Commit | Line | Data |
---|---|---|
0b86a832 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8a4b83cc | 21 | #include <linux/buffer_head.h> |
f2d8d74d | 22 | #include <linux/blkdev.h> |
788f20eb | 23 | #include <linux/random.h> |
b765ead5 | 24 | #include <linux/iocontext.h> |
6f88a440 | 25 | #include <linux/capability.h> |
442a4f63 | 26 | #include <linux/ratelimit.h> |
59641015 | 27 | #include <linux/kthread.h> |
53b381b3 | 28 | #include <linux/raid/pq.h> |
803b2f54 | 29 | #include <linux/semaphore.h> |
53b381b3 | 30 | #include <asm/div64.h> |
0b86a832 CM |
31 | #include "ctree.h" |
32 | #include "extent_map.h" | |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "print-tree.h" | |
36 | #include "volumes.h" | |
53b381b3 | 37 | #include "raid56.h" |
8b712842 | 38 | #include "async-thread.h" |
21adbd5c | 39 | #include "check-integrity.h" |
606686ee | 40 | #include "rcu-string.h" |
3fed40cc | 41 | #include "math.h" |
8dabb742 | 42 | #include "dev-replace.h" |
99994cde | 43 | #include "sysfs.h" |
0b86a832 | 44 | |
2b82032c YZ |
45 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
46 | struct btrfs_root *root, | |
47 | struct btrfs_device *device); | |
48 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 49 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 50 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 51 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 52 | |
67a2c45e | 53 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 54 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
55 | struct list_head *btrfs_get_fs_uuids(void) |
56 | { | |
57 | return &fs_uuids; | |
58 | } | |
8a4b83cc | 59 | |
2208a378 ID |
60 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
61 | { | |
62 | struct btrfs_fs_devices *fs_devs; | |
63 | ||
64 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS); | |
65 | if (!fs_devs) | |
66 | return ERR_PTR(-ENOMEM); | |
67 | ||
68 | mutex_init(&fs_devs->device_list_mutex); | |
69 | ||
70 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 71 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
72 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
73 | INIT_LIST_HEAD(&fs_devs->list); | |
74 | ||
75 | return fs_devs; | |
76 | } | |
77 | ||
78 | /** | |
79 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
80 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
81 | * generated. | |
82 | * | |
83 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
84 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
85 | * can be destroyed with kfree() right away. | |
86 | */ | |
87 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
88 | { | |
89 | struct btrfs_fs_devices *fs_devs; | |
90 | ||
91 | fs_devs = __alloc_fs_devices(); | |
92 | if (IS_ERR(fs_devs)) | |
93 | return fs_devs; | |
94 | ||
95 | if (fsid) | |
96 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
97 | else | |
98 | generate_random_uuid(fs_devs->fsid); | |
99 | ||
100 | return fs_devs; | |
101 | } | |
102 | ||
e4404d6e YZ |
103 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
104 | { | |
105 | struct btrfs_device *device; | |
106 | WARN_ON(fs_devices->opened); | |
107 | while (!list_empty(&fs_devices->devices)) { | |
108 | device = list_entry(fs_devices->devices.next, | |
109 | struct btrfs_device, dev_list); | |
110 | list_del(&device->dev_list); | |
606686ee | 111 | rcu_string_free(device->name); |
e4404d6e YZ |
112 | kfree(device); |
113 | } | |
114 | kfree(fs_devices); | |
115 | } | |
116 | ||
b8b8ff59 LC |
117 | static void btrfs_kobject_uevent(struct block_device *bdev, |
118 | enum kobject_action action) | |
119 | { | |
120 | int ret; | |
121 | ||
122 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
123 | if (ret) | |
efe120a0 | 124 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
125 | action, |
126 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
127 | &disk_to_dev(bdev->bd_disk)->kobj); | |
128 | } | |
129 | ||
143bede5 | 130 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
131 | { |
132 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 133 | |
2b82032c YZ |
134 | while (!list_empty(&fs_uuids)) { |
135 | fs_devices = list_entry(fs_uuids.next, | |
136 | struct btrfs_fs_devices, list); | |
137 | list_del(&fs_devices->list); | |
e4404d6e | 138 | free_fs_devices(fs_devices); |
8a4b83cc | 139 | } |
8a4b83cc CM |
140 | } |
141 | ||
12bd2fc0 ID |
142 | static struct btrfs_device *__alloc_device(void) |
143 | { | |
144 | struct btrfs_device *dev; | |
145 | ||
146 | dev = kzalloc(sizeof(*dev), GFP_NOFS); | |
147 | if (!dev) | |
148 | return ERR_PTR(-ENOMEM); | |
149 | ||
150 | INIT_LIST_HEAD(&dev->dev_list); | |
151 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 152 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
153 | |
154 | spin_lock_init(&dev->io_lock); | |
155 | ||
156 | spin_lock_init(&dev->reada_lock); | |
157 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 158 | atomic_set(&dev->dev_stats_ccnt, 0); |
12bd2fc0 ID |
159 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT); |
160 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
161 | ||
162 | return dev; | |
163 | } | |
164 | ||
a1b32a59 CM |
165 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
166 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
167 | { |
168 | struct btrfs_device *dev; | |
8a4b83cc | 169 | |
c6e30871 | 170 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 171 | if (dev->devid == devid && |
8f18cf13 | 172 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 173 | return dev; |
a443755f | 174 | } |
8a4b83cc CM |
175 | } |
176 | return NULL; | |
177 | } | |
178 | ||
a1b32a59 | 179 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 180 | { |
8a4b83cc CM |
181 | struct btrfs_fs_devices *fs_devices; |
182 | ||
c6e30871 | 183 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
184 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
185 | return fs_devices; | |
186 | } | |
187 | return NULL; | |
188 | } | |
189 | ||
beaf8ab3 SB |
190 | static int |
191 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
192 | int flush, struct block_device **bdev, | |
193 | struct buffer_head **bh) | |
194 | { | |
195 | int ret; | |
196 | ||
197 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
198 | ||
199 | if (IS_ERR(*bdev)) { | |
200 | ret = PTR_ERR(*bdev); | |
efe120a0 | 201 | printk(KERN_INFO "BTRFS: open %s failed\n", device_path); |
beaf8ab3 SB |
202 | goto error; |
203 | } | |
204 | ||
205 | if (flush) | |
206 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
207 | ret = set_blocksize(*bdev, 4096); | |
208 | if (ret) { | |
209 | blkdev_put(*bdev, flags); | |
210 | goto error; | |
211 | } | |
212 | invalidate_bdev(*bdev); | |
213 | *bh = btrfs_read_dev_super(*bdev); | |
214 | if (!*bh) { | |
215 | ret = -EINVAL; | |
216 | blkdev_put(*bdev, flags); | |
217 | goto error; | |
218 | } | |
219 | ||
220 | return 0; | |
221 | ||
222 | error: | |
223 | *bdev = NULL; | |
224 | *bh = NULL; | |
225 | return ret; | |
226 | } | |
227 | ||
ffbd517d CM |
228 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
229 | struct bio *head, struct bio *tail) | |
230 | { | |
231 | ||
232 | struct bio *old_head; | |
233 | ||
234 | old_head = pending_bios->head; | |
235 | pending_bios->head = head; | |
236 | if (pending_bios->tail) | |
237 | tail->bi_next = old_head; | |
238 | else | |
239 | pending_bios->tail = tail; | |
240 | } | |
241 | ||
8b712842 CM |
242 | /* |
243 | * we try to collect pending bios for a device so we don't get a large | |
244 | * number of procs sending bios down to the same device. This greatly | |
245 | * improves the schedulers ability to collect and merge the bios. | |
246 | * | |
247 | * But, it also turns into a long list of bios to process and that is sure | |
248 | * to eventually make the worker thread block. The solution here is to | |
249 | * make some progress and then put this work struct back at the end of | |
250 | * the list if the block device is congested. This way, multiple devices | |
251 | * can make progress from a single worker thread. | |
252 | */ | |
143bede5 | 253 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
254 | { |
255 | struct bio *pending; | |
256 | struct backing_dev_info *bdi; | |
b64a2851 | 257 | struct btrfs_fs_info *fs_info; |
ffbd517d | 258 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
259 | struct bio *tail; |
260 | struct bio *cur; | |
261 | int again = 0; | |
ffbd517d | 262 | unsigned long num_run; |
d644d8a1 | 263 | unsigned long batch_run = 0; |
b64a2851 | 264 | unsigned long limit; |
b765ead5 | 265 | unsigned long last_waited = 0; |
d84275c9 | 266 | int force_reg = 0; |
0e588859 | 267 | int sync_pending = 0; |
211588ad CM |
268 | struct blk_plug plug; |
269 | ||
270 | /* | |
271 | * this function runs all the bios we've collected for | |
272 | * a particular device. We don't want to wander off to | |
273 | * another device without first sending all of these down. | |
274 | * So, setup a plug here and finish it off before we return | |
275 | */ | |
276 | blk_start_plug(&plug); | |
8b712842 | 277 | |
bedf762b | 278 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
279 | fs_info = device->dev_root->fs_info; |
280 | limit = btrfs_async_submit_limit(fs_info); | |
281 | limit = limit * 2 / 3; | |
282 | ||
8b712842 CM |
283 | loop: |
284 | spin_lock(&device->io_lock); | |
285 | ||
a6837051 | 286 | loop_lock: |
d84275c9 | 287 | num_run = 0; |
ffbd517d | 288 | |
8b712842 CM |
289 | /* take all the bios off the list at once and process them |
290 | * later on (without the lock held). But, remember the | |
291 | * tail and other pointers so the bios can be properly reinserted | |
292 | * into the list if we hit congestion | |
293 | */ | |
d84275c9 | 294 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 295 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
296 | force_reg = 1; |
297 | } else { | |
ffbd517d | 298 | pending_bios = &device->pending_bios; |
d84275c9 CM |
299 | force_reg = 0; |
300 | } | |
ffbd517d CM |
301 | |
302 | pending = pending_bios->head; | |
303 | tail = pending_bios->tail; | |
8b712842 | 304 | WARN_ON(pending && !tail); |
8b712842 CM |
305 | |
306 | /* | |
307 | * if pending was null this time around, no bios need processing | |
308 | * at all and we can stop. Otherwise it'll loop back up again | |
309 | * and do an additional check so no bios are missed. | |
310 | * | |
311 | * device->running_pending is used to synchronize with the | |
312 | * schedule_bio code. | |
313 | */ | |
ffbd517d CM |
314 | if (device->pending_sync_bios.head == NULL && |
315 | device->pending_bios.head == NULL) { | |
8b712842 CM |
316 | again = 0; |
317 | device->running_pending = 0; | |
ffbd517d CM |
318 | } else { |
319 | again = 1; | |
320 | device->running_pending = 1; | |
8b712842 | 321 | } |
ffbd517d CM |
322 | |
323 | pending_bios->head = NULL; | |
324 | pending_bios->tail = NULL; | |
325 | ||
8b712842 CM |
326 | spin_unlock(&device->io_lock); |
327 | ||
d397712b | 328 | while (pending) { |
ffbd517d CM |
329 | |
330 | rmb(); | |
d84275c9 CM |
331 | /* we want to work on both lists, but do more bios on the |
332 | * sync list than the regular list | |
333 | */ | |
334 | if ((num_run > 32 && | |
335 | pending_bios != &device->pending_sync_bios && | |
336 | device->pending_sync_bios.head) || | |
337 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
338 | device->pending_bios.head)) { | |
ffbd517d CM |
339 | spin_lock(&device->io_lock); |
340 | requeue_list(pending_bios, pending, tail); | |
341 | goto loop_lock; | |
342 | } | |
343 | ||
8b712842 CM |
344 | cur = pending; |
345 | pending = pending->bi_next; | |
346 | cur->bi_next = NULL; | |
b64a2851 | 347 | |
66657b31 | 348 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
349 | waitqueue_active(&fs_info->async_submit_wait)) |
350 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
351 | |
352 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 353 | |
2ab1ba68 CM |
354 | /* |
355 | * if we're doing the sync list, record that our | |
356 | * plug has some sync requests on it | |
357 | * | |
358 | * If we're doing the regular list and there are | |
359 | * sync requests sitting around, unplug before | |
360 | * we add more | |
361 | */ | |
362 | if (pending_bios == &device->pending_sync_bios) { | |
363 | sync_pending = 1; | |
364 | } else if (sync_pending) { | |
365 | blk_finish_plug(&plug); | |
366 | blk_start_plug(&plug); | |
367 | sync_pending = 0; | |
368 | } | |
369 | ||
21adbd5c | 370 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
371 | num_run++; |
372 | batch_run++; | |
853d8ec4 DS |
373 | |
374 | cond_resched(); | |
8b712842 CM |
375 | |
376 | /* | |
377 | * we made progress, there is more work to do and the bdi | |
378 | * is now congested. Back off and let other work structs | |
379 | * run instead | |
380 | */ | |
57fd5a5f | 381 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 382 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 383 | struct io_context *ioc; |
8b712842 | 384 | |
b765ead5 CM |
385 | ioc = current->io_context; |
386 | ||
387 | /* | |
388 | * the main goal here is that we don't want to | |
389 | * block if we're going to be able to submit | |
390 | * more requests without blocking. | |
391 | * | |
392 | * This code does two great things, it pokes into | |
393 | * the elevator code from a filesystem _and_ | |
394 | * it makes assumptions about how batching works. | |
395 | */ | |
396 | if (ioc && ioc->nr_batch_requests > 0 && | |
397 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
398 | (last_waited == 0 || | |
399 | ioc->last_waited == last_waited)) { | |
400 | /* | |
401 | * we want to go through our batch of | |
402 | * requests and stop. So, we copy out | |
403 | * the ioc->last_waited time and test | |
404 | * against it before looping | |
405 | */ | |
406 | last_waited = ioc->last_waited; | |
853d8ec4 | 407 | cond_resched(); |
b765ead5 CM |
408 | continue; |
409 | } | |
8b712842 | 410 | spin_lock(&device->io_lock); |
ffbd517d | 411 | requeue_list(pending_bios, pending, tail); |
a6837051 | 412 | device->running_pending = 1; |
8b712842 CM |
413 | |
414 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
415 | btrfs_queue_work(fs_info->submit_workers, |
416 | &device->work); | |
8b712842 CM |
417 | goto done; |
418 | } | |
d85c8a6f CM |
419 | /* unplug every 64 requests just for good measure */ |
420 | if (batch_run % 64 == 0) { | |
421 | blk_finish_plug(&plug); | |
422 | blk_start_plug(&plug); | |
423 | sync_pending = 0; | |
424 | } | |
8b712842 | 425 | } |
ffbd517d | 426 | |
51684082 CM |
427 | cond_resched(); |
428 | if (again) | |
429 | goto loop; | |
430 | ||
431 | spin_lock(&device->io_lock); | |
432 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
433 | goto loop_lock; | |
434 | spin_unlock(&device->io_lock); | |
435 | ||
8b712842 | 436 | done: |
211588ad | 437 | blk_finish_plug(&plug); |
8b712842 CM |
438 | } |
439 | ||
b2950863 | 440 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
441 | { |
442 | struct btrfs_device *device; | |
443 | ||
444 | device = container_of(work, struct btrfs_device, work); | |
445 | run_scheduled_bios(device); | |
446 | } | |
447 | ||
60999ca4 DS |
448 | /* |
449 | * Add new device to list of registered devices | |
450 | * | |
451 | * Returns: | |
452 | * 1 - first time device is seen | |
453 | * 0 - device already known | |
454 | * < 0 - error | |
455 | */ | |
a1b32a59 | 456 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
457 | struct btrfs_super_block *disk_super, |
458 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
459 | { | |
460 | struct btrfs_device *device; | |
461 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 462 | struct rcu_string *name; |
