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