
5045 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5046 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5047 | if (dev->bdev) { |
5048 | if (bio->bi_rw & WRITE) | |
5049 | btrfs_dev_stat_inc(dev, | |
5050 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5051 | else | |
5052 | btrfs_dev_stat_inc(dev, | |
5053 | BTRFS_DEV_STAT_READ_ERRS); | |
5054 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5055 | btrfs_dev_stat_inc(dev, | |
5056 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5057 | btrfs_dev_stat_print_on_error(dev); | |
5058 | } | |
442a4f63 SB |
5059 | } |
5060 | } | |
8790d502 | 5061 | |
a1d3c478 | 5062 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5063 | is_orig_bio = 1; |
5064 | ||
a1d3c478 | 5065 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5066 | if (!is_orig_bio) { |
5067 | bio_put(bio); | |
a1d3c478 | 5068 | bio = bbio->orig_bio; |
7d2b4daa | 5069 | } |
a1d3c478 JS |
5070 | bio->bi_private = bbio->private; |
5071 | bio->bi_end_io = bbio->end_io; | |
2774b2ca JS |
5072 | bio->bi_bdev = (struct block_device *) |
5073 | (unsigned long)bbio->mirror_num; | |
a236aed1 | 5074 | /* only send an error to the higher layers if it is |
53b381b3 | 5075 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5076 | */ |
a1d3c478 | 5077 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 5078 | err = -EIO; |
5dbc8fca | 5079 | } else { |
1259ab75 CM |
5080 | /* |
5081 | * this bio is actually up to date, we didn't | |
5082 | * go over the max number of errors | |
5083 | */ | |
5084 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 5085 | err = 0; |
1259ab75 | 5086 | } |
a1d3c478 | 5087 | kfree(bbio); |
8790d502 CM |
5088 | |
5089 | bio_endio(bio, err); | |
7d2b4daa | 5090 | } else if (!is_orig_bio) { |
8790d502 CM |
5091 | bio_put(bio); |
5092 | } | |
8790d502 CM |
5093 | } |
5094 | ||
8b712842 CM |
5095 | struct async_sched { |
5096 | struct bio *bio; | |
5097 | int rw; | |
5098 | struct btrfs_fs_info *info; | |
5099 | struct btrfs_work work; | |
5100 | }; | |
5101 | ||
5102 | /* | |
5103 | * see run_scheduled_bios for a description of why bios are collected for | |
5104 | * async submit. | |
5105 | * | |
5106 | * This will add one bio to the pending list for a device and make sure | |
5107 | * the work struct is scheduled. | |
5108 | */ | |
53b381b3 | 5109 | noinline void btrfs_schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
5110 | struct btrfs_device *device, |
5111 | int rw, struct bio *bio) | |
8b712842 CM |
5112 | { |
5113 | int should_queue = 1; | |
ffbd517d | 5114 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5115 | |
53b381b3 DW |
5116 | if (device->missing || !device->bdev) { |
5117 | bio_endio(bio, -EIO); | |
5118 | return; | |
5119 | } | |
5120 | ||
8b712842 | 5121 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5122 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5123 | bio_get(bio); |
21adbd5c | 5124 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5125 | bio_put(bio); |
143bede5 | 5126 | return; |
8b712842 CM |
5127 | } |
5128 | ||
5129 | /* | |
0986fe9e | 5130 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5131 | * higher layers. Otherwise, the async bio makes it appear we have |
5132 | * made progress against dirty pages when we've really just put it | |
5133 | * on a queue for later | |
5134 | */ | |
0986fe9e | 5135 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5136 | WARN_ON(bio->bi_next); |
8b712842 CM |
5137 | bio->bi_next = NULL; |
5138 | bio->bi_rw |= rw; | |
5139 | ||
5140 | spin_lock(&device->io_lock); | |
7b6d91da | 5141 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5142 | pending_bios = &device->pending_sync_bios; |
5143 | else | |
5144 | pending_bios = &device->pending_bios; | |
8b712842 | 5145 | |
ffbd517d CM |
5146 | if (pending_bios->tail) |
5147 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5148 | |
ffbd517d CM |
5149 | pending_bios->tail = bio; |
5150 | if (!pending_bios->head) | |
5151 | pending_bios->head = bio; | |
8b712842 CM |
5152 | if (device->running_pending) |
5153 | should_queue = 0; | |
5154 | ||
5155 | spin_unlock(&device->io_lock); | |
5156 | ||
5157 | if (should_queue) | |
1cc127b5 CM |
5158 | btrfs_queue_worker(&root->fs_info->submit_workers, |
5159 | &device->work); | |
8b712842 CM |
5160 | } |
5161 | ||
de1ee92a JB |
5162 | static int bio_size_ok(struct block_device *bdev, struct bio *bio, |
5163 | sector_t sector) | |
5164 | { | |