60999ca4 | 463 | int ret = 0; |
8a4b83cc CM |
464 | u64 found_transid = btrfs_super_generation(disk_super); |
465 | ||
466 | fs_devices = find_fsid(disk_super->fsid); | |
467 | if (!fs_devices) { | |
2208a378 ID |
468 | fs_devices = alloc_fs_devices(disk_super->fsid); |
469 | if (IS_ERR(fs_devices)) | |
470 | return PTR_ERR(fs_devices); | |
471 | ||
8a4b83cc | 472 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 473 | |
8a4b83cc CM |
474 | device = NULL; |
475 | } else { | |
a443755f CM |
476 | device = __find_device(&fs_devices->devices, devid, |
477 | disk_super->dev_item.uuid); | |
8a4b83cc | 478 | } |
443f24fe | 479 | |
8a4b83cc | 480 | if (!device) { |
2b82032c YZ |
481 | if (fs_devices->opened) |
482 | return -EBUSY; | |
483 | ||
12bd2fc0 ID |
484 | device = btrfs_alloc_device(NULL, &devid, |
485 | disk_super->dev_item.uuid); | |
486 | if (IS_ERR(device)) { | |
8a4b83cc | 487 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 488 | return PTR_ERR(device); |
8a4b83cc | 489 | } |
606686ee JB |
490 | |
491 | name = rcu_string_strdup(path, GFP_NOFS); | |
492 | if (!name) { | |
8a4b83cc CM |
493 | kfree(device); |
494 | return -ENOMEM; | |
495 | } | |
606686ee | 496 | rcu_assign_pointer(device->name, name); |
90519d66 | 497 | |
e5e9a520 | 498 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 499 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 500 | fs_devices->num_devices++; |
e5e9a520 CM |
501 | mutex_unlock(&fs_devices->device_list_mutex); |
502 | ||
60999ca4 | 503 | ret = 1; |
2b82032c | 504 | device->fs_devices = fs_devices; |
606686ee | 505 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
506 | /* |
507 | * When FS is already mounted. | |
508 | * 1. If you are here and if the device->name is NULL that | |
509 | * means this device was missing at time of FS mount. | |
510 | * 2. If you are here and if the device->name is different | |
511 | * from 'path' that means either | |
512 | * a. The same device disappeared and reappeared with | |
513 | * different name. or | |
514 | * b. The missing-disk-which-was-replaced, has | |
515 | * reappeared now. | |
516 | * | |
517 | * We must allow 1 and 2a above. But 2b would be a spurious | |
518 | * and unintentional. | |
519 | * | |
520 | * Further in case of 1 and 2a above, the disk at 'path' | |
521 | * would have missed some transaction when it was away and | |
522 | * in case of 2a the stale bdev has to be updated as well. | |
523 | * 2b must not be allowed at all time. | |
524 | */ | |
525 | ||
526 | /* | |
0f23ae74 CM |
527 | * For now, we do allow update to btrfs_fs_device through the |
528 | * btrfs dev scan cli after FS has been mounted. We're still | |
529 | * tracking a problem where systems fail mount by subvolume id | |
530 | * when we reject replacement on a mounted FS. | |
b96de000 | 531 | */ |
0f23ae74 | 532 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
533 | /* |
534 | * That is if the FS is _not_ mounted and if you | |
535 | * are here, that means there is more than one | |
536 | * disk with same uuid and devid.We keep the one | |
537 | * with larger generation number or the last-in if | |
538 | * generation are equal. | |
539 | */ | |
0f23ae74 | 540 | return -EEXIST; |
77bdae4d | 541 | } |
b96de000 | 542 | |
606686ee | 543 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
544 | if (!name) |
545 | return -ENOMEM; | |
606686ee JB |
546 | rcu_string_free(device->name); |
547 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
548 | if (device->missing) { |
549 | fs_devices->missing_devices--; | |
550 | device->missing = 0; | |
551 | } | |
8a4b83cc CM |
552 | } |
553 | ||
77bdae4d AJ |
554 | /* |
555 | * Unmount does not free the btrfs_device struct but would zero | |
556 | * generation along with most of the other members. So just update | |
557 | * it back. We need it to pick the disk with largest generation | |
558 | * (as above). | |
559 | */ | |
560 | if (!fs_devices->opened) | |
561 | device->generation = found_transid; | |
562 | ||
8a4b83cc | 563 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
564 | |
565 | return ret; | |
8a4b83cc CM |
566 | } |
567 | ||
e4404d6e YZ |
568 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
569 | { | |
570 | struct btrfs_fs_devices *fs_devices; | |
571 | struct btrfs_device *device; | |
572 | struct btrfs_device *orig_dev; | |
573 | ||
2208a378 ID |
574 | fs_devices = alloc_fs_devices(orig->fsid); |
575 | if (IS_ERR(fs_devices)) | |
576 | return fs_devices; | |
e4404d6e | 577 | |
adbbb863 | 578 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 579 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 580 | |
46224705 | 581 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 582 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
583 | struct rcu_string *name; |
584 | ||
12bd2fc0 ID |
585 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
586 | orig_dev->uuid); | |
587 | if (IS_ERR(device)) | |
e4404d6e YZ |
588 | goto error; |
589 | ||
606686ee JB |
590 | /* |
591 | * This is ok to do without rcu read locked because we hold the | |
592 | * uuid mutex so nothing we touch in here is going to disappear. | |
593 | */ | |
e755f780 AJ |
594 | if (orig_dev->name) { |
595 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
596 | if (!name) { | |
597 | kfree(device); | |
598 | goto error; | |
599 | } | |
600 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 601 | } |
e4404d6e | 602 | |
e4404d6e YZ |
603 | list_add(&device->dev_list, &fs_devices->devices); |
604 | device->fs_devices = fs_devices; | |
605 | fs_devices->num_devices++; | |
606 | } | |
adbbb863 | 607 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
608 | return fs_devices; |
609 | error: | |
adbbb863 | 610 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
611 | free_fs_devices(fs_devices); |
612 | return ERR_PTR(-ENOMEM); | |
613 | } | |
614 | ||
9eaed21e | 615 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 616 | { |
c6e30871 | 617 | struct btrfs_device *device, *next; |
443f24fe | 618 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 619 | |
dfe25020 CM |
620 | mutex_lock(&uuid_mutex); |
621 | again: | |
46224705 | 622 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 623 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 624 | if (device->in_fs_metadata) { |
63a212ab | 625 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
626 | (!latest_dev || |
627 | device->generation > latest_dev->generation)) { | |
628 | latest_dev = device; | |
a6b0d5c8 | 629 | } |
2b82032c | 630 | continue; |
a6b0d5c8 | 631 | } |
2b82032c | 632 | |
8dabb742 SB |
633 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
634 | /* | |
635 | * In the first step, keep the device which has | |
636 | * the correct fsid and the devid that is used | |
637 | * for the dev_replace procedure. | |
638 | * In the second step, the dev_replace state is | |
639 | * read from the device tree and it is known | |
640 | * whether the procedure is really active or | |
641 | * not, which means whether this device is | |
642 | * used or whether it should be removed. | |
643 | */ | |
644 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
645 | continue; | |
646 | } | |
647 | } | |
2b82032c | 648 | if (device->bdev) { |
d4d77629 | 649 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
650 | device->bdev = NULL; |
651 | fs_devices->open_devices--; | |
652 | } | |
653 | if (device->writeable) { | |
654 | list_del_init(&device->dev_alloc_list); | |
655 | device->writeable = 0; | |
8dabb742 SB |
656 | if (!device->is_tgtdev_for_dev_replace) |
657 | fs_devices->rw_devices--; | |
2b82032c | 658 | } |
e4404d6e YZ |
659 | list_del_init(&device->dev_list); |
660 | fs_devices->num_devices--; | |
606686ee | 661 | rcu_string_free(device->name); |
e4404d6e | 662 | kfree(device); |
dfe25020 | 663 | } |
2b82032c YZ |
664 | |
665 | if (fs_devices->seed) { | |
666 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
667 | goto again; |
668 | } | |
669 | ||
443f24fe | 670 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 671 | |
dfe25020 | 672 | mutex_unlock(&uuid_mutex); |
dfe25020 | 673 | } |
a0af469b | 674 | |
1f78160c XG |
675 | static void __free_device(struct work_struct *work) |
676 | { | |
677 | struct btrfs_device *device; | |
678 | ||
679 | device = container_of(work, struct btrfs_device, rcu_work); | |
680 | ||
681 | if (device->bdev) | |
682 | blkdev_put(device->bdev, device->mode); | |
683 | ||
606686ee | 684 | rcu_string_free(device->name); |
1f78160c XG |
685 | kfree(device); |
686 | } | |
687 | ||
688 | static void free_device(struct rcu_head *head) | |
689 | { | |
690 | struct btrfs_device *device; | |
691 | ||
692 | device = container_of(head, struct btrfs_device, rcu); | |
693 | ||
694 | INIT_WORK(&device->rcu_work, __free_device); | |
695 | schedule_work(&device->rcu_work); | |
696 | } | |
697 | ||
2b82032c | 698 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 699 | { |
8a4b83cc | 700 | struct btrfs_device *device; |
e4404d6e | 701 | |
2b82032c YZ |
702 | if (--fs_devices->opened > 0) |
703 | return 0; | |
8a4b83cc | 704 | |
c9513edb | 705 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 706 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c | 707 | struct btrfs_device *new_device; |
606686ee | 708 | struct rcu_string *name; |
1f78160c XG |
709 | |
710 | if (device->bdev) | |
a0af469b | 711 | fs_devices->open_devices--; |
1f78160c | 712 | |
f747cab7 ID |
713 | if (device->writeable && |
714 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
715 | list_del_init(&device->dev_alloc_list); |
716 | fs_devices->rw_devices--; | |
717 | } | |
718 | ||
726551eb JB |
719 | if (device->missing) |
720 | fs_devices->missing_devices--; | |
d5e2003c | 721 | |
a1e8780a ID |
722 | new_device = btrfs_alloc_device(NULL, &device->devid, |
723 | device->uuid); | |
724 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
606686ee JB |
725 | |
726 | /* Safe because we are under uuid_mutex */ | |
99f5944b JB |
727 | if (device->name) { |
728 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
a1e8780a | 729 | BUG_ON(!name); /* -ENOMEM */ |
99f5944b JB |
730 | rcu_assign_pointer(new_device->name, name); |
731 | } | |
a1e8780a | 732 | |
1f78160c | 733 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
a1e8780a | 734 | new_device->fs_devices = device->fs_devices; |
1f78160c XG |
735 | |
736 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 737 | } |
c9513edb XG |
738 | mutex_unlock(&fs_devices->device_list_mutex); |
739 | ||
e4404d6e YZ |
740 | WARN_ON(fs_devices->open_devices); |
741 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
742 | fs_devices->opened = 0; |
743 | fs_devices->seeding = 0; | |
2b82032c | 744 | |
8a4b83cc CM |
745 | return 0; |
746 | } | |
747 | ||
2b82032c YZ |
748 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
749 | { | |
e4404d6e | 750 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
751 | int ret; |
752 | ||
753 | mutex_lock(&uuid_mutex); | |
754 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
755 | if (!fs_devices->opened) { |
756 | seed_devices = fs_devices->seed; | |
757 | fs_devices->seed = NULL; | |
758 | } | |
2b82032c | 759 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
760 | |
761 | while (seed_devices) { | |
762 | fs_devices = seed_devices; | |
763 | seed_devices = fs_devices->seed; | |
764 | __btrfs_close_devices(fs_devices); | |
765 | free_fs_devices(fs_devices); | |
766 | } | |
bc178622 ES |
767 | /* |
768 | * Wait for rcu kworkers under __btrfs_close_devices | |
769 | * to finish all blkdev_puts so device is really | |
770 | * free when umount is done. | |
771 | */ | |
772 | rcu_barrier(); | |
2b82032c YZ |
773 | return ret; |
774 | } | |
775 | ||
e4404d6e YZ |
776 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
777 | fmode_t flags, void *holder) | |
8a4b83cc | 778 | { |
d5e2003c | 779 | struct request_queue *q; |
8a4b83cc CM |
780 | struct block_device *bdev; |
781 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 782 | struct btrfs_device *device; |
443f24fe | 783 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
784 | struct buffer_head *bh; |
785 | struct btrfs_super_block *disk_super; | |
a0af469b | 786 | u64 devid; |
2b82032c | 787 | int seeding = 1; |
a0af469b | 788 | int ret = 0; |
8a4b83cc | 789 | |
d4d77629 TH |
790 | flags |= FMODE_EXCL; |
791 | ||
c6e30871 | 792 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
793 | if (device->bdev) |
794 | continue; | |
dfe25020 CM |
795 | if (!device->name) |
796 | continue; | |
797 | ||
f63e0cca ES |
798 | /* Just open everything we can; ignore failures here */ |
799 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
800 | &bdev, &bh)) | |
beaf8ab3 | 801 | continue; |
a0af469b CM |
802 | |
803 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 804 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
805 | if (devid != device->devid) |
806 | goto error_brelse; | |
807 | ||
2b82032c YZ |
808 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
809 | BTRFS_UUID_SIZE)) | |
810 | goto error_brelse; | |
811 | ||
812 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
813 | if (!latest_dev || |
814 | device->generation > latest_dev->generation) | |
815 | latest_dev = device; | |
a0af469b | 816 | |
2b82032c YZ |
817 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
818 | device->writeable = 0; | |
819 | } else { | |
820 | device->writeable = !bdev_read_only(bdev); | |
821 | seeding = 0; | |
822 | } | |
823 | ||
d5e2003c | 824 | q = bdev_get_queue(bdev); |
90180da4 | 825 | if (blk_queue_discard(q)) |
d5e2003c | 826 | device->can_discard = 1; |
d5e2003c | 827 | |
8a4b83cc | 828 | device->bdev = bdev; |
dfe25020 | 829 | device->in_fs_metadata = 0; |
15916de8 CM |
830 | device->mode = flags; |
831 | ||
c289811c CM |
832 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
833 | fs_devices->rotating = 1; | |
834 | ||
a0af469b | 835 | fs_devices->open_devices++; |
55e50e45 ID |
836 | if (device->writeable && |
837 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
838 | fs_devices->rw_devices++; |
839 | list_add(&device->dev_alloc_list, | |
840 | &fs_devices->alloc_list); | |
841 | } | |
4f6c9328 | 842 | brelse(bh); |
a0af469b | 843 | continue; |
a061fc8d | 844 | |
a0af469b CM |
845 | error_brelse: |
846 | brelse(bh); | |
d4d77629 | 847 | blkdev_put(bdev, flags); |
a0af469b | 848 | continue; |
8a4b83cc | 849 | } |
a0af469b | 850 | if (fs_devices->open_devices == 0) { |
20bcd649 | 851 | ret = -EINVAL; |
a0af469b CM |
852 | goto out; |
853 | } | |
2b82032c YZ |
854 | fs_devices->seeding = seeding; |
855 | fs_devices->opened = 1; | |
443f24fe | 856 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 857 | fs_devices->total_rw_bytes = 0; |
a0af469b | 858 | out: |
2b82032c YZ |
859 | return ret; |
860 | } | |
861 | ||
862 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 863 | fmode_t flags, void *holder) |
2b82032c YZ |
864 | { |
865 | int ret; | |
866 | ||
867 | mutex_lock(&uuid_mutex); | |
868 | if (fs_devices->opened) { | |
e4404d6e YZ |
869 | fs_devices->opened++; |
870 | ret = 0; | |
2b82032c | 871 | } else { |
15916de8 | 872 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 873 | } |
8a4b83cc | 874 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
875 | return ret; |
876 | } | |
877 | ||
6f60cbd3 DS |
878 | /* |
879 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
880 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
881 | * is read via pagecache | |
882 | */ | |
97288f2c | 883 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
884 | struct btrfs_fs_devices **fs_devices_ret) |
885 | { | |