5165 | struct bio_vec *prev; | |
5166 | struct request_queue *q = bdev_get_queue(bdev); | |
5167 | unsigned short max_sectors = queue_max_sectors(q); | |
5168 | struct bvec_merge_data bvm = { | |
5169 | .bi_bdev = bdev, | |
5170 | .bi_sector = sector, | |
5171 | .bi_rw = bio->bi_rw, | |
5172 | }; | |
5173 | ||
5174 | if (bio->bi_vcnt == 0) { | |
5175 | WARN_ON(1); | |
5176 | return 1; | |
5177 | } | |
5178 | ||
5179 | prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
5180 | if ((bio->bi_size >> 9) > max_sectors) | |
5181 | return 0; | |
5182 | ||
5183 | if (!q->merge_bvec_fn) | |
5184 | return 1; | |
5185 | ||
5186 | bvm.bi_size = bio->bi_size - prev->bv_len; | |
5187 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) | |
5188 | return 0; | |
5189 | return 1; | |
5190 | } | |
5191 | ||
5192 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5193 | struct bio *bio, u64 physical, int dev_nr, | |
5194 | int rw, int async) | |
5195 | { | |
5196 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5197 | ||
5198 | bio->bi_private = bbio; | |
5199 | bio->bi_private = merge_stripe_index_into_bio_private( | |
5200 | bio->bi_private, (unsigned int)dev_nr); | |
5201 | bio->bi_end_io = btrfs_end_bio; | |
5202 | bio->bi_sector = physical >> 9; | |
5203 | #ifdef DEBUG | |
5204 | { | |
5205 | struct rcu_string *name; | |
5206 | ||
5207 | rcu_read_lock(); | |
5208 | name = rcu_dereference(dev->name); | |
d1423248 | 5209 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a JB |
5210 | "(%s id %llu), size=%u\n", rw, |
5211 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
5212 | name->str, dev->devid, bio->bi_size); | |
5213 | rcu_read_unlock(); | |
5214 | } | |
5215 | #endif | |
5216 | bio->bi_bdev = dev->bdev; | |
5217 | if (async) | |
53b381b3 | 5218 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
5219 | else |
5220 | btrfsic_submit_bio(rw, bio); | |
5221 | } | |
5222 | ||
5223 | static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5224 | struct bio *first_bio, struct btrfs_device *dev, | |
5225 | int dev_nr, int rw, int async) | |
5226 | { | |
5227 | struct bio_vec *bvec = first_bio->bi_io_vec; | |
5228 | struct bio *bio; | |
5229 | int nr_vecs = bio_get_nr_vecs(dev->bdev); | |
5230 | u64 physical = bbio->stripes[dev_nr].physical; | |
5231 | ||
5232 | again: | |
5233 | bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS); | |
5234 | if (!bio) | |
5235 | return -ENOMEM; | |
5236 | ||
5237 | while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) { | |
5238 | if (bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5239 | bvec->bv_offset) < bvec->bv_len) { | |
5240 | u64 len = bio->bi_size; | |
5241 | ||
5242 | atomic_inc(&bbio->stripes_pending); | |
5243 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, | |
5244 | rw, async); | |
5245 | physical += len; | |
5246 | goto again; | |
5247 | } | |
5248 | bvec++; | |
5249 | } | |
5250 | ||
5251 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async); | |
5252 | return 0; | |
5253 | } | |
5254 | ||
5255 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
5256 | { | |
5257 | atomic_inc(&bbio->error); | |
5258 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
5259 | bio->bi_private = bbio->private; | |
5260 | bio->bi_end_io = bbio->end_io; | |
5261 | bio->bi_bdev = (struct block_device *) | |
5262 | (unsigned long)bbio->mirror_num; | |
5263 | bio->bi_sector = logical >> 9; | |
5264 | kfree(bbio); | |
5265 | bio_endio(bio, -EIO); | |
5266 | } | |
5267 | } | |
5268 | ||
f188591e | 5269 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 5270 | int mirror_num, int async_submit) |
0b86a832 | 5271 | { |
0b86a832 | 5272 | struct btrfs_device *dev; |
8790d502 | 5273 | struct bio *first_bio = bio; |
a62b9401 | 5274 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
5275 | u64 length = 0; |
5276 | u64 map_length; | |
53b381b3 | 5277 | u64 *raid_map = NULL; |
0b86a832 | 5278 | int ret; |
8790d502 CM |
5279 | int dev_nr = 0; |
5280 | int total_devs = 1; | |
a1d3c478 | 5281 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 5282 | |
f2d8d74d | 5283 | length = bio->bi_size; |
0b86a832 | 5284 | map_length = length; |
cea9e445 | 5285 | |
53b381b3 DW |
5286 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
5287 | mirror_num, &raid_map); | |
5288 | if (ret) /* -ENOMEM */ | |
79787eaa | 5289 | return ret; |
cea9e445 | 5290 | |
a1d3c478 | 5291 | total_devs = bbio->num_stripes; |
53b381b3 DW |
5292 | bbio->orig_bio = first_bio; |
5293 | bbio->private = first_bio->bi_private; | |
5294 | bbio->end_io = first_bio->bi_end_io; | |
5295 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
5296 | ||
5297 | if (raid_map) { | |
5298 | /* In this case, map_length has been set to the length of | |
5299 | a single stripe; not the whole write */ | |
5300 | if (rw & WRITE) { | |
5301 | return raid56_parity_write(root, bio, bbio, | |
5302 | raid_map, map_length); | |
5303 | } else { | |
5304 | return raid56_parity_recover(root, bio, bbio, | |
5305 | raid_map, map_length, | |
5306 | mirror_num); | |
5307 | } | |
5308 | } | |
5309 | ||
cea9e445 | 5310 | if (map_length < length) { |
48940662 | 5311 | printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu " |
d397712b CM |
5312 | "len %llu\n", (unsigned long long)logical, |
5313 | (unsigned long long)length, | |
5314 | (unsigned long long)map_length); | |
cea9e445 CM |
5315 | BUG(); |
5316 | } | |
a1d3c478 | 5317 | |
d397712b | 5318 | while (dev_nr < total_devs) { |
de1ee92a JB |
5319 | dev = bbio->stripes[dev_nr].dev; |
5320 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
5321 | bbio_error(bbio, first_bio, logical); | |
5322 | dev_nr++; | |
5323 | continue; | |
5324 | } | |
5325 | ||
5326 | /* | |
5327 | * Check and see if we're ok with this bio based on it's size | |
5328 | * and offset with the given device. | |
5329 | */ | |
5330 | if (!bio_size_ok(dev->bdev, first_bio, | |
5331 | bbio->stripes[dev_nr].physical >> 9)) { | |
5332 | ret = breakup_stripe_bio(root, bbio, first_bio, dev, | |
5333 | dev_nr, rw, async_submit); | |
5334 | BUG_ON(ret); | |
5335 | dev_nr++; | |
5336 | continue; | |
5337 | } | |
5338 | ||
a1d3c478 JS |
5339 | if (dev_nr < total_devs - 1) { |
5340 | bio = bio_clone(first_bio, GFP_NOFS); | |
79787eaa | 5341 | BUG_ON(!bio); /* -ENOMEM */ |
a1d3c478 JS |
5342 | } else { |
5343 | bio = first_bio; | |
8790d502 | 5344 | } |
de1ee92a JB |
5345 | |
5346 | submit_stripe_bio(root, bbio, bio, | |
5347 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
5348 | async_submit); | |
8790d502 CM |
5349 | dev_nr++; |
5350 | } | |
0b86a832 CM |
5351 | return 0; |
5352 | } | |
5353 | ||
aa1b8cd4 | 5354 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 5355 | u8 *uuid, u8 *fsid) |
0b86a832 | 5356 | { |
2b82032c YZ |
5357 | struct btrfs_device *device; |
5358 | struct btrfs_fs_devices *cur_devices; | |
5359 | ||
aa1b8cd4 | 5360 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
5361 | while (cur_devices) { |
5362 | if (!fsid || | |
5363 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
5364 | device = __find_device(&cur_devices->devices, | |
5365 | devid, uuid); | |
5366 | if (device) | |
5367 | return device; | |
5368 | } | |
5369 | cur_devices = cur_devices->seed; | |
5370 | } | |
5371 | return NULL; | |
0b86a832 CM |
5372 | } |
5373 | ||
dfe25020 CM |
5374 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5375 | u64 devid, u8 *dev_uuid) | |
5376 | { | |
5377 | struct btrfs_device *device; | |
5378 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
5379 | ||
5380 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 5381 | if (!device) |
5382 | return NULL; | |
dfe25020 CM |
5383 | list_add(&device->dev_list, |
5384 | &fs_devices->devices); | |
dfe25020 CM |
5385 | device->dev_root = root->fs_info->dev_root; |
5386 | device->devid = devid; | |
8b712842 | 5387 | device->work.func = pending_bios_fn; |
e4404d6e | 5388 | device->fs_devices = fs_devices; |
cd02dca5 | 5389 | device->missing = 1; |
dfe25020 | 5390 | fs_devices->num_devices++; |
cd02dca5 | 5391 | fs_devices->missing_devices++; |
dfe25020 | 5392 | spin_lock_init(&device->io_lock); |
d20f7043 | 5393 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
5394 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
5395 | return device; | |
5396 | } | |
5397 | ||
0b86a832 CM |
5398 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
5399 | struct extent_buffer *leaf, | |
5400 | struct btrfs_chunk *chunk) | |
5401 | { | |
5402 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5403 | struct map_lookup *map; | |
5404 | struct extent_map *em; | |
5405 | u64 logical; | |
5406 | u64 length; | |
5407 | u64 devid; | |
a443755f | 5408 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 5409 | int num_stripes; |
0b86a832 | 5410 | int ret; |
593060d7 | 5411 | int i; |
0b86a832 | 5412 | |
e17cade2 CM |
5413 | logical = key->offset; |
5414 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 5415 | |