886 | struct btrfs_super_block *disk_super; | |
887 | struct block_device *bdev; | |
6f60cbd3 DS |
888 | struct page *page; |
889 | void *p; | |
890 | int ret = -EINVAL; | |
8a4b83cc | 891 | u64 devid; |
f2984462 | 892 | u64 transid; |
02db0844 | 893 | u64 total_devices; |
6f60cbd3 DS |
894 | u64 bytenr; |
895 | pgoff_t index; | |
8a4b83cc | 896 | |
6f60cbd3 DS |
897 | /* |
898 | * we would like to check all the supers, but that would make | |
899 | * a btrfs mount succeed after a mkfs from a different FS. | |
900 | * So, we need to add a special mount option to scan for | |
901 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
902 | */ | |
903 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 904 | flags |= FMODE_EXCL; |
10f6327b | 905 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
906 | |
907 | bdev = blkdev_get_by_path(path, flags, holder); | |
908 | ||
909 | if (IS_ERR(bdev)) { | |
910 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 911 | goto error; |
6f60cbd3 DS |
912 | } |
913 | ||
914 | /* make sure our super fits in the device */ | |
915 | if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode)) | |
916 | goto error_bdev_put; | |
917 | ||
918 | /* make sure our super fits in the page */ | |
919 | if (sizeof(*disk_super) > PAGE_CACHE_SIZE) | |
920 | goto error_bdev_put; | |
921 | ||
922 | /* make sure our super doesn't straddle pages on disk */ | |
923 | index = bytenr >> PAGE_CACHE_SHIFT; | |
924 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index) | |
925 | goto error_bdev_put; | |
926 | ||
927 | /* pull in the page with our super */ | |
928 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
929 | index, GFP_NOFS); | |
930 | ||
931 | if (IS_ERR_OR_NULL(page)) | |
932 | goto error_bdev_put; | |
933 | ||
934 | p = kmap(page); | |
935 | ||
936 | /* align our pointer to the offset of the super block */ | |
937 | disk_super = p + (bytenr & ~PAGE_CACHE_MASK); | |
938 | ||
939 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
3cae210f | 940 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) |
6f60cbd3 DS |
941 | goto error_unmap; |
942 | ||
a343832f | 943 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 944 | transid = btrfs_super_generation(disk_super); |
02db0844 | 945 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 946 | |
8a4b83cc | 947 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
948 | if (ret > 0) { |
949 | if (disk_super->label[0]) { | |
950 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
951 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
952 | printk(KERN_INFO "BTRFS: device label %s ", disk_super->label); | |
953 | } else { | |
954 | printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid); | |
955 | } | |
956 | ||
957 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); | |
958 | ret = 0; | |
959 | } | |
02db0844 JB |
960 | if (!ret && fs_devices_ret) |
961 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 DS |
962 | |
963 | error_unmap: | |
964 | kunmap(page); | |
965 | page_cache_release(page); | |
966 | ||
967 | error_bdev_put: | |
d4d77629 | 968 | blkdev_put(bdev, flags); |
8a4b83cc | 969 | error: |
beaf8ab3 | 970 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
971 | return ret; |
972 | } | |
0b86a832 | 973 | |
6d07bcec MX |
974 | /* helper to account the used device space in the range */ |
975 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
976 | u64 end, u64 *length) | |
977 | { | |
978 | struct btrfs_key key; | |
979 | struct btrfs_root *root = device->dev_root; | |
980 | struct btrfs_dev_extent *dev_extent; | |
981 | struct btrfs_path *path; | |
982 | u64 extent_end; | |
983 | int ret; | |
984 | int slot; | |
985 | struct extent_buffer *l; | |
986 | ||
987 | *length = 0; | |
988 | ||
63a212ab | 989 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
990 | return 0; |
991 | ||
992 | path = btrfs_alloc_path(); | |
993 | if (!path) | |
994 | return -ENOMEM; | |
995 | path->reada = 2; | |
996 | ||
997 | key.objectid = device->devid; | |
998 | key.offset = start; | |
999 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1000 | ||
1001 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1002 | if (ret < 0) | |
1003 | goto out; | |
1004 | if (ret > 0) { | |
1005 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1006 | if (ret < 0) | |
1007 | goto out; | |
1008 | } | |
1009 | ||
1010 | while (1) { | |
1011 | l = path->nodes[0]; | |
1012 | slot = path->slots[0]; | |
1013 | if (slot >= btrfs_header_nritems(l)) { | |
1014 | ret = btrfs_next_leaf(root, path); | |
1015 | if (ret == 0) | |
1016 | continue; | |
1017 | if (ret < 0) | |
1018 | goto out; | |
1019 | ||
1020 | break; | |
1021 | } | |
1022 | btrfs_item_key_to_cpu(l, &key, slot); | |
1023 | ||
1024 | if (key.objectid < device->devid) | |
1025 | goto next; | |
1026 | ||
1027 | if (key.objectid > device->devid) | |
1028 | break; | |
1029 | ||
962a298f | 1030 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1031 | goto next; |
1032 | ||
1033 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1034 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1035 | dev_extent); | |
1036 | if (key.offset <= start && extent_end > end) { | |
1037 | *length = end - start + 1; | |
1038 | break; | |
1039 | } else if (key.offset <= start && extent_end > start) | |
1040 | *length += extent_end - start; | |
1041 | else if (key.offset > start && extent_end <= end) | |
1042 | *length += extent_end - key.offset; | |
1043 | else if (key.offset > start && key.offset <= end) { | |
1044 | *length += end - key.offset + 1; | |
1045 | break; | |
1046 | } else if (key.offset > end) | |
1047 | break; | |
1048 | ||
1049 | next: | |
1050 | path->slots[0]++; | |
1051 | } | |
1052 | ret = 0; | |
1053 | out: | |
1054 | btrfs_free_path(path); | |
1055 | return ret; | |
1056 | } | |
1057 | ||
6df9a95e JB |
1058 | static int contains_pending_extent(struct btrfs_trans_handle *trans, |
1059 | struct btrfs_device *device, | |
1060 | u64 *start, u64 len) | |
1061 | { | |
1062 | struct extent_map *em; | |
04216820 | 1063 | struct list_head *search_list = &trans->transaction->pending_chunks; |
6df9a95e | 1064 | int ret = 0; |
1b984508 | 1065 | u64 physical_start = *start; |
6df9a95e | 1066 | |
04216820 FM |
1067 | again: |
1068 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1069 | struct map_lookup *map; |
1070 | int i; | |
1071 | ||
1072 | map = (struct map_lookup *)em->bdev; | |
1073 | for (i = 0; i < map->num_stripes; i++) { | |
1074 | if (map->stripes[i].dev != device) | |
1075 | continue; | |
1b984508 | 1076 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1077 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1078 | physical_start) |
6df9a95e JB |
1079 | continue; |
1080 | *start = map->stripes[i].physical + | |
1081 | em->orig_block_len; | |
1082 | ret = 1; | |
1083 | } | |
1084 | } | |
04216820 FM |
1085 | if (search_list == &trans->transaction->pending_chunks) { |
1086 | search_list = &trans->root->fs_info->pinned_chunks; | |
1087 | goto again; | |
1088 | } | |
6df9a95e JB |
1089 | |
1090 | return ret; | |
1091 | } | |
1092 | ||
1093 | ||
0b86a832 | 1094 | /* |
7bfc837d | 1095 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
1096 | * @device: the device which we search the free space in |
1097 | * @num_bytes: the size of the free space that we need | |
1098 | * @start: store the start of the free space. | |
1099 | * @len: the size of the free space. that we find, or the size of the max | |
1100 | * free space if we don't find suitable free space | |
1101 | * | |
0b86a832 CM |
1102 | * this uses a pretty simple search, the expectation is that it is |
1103 | * called very infrequently and that a given device has a small number | |
1104 | * of extents | |
7bfc837d MX |
1105 | * |
1106 | * @start is used to store the start of the free space if we find. But if we | |
1107 | * don't find suitable free space, it will be used to store the start position | |
1108 | * of the max free space. | |
1109 | * | |
1110 | * @len is used to store the size of the free space that we find. | |
1111 | * But if we don't find suitable free space, it is used to store the size of | |
1112 | * the max free space. | |
0b86a832 | 1113 | */ |
6df9a95e JB |
1114 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1115 | struct btrfs_device *device, u64 num_bytes, | |
7bfc837d | 1116 | u64 *start, u64 *len) |
0b86a832 CM |
1117 | { |
1118 | struct btrfs_key key; | |
1119 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1120 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1121 | struct btrfs_path *path; |
7bfc837d MX |
1122 | u64 hole_size; |
1123 | u64 max_hole_start; | |
1124 | u64 max_hole_size; | |
1125 | u64 extent_end; | |
1126 | u64 search_start; | |
0b86a832 CM |
1127 | u64 search_end = device->total_bytes; |
1128 | int ret; | |
7bfc837d | 1129 | int slot; |
0b86a832 CM |
1130 | struct extent_buffer *l; |
1131 | ||
0b86a832 CM |
1132 | /* FIXME use last free of some kind */ |
1133 | ||
8a4b83cc CM |
1134 | /* we don't want to overwrite the superblock on the drive, |
1135 | * so we make sure to start at an offset of at least 1MB | |
1136 | */ | |
a9c9bf68 | 1137 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 1138 | |
6df9a95e JB |
1139 | path = btrfs_alloc_path(); |
1140 | if (!path) | |
1141 | return -ENOMEM; | |
f2ab7618 | 1142 | |
7bfc837d MX |
1143 | max_hole_start = search_start; |
1144 | max_hole_size = 0; | |
1145 | ||
f2ab7618 | 1146 | again: |
63a212ab | 1147 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1148 | ret = -ENOSPC; |
6df9a95e | 1149 | goto out; |
7bfc837d MX |
1150 | } |
1151 | ||
7bfc837d | 1152 | path->reada = 2; |
6df9a95e JB |
1153 | path->search_commit_root = 1; |
1154 | path->skip_locking = 1; | |
7bfc837d | 1155 | |
0b86a832 CM |
1156 | key.objectid = device->devid; |
1157 | key.offset = search_start; | |
1158 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1159 | |
125ccb0a | 1160 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1161 | if (ret < 0) |
7bfc837d | 1162 | goto out; |
1fcbac58 YZ |
1163 | if (ret > 0) { |
1164 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1165 | if (ret < 0) | |
7bfc837d | 1166 | goto out; |
1fcbac58 | 1167 | } |
7bfc837d | 1168 | |
0b86a832 CM |
1169 | while (1) { |
1170 | l = path->nodes[0]; | |
1171 | slot = path->slots[0]; | |
1172 | if (slot >= btrfs_header_nritems(l)) { | |
1173 | ret = btrfs_next_leaf(root, path); | |
1174 | if (ret == 0) | |
1175 | continue; | |
1176 | if (ret < 0) | |
7bfc837d MX |
1177 | goto out; |
1178 | ||
1179 | break; | |
0b86a832 CM |
1180 | } |
1181 | btrfs_item_key_to_cpu(l, &key, slot); | |
1182 | ||
1183 | if (key.objectid < device->devid) | |
1184 | goto next; | |
1185 | ||
1186 | if (key.objectid > device->devid) | |
7bfc837d | 1187 | break; |
0b86a832 | 1188 | |
962a298f | 1189 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1190 | goto next; |
9779b72f | 1191 | |
7bfc837d MX |
1192 | if (key.offset > search_start) { |
1193 | hole_size = key.offset - search_start; | |
9779b72f | 1194 | |
6df9a95e JB |
1195 | /* |
1196 | * Have to check before we set max_hole_start, otherwise | |
1197 | * we could end up sending back this offset anyway. | |
1198 | */ | |
1199 | if (contains_pending_extent(trans, device, | |
1200 | &search_start, | |
1b984508 FL |
1201 | hole_size)) { |
1202 | if (key.offset >= search_start) { | |
1203 | hole_size = key.offset - search_start; | |
1204 | } else { | |
1205 | WARN_ON_ONCE(1); | |
1206 | hole_size = 0; | |
1207 | } | |
1208 | } | |
6df9a95e | 1209 | |
7bfc837d MX |
1210 | if (hole_size > max_hole_size) { |
1211 | max_hole_start = search_start; | |
1212 | max_hole_size = hole_size; | |
1213 | } | |
9779b72f | 1214 | |
7bfc837d MX |
1215 | /* |
1216 | * If this free space is greater than which we need, | |
1217 | * it must be the max free space that we have found | |
1218 | * until now, so max_hole_start must point to the start | |
1219 | * of this free space and the length of this free space | |
1220 | * is stored in max_hole_size. Thus, we return | |
1221 | * max_hole_start and max_hole_size and go back to the | |
1222 | * caller. | |
1223 | */ | |
1224 | if (hole_size >= num_bytes) { | |
1225 | ret = 0; | |
1226 | goto out; | |
0b86a832 CM |
1227 | } |
1228 | } | |
0b86a832 | 1229 | |
0b86a832 | 1230 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1231 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1232 | dev_extent); | |
1233 | if (extent_end > search_start) | |
1234 | search_start = extent_end; | |
0b86a832 CM |
1235 | next: |
1236 | path->slots[0]++; | |
1237 | cond_resched(); | |
1238 | } | |
0b86a832 | 1239 | |
38c01b96 | 1240 | /* |
1241 | * At this point, search_start should be the end of | |
1242 | * allocated dev extents, and when shrinking the device, | |
1243 | * search_end may be smaller than search_start. | |
1244 | */ | |
f2ab7618 | 1245 | if (search_end > search_start) { |
38c01b96 | 1246 | hole_size = search_end - search_start; |
1247 | ||
f2ab7618 ZL |
1248 | if (contains_pending_extent(trans, device, &search_start, |
1249 | hole_size)) { | |
1250 | btrfs_release_path(path); | |
1251 | goto again; | |
1252 | } | |
0b86a832 | 1253 | |
f2ab7618 ZL |
1254 | if (hole_size > max_hole_size) { |
1255 | max_hole_start = search_start; | |
1256 | max_hole_size = hole_size; | |
1257 | } | |
6df9a95e JB |
1258 | } |
1259 | ||
7bfc837d | 1260 | /* See above. */ |
f2ab7618 | 1261 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1262 | ret = -ENOSPC; |
1263 | else | |
1264 | ret = 0; | |
1265 | ||
1266 | out: | |
2b82032c | 1267 | btrfs_free_path(path); |
7bfc837d | 1268 | *start = max_hole_start; |
b2117a39 | 1269 | if (len) |
7bfc837d | 1270 | *len = max_hole_size; |
0b86a832 CM |
1271 | return ret; |
1272 | } | |
1273 | ||
b2950863 | 1274 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1275 | struct btrfs_device *device, |
2196d6e8 | 1276 | u64 start, u64 *dev_extent_len) |
8f18cf13 CM |
1277 | { |
1278 | int ret; | |
1279 | struct btrfs_path *path; | |
1280 | struct btrfs_root *root = device->dev_root; | |
1281 | struct btrfs_key key; | |
a061fc8d CM |
1282 | struct btrfs_key found_key; |
1283 | struct extent_buffer *leaf = NULL; | |
1284 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1285 | |
1286 | path = btrfs_alloc_path(); | |
1287 | if (!path) | |
1288 | return -ENOMEM; | |
1289 | ||
1290 | key.objectid = device->devid; | |
1291 | key.offset = start; | |
1292 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1293 | again: |
8f18cf13 | 1294 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1295 | if (ret > 0) { |
1296 | ret = btrfs_previous_item(root, path, key.objectid, | |
1297 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1298 | if (ret) |
1299 | goto out; | |
a061fc8d CM |
1300 | leaf = path->nodes[0]; |
1301 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1302 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1303 | struct btrfs_dev_extent); | |
1304 | BUG_ON(found_key.offset > start || found_key.offset + | |
1305 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1306 | key = found_key; |
1307 | btrfs_release_path(path); | |
1308 | goto again; | |
a061fc8d CM |
1309 | } else if (ret == 0) { |
1310 | leaf = path->nodes[0]; | |
1311 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1312 | struct btrfs_dev_extent); | |
79787eaa JM |
1313 | } else { |
1314 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1315 | goto out; | |
a061fc8d | 1316 | } |
8f18cf13 | 1317 | |
2196d6e8 MX |
1318 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1319 | ||
8f18cf13 | 1320 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1321 | if (ret) { |
1322 | btrfs_error(root->fs_info, ret, | |