890871be | 5416 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 5417 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 5418 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
5419 | |
5420 | /* already mapped? */ | |
5421 | if (em && em->start <= logical && em->start + em->len > logical) { | |
5422 | free_extent_map(em); | |
0b86a832 CM |
5423 | return 0; |
5424 | } else if (em) { | |
5425 | free_extent_map(em); | |
5426 | } | |
0b86a832 | 5427 | |
172ddd60 | 5428 | em = alloc_extent_map(); |
0b86a832 CM |
5429 | if (!em) |
5430 | return -ENOMEM; | |
593060d7 CM |
5431 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
5432 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
5433 | if (!map) { |
5434 | free_extent_map(em); | |
5435 | return -ENOMEM; | |
5436 | } | |
5437 | ||
5438 | em->bdev = (struct block_device *)map; | |
5439 | em->start = logical; | |
5440 | em->len = length; | |
70c8a91c | 5441 | em->orig_start = 0; |
0b86a832 | 5442 | em->block_start = 0; |
c8b97818 | 5443 | em->block_len = em->len; |
0b86a832 | 5444 | |
593060d7 CM |
5445 | map->num_stripes = num_stripes; |
5446 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
5447 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
5448 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
5449 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
5450 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 5451 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
5452 | for (i = 0; i < num_stripes; i++) { |
5453 | map->stripes[i].physical = | |
5454 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
5455 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
5456 | read_extent_buffer(leaf, uuid, (unsigned long) |
5457 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
5458 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
5459 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
5460 | uuid, NULL); | |
dfe25020 | 5461 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
5462 | kfree(map); |
5463 | free_extent_map(em); | |
5464 | return -EIO; | |
5465 | } | |
dfe25020 CM |
5466 | if (!map->stripes[i].dev) { |
5467 | map->stripes[i].dev = | |
5468 | add_missing_dev(root, devid, uuid); | |
5469 | if (!map->stripes[i].dev) { | |
5470 | kfree(map); | |
5471 | free_extent_map(em); | |
5472 | return -EIO; | |
5473 | } | |
5474 | } | |
5475 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
5476 | } |
5477 | ||
890871be | 5478 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 5479 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 5480 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 5481 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
5482 | free_extent_map(em); |
5483 | ||
5484 | return 0; | |
5485 | } | |
5486 | ||
143bede5 | 5487 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
5488 | struct btrfs_dev_item *dev_item, |
5489 | struct btrfs_device *device) | |
5490 | { | |
5491 | unsigned long ptr; | |
0b86a832 CM |
5492 | |
5493 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
5494 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
5495 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
5496 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
5497 | device->type = btrfs_device_type(leaf, dev_item); | |
5498 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
5499 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
5500 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 5501 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 5502 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 CM |
5503 | |
5504 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 5505 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
5506 | } |
5507 | ||
2b82032c YZ |
5508 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
5509 | { | |
5510 | struct btrfs_fs_devices *fs_devices; | |
5511 | int ret; | |
5512 | ||
b367e47f | 5513 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
5514 | |
5515 | fs_devices = root->fs_info->fs_devices->seed; | |
5516 | while (fs_devices) { | |
5517 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
5518 | ret = 0; | |
5519 | goto out; | |
5520 | } | |
5521 | fs_devices = fs_devices->seed; | |
5522 | } | |
5523 | ||
5524 | fs_devices = find_fsid(fsid); | |
5525 | if (!fs_devices) { | |
5526 | ret = -ENOENT; | |
5527 | goto out; | |
5528 | } | |
e4404d6e YZ |
5529 | |
5530 | fs_devices = clone_fs_devices(fs_devices); | |
5531 | if (IS_ERR(fs_devices)) { | |
5532 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
5533 | goto out; |
5534 | } | |
5535 | ||
97288f2c | 5536 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 5537 | root->fs_info->bdev_holder); |
48d28232 JL |
5538 | if (ret) { |
5539 | free_fs_devices(fs_devices); | |
2b82032c | 5540 | goto out; |
48d28232 | 5541 | } |
2b82032c YZ |
5542 | |
5543 | if (!fs_devices->seeding) { | |
5544 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 5545 | free_fs_devices(fs_devices); |
2b82032c YZ |
5546 | ret = -EINVAL; |
5547 | goto out; | |
5548 | } | |
5549 | ||
5550 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
5551 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 5552 | out: |
2b82032c YZ |
5553 | return ret; |
5554 | } | |
5555 | ||
0d81ba5d | 5556 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
5557 | struct extent_buffer *leaf, |
5558 | struct btrfs_dev_item *dev_item) | |
5559 | { | |
5560 | struct btrfs_device *device; | |
5561 | u64 devid; | |
5562 | int ret; | |
2b82032c | 5563 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
5564 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
5565 | ||
0b86a832 | 5566 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
5567 | read_extent_buffer(leaf, dev_uuid, |
5568 | (unsigned long)btrfs_device_uuid(dev_item), | |
5569 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
5570 | read_extent_buffer(leaf, fs_uuid, |
5571 | (unsigned long)btrfs_device_fsid(dev_item), | |
5572 | BTRFS_UUID_SIZE); | |
5573 | ||
5574 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5575 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 5576 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 5577 | return ret; |
2b82032c YZ |
5578 | } |
5579 | ||
aa1b8cd4 | 5580 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
2b82032c | 5581 | if (!device || !device->bdev) { |
e4404d6e | 5582 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
5583 | return -EIO; |
5584 | ||
5585 | if (!device) { | |
d397712b CM |
5586 | printk(KERN_WARNING "warning devid %llu missing\n", |
5587 | (unsigned long long)devid); | |
2b82032c YZ |
5588 | device = add_missing_dev(root, devid, dev_uuid); |
5589 | if (!device) | |
5590 | return -ENOMEM; | |
cd02dca5 CM |
5591 | } else if (!device->missing) { |
5592 | /* | |
5593 | * this happens when a device that was properly setup | |
5594 | * in the device info lists suddenly goes bad. | |
5595 | * device->bdev is NULL, and so we have to set | |
5596 | * device->missing to one here | |
5597 | */ | |
5598 | root->fs_info->fs_devices->missing_devices++; | |
5599 | device->missing = 1; | |
2b82032c YZ |
5600 | } |
5601 | } | |
5602 | ||
5603 | if (device->fs_devices != root->fs_info->fs_devices) { | |
5604 | BUG_ON(device->writeable); | |
5605 | if (device->generation != | |
5606 | btrfs_device_generation(leaf, dev_item)) | |
5607 | return -EINVAL; | |
6324fbf3 | 5608 | } |
0b86a832 CM |
5609 | |
5610 | fill_device_from_item(leaf, dev_item, device); | |
5611 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 5612 | device->in_fs_metadata = 1; |
63a212ab | 5613 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 5614 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
5615 | spin_lock(&root->fs_info->free_chunk_lock); |
5616 | root->fs_info->free_chunk_space += device->total_bytes - | |
5617 | device->bytes_used; | |
5618 | spin_unlock(&root->fs_info->free_chunk_lock); | |
5619 | } | |
0b86a832 | 5620 | ret = 0; |
0b86a832 CM |
5621 | return ret; |
5622 | } | |
5623 | ||
e4404d6e | 5624 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 5625 | { |
6c41761f | 5626 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 5627 | struct extent_buffer *sb; |
0b86a832 | 5628 | struct btrfs_disk_key *disk_key; |
0b86a832 | 5629 | struct btrfs_chunk *chunk; |
84eed90f CM |
5630 | u8 *ptr; |
5631 | unsigned long sb_ptr; | |
5632 | int ret = 0; | |
0b86a832 CM |
5633 | u32 num_stripes; |
5634 | u32 array_size; | |
5635 | u32 len = 0; | |
0b86a832 | 5636 | u32 cur; |
84eed90f | 5637 | struct btrfs_key key; |
0b86a832 | 5638 | |
e4404d6e | 5639 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
5640 | BTRFS_SUPER_INFO_SIZE); |
5641 | if (!sb) | |
5642 | return -ENOMEM; | |
5643 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 5644 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
5645 | /* |
5646 | * The sb extent buffer is artifical and just used to read the system array. | |