1323 | "Failed to remove dev extent item"); | |
13212b54 ZL |
1324 | } else { |
1325 | trans->transaction->have_free_bgs = 1; | |
79787eaa | 1326 | } |
b0b802d7 | 1327 | out: |
8f18cf13 CM |
1328 | btrfs_free_path(path); |
1329 | return ret; | |
1330 | } | |
1331 | ||
48a3b636 ES |
1332 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1333 | struct btrfs_device *device, | |
1334 | u64 chunk_tree, u64 chunk_objectid, | |
1335 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1336 | { |
1337 | int ret; | |
1338 | struct btrfs_path *path; | |
1339 | struct btrfs_root *root = device->dev_root; | |
1340 | struct btrfs_dev_extent *extent; | |
1341 | struct extent_buffer *leaf; | |
1342 | struct btrfs_key key; | |
1343 | ||
dfe25020 | 1344 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1345 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1346 | path = btrfs_alloc_path(); |
1347 | if (!path) | |
1348 | return -ENOMEM; | |
1349 | ||
0b86a832 | 1350 | key.objectid = device->devid; |
2b82032c | 1351 | key.offset = start; |
0b86a832 CM |
1352 | key.type = BTRFS_DEV_EXTENT_KEY; |
1353 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1354 | sizeof(*extent)); | |
2cdcecbc MF |
1355 | if (ret) |
1356 | goto out; | |
0b86a832 CM |
1357 | |
1358 | leaf = path->nodes[0]; | |
1359 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1360 | struct btrfs_dev_extent); | |
e17cade2 CM |
1361 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1362 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1363 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1364 | ||
1365 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1366 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1367 | |
0b86a832 CM |
1368 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1369 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1370 | out: |
0b86a832 CM |
1371 | btrfs_free_path(path); |
1372 | return ret; | |
1373 | } | |
1374 | ||
6df9a95e | 1375 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1376 | { |
6df9a95e JB |
1377 | struct extent_map_tree *em_tree; |
1378 | struct extent_map *em; | |
1379 | struct rb_node *n; | |
1380 | u64 ret = 0; | |
0b86a832 | 1381 | |
6df9a95e JB |
1382 | em_tree = &fs_info->mapping_tree.map_tree; |
1383 | read_lock(&em_tree->lock); | |
1384 | n = rb_last(&em_tree->map); | |
1385 | if (n) { | |
1386 | em = rb_entry(n, struct extent_map, rb_node); | |
1387 | ret = em->start + em->len; | |
0b86a832 | 1388 | } |
6df9a95e JB |
1389 | read_unlock(&em_tree->lock); |
1390 | ||
0b86a832 CM |
1391 | return ret; |
1392 | } | |
1393 | ||
53f10659 ID |
1394 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1395 | u64 *devid_ret) | |
0b86a832 CM |
1396 | { |
1397 | int ret; | |
1398 | struct btrfs_key key; | |
1399 | struct btrfs_key found_key; | |
2b82032c YZ |
1400 | struct btrfs_path *path; |
1401 | ||
2b82032c YZ |
1402 | path = btrfs_alloc_path(); |
1403 | if (!path) | |
1404 | return -ENOMEM; | |
0b86a832 CM |
1405 | |
1406 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1407 | key.type = BTRFS_DEV_ITEM_KEY; | |
1408 | key.offset = (u64)-1; | |
1409 | ||
53f10659 | 1410 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1411 | if (ret < 0) |
1412 | goto error; | |
1413 | ||
79787eaa | 1414 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1415 | |
53f10659 ID |
1416 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1417 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1418 | BTRFS_DEV_ITEM_KEY); |
1419 | if (ret) { | |
53f10659 | 1420 | *devid_ret = 1; |
0b86a832 CM |
1421 | } else { |
1422 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1423 | path->slots[0]); | |
53f10659 | 1424 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1425 | } |
1426 | ret = 0; | |
1427 | error: | |
2b82032c | 1428 | btrfs_free_path(path); |
0b86a832 CM |
1429 | return ret; |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * the device information is stored in the chunk root | |
1434 | * the btrfs_device struct should be fully filled in | |
1435 | */ | |
48a3b636 ES |
1436 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1437 | struct btrfs_root *root, | |
1438 | struct btrfs_device *device) | |
0b86a832 CM |
1439 | { |
1440 | int ret; | |
1441 | struct btrfs_path *path; | |
1442 | struct btrfs_dev_item *dev_item; | |
1443 | struct extent_buffer *leaf; | |
1444 | struct btrfs_key key; | |
1445 | unsigned long ptr; | |
0b86a832 CM |
1446 | |
1447 | root = root->fs_info->chunk_root; | |
1448 | ||
1449 | path = btrfs_alloc_path(); | |
1450 | if (!path) | |
1451 | return -ENOMEM; | |
1452 | ||
0b86a832 CM |
1453 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1454 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1455 | key.offset = device->devid; |
0b86a832 CM |
1456 | |
1457 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1458 | sizeof(*dev_item)); |
0b86a832 CM |
1459 | if (ret) |
1460 | goto out; | |
1461 | ||
1462 | leaf = path->nodes[0]; | |
1463 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1464 | ||
1465 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1466 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1467 | btrfs_set_device_type(leaf, dev_item, device->type); |
1468 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1469 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1470 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1471 | btrfs_set_device_total_bytes(leaf, dev_item, |
1472 | btrfs_device_get_disk_total_bytes(device)); | |
1473 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1474 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1475 | btrfs_set_device_group(leaf, dev_item, 0); |
1476 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1477 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1478 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1479 | |
410ba3a2 | 1480 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1481 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1482 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1483 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1484 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1485 | |
2b82032c | 1486 | ret = 0; |
0b86a832 CM |
1487 | out: |
1488 | btrfs_free_path(path); | |
1489 | return ret; | |
1490 | } | |
8f18cf13 | 1491 | |
5a1972bd QW |
1492 | /* |
1493 | * Function to update ctime/mtime for a given device path. | |
1494 | * Mainly used for ctime/mtime based probe like libblkid. | |
1495 | */ | |
1496 | static void update_dev_time(char *path_name) | |
1497 | { | |
1498 | struct file *filp; | |
1499 | ||
1500 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1501 | if (IS_ERR(filp)) |
5a1972bd QW |
1502 | return; |
1503 | file_update_time(filp); | |
1504 | filp_close(filp, NULL); | |
1505 | return; | |
1506 | } | |
1507 | ||
a061fc8d CM |
1508 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1509 | struct btrfs_device *device) | |
1510 | { | |
1511 | int ret; | |
1512 | struct btrfs_path *path; | |
a061fc8d | 1513 | struct btrfs_key key; |
a061fc8d CM |
1514 | struct btrfs_trans_handle *trans; |
1515 | ||
1516 | root = root->fs_info->chunk_root; | |
1517 | ||
1518 | path = btrfs_alloc_path(); | |
1519 | if (!path) | |
1520 | return -ENOMEM; | |
1521 | ||
a22285a6 | 1522 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1523 | if (IS_ERR(trans)) { |
1524 | btrfs_free_path(path); | |
1525 | return PTR_ERR(trans); | |
1526 | } | |
a061fc8d CM |
1527 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1528 | key.type = BTRFS_DEV_ITEM_KEY; | |
1529 | key.offset = device->devid; | |
1530 | ||
1531 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1532 | if (ret < 0) | |
1533 | goto out; | |
1534 | ||
1535 | if (ret > 0) { | |
1536 | ret = -ENOENT; | |
1537 | goto out; | |
1538 | } | |
1539 | ||
1540 | ret = btrfs_del_item(trans, root, path); | |
1541 | if (ret) | |
1542 | goto out; | |
a061fc8d CM |
1543 | out: |
1544 | btrfs_free_path(path); | |
1545 | btrfs_commit_transaction(trans, root); | |
1546 | return ret; | |
1547 | } | |
1548 | ||
1549 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1550 | { | |
1551 | struct btrfs_device *device; | |
2b82032c | 1552 | struct btrfs_device *next_device; |
a061fc8d | 1553 | struct block_device *bdev; |
dfe25020 | 1554 | struct buffer_head *bh = NULL; |
a061fc8d | 1555 | struct btrfs_super_block *disk_super; |
1f78160c | 1556 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1557 | u64 all_avail; |
1558 | u64 devid; | |
2b82032c YZ |
1559 | u64 num_devices; |
1560 | u8 *dev_uuid; | |
de98ced9 | 1561 | unsigned seq; |
a061fc8d | 1562 | int ret = 0; |
1f78160c | 1563 | bool clear_super = false; |
a061fc8d | 1564 | |
a061fc8d CM |
1565 | mutex_lock(&uuid_mutex); |
1566 | ||
de98ced9 MX |
1567 | do { |
1568 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
1569 | ||
1570 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1571 | root->fs_info->avail_system_alloc_bits | | |
1572 | root->fs_info->avail_metadata_alloc_bits; | |
1573 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
a061fc8d | 1574 | |
8dabb742 SB |
1575 | num_devices = root->fs_info->fs_devices->num_devices; |
1576 | btrfs_dev_replace_lock(&root->fs_info->dev_replace); | |
1577 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1578 | WARN_ON(num_devices < 1); | |
1579 | num_devices--; | |
1580 | } | |
1581 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace); | |
1582 | ||
1583 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) { | |
183860f6 | 1584 | ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET; |
a061fc8d CM |
1585 | goto out; |
1586 | } | |
1587 | ||
8dabb742 | 1588 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) { |
183860f6 | 1589 | ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET; |
a061fc8d CM |
1590 | goto out; |
1591 | } | |
1592 | ||
53b381b3 DW |
1593 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && |
1594 | root->fs_info->fs_devices->rw_devices <= 2) { | |
183860f6 | 1595 | ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET; |
53b381b3 DW |
1596 | goto out; |
1597 | } | |
1598 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1599 | root->fs_info->fs_devices->rw_devices <= 3) { | |
183860f6 | 1600 | ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET; |
53b381b3 DW |
1601 | goto out; |
1602 | } | |
1603 | ||
dfe25020 | 1604 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1605 | struct list_head *devices; |
1606 | struct btrfs_device *tmp; | |
a061fc8d | 1607 | |
dfe25020 CM |
1608 | device = NULL; |
1609 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1610 | /* |
1611 | * It is safe to read the devices since the volume_mutex | |
1612 | * is held. | |
1613 | */ | |
c6e30871 | 1614 | list_for_each_entry(tmp, devices, dev_list) { |
63a212ab SB |
1615 | if (tmp->in_fs_metadata && |
1616 | !tmp->is_tgtdev_for_dev_replace && | |
1617 | !tmp->bdev) { | |
dfe25020 CM |
1618 | device = tmp; |
1619 | break; | |
1620 | } | |
1621 | } | |
1622 | bdev = NULL; | |
1623 | bh = NULL; | |
1624 | disk_super = NULL; | |
1625 | if (!device) { | |
183860f6 | 1626 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
dfe25020 CM |
1627 | goto out; |
1628 | } | |
dfe25020 | 1629 | } else { |
beaf8ab3 | 1630 | ret = btrfs_get_bdev_and_sb(device_path, |
cc975eb4 | 1631 | FMODE_WRITE | FMODE_EXCL, |
beaf8ab3 SB |
1632 | root->fs_info->bdev_holder, 0, |
1633 | &bdev, &bh); | |
1634 | if (ret) | |
dfe25020 | 1635 | goto out; |
dfe25020 | 1636 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1637 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c | 1638 | dev_uuid = disk_super->dev_item.uuid; |
aa1b8cd4 | 1639 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2b82032c | 1640 | disk_super->fsid); |
dfe25020 CM |
1641 | if (!device) { |
1642 | ret = -ENOENT; | |
1643 | goto error_brelse; | |
1644 | } | |
2b82032c | 1645 | } |
dfe25020 | 1646 | |
63a212ab | 1647 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1648 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
63a212ab SB |
1649 | goto error_brelse; |
1650 | } | |
1651 | ||
2b82032c | 1652 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1653 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
2b82032c YZ |
1654 | goto error_brelse; |
1655 | } | |
1656 | ||
1657 | if (device->writeable) { | |
0c1daee0 | 1658 | lock_chunks(root); |
2b82032c | 1659 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1660 | device->fs_devices->rw_devices--; |
0c1daee0 | 1661 | unlock_chunks(root); |
1f78160c | 1662 | clear_super = true; |
dfe25020 | 1663 | } |
a061fc8d | 1664 | |
d7901554 | 1665 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1666 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1667 | mutex_lock(&uuid_mutex); |
a061fc8d | 1668 | if (ret) |
9b3517e9 | 1669 | goto error_undo; |
a061fc8d | 1670 | |
63a212ab SB |
1671 | /* |
1672 | * TODO: the superblock still includes this device in its num_devices | |
1673 | * counter although write_all_supers() is not locked out. This | |
1674 | * could give a filesystem state which requires a degraded mount. | |
1675 | */ | |
a061fc8d CM |
1676 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1677 | if (ret) | |
9b3517e9 | 1678 | goto error_undo; |
a061fc8d | 1679 | |
2b82032c | 1680 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1681 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1682 | |
1683 | /* | |
1684 | * the device list mutex makes sure that we don't change | |
1685 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1686 | * the device supers. Whoever is writing all supers, should |
1687 | * lock the device list mutex before getting the number of | |
1688 | * devices in the super block (super_copy). Conversely, | |
1689 | * whoever updates the number of devices in the super block | |
1690 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1691 | */ |
1f78160c XG |
1692 | |
1693 | cur_devices = device->fs_devices; | |
e5e9a520 | 1694 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1695 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1696 | |
e4404d6e | 1697 | device->fs_devices->num_devices--; |
02db0844 | 1698 | device->fs_devices->total_devices--; |
2b82032c | 1699 | |
cd02dca5 | 1700 | if (device->missing) |
3a7d55c8 | 1701 | device->fs_devices->missing_devices--; |
cd02dca5 | 1702 | |
2b82032c YZ |
1703 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1704 | struct btrfs_device, dev_list); | |
1705 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1706 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1707 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1708 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1709 | ||
0bfaa9c5 | 1710 | if (device->bdev) { |
e4404d6e | 1711 | device->fs_devices->open_devices--; |
0bfaa9c5 ES |
1712 | /* remove sysfs entry */ |
1713 | btrfs_kobj_rm_device(root->fs_info, device); | |
1714 | } | |
99994cde | 1715 | |
1f78160c | 1716 | call_rcu(&device->rcu, free_device); |
e4404d6e | 1717 | |
6c41761f DS |
1718 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1719 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
d7306801 | 1720 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2b82032c | 1721 | |
1f78160c | 1722 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1723 | struct btrfs_fs_devices *fs_devices; |
1724 | fs_devices = root->fs_info->fs_devices; | |
1725 | while (fs_devices) { | |
8321cf25 RS |
1726 | if (fs_devices->seed == cur_devices) { |
1727 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1728 | break; |
8321cf25 | 1729 | } |
e4404d6e | 1730 | fs_devices = fs_devices->seed; |
2b82032c | 1731 | } |
1f78160c | 1732 | cur_devices->seed = NULL; |