5647 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
5648 | * pages up-to-date when the page is larger: extent does not cover the | |
5649 | * whole page and consequently check_page_uptodate does not find all | |
5650 | * the page's extents up-to-date (the hole beyond sb), | |
5651 | * write_extent_buffer then triggers a WARN_ON. | |
5652 | * | |
5653 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
5654 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
5655 | * to silence the warning eg. on PowerPC 64. | |
5656 | */ | |
5657 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 5658 | SetPageUptodate(sb->pages[0]); |
4008c04a | 5659 | |
a061fc8d | 5660 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
5661 | array_size = btrfs_super_sys_array_size(super_copy); |
5662 | ||
0b86a832 CM |
5663 | ptr = super_copy->sys_chunk_array; |
5664 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
5665 | cur = 0; | |
5666 | ||
5667 | while (cur < array_size) { | |
5668 | disk_key = (struct btrfs_disk_key *)ptr; | |
5669 | btrfs_disk_key_to_cpu(&key, disk_key); | |
5670 | ||
a061fc8d | 5671 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
5672 | sb_ptr += len; |
5673 | cur += len; | |
5674 | ||
0d81ba5d | 5675 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 5676 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 5677 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
5678 | if (ret) |
5679 | break; | |
0b86a832 CM |
5680 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
5681 | len = btrfs_chunk_item_size(num_stripes); | |
5682 | } else { | |
84eed90f CM |
5683 | ret = -EIO; |
5684 | break; | |
0b86a832 CM |
5685 | } |
5686 | ptr += len; | |
5687 | sb_ptr += len; | |
5688 | cur += len; | |
5689 | } | |
a061fc8d | 5690 | free_extent_buffer(sb); |
84eed90f | 5691 | return ret; |
0b86a832 CM |
5692 | } |
5693 | ||
5694 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
5695 | { | |
5696 | struct btrfs_path *path; | |
5697 | struct extent_buffer *leaf; | |
5698 | struct btrfs_key key; | |
5699 | struct btrfs_key found_key; | |
5700 | int ret; | |
5701 | int slot; | |
5702 | ||
5703 | root = root->fs_info->chunk_root; | |
5704 | ||
5705 | path = btrfs_alloc_path(); | |
5706 | if (!path) | |
5707 | return -ENOMEM; | |
5708 | ||
b367e47f LZ |
5709 | mutex_lock(&uuid_mutex); |
5710 | lock_chunks(root); | |
5711 | ||
0b86a832 CM |
5712 | /* first we search for all of the device items, and then we |
5713 | * read in all of the chunk items. This way we can create chunk | |
5714 | * mappings that reference all of the devices that are afound | |
5715 | */ | |
5716 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
5717 | key.offset = 0; | |
5718 | key.type = 0; | |
5719 | again: | |
5720 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
5721 | if (ret < 0) |
5722 | goto error; | |
d397712b | 5723 | while (1) { |
0b86a832 CM |
5724 | leaf = path->nodes[0]; |
5725 | slot = path->slots[0]; | |
5726 | if (slot >= btrfs_header_nritems(leaf)) { | |
5727 | ret = btrfs_next_leaf(root, path); | |
5728 | if (ret == 0) | |
5729 | continue; | |
5730 | if (ret < 0) | |
5731 | goto error; | |
5732 | break; | |
5733 | } | |
5734 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5735 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
5736 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
5737 | break; | |
5738 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
5739 | struct btrfs_dev_item *dev_item; | |
5740 | dev_item = btrfs_item_ptr(leaf, slot, | |
5741 | struct btrfs_dev_item); | |
0d81ba5d | 5742 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
5743 | if (ret) |
5744 | goto error; | |
0b86a832 CM |
5745 | } |
5746 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
5747 | struct btrfs_chunk *chunk; | |
5748 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
5749 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
5750 | if (ret) |
5751 | goto error; | |
0b86a832 CM |
5752 | } |
5753 | path->slots[0]++; | |
5754 | } | |
5755 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
5756 | key.objectid = 0; | |
b3b4aa74 | 5757 | btrfs_release_path(path); |
0b86a832 CM |
5758 | goto again; |
5759 | } | |
0b86a832 CM |
5760 | ret = 0; |
5761 | error: | |
b367e47f LZ |
5762 | unlock_chunks(root); |
5763 | mutex_unlock(&uuid_mutex); | |
5764 | ||
2b82032c | 5765 | btrfs_free_path(path); |
0b86a832 CM |
5766 | return ret; |
5767 | } | |
442a4f63 | 5768 | |