1f78160c | 1733 | __btrfs_close_devices(cur_devices); |
1f78160c | 1734 | free_fs_devices(cur_devices); |
2b82032c YZ |
1735 | } |
1736 | ||
5af3e8cc SB |
1737 | root->fs_info->num_tolerated_disk_barrier_failures = |
1738 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1739 | ||
2b82032c YZ |
1740 | /* |
1741 | * at this point, the device is zero sized. We want to | |
1742 | * remove it from the devices list and zero out the old super | |
1743 | */ | |
aa1b8cd4 | 1744 | if (clear_super && disk_super) { |
4d90d28b AJ |
1745 | u64 bytenr; |
1746 | int i; | |
1747 | ||
dfe25020 CM |
1748 | /* make sure this device isn't detected as part of |
1749 | * the FS anymore | |
1750 | */ | |
1751 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1752 | set_buffer_dirty(bh); | |
1753 | sync_dirty_buffer(bh); | |
4d90d28b AJ |
1754 | |
1755 | /* clear the mirror copies of super block on the disk | |
1756 | * being removed, 0th copy is been taken care above and | |
1757 | * the below would take of the rest | |
1758 | */ | |
1759 | for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1760 | bytenr = btrfs_sb_offset(i); | |
1761 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
1762 | i_size_read(bdev->bd_inode)) | |
1763 | break; | |
1764 | ||
1765 | brelse(bh); | |
1766 | bh = __bread(bdev, bytenr / 4096, | |
1767 | BTRFS_SUPER_INFO_SIZE); | |
1768 | if (!bh) | |
1769 | continue; | |
1770 | ||
1771 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1772 | ||
1773 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
1774 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { | |
1775 | continue; | |
1776 | } | |
1777 | memset(&disk_super->magic, 0, | |
1778 | sizeof(disk_super->magic)); | |
1779 | set_buffer_dirty(bh); | |
1780 | sync_dirty_buffer(bh); | |
1781 | } | |
dfe25020 | 1782 | } |
a061fc8d | 1783 | |
a061fc8d | 1784 | ret = 0; |
a061fc8d | 1785 | |
5a1972bd QW |
1786 | if (bdev) { |
1787 | /* Notify udev that device has changed */ | |
3c911608 | 1788 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); |
b8b8ff59 | 1789 | |
5a1972bd QW |
1790 | /* Update ctime/mtime for device path for libblkid */ |
1791 | update_dev_time(device_path); | |
1792 | } | |
1793 | ||
a061fc8d CM |
1794 | error_brelse: |
1795 | brelse(bh); | |
dfe25020 | 1796 | if (bdev) |
e525fd89 | 1797 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1798 | out: |
1799 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1800 | return ret; |
9b3517e9 ID |
1801 | error_undo: |
1802 | if (device->writeable) { | |
0c1daee0 | 1803 | lock_chunks(root); |
9b3517e9 ID |
1804 | list_add(&device->dev_alloc_list, |
1805 | &root->fs_info->fs_devices->alloc_list); | |
c3929c36 | 1806 | device->fs_devices->rw_devices++; |
0c1daee0 | 1807 | unlock_chunks(root); |
9b3517e9 ID |
1808 | } |
1809 | goto error_brelse; | |
a061fc8d CM |
1810 | } |
1811 | ||
084b6e7c QW |
1812 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1813 | struct btrfs_device *srcdev) | |
e93c89c1 | 1814 | { |
d51908ce AJ |
1815 | struct btrfs_fs_devices *fs_devices; |
1816 | ||
e93c89c1 | 1817 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 1818 | |
25e8e911 AJ |
1819 | /* |
1820 | * in case of fs with no seed, srcdev->fs_devices will point | |
1821 | * to fs_devices of fs_info. However when the dev being replaced is | |
1822 | * a seed dev it will point to the seed's local fs_devices. In short | |
1823 | * srcdev will have its correct fs_devices in both the cases. | |
1824 | */ | |
1825 | fs_devices = srcdev->fs_devices; | |
d51908ce | 1826 | |
e93c89c1 SB |
1827 | list_del_rcu(&srcdev->dev_list); |
1828 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 1829 | fs_devices->num_devices--; |
82372bc8 | 1830 | if (srcdev->missing) |
d51908ce | 1831 | fs_devices->missing_devices--; |
e93c89c1 | 1832 | |
82372bc8 MX |
1833 | if (srcdev->writeable) { |
1834 | fs_devices->rw_devices--; | |
1835 | /* zero out the old super if it is writable */ | |
1836 | btrfs_scratch_superblock(srcdev); | |
1357272f ID |
1837 | } |
1838 | ||
82372bc8 | 1839 | if (srcdev->bdev) |
d51908ce | 1840 | fs_devices->open_devices--; |
084b6e7c QW |
1841 | } |
1842 | ||
1843 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
1844 | struct btrfs_device *srcdev) | |
1845 | { | |
1846 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 SB |
1847 | |
1848 | call_rcu(&srcdev->rcu, free_device); | |
94d5f0c2 AJ |
1849 | |
1850 | /* | |
1851 | * unless fs_devices is seed fs, num_devices shouldn't go | |
1852 | * zero | |
1853 | */ | |
1854 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
1855 | ||
1856 | /* if this is no devs we rather delete the fs_devices */ | |
1857 | if (!fs_devices->num_devices) { | |
1858 | struct btrfs_fs_devices *tmp_fs_devices; | |
1859 | ||
1860 | tmp_fs_devices = fs_info->fs_devices; | |
1861 | while (tmp_fs_devices) { | |
1862 | if (tmp_fs_devices->seed == fs_devices) { | |
1863 | tmp_fs_devices->seed = fs_devices->seed; | |
1864 | break; | |
1865 | } | |
1866 | tmp_fs_devices = tmp_fs_devices->seed; | |
1867 | } | |
1868 | fs_devices->seed = NULL; | |
8bef8401 AJ |
1869 | __btrfs_close_devices(fs_devices); |
1870 | free_fs_devices(fs_devices); | |
94d5f0c2 | 1871 | } |
e93c89c1 SB |
1872 | } |
1873 | ||
1874 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
1875 | struct btrfs_device *tgtdev) | |
1876 | { | |
1877 | struct btrfs_device *next_device; | |
1878 | ||
67a2c45e | 1879 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
1880 | WARN_ON(!tgtdev); |
1881 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
1882 | if (tgtdev->bdev) { | |
1883 | btrfs_scratch_superblock(tgtdev); | |
1884 | fs_info->fs_devices->open_devices--; | |
1885 | } | |
1886 | fs_info->fs_devices->num_devices--; | |
e93c89c1 SB |
1887 | |
1888 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
1889 | struct btrfs_device, dev_list); | |
1890 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
1891 | fs_info->sb->s_bdev = next_device->bdev; | |
1892 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
1893 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1894 | list_del_rcu(&tgtdev->dev_list); | |
1895 | ||
1896 | call_rcu(&tgtdev->rcu, free_device); | |
1897 | ||
1898 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 1899 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
1900 | } |
1901 | ||
48a3b636 ES |
1902 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
1903 | struct btrfs_device **device) | |
7ba15b7d SB |
1904 | { |
1905 | int ret = 0; | |
1906 | struct btrfs_super_block *disk_super; | |
1907 | u64 devid; | |
1908 | u8 *dev_uuid; | |
1909 | struct block_device *bdev; | |
1910 | struct buffer_head *bh; | |
1911 | ||
1912 | *device = NULL; | |
1913 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
1914 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
1915 | if (ret) | |
1916 | return ret; | |
1917 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1918 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
1919 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 1920 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
1921 | disk_super->fsid); |
1922 | brelse(bh); | |
1923 | if (!*device) | |
1924 | ret = -ENOENT; | |
1925 | blkdev_put(bdev, FMODE_READ); | |
1926 | return ret; | |
1927 | } | |
1928 | ||
1929 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
1930 | char *device_path, | |
1931 | struct btrfs_device **device) | |
1932 | { | |
1933 | *device = NULL; | |
1934 | if (strcmp(device_path, "missing") == 0) { | |
1935 | struct list_head *devices; | |
1936 | struct btrfs_device *tmp; | |
1937 | ||
1938 | devices = &root->fs_info->fs_devices->devices; | |
1939 | /* | |
1940 | * It is safe to read the devices since the volume_mutex | |
1941 | * is held by the caller. | |
1942 | */ | |
1943 | list_for_each_entry(tmp, devices, dev_list) { | |
1944 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
1945 | *device = tmp; | |
1946 | break; | |
1947 | } | |
1948 | } | |
1949 | ||
1950 | if (!*device) { | |
efe120a0 | 1951 | btrfs_err(root->fs_info, "no missing device found"); |
7ba15b7d SB |
1952 | return -ENOENT; |
1953 | } | |
1954 | ||
1955 | return 0; | |
1956 | } else { | |
1957 | return btrfs_find_device_by_path(root, device_path, device); | |
1958 | } | |
1959 | } | |
1960 | ||
2b82032c YZ |
1961 | /* |
1962 | * does all the dirty work required for changing file system's UUID. | |
1963 | */ | |
125ccb0a | 1964 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1965 | { |
1966 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1967 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1968 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1969 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1970 | struct btrfs_device *device; |
1971 | u64 super_flags; | |
1972 | ||
1973 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1974 | if (!fs_devices->seeding) |
2b82032c YZ |
1975 | return -EINVAL; |
1976 | ||
2208a378 ID |
1977 | seed_devices = __alloc_fs_devices(); |
1978 | if (IS_ERR(seed_devices)) | |
1979 | return PTR_ERR(seed_devices); | |
2b82032c | 1980 | |
e4404d6e YZ |
1981 | old_devices = clone_fs_devices(fs_devices); |
1982 | if (IS_ERR(old_devices)) { | |
1983 | kfree(seed_devices); | |
1984 | return PTR_ERR(old_devices); | |
2b82032c | 1985 | } |
e4404d6e | 1986 | |
2b82032c YZ |
1987 | list_add(&old_devices->list, &fs_uuids); |
1988 | ||
e4404d6e YZ |
1989 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1990 | seed_devices->opened = 1; | |
1991 | INIT_LIST_HEAD(&seed_devices->devices); | |
1992 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1993 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1994 | |
1995 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1996 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1997 | synchronize_rcu); | |
2196d6e8 MX |
1998 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
1999 | device->fs_devices = seed_devices; | |
c9513edb | 2000 | |
2196d6e8 | 2001 | lock_chunks(root); |
e4404d6e | 2002 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
2196d6e8 | 2003 | unlock_chunks(root); |
e4404d6e | 2004 | |
2b82032c YZ |
2005 | fs_devices->seeding = 0; |
2006 | fs_devices->num_devices = 0; | |
2007 | fs_devices->open_devices = 0; | |
69611ac8 | 2008 | fs_devices->missing_devices = 0; |
69611ac8 | 2009 | fs_devices->rotating = 0; |
e4404d6e | 2010 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2011 | |
2012 | generate_random_uuid(fs_devices->fsid); | |
2013 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2014 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
f7171750 FDBM |
2015 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2016 | ||
2b82032c YZ |
2017 | super_flags = btrfs_super_flags(disk_super) & |
2018 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2019 | btrfs_set_super_flags(disk_super, super_flags); | |
2020 | ||
2021 | return 0; | |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * strore the expected generation for seed devices in device items. | |
2026 | */ | |
2027 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2028 | struct btrfs_root *root) | |
2029 | { | |
2030 | struct btrfs_path *path; | |
2031 | struct extent_buffer *leaf; | |
2032 | struct btrfs_dev_item *dev_item; | |
2033 | struct btrfs_device *device; | |
2034 | struct btrfs_key key; | |
2035 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2036 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2037 | u64 devid; | |
2038 | int ret; | |
2039 | ||
2040 | path = btrfs_alloc_path(); | |
2041 | if (!path) | |
2042 | return -ENOMEM; | |
2043 | ||
2044 | root = root->fs_info->chunk_root; | |
2045 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2046 | key.offset = 0; | |
2047 | key.type = BTRFS_DEV_ITEM_KEY; | |
2048 | ||
2049 | while (1) { | |
2050 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2051 | if (ret < 0) | |
2052 | goto error; | |
2053 | ||
2054 | leaf = path->nodes[0]; | |
2055 | next_slot: | |
2056 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2057 | ret = btrfs_next_leaf(root, path); | |
2058 | if (ret > 0) | |
2059 | break; | |
2060 | if (ret < 0) | |
2061 | goto error; | |
2062 | leaf = path->nodes[0]; | |
2063 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2064 | btrfs_release_path(path); |
2b82032c YZ |
2065 | continue; |
2066 | } | |
2067 | ||
2068 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2069 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2070 | key.type != BTRFS_DEV_ITEM_KEY) | |
2071 | break; | |
2072 | ||
2073 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2074 | struct btrfs_dev_item); | |
2075 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2076 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2077 | BTRFS_UUID_SIZE); |
1473b24e | 2078 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2079 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
2080 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2081 | fs_uuid); | |
79787eaa | 2082 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2083 | |
2084 | if (device->fs_devices->seeding) { | |
2085 | btrfs_set_device_generation(leaf, dev_item, | |
2086 | device->generation); | |
2087 | btrfs_mark_buffer_dirty(leaf); | |
2088 | } | |
2089 | ||
2090 | path->slots[0]++; | |
2091 | goto next_slot; | |
2092 | } | |
2093 | ret = 0; | |
2094 | error: | |
2095 | btrfs_free_path(path); | |
2096 | return ret; | |
2097 | } | |
2098 | ||
788f20eb CM |
2099 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
2100 | { | |
d5e2003c | 2101 | struct request_queue *q; |
788f20eb CM |
2102 | struct btrfs_trans_handle *trans; |
2103 | struct btrfs_device *device; | |
2104 | struct block_device *bdev; | |
788f20eb | 2105 | struct list_head *devices; |
2b82032c | 2106 | struct super_block *sb = root->fs_info->sb; |
606686ee | 2107 | struct rcu_string *name; |
3c1dbdf5 | 2108 | u64 tmp; |
2b82032c | 2109 | int seeding_dev = 0; |
788f20eb CM |
2110 | int ret = 0; |
2111 | ||
2b82032c | 2112 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 2113 | return -EROFS; |
788f20eb | 2114 | |
a5d16333 | 2115 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 2116 | root->fs_info->bdev_holder); |
7f59203a JB |
2117 | if (IS_ERR(bdev)) |
2118 | return PTR_ERR(bdev); | |
a2135011 | 2119 | |
2b82032c YZ |
2120 | if (root->fs_info->fs_devices->seeding) { |
2121 | seeding_dev = 1; | |
2122 | down_write(&sb->s_umount); | |
2123 | mutex_lock(&uuid_mutex); | |
2124 | } | |
2125 | ||
8c8bee1d | 2126 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2127 | |
788f20eb | 2128 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
2129 | |
2130 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 2131 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2132 | if (device->bdev == bdev) { |
2133 | ret = -EEXIST; | |
d25628bd LB |
2134 | mutex_unlock( |
2135 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2136 | goto error; |
788f20eb CM |
2137 | } |
2138 | } | |
d25628bd | 2139 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2140 | |
12bd2fc0 ID |
2141 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
2142 | if (IS_ERR(device)) { | |
788f20eb | 2143 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2144 | ret = PTR_ERR(device); |
2b82032c | 2145 | goto error; |
788f20eb CM |
2146 | } |
2147 | ||
606686ee JB |
2148 | name = rcu_string_strdup(device_path, GFP_NOFS); |
2149 | if (!name) { | |
788f20eb | 2150 | kfree(device); |
2b82032c YZ |
2151 | ret = -ENOMEM; |
2152 | goto error; | |
788f20eb | 2153 | } |
606686ee | 2154 | rcu_assign_pointer(device->name, name); |
2b82032c | 2155 | |