733f4fbb SB |
5769 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
5770 | { | |
5771 | int i; | |
5772 | ||
5773 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
5774 | btrfs_dev_stat_reset(dev, i); | |
5775 | } | |
5776 | ||
5777 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
5778 | { | |
5779 | struct btrfs_key key; | |
5780 | struct btrfs_key found_key; | |
5781 | struct btrfs_root *dev_root = fs_info->dev_root; | |
5782 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
5783 | struct extent_buffer *eb; | |
5784 | int slot; | |
5785 | int ret = 0; | |
5786 | struct btrfs_device *device; | |
5787 | struct btrfs_path *path = NULL; | |
5788 | int i; | |
5789 | ||
5790 | path = btrfs_alloc_path(); | |
5791 | if (!path) { | |
5792 | ret = -ENOMEM; | |
5793 | goto out; | |
5794 | } | |
5795 | ||
5796 | mutex_lock(&fs_devices->device_list_mutex); | |
5797 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
5798 | int item_size; | |
5799 | struct btrfs_dev_stats_item *ptr; | |
5800 | ||
5801 | key.objectid = 0; | |
5802 | key.type = BTRFS_DEV_STATS_KEY; | |
5803 | key.offset = device->devid; | |
5804 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
5805 | if (ret) { | |
733f4fbb SB |
5806 | __btrfs_reset_dev_stats(device); |
5807 | device->dev_stats_valid = 1; | |
5808 | btrfs_release_path(path); | |
5809 | continue; | |
5810 | } | |
5811 | slot = path->slots[0]; | |
5812 | eb = path->nodes[0]; | |
5813 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5814 | item_size = btrfs_item_size_nr(eb, slot); | |
5815 | ||
5816 | ptr = btrfs_item_ptr(eb, slot, | |
5817 | struct btrfs_dev_stats_item); | |
5818 | ||
5819 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
5820 | if (item_size >= (1 + i) * sizeof(__le64)) | |
5821 | btrfs_dev_stat_set(device, i, | |
5822 | btrfs_dev_stats_value(eb, ptr, i)); | |
5823 | else | |
5824 | btrfs_dev_stat_reset(device, i); | |
5825 | } | |
5826 | ||
5827 | device->dev_stats_valid = 1; | |
5828 | btrfs_dev_stat_print_on_load(device); | |
5829 | btrfs_release_path(path); | |
5830 | } | |
5831 | mutex_unlock(&fs_devices->device_list_mutex); | |
5832 | ||
5833 | out: | |
5834 | btrfs_free_path(path); | |
5835 | return ret < 0 ? ret : 0; | |
5836 | } | |
5837 | ||
5838 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
5839 | struct btrfs_root *dev_root, | |
5840 | struct btrfs_device *device) | |
5841 | { | |
5842 | struct btrfs_path *path; | |
5843 | struct btrfs_key key; | |
5844 | struct extent_buffer *eb; | |
5845 | struct btrfs_dev_stats_item *ptr; | |
5846 | int ret; | |
5847 | int i; | |
5848 | ||
5849 | key.objectid = 0; | |
5850 | key.type = BTRFS_DEV_STATS_KEY; | |
5851 | key.offset = device->devid; | |
5852 | ||
5853 | path = btrfs_alloc_path(); | |
5854 | BUG_ON(!path); | |
5855 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
5856 | if (ret < 0) { | |
606686ee JB |
5857 | printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n", |
5858 | ret, rcu_str_deref(device->name)); | |
733f4fbb SB |
5859 | goto out; |
5860 | } | |
5861 | ||
5862 | if (ret == 0 && | |
5863 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
5864 | /* need to delete old one and insert a new one */ | |
5865 | ret = btrfs_del_item(trans, dev_root, path); | |
5866 | if (ret != 0) { | |
606686ee JB |
5867 | printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n", |
5868 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
5869 | goto out; |
5870 | } | |
5871 | ret = 1; | |
5872 | } | |
5873 | ||
5874 | if (ret == 1) { | |
5875 | /* need to insert a new item */ | |
5876 | btrfs_release_path(path); | |
5877 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
5878 | &key, sizeof(*ptr)); | |
5879 | if (ret < 0) { | |
606686ee JB |
5880 | printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n", |
5881 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
5882 | goto out; |
5883 | } | |
5884 | } | |
5885 | ||
5886 | eb = path->nodes[0]; | |
5887 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
5888 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
5889 | btrfs_set_dev_stats_value(eb, ptr, i, | |
5890 | btrfs_dev_stat_read(device, i)); | |
5891 | btrfs_mark_buffer_dirty(eb); | |
5892 | ||
5893 | out: | |
5894 | btrfs_free_path(path); | |
5895 | return ret; | |
5896 | } | |
5897 | ||
5898 | /* | |
5899 | * called from commit_transaction. Writes all changed device stats to disk. | |
5900 | */ | |
5901 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