a22285a6 | 2156 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2157 | if (IS_ERR(trans)) { |
606686ee | 2158 | rcu_string_free(device->name); |
98d5dc13 TI |
2159 | kfree(device); |
2160 | ret = PTR_ERR(trans); | |
2161 | goto error; | |
2162 | } | |
2163 | ||
d5e2003c JB |
2164 | q = bdev_get_queue(bdev); |
2165 | if (blk_queue_discard(q)) | |
2166 | device->can_discard = 1; | |
2b82032c | 2167 | device->writeable = 1; |
2b82032c | 2168 | device->generation = trans->transid; |
788f20eb CM |
2169 | device->io_width = root->sectorsize; |
2170 | device->io_align = root->sectorsize; | |
2171 | device->sector_size = root->sectorsize; | |
2172 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2173 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2174 | device->commit_total_bytes = device->total_bytes; |
788f20eb CM |
2175 | device->dev_root = root->fs_info->dev_root; |
2176 | device->bdev = bdev; | |
dfe25020 | 2177 | device->in_fs_metadata = 1; |
63a212ab | 2178 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2179 | device->mode = FMODE_EXCL; |
27087f37 | 2180 | device->dev_stats_valid = 1; |
2b82032c | 2181 | set_blocksize(device->bdev, 4096); |
788f20eb | 2182 | |
2b82032c YZ |
2183 | if (seeding_dev) { |
2184 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2185 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2186 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2187 | } |
788f20eb | 2188 | |
2b82032c | 2189 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2190 | |
e5e9a520 | 2191 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
2196d6e8 | 2192 | lock_chunks(root); |
1f78160c | 2193 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2194 | list_add(&device->dev_alloc_list, |
2195 | &root->fs_info->fs_devices->alloc_list); | |
2196 | root->fs_info->fs_devices->num_devices++; | |
2197 | root->fs_info->fs_devices->open_devices++; | |
2198 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2199 | root->fs_info->fs_devices->total_devices++; |
2b82032c | 2200 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2201 | |
2bf64758 JB |
2202 | spin_lock(&root->fs_info->free_chunk_lock); |
2203 | root->fs_info->free_chunk_space += device->total_bytes; | |
2204 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2205 | ||
c289811c CM |
2206 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2207 | root->fs_info->fs_devices->rotating = 1; | |
2208 | ||
3c1dbdf5 | 2209 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); |
6c41761f | 2210 | btrfs_set_super_total_bytes(root->fs_info->super_copy, |
3c1dbdf5 | 2211 | tmp + device->total_bytes); |
788f20eb | 2212 | |
3c1dbdf5 | 2213 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); |
6c41761f | 2214 | btrfs_set_super_num_devices(root->fs_info->super_copy, |
3c1dbdf5 | 2215 | tmp + 1); |
0d39376a AJ |
2216 | |
2217 | /* add sysfs device entry */ | |
2218 | btrfs_kobj_add_device(root->fs_info, device); | |
2219 | ||
2196d6e8 MX |
2220 | /* |
2221 | * we've got more storage, clear any full flags on the space | |
2222 | * infos | |
2223 | */ | |
2224 | btrfs_clear_space_info_full(root->fs_info); | |
2225 | ||
2226 | unlock_chunks(root); | |
e5e9a520 | 2227 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2228 | |
2b82032c | 2229 | if (seeding_dev) { |
2196d6e8 | 2230 | lock_chunks(root); |
2b82032c | 2231 | ret = init_first_rw_device(trans, root, device); |
2196d6e8 | 2232 | unlock_chunks(root); |
005d6427 DS |
2233 | if (ret) { |
2234 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2235 | goto error_trans; |
005d6427 | 2236 | } |
2196d6e8 MX |
2237 | } |
2238 | ||
2239 | ret = btrfs_add_device(trans, root, device); | |
2240 | if (ret) { | |
2241 | btrfs_abort_transaction(trans, root, ret); | |
2242 | goto error_trans; | |
2243 | } | |
2244 | ||
2245 | if (seeding_dev) { | |
2246 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2247 | ||
2b82032c | 2248 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2249 | if (ret) { |
2250 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2251 | goto error_trans; |
005d6427 | 2252 | } |
b2373f25 AJ |
2253 | |
2254 | /* Sprouting would change fsid of the mounted root, | |
2255 | * so rename the fsid on the sysfs | |
2256 | */ | |
2257 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2258 | root->fs_info->fsid); | |
2e7910d6 AJ |
2259 | if (kobject_rename(&root->fs_info->fs_devices->super_kobj, |
2260 | fsid_buf)) | |
b2373f25 | 2261 | goto error_trans; |
2b82032c YZ |
2262 | } |
2263 | ||
5af3e8cc SB |
2264 | root->fs_info->num_tolerated_disk_barrier_failures = |
2265 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2266 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2267 | |
2b82032c YZ |
2268 | if (seeding_dev) { |
2269 | mutex_unlock(&uuid_mutex); | |
2270 | up_write(&sb->s_umount); | |
788f20eb | 2271 | |
79787eaa JM |
2272 | if (ret) /* transaction commit */ |
2273 | return ret; | |
2274 | ||
2b82032c | 2275 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
2276 | if (ret < 0) |
2277 | btrfs_error(root->fs_info, ret, | |
2278 | "Failed to relocate sys chunks after " | |
2279 | "device initialization. This can be fixed " | |
2280 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2281 | trans = btrfs_attach_transaction(root); |
2282 | if (IS_ERR(trans)) { | |
2283 | if (PTR_ERR(trans) == -ENOENT) | |
2284 | return 0; | |
2285 | return PTR_ERR(trans); | |
2286 | } | |
2287 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2288 | } |
c9e9f97b | 2289 | |
5a1972bd QW |
2290 | /* Update ctime/mtime for libblkid */ |
2291 | update_dev_time(device_path); | |
2b82032c | 2292 | return ret; |
79787eaa JM |
2293 | |
2294 | error_trans: | |
79787eaa | 2295 | btrfs_end_transaction(trans, root); |
606686ee | 2296 | rcu_string_free(device->name); |
0d39376a | 2297 | btrfs_kobj_rm_device(root->fs_info, device); |
79787eaa | 2298 | kfree(device); |
2b82032c | 2299 | error: |
e525fd89 | 2300 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2301 | if (seeding_dev) { |
2302 | mutex_unlock(&uuid_mutex); | |
2303 | up_write(&sb->s_umount); | |
2304 | } | |
c9e9f97b | 2305 | return ret; |
788f20eb CM |
2306 | } |
2307 | ||
e93c89c1 | 2308 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2309 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2310 | struct btrfs_device **device_out) |
2311 | { | |
2312 | struct request_queue *q; | |
2313 | struct btrfs_device *device; | |
2314 | struct block_device *bdev; | |
2315 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2316 | struct list_head *devices; | |
2317 | struct rcu_string *name; | |
12bd2fc0 | 2318 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2319 | int ret = 0; |
2320 | ||
2321 | *device_out = NULL; | |
1c43366d MX |
2322 | if (fs_info->fs_devices->seeding) { |
2323 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2324 | return -EINVAL; |
1c43366d | 2325 | } |
e93c89c1 SB |
2326 | |
2327 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2328 | fs_info->bdev_holder); | |
1c43366d MX |
2329 | if (IS_ERR(bdev)) { |
2330 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2331 | return PTR_ERR(bdev); |
1c43366d | 2332 | } |
e93c89c1 SB |
2333 | |
2334 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2335 | ||
2336 | devices = &fs_info->fs_devices->devices; | |
2337 | list_for_each_entry(device, devices, dev_list) { | |
2338 | if (device->bdev == bdev) { | |
1c43366d | 2339 | btrfs_err(fs_info, "target device is in the filesystem!"); |
e93c89c1 SB |
2340 | ret = -EEXIST; |
2341 | goto error; | |
2342 | } | |
2343 | } | |
2344 | ||
1c43366d | 2345 | |
7cc8e58d MX |
2346 | if (i_size_read(bdev->bd_inode) < |
2347 | btrfs_device_get_total_bytes(srcdev)) { | |
1c43366d MX |
2348 | btrfs_err(fs_info, "target device is smaller than source device!"); |
2349 | ret = -EINVAL; | |
2350 | goto error; | |
2351 | } | |
2352 | ||
2353 | ||
12bd2fc0 ID |
2354 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2355 | if (IS_ERR(device)) { | |
2356 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2357 | goto error; |
2358 | } | |
2359 | ||
2360 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2361 | if (!name) { | |
2362 | kfree(device); | |
2363 | ret = -ENOMEM; | |
2364 | goto error; | |
2365 | } | |
2366 | rcu_assign_pointer(device->name, name); | |
2367 | ||
2368 | q = bdev_get_queue(bdev); | |
2369 | if (blk_queue_discard(q)) | |
2370 | device->can_discard = 1; | |
2371 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2372 | device->writeable = 1; | |
e93c89c1 SB |
2373 | device->generation = 0; |
2374 | device->io_width = root->sectorsize; | |
2375 | device->io_align = root->sectorsize; | |
2376 | device->sector_size = root->sectorsize; | |
7cc8e58d MX |
2377 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2378 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2379 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2380 | ASSERT(list_empty(&srcdev->resized_list)); |
2381 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2382 | device->commit_bytes_used = device->bytes_used; |
e93c89c1 SB |
2383 | device->dev_root = fs_info->dev_root; |
2384 | device->bdev = bdev; | |
2385 | device->in_fs_metadata = 1; | |
2386 | device->is_tgtdev_for_dev_replace = 1; | |
2387 | device->mode = FMODE_EXCL; | |
27087f37 | 2388 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2389 | set_blocksize(device->bdev, 4096); |
2390 | device->fs_devices = fs_info->fs_devices; | |
2391 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2392 | fs_info->fs_devices->num_devices++; | |
2393 | fs_info->fs_devices->open_devices++; | |
e93c89c1 SB |
2394 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2395 | ||
2396 | *device_out = device; | |
2397 | return ret; | |
2398 | ||
2399 | error: | |
2400 | blkdev_put(bdev, FMODE_EXCL); | |
2401 | return ret; | |
2402 | } | |
2403 | ||
2404 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2405 | struct btrfs_device *tgtdev) | |
2406 | { | |
2407 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2408 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2409 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2410 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2411 | tgtdev->dev_root = fs_info->dev_root; | |
2412 | tgtdev->in_fs_metadata = 1; | |
2413 | } | |
2414 | ||
d397712b CM |
2415 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2416 | struct btrfs_device *device) | |
0b86a832 CM |
2417 | { |
2418 | int ret; | |
2419 | struct btrfs_path *path; | |
2420 | struct btrfs_root *root; | |
2421 | struct btrfs_dev_item *dev_item; | |
2422 | struct extent_buffer *leaf; | |
2423 | struct btrfs_key key; | |
2424 | ||
2425 | root = device->dev_root->fs_info->chunk_root; | |
2426 | ||
2427 | path = btrfs_alloc_path(); | |
2428 | if (!path) | |
2429 | return -ENOMEM; | |
2430 | ||
2431 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2432 | key.type = BTRFS_DEV_ITEM_KEY; | |
2433 | key.offset = device->devid; | |
2434 | ||
2435 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2436 | if (ret < 0) | |
2437 | goto out; | |
2438 | ||
2439 | if (ret > 0) { | |
2440 | ret = -ENOENT; | |
2441 | goto out; | |
2442 | } | |
2443 | ||
2444 | leaf = path->nodes[0]; | |
2445 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2446 | ||
2447 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2448 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2449 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2450 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2451 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2452 | btrfs_set_device_total_bytes(leaf, dev_item, |
2453 | btrfs_device_get_disk_total_bytes(device)); | |
2454 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2455 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2456 | btrfs_mark_buffer_dirty(leaf); |
2457 | ||
2458 | out: | |
2459 | btrfs_free_path(path); | |
2460 | return ret; | |
2461 | } | |
2462 | ||
2196d6e8 | 2463 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2464 | struct btrfs_device *device, u64 new_size) |
2465 | { | |
2466 | struct btrfs_super_block *super_copy = | |
6c41761f | 2467 | device->dev_root->fs_info->super_copy; |
935e5cc9 | 2468 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2469 | u64 old_total; |
2470 | u64 diff; | |
8f18cf13 | 2471 | |
2b82032c YZ |
2472 | if (!device->writeable) |
2473 | return -EACCES; | |
2196d6e8 MX |
2474 | |
2475 | lock_chunks(device->dev_root); | |
2476 | old_total = btrfs_super_total_bytes(super_copy); | |
2477 | diff = new_size - device->total_bytes; | |
2478 | ||
63a212ab | 2479 | if (new_size <= device->total_bytes || |
2196d6e8 MX |
2480 | device->is_tgtdev_for_dev_replace) { |
2481 | unlock_chunks(device->dev_root); | |
2b82032c | 2482 | return -EINVAL; |
2196d6e8 | 2483 | } |
2b82032c | 2484 | |
935e5cc9 | 2485 | fs_devices = device->dev_root->fs_info->fs_devices; |
2b82032c | 2486 | |
8f18cf13 | 2487 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2488 | device->fs_devices->total_rw_bytes += diff; |
2489 | ||
7cc8e58d MX |
2490 | btrfs_device_set_total_bytes(device, new_size); |
2491 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4184ea7f | 2492 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
935e5cc9 MX |
2493 | if (list_empty(&device->resized_list)) |
2494 | list_add_tail(&device->resized_list, | |
2495 | &fs_devices->resized_devices); | |
2196d6e8 | 2496 | unlock_chunks(device->dev_root); |
4184ea7f | 2497 | |
8f18cf13 CM |
2498 | return btrfs_update_device(trans, device); |
2499 | } | |
2500 | ||
2501 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
a688a04a | 2502 | struct btrfs_root *root, u64 chunk_objectid, |
8f18cf13 CM |
2503 | u64 chunk_offset) |
2504 | { | |
2505 | int ret; | |
2506 | struct btrfs_path *path; | |
2507 | struct btrfs_key key; | |
2508 | ||
2509 | root = root->fs_info->chunk_root; | |
2510 | path = btrfs_alloc_path(); | |
2511 | if (!path) | |
2512 | return -ENOMEM; | |
2513 | ||
2514 | key.objectid = chunk_objectid; | |
2515 | key.offset = chunk_offset; | |
2516 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2517 | ||
2518 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2519 | if (ret < 0) |
2520 | goto out; | |
2521 | else if (ret > 0) { /* Logic error or corruption */ | |
2522 | btrfs_error(root->fs_info, -ENOENT, | |
2523 | "Failed lookup while freeing chunk."); | |
2524 | ret = -ENOENT; | |
2525 | goto out; | |
2526 | } | |
8f18cf13 CM |
2527 | |
2528 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
2529 | if (ret < 0) |
2530 | btrfs_error(root->fs_info, ret, | |
2531 | "Failed to delete chunk item."); | |
2532 | out: | |
8f18cf13 | 2533 | btrfs_free_path(path); |
65a246c5 | 2534 | return ret; |
8f18cf13 CM |
2535 | } |
2536 | ||
b2950863 | 2537 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2538 | chunk_offset) |
2539 | { | |
6c41761f | 2540 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2541 | struct btrfs_disk_key *disk_key; |
2542 | struct btrfs_chunk *chunk; | |
2543 | u8 *ptr; | |
2544 | int ret = 0; | |
2545 | u32 num_stripes; | |
2546 | u32 array_size; | |
2547 | u32 len = 0; | |
2548 | u32 cur; | |
2549 | struct btrfs_key key; | |
2550 | ||
2196d6e8 | 2551 | lock_chunks(root); |
8f18cf13 CM |
2552 | array_size = btrfs_super_sys_array_size(super_copy); |
2553 | ||
2554 | ptr = super_copy->sys_chunk_array; | |
2555 | cur = 0; | |
2556 | ||
2557 | while (cur < array_size) { | |
2558 | disk_key = (struct btrfs_disk_key *)ptr; | |
2559 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2560 | ||
2561 | len = sizeof(*disk_key); | |
2562 | ||
2563 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2564 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2565 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2566 | len += btrfs_chunk_item_size(num_stripes); | |
2567 | } else { | |
2568 | ret = -EIO; | |
2569 | break; | |
2570 | } | |
2571 | if (key.objectid == chunk_objectid && | |
2572 | key.offset == chunk_offset) { | |
2573 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2574 | array_size -= len; | |
2575 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2576 | } else { | |
2577 | ptr += len; | |
2578 | cur += len; | |
2579 | } | |
2580 | } | |
2196d6e8 | 2581 | unlock_chunks(root); |
8f18cf13 CM |
2582 | return ret; |
2583 | } | |
2584 | ||
47ab2a6c JB |
2585 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
2586 | struct btrfs_root *root, u64 chunk_offset) | |
8f18cf13 CM |
2587 | { |
2588 | struct extent_map_tree *em_tree; | |
8f18cf13 | 2589 | struct extent_map *em; |
47ab2a6c | 2590 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
8f18cf13 | 2591 | struct map_lookup *map; |
2196d6e8 | 2592 | u64 dev_extent_len = 0; |
47ab2a6c | 2593 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2594 | int i, ret = 0; |
8f18cf13 | 2595 | |
47ab2a6c | 2596 | /* Just in case */ |
8f18cf13 | 2597 | root = root->fs_info->chunk_root; |
8f18cf13 CM |
2598 | em_tree = &root->fs_info->mapping_tree.map_tree; |
2599 | ||
890871be | 2600 | read_lock(&em_tree->lock); |
8f18cf13 | 2601 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2602 | read_unlock(&em_tree->lock); |
8f18cf13 | 2603 | |
47ab2a6c JB |
2604 | if (!em || em->start > chunk_offset || |
2605 | em->start + em->len < chunk_offset) { | |
2606 | /* | |
2607 | * This is a logic error, but we don't want to just rely on the | |
2608 | * user having built with ASSERT enabled, so if ASSERT doens't | |
2609 | * do anything we still error out. | |
2610 | */ | |
2611 | ASSERT(0); | |
2612 | if (em) | |
2613 | free_extent_map(em); | |
2614 | return -EINVAL; | |
2615 | } | |
8f18cf13 CM |
2616 | map = (struct map_lookup *)em->bdev; |
2617 | ||
2618 | for (i = 0; i < map->num_stripes; i++) { | |
47ab2a6c | 2619 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2620 | ret = btrfs_free_dev_extent(trans, device, |
2621 | map->stripes[i].physical, | |
2622 | &dev_extent_len); | |
47ab2a6c JB |
2623 | if (ret) { |
2624 | btrfs_abort_transaction(trans, root, ret); | |
2625 | goto out; | |
2626 | } | |
a061fc8d | 2627 | |
2196d6e8 MX |
2628 | if (device->bytes_used > 0) { |
2629 | lock_chunks(root); | |
2630 | btrfs_device_set_bytes_used(device, | |
2631 | device->bytes_used - dev_extent_len); | |
2632 | spin_lock(&root->fs_info->free_chunk_lock); | |
2633 | root->fs_info->free_chunk_space += dev_extent_len; | |
2634 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2635 | btrfs_clear_space_info_full(root->fs_info); | |
2636 | unlock_chunks(root); | |
2637 | } | |
a061fc8d | 2638 | |
dfe25020 CM |
2639 | if (map->stripes[i].dev) { |
2640 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c JB |
2641 | if (ret) { |
2642 | btrfs_abort_transaction(trans, root, ret); | |
2643 | goto out; | |
2644 | } | |
dfe25020 | 2645 | } |
8f18cf13 | 2646 | } |
a688a04a | 2647 | ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset); |
47ab2a6c JB |
2648 | if (ret) { |
2649 | btrfs_abort_transaction(trans, root, ret); | |
2650 | goto out; | |
2651 | } | |
8f18cf13 | 2652 | |
1abe9b8a | 2653 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2654 | ||
8f18cf13 CM |
2655 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2656 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
47ab2a6c JB |
2657 | if (ret) { |
2658 | btrfs_abort_transaction(trans, root, ret); | |
2659 | goto out; | |
2660 | } | |
8f18cf13 CM |
2661 | } |
2662 | ||
04216820 | 2663 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em); |
47ab2a6c JB |
2664 | if (ret) { |
2665 | btrfs_abort_transaction(trans, extent_root, ret); | |
2666 | goto out; | |
2667 | } | |
2b82032c | 2668 | |
47ab2a6c | 2669 | out: |
2b82032c YZ |
2670 | /* once for us */ |
2671 | free_extent_map(em); | |
47ab2a6c JB |
2672 | return ret; |
2673 | } | |
2b82032c | 2674 | |
47ab2a6c | 2675 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
a688a04a ZL |
2676 | u64 chunk_objectid, |
2677 | u64 chunk_offset) | |
47ab2a6c JB |
2678 | { |
2679 | struct btrfs_root *extent_root; | |
2680 | struct btrfs_trans_handle *trans; | |
2681 | int ret; | |
2b82032c | 2682 | |
47ab2a6c JB |
2683 | root = root->fs_info->chunk_root; |
2684 | extent_root = root->fs_info->extent_root; | |
2685 | ||
2686 | ret = btrfs_can_relocate(extent_root, chunk_offset); | |
2687 | if (ret) | |
2688 | return -ENOSPC; | |
2689 | ||
2690 | /* step one, relocate all the extents inside this chunk */ | |
2691 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); | |
2692 | if (ret) | |
2693 | return ret; | |
2694 | ||
2695 | trans = btrfs_start_transaction(root, 0); | |
2696 | if (IS_ERR(trans)) { | |
2697 | ret = PTR_ERR(trans); | |
2698 | btrfs_std_error(root->fs_info, ret); | |
2699 | return ret; | |
2700 | } | |
2701 | ||
2702 | /* | |
2703 | * step two, delete the device extents and the | |
2704 | * chunk tree entries | |
2705 | */ | |
2706 | ret = btrfs_remove_chunk(trans, root, chunk_offset); | |
2b82032c | 2707 | btrfs_end_transaction(trans, root); |
47ab2a6c | 2708 | return ret; |
2b82032c YZ |
2709 | } |
2710 | ||
2711 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2712 | { | |
2713 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2714 | struct btrfs_path *path; | |
2715 | struct extent_buffer *leaf; | |
2716 | struct btrfs_chunk *chunk; | |
2717 | struct btrfs_key key; | |
2718 | struct btrfs_key found_key; | |
2b82032c | 2719 | u64 chunk_type; |
ba1bf481 JB |
2720 | bool retried = false; |
2721 | int failed = 0; | |
2b82032c YZ |
2722 | int ret; |
2723 | ||
2724 | path = btrfs_alloc_path(); | |
2725 | if (!path) | |
2726 | return -ENOMEM; | |
2727 | ||
ba1bf481 | 2728 | again: |
2b82032c YZ |
2729 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2730 | key.offset = (u64)-1; | |
2731 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2732 | ||
2733 | while (1) { | |
2734 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2735 | if (ret < 0) | |
2736 | goto error; | |
79787eaa | 2737 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2738 | |
2739 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2740 | key.type); | |
2741 | if (ret < 0) | |
2742 | goto error; | |
2743 | if (ret > 0) | |
2744 | break; | |
1a40e23b | 2745 | |
2b82032c YZ |
2746 | leaf = path->nodes[0]; |
2747 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2748 | |
2b82032c YZ |
2749 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2750 | struct btrfs_chunk); | |
2751 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2752 | btrfs_release_path(path); |
8f18cf13 | 2753 | |
2b82032c | 2754 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
a688a04a | 2755 | ret = btrfs_relocate_chunk(chunk_root, |
2b82032c YZ |
2756 | found_key.objectid, |
2757 | found_key.offset); | |
ba1bf481 JB |
2758 | if (ret == -ENOSPC) |
2759 | failed++; | |
14586651 HS |
2760 | else |
2761 | BUG_ON(ret); | |
2b82032c | 2762 | } |
8f18cf13 | 2763 | |
2b82032c YZ |
2764 | if (found_key.offset == 0) |
2765 | break; | |
2766 | key.offset = found_key.offset - 1; | |
2767 | } | |
2768 | ret = 0; | |
ba1bf481 JB |
2769 | if (failed && !retried) { |
2770 | failed = 0; | |
2771 | retried = true; | |
2772 | goto again; | |
fae7f21c | 2773 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2774 | ret = -ENOSPC; |
2775 | } | |
2b82032c YZ |
2776 | error: |
2777 | btrfs_free_path(path); | |
2778 | return ret; | |
8f18cf13 CM |
2779 | } |
2780 | ||
0940ebf6 ID |
2781 | static int insert_balance_item(struct btrfs_root *root, |
2782 | struct btrfs_balance_control *bctl) | |
2783 | { | |
2784 | struct btrfs_trans_handle *trans; | |
2785 | struct btrfs_balance_item *item; | |
2786 | struct btrfs_disk_balance_args disk_bargs; | |
2787 | struct btrfs_path *path; | |
2788 | struct extent_buffer *leaf; | |
2789 | struct btrfs_key key; | |
2790 | int ret, err; | |
2791 | ||
2792 | path = btrfs_alloc_path(); | |
2793 | if (!path) | |
2794 | return -ENOMEM; | |
2795 | ||
2796 | trans = btrfs_start_transaction(root, 0); | |
2797 | if (IS_ERR(trans)) { | |
2798 | btrfs_free_path(path); | |
2799 | return PTR_ERR(trans); | |
2800 | } | |
2801 | ||
2802 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2803 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2804 | key.offset = 0; | |
2805 | ||
2806 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2807 | sizeof(*item)); | |
2808 | if (ret) | |
2809 | goto out; | |
2810 | ||
2811 | leaf = path->nodes[0]; | |
2812 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2813 | ||
2814 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2815 | ||
2816 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2817 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2818 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2819 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2820 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2821 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2822 | ||
2823 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2824 | ||
2825 | btrfs_mark_buffer_dirty(leaf); | |
2826 | out: | |
2827 | btrfs_free_path(path); | |
2828 | err = btrfs_commit_transaction(trans, root); | |
2829 | if (err && !ret) | |
2830 | ret = err; | |
2831 | return ret; | |
2832 | } | |
2833 | ||
2834 | static int del_balance_item(struct btrfs_root *root) | |
2835 | { | |
2836 | struct btrfs_trans_handle *trans; | |
2837 | struct btrfs_path *path; | |
2838 | struct btrfs_key key; | |
2839 | int ret, err; | |
2840 | ||
2841 | path = btrfs_alloc_path(); | |
2842 | if (!path) | |
2843 | return -ENOMEM; | |
2844 | ||
2845 | trans = btrfs_start_transaction(root, 0); | |
2846 | if (IS_ERR(trans)) { | |
2847 | btrfs_free_path(path); | |
2848 | return PTR_ERR(trans); | |
2849 | } | |
2850 | ||
2851 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2852 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2853 | key.offset = 0; | |
2854 | ||
2855 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2856 | if (ret < 0) | |
2857 | goto out; | |
2858 | if (ret > 0) { | |
2859 | ret = -ENOENT; | |
2860 | goto out; | |
2861 | } | |
2862 | ||
2863 | ret = btrfs_del_item(trans, root, path); | |
2864 | out: | |
2865 | btrfs_free_path(path); | |
2866 | err = btrfs_commit_transaction(trans, root); | |
2867 | if (err && !ret) | |
2868 | ret = err; | |
2869 | return ret; | |
2870 | } | |
2871 | ||
59641015 ID |
2872 | /* |
2873 | * This is a heuristic used to reduce the number of chunks balanced on | |
2874 | * resume after balance was interrupted. | |
2875 | */ | |
2876 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2877 | { | |
2878 | /* | |
2879 | * Turn on soft mode for chunk types that were being converted. | |
2880 | */ | |
2881 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2882 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2883 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2884 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2885 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2886 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2887 | ||
2888 | /* | |
2889 | * Turn on usage filter if is not already used. The idea is | |
2890 | * that chunks that we have already balanced should be | |
2891 | * reasonably full. Don't do it for chunks that are being | |
2892 | * converted - that will keep us from relocating unconverted | |
2893 | * (albeit full) chunks. | |
2894 | */ | |
2895 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2896 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2897 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2898 | bctl->data.usage = 90; | |
2899 | } | |
2900 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2901 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2902 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2903 | bctl->sys.usage = 90; | |
2904 | } | |
2905 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2906 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2907 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2908 | bctl->meta.usage = 90; | |
2909 | } | |
2910 | } | |
2911 | ||
c9e9f97b ID |
2912 | /* |
2913 | * Should be called with both balance and volume mutexes held to | |
2914 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2915 | * restriper. Same goes for unset_balance_control. | |
2916 | */ | |
2917 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2918 | { | |
2919 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2920 | ||
2921 | BUG_ON(fs_info->balance_ctl); | |
2922 | ||
2923 | spin_lock(&fs_info->balance_lock); | |
2924 | fs_info->balance_ctl = bctl; | |
2925 | spin_unlock(&fs_info->balance_lock); | |
2926 | } | |
2927 | ||
2928 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2929 | { | |
2930 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2931 | ||
2932 | BUG_ON(!fs_info->balance_ctl); | |
2933 | ||
2934 | spin_lock(&fs_info->balance_lock); | |
2935 | fs_info->balance_ctl = NULL; | |
2936 | spin_unlock(&fs_info->balance_lock); | |
2937 | ||
2938 | kfree(bctl); | |
2939 | } | |
2940 | ||
ed25e9b2 ID |
2941 | /* |
2942 | * Balance filters. Return 1 if chunk should be filtered out | |
2943 | * (should not be balanced). | |
2944 | */ | |
899c81ea | 2945 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2946 | struct btrfs_balance_args *bargs) |
2947 | { | |
899c81ea ID |
2948 | chunk_type = chunk_to_extended(chunk_type) & |
2949 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2950 | |
899c81ea | 2951 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2952 | return 0; |
2953 | ||
2954 | return 1; | |
2955 | } | |
2956 | ||
5ce5b3c0 ID |
2957 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
2958 | struct btrfs_balance_args *bargs) | |
2959 | { | |
2960 | struct btrfs_block_group_cache *cache; | |
2961 | u64 chunk_used, user_thresh; | |
2962 | int ret = 1; | |
2963 | ||
2964 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2965 | chunk_used = btrfs_block_group_used(&cache->item); | |
2966 | ||
a105bb88 | 2967 | if (bargs->usage == 0) |
3e39cea6 | 2968 | user_thresh = 1; |
a105bb88 ID |
2969 | else if (bargs->usage > 100) |
2970 | user_thresh = cache->key.offset; | |
2971 | else | |
2972 | user_thresh = div_factor_fine(cache->key.offset, | |
2973 | bargs->usage); | |
2974 | ||
5ce5b3c0 ID |
2975 | if (chunk_used < user_thresh) |
2976 | ret = 0; | |
2977 | ||
2978 | btrfs_put_block_group(cache); | |
2979 | return ret; | |
2980 | } | |
2981 | ||
409d404b ID |
2982 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2983 | struct btrfs_chunk *chunk, | |
2984 | struct btrfs_balance_args *bargs) | |
2985 | { | |
2986 | struct btrfs_stripe *stripe; | |
2987 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2988 | int i; | |
2989 | ||
2990 | for (i = 0; i < num_stripes; i++) { | |
2991 | stripe = btrfs_stripe_nr(chunk, i); | |
2992 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2993 | return 0; | |
2994 | } | |
2995 | ||
2996 | return 1; | |
2997 | } | |
2998 | ||
94e60d5a ID |
2999 | /* [pstart, pend) */ |
3000 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3001 | struct btrfs_chunk *chunk, | |
3002 | u64 chunk_offset, | |
3003 | struct btrfs_balance_args *bargs) | |
3004 | { | |
3005 | struct btrfs_stripe *stripe; | |
3006 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3007 | u64 stripe_offset; | |
3008 | u64 stripe_length; | |
3009 | int factor; | |
3010 | int i; | |
3011 | ||
3012 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3013 | return 0; | |