5902 | struct btrfs_fs_info *fs_info) | |
5903 | { | |
5904 | struct btrfs_root *dev_root = fs_info->dev_root; | |
5905 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
5906 | struct btrfs_device *device; | |
5907 | int ret = 0; | |
5908 | ||
5909 | mutex_lock(&fs_devices->device_list_mutex); | |
5910 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
5911 | if (!device->dev_stats_valid || !device->dev_stats_dirty) | |
5912 | continue; | |
5913 | ||
5914 | ret = update_dev_stat_item(trans, dev_root, device); | |
5915 | if (!ret) | |
5916 | device->dev_stats_dirty = 0; | |
5917 | } | |
5918 | mutex_unlock(&fs_devices->device_list_mutex); | |
5919 | ||
5920 | return ret; | |
5921 | } | |
5922 | ||
442a4f63 SB |
5923 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
5924 | { | |
5925 | btrfs_dev_stat_inc(dev, index); | |
5926 | btrfs_dev_stat_print_on_error(dev); | |
5927 | } | |
5928 | ||
5929 | void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) | |
5930 | { | |
733f4fbb SB |
5931 | if (!dev->dev_stats_valid) |
5932 | return; | |
606686ee | 5933 | printk_ratelimited_in_rcu(KERN_ERR |
442a4f63 | 5934 | "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
606686ee | 5935 | rcu_str_deref(dev->name), |
442a4f63 SB |
5936 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
5937 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
5938 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
5939 | btrfs_dev_stat_read(dev, | |
5940 | BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
5941 | btrfs_dev_stat_read(dev, | |
5942 | BTRFS_DEV_STAT_GENERATION_ERRS)); | |
5943 | } | |
c11d2c23 | 5944 | |
733f4fbb SB |
5945 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
5946 | { | |
a98cdb85 SB |
5947 | int i; |
5948 | ||
5949 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
5950 | if (btrfs_dev_stat_read(dev, i) != 0) | |
5951 | break; | |
5952 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
5953 | return; /* all values == 0, suppress message */ | |
5954 | ||
606686ee JB |
5955 | printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
5956 | rcu_str_deref(dev->name), | |
733f4fbb SB |
5957 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
5958 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
5959 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
5960 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
5961 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
5962 | } | |
5963 | ||
c11d2c23 | 5964 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 5965 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
5966 | { |
5967 | struct btrfs_device *dev; | |
5968 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
5969 | int i; | |
5970 | ||
5971 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 5972 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
5973 | mutex_unlock(&fs_devices->device_list_mutex); |
5974 | ||
5975 | if (!dev) { | |
5976 | printk(KERN_WARNING | |
5977 | "btrfs: get dev_stats failed, device not found\n"); | |
5978 | return -ENODEV; | |
733f4fbb SB |
5979 | } else if (!dev->dev_stats_valid) { |
5980 | printk(KERN_WARNING | |
5981 | "btrfs: get dev_stats failed, not yet valid\n"); | |
5982 | return -ENODEV; | |
b27f7c0c | 5983 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
5984 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
5985 | if (stats->nr_items > i) | |
5986 | stats->values[i] = | |
5987 | btrfs_dev_stat_read_and_reset(dev, i); | |
5988 | else | |
5989 | btrfs_dev_stat_reset(dev, i); | |
5990 | } | |
5991 | } else { | |
5992 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
5993 | if (stats->nr_items > i) | |
5994 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
5995 | } | |
5996 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
5997 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
5998 | return 0; | |
5999 | } | |
a8a6dab7 SB |
6000 | |
6001 | int btrfs_scratch_superblock(struct btrfs_device *device) | |
6002 | { | |
6003 | struct buffer_head *bh; | |
6004 | struct btrfs_super_block *disk_super; | |
6005 | ||
6006 | bh = btrfs_read_dev_super(device->bdev); | |
6007 | if (!bh) | |
6008 | return -EINVAL; | |
6009 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6010 | ||
6011 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6012 | set_buffer_dirty(bh); | |
6013 | sync_dirty_buffer(bh); | |
6014 | brelse(bh); | |
6015 | ||
6016 | return 0; | |
6017 | } |