3014 | ||
3015 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3016 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3017 | factor = num_stripes / 2; | |
3018 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3019 | factor = num_stripes - 1; | |
3020 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3021 | factor = num_stripes - 2; | |
3022 | } else { | |
3023 | factor = num_stripes; | |
3024 | } | |
94e60d5a ID |
3025 | |
3026 | for (i = 0; i < num_stripes; i++) { | |
3027 | stripe = btrfs_stripe_nr(chunk, i); | |
3028 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3029 | continue; | |
3030 | ||
3031 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3032 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3033 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3034 | |
3035 | if (stripe_offset < bargs->pend && | |
3036 | stripe_offset + stripe_length > bargs->pstart) | |
3037 | return 0; | |
3038 | } | |
3039 | ||
3040 | return 1; | |
3041 | } | |
3042 | ||
ea67176a ID |
3043 | /* [vstart, vend) */ |
3044 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3045 | struct btrfs_chunk *chunk, | |
3046 | u64 chunk_offset, | |
3047 | struct btrfs_balance_args *bargs) | |
3048 | { | |
3049 | if (chunk_offset < bargs->vend && | |
3050 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3051 | /* at least part of the chunk is inside this vrange */ | |
3052 | return 0; | |
3053 | ||
3054 | return 1; | |
3055 | } | |
3056 | ||
899c81ea | 3057 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3058 | struct btrfs_balance_args *bargs) |
3059 | { | |
3060 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3061 | return 0; | |
3062 | ||
899c81ea ID |
3063 | chunk_type = chunk_to_extended(chunk_type) & |
3064 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3065 | |
899c81ea | 3066 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3067 | return 1; |
3068 | ||
3069 | return 0; | |
3070 | } | |
3071 | ||
f43ffb60 ID |
3072 | static int should_balance_chunk(struct btrfs_root *root, |
3073 | struct extent_buffer *leaf, | |
3074 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3075 | { | |
3076 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3077 | struct btrfs_balance_args *bargs = NULL; | |
3078 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3079 | ||
3080 | /* type filter */ | |
3081 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3082 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3083 | return 0; | |
3084 | } | |
3085 | ||
3086 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3087 | bargs = &bctl->data; | |
3088 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3089 | bargs = &bctl->sys; | |
3090 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3091 | bargs = &bctl->meta; | |
3092 | ||
ed25e9b2 ID |
3093 | /* profiles filter */ |
3094 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3095 | chunk_profiles_filter(chunk_type, bargs)) { | |
3096 | return 0; | |
5ce5b3c0 ID |
3097 | } |
3098 | ||
3099 | /* usage filter */ | |
3100 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3101 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3102 | return 0; | |
409d404b ID |
3103 | } |
3104 | ||
3105 | /* devid filter */ | |
3106 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3107 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3108 | return 0; | |
94e60d5a ID |
3109 | } |
3110 | ||
3111 | /* drange filter, makes sense only with devid filter */ | |
3112 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3113 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3114 | return 0; | |
ea67176a ID |
3115 | } |
3116 | ||
3117 | /* vrange filter */ | |
3118 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3119 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3120 | return 0; | |
ed25e9b2 ID |
3121 | } |
3122 | ||
cfa4c961 ID |
3123 | /* soft profile changing mode */ |
3124 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3125 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3126 | return 0; | |
3127 | } | |
3128 | ||
7d824b6f DS |
3129 | /* |
3130 | * limited by count, must be the last filter | |
3131 | */ | |
3132 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3133 | if (bargs->limit == 0) | |
3134 | return 0; | |
3135 | else | |
3136 | bargs->limit--; | |
3137 | } | |
3138 | ||
f43ffb60 ID |
3139 | return 1; |
3140 | } | |
3141 | ||
c9e9f97b | 3142 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3143 | { |
19a39dce | 3144 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3145 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3146 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3147 | struct list_head *devices; | |
ec44a35c CM |
3148 | struct btrfs_device *device; |
3149 | u64 old_size; | |
3150 | u64 size_to_free; | |
f43ffb60 | 3151 | struct btrfs_chunk *chunk; |
ec44a35c CM |
3152 | struct btrfs_path *path; |
3153 | struct btrfs_key key; | |
ec44a35c | 3154 | struct btrfs_key found_key; |
c9e9f97b | 3155 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3156 | struct extent_buffer *leaf; |
3157 | int slot; | |
c9e9f97b ID |
3158 | int ret; |
3159 | int enospc_errors = 0; | |
19a39dce | 3160 | bool counting = true; |
7d824b6f DS |
3161 | u64 limit_data = bctl->data.limit; |
3162 | u64 limit_meta = bctl->meta.limit; | |
3163 | u64 limit_sys = bctl->sys.limit; | |
ec44a35c | 3164 | |
ec44a35c | 3165 | /* step one make some room on all the devices */ |
c9e9f97b | 3166 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3167 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3168 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c CM |
3169 | size_to_free = div_factor(old_size, 1); |
3170 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c | 3171 | if (!device->writeable || |
7cc8e58d MX |
3172 | btrfs_device_get_total_bytes(device) - |
3173 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3174 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3175 | continue; |
3176 | ||
3177 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3178 | if (ret == -ENOSPC) |
3179 | break; | |
ec44a35c CM |
3180 | BUG_ON(ret); |
3181 | ||
a22285a6 | 3182 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 3183 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
3184 | |
3185 | ret = btrfs_grow_device(trans, device, old_size); | |
3186 | BUG_ON(ret); | |
3187 | ||
3188 | btrfs_end_transaction(trans, dev_root); | |
3189 | } | |
3190 | ||
3191 | /* step two, relocate all the chunks */ | |
3192 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3193 | if (!path) { |
3194 | ret = -ENOMEM; | |
3195 | goto error; | |
3196 | } | |
19a39dce ID |
3197 | |
3198 | /* zero out stat counters */ | |
3199 | spin_lock(&fs_info->balance_lock); | |
3200 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3201 | spin_unlock(&fs_info->balance_lock); | |
3202 | again: | |
7d824b6f DS |
3203 | if (!counting) { |
3204 | bctl->data.limit = limit_data; | |
3205 | bctl->meta.limit = limit_meta; | |
3206 | bctl->sys.limit = limit_sys; | |
3207 | } | |
ec44a35c CM |
3208 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3209 | key.offset = (u64)-1; | |
3210 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3211 | ||
d397712b | 3212 | while (1) { |
19a39dce | 3213 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3214 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3215 | ret = -ECANCELED; |
3216 | goto error; | |
3217 | } | |
3218 | ||
ec44a35c CM |
3219 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
3220 | if (ret < 0) | |
3221 | goto error; | |
3222 | ||
3223 | /* | |
3224 | * this shouldn't happen, it means the last relocate | |
3225 | * failed | |
3226 | */ | |
3227 | if (ret == 0) | |
c9e9f97b | 3228 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3229 | |
3230 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3231 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
3232 | if (ret) { |
3233 | ret = 0; | |
ec44a35c | 3234 | break; |
c9e9f97b | 3235 | } |
7d9eb12c | 3236 | |
f43ffb60 ID |
3237 | leaf = path->nodes[0]; |
3238 | slot = path->slots[0]; | |
3239 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3240 | |
ec44a35c CM |
3241 | if (found_key.objectid != key.objectid) |
3242 | break; | |
7d9eb12c | 3243 | |
f43ffb60 ID |
3244 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
3245 | ||
19a39dce ID |
3246 | if (!counting) { |
3247 | spin_lock(&fs_info->balance_lock); | |
3248 | bctl->stat.considered++; | |
3249 | spin_unlock(&fs_info->balance_lock); | |
3250 | } | |
3251 | ||
f43ffb60 ID |
3252 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3253 | found_key.offset); | |
b3b4aa74 | 3254 | btrfs_release_path(path); |
f43ffb60 ID |
3255 | if (!ret) |
3256 | goto loop; | |
3257 | ||
19a39dce ID |
3258 | if (counting) { |
3259 | spin_lock(&fs_info->balance_lock); | |
3260 | bctl->stat.expected++; | |
3261 | spin_unlock(&fs_info->balance_lock); | |
3262 | goto loop; | |
3263 | } | |
3264 | ||
ec44a35c | 3265 | ret = btrfs_relocate_chunk(chunk_root, |
ec44a35c CM |
3266 | found_key.objectid, |
3267 | found_key.offset); | |
508794eb JB |
3268 | if (ret && ret != -ENOSPC) |
3269 | goto error; | |
19a39dce | 3270 | if (ret == -ENOSPC) { |
c9e9f97b | 3271 | enospc_errors++; |
19a39dce ID |
3272 | } else { |
3273 | spin_lock(&fs_info->balance_lock); | |
3274 | bctl->stat.completed++; | |
3275 | spin_unlock(&fs_info->balance_lock); | |
3276 | } | |
f43ffb60 | 3277 | loop: |
795a3321 ID |
3278 | if (found_key.offset == 0) |
3279 | break; | |
ba1bf481 | 3280 | key.offset = found_key.offset - 1; |
ec44a35c | 3281 | } |
c9e9f97b | 3282 | |
19a39dce ID |
3283 | if (counting) { |
3284 | btrfs_release_path(path); | |
3285 | counting = false; | |
3286 | goto again; | |
3287 | } | |
ec44a35c CM |
3288 | error: |
3289 | btrfs_free_path(path); | |
c9e9f97b | 3290 | if (enospc_errors) { |
efe120a0 | 3291 | btrfs_info(fs_info, "%d enospc errors during balance", |
c9e9f97b ID |
3292 | enospc_errors); |
3293 | if (!ret) | |
3294 | ret = -ENOSPC; | |
3295 | } | |
3296 | ||
ec44a35c CM |
3297 | return ret; |
3298 | } | |
3299 | ||
0c460c0d ID |
3300 | /** |
3301 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3302 | * @flags: profile to validate | |
3303 | * @extended: if true @flags is treated as an extended profile | |
3304 | */ | |
3305 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3306 | { | |
3307 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3308 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3309 | ||
3310 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3311 | ||
3312 | /* 1) check that all other bits are zeroed */ | |
3313 | if (flags & ~mask) | |
3314 | return 0; | |
3315 | ||
3316 | /* 2) see if profile is reduced */ | |
3317 | if (flags == 0) | |
3318 | return !extended; /* "0" is valid for usual profiles */ | |
3319 | ||
3320 | /* true if exactly one bit set */ | |
3321 | return (flags & (flags - 1)) == 0; | |
3322 | } | |
3323 | ||
837d5b6e ID |
3324 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3325 | { | |
a7e99c69 ID |
3326 | /* cancel requested || normal exit path */ |
3327 | return atomic_read(&fs_info->balance_cancel_req) || | |
3328 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3329 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3330 | } |
3331 | ||
c9e9f97b ID |
3332 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3333 | { | |
0940ebf6 ID |
3334 | int ret; |
3335 | ||
c9e9f97b | 3336 | unset_balance_control(fs_info); |
0940ebf6 | 3337 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 LB |
3338 | if (ret) |
3339 | btrfs_std_error(fs_info, ret); | |
ed0fb78f ID |
3340 | |
3341 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3342 | } |
3343 | ||
c9e9f97b ID |
3344 | /* |
3345 | * Should be called with both balance and volume mutexes held | |
3346 | */ | |
3347 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3348 | struct btrfs_ioctl_balance_args *bargs) | |
3349 | { | |
3350 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3351 | u64 allowed; |
e4837f8f | 3352 | int mixed = 0; |
c9e9f97b | 3353 | int ret; |
8dabb742 | 3354 | u64 num_devices; |
de98ced9 | 3355 | unsigned seq; |
c9e9f97b | 3356 | |
837d5b6e | 3357 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3358 | atomic_read(&fs_info->balance_pause_req) || |
3359 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3360 | ret = -EINVAL; |
3361 | goto out; | |
3362 | } | |
3363 | ||
e4837f8f ID |
3364 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3365 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3366 | mixed = 1; | |
3367 | ||
f43ffb60 ID |
3368 | /* |
3369 | * In case of mixed groups both data and meta should be picked, | |
3370 | * and identical options should be given for both of them. | |
3371 | */ | |
e4837f8f ID |
3372 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3373 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3374 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3375 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3376 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
efe120a0 FH |
3377 | btrfs_err(fs_info, "with mixed groups data and " |
3378 | "metadata balance options must be the same"); | |
f43ffb60 ID |
3379 | ret = -EINVAL; |
3380 | goto out; | |
3381 | } | |
3382 | } | |
3383 | ||
8dabb742 SB |
3384 | num_devices = fs_info->fs_devices->num_devices; |
3385 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
3386 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3387 | BUG_ON(num_devices < 1); | |
3388 | num_devices--; | |
3389 | } | |
3390 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
e4d8ec0f | 3391 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3392 | if (num_devices == 1) |
e4d8ec0f | 3393 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8250dabe | 3394 | else if (num_devices > 1) |
e4d8ec0f | 3395 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3396 | if (num_devices > 2) |
3397 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3398 | if (num_devices > 3) | |
3399 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3400 | BTRFS_BLOCK_GROUP_RAID6); | |
6728b198 ID |
3401 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3402 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
3403 | (bctl->data.target & ~allowed))) { | |
efe120a0 FH |
3404 | btrfs_err(fs_info, "unable to start balance with target " |
3405 | "data profile %llu", | |
c1c9ff7c | 3406 | bctl->data.target); |
e4d8ec0f ID |
3407 | ret = -EINVAL; |
3408 | goto out; | |
3409 | } | |
6728b198 ID |
3410 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3411 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
3412 | (bctl->meta.target & ~allowed))) { | |
efe120a0 FH |
3413 | btrfs_err(fs_info, |
3414 | "unable to start balance with target metadata profile %llu", | |
c1c9ff7c | 3415 | bctl->meta.target); |
e4d8ec0f ID |
3416 | ret = -EINVAL; |
3417 | goto out; | |
3418 | } | |
6728b198 ID |
3419 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3420 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
3421 | (bctl->sys.target & ~allowed))) { | |
efe120a0 FH |
3422 | btrfs_err(fs_info, |
3423 | "unable to start balance with target system profile %llu", | |
c1c9ff7c | 3424 | bctl->sys.target); |
e4d8ec0f ID |
3425 | ret = -EINVAL; |
3426 | goto out; | |
3427 | } | |
3428 | ||
e4837f8f ID |
3429 | /* allow dup'ed data chunks only in mixed mode */ |
3430 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 3431 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
efe120a0 | 3432 | btrfs_err(fs_info, "dup for data is not allowed"); |
e4d8ec0f ID |
3433 | ret = -EINVAL; |
3434 | goto out; | |
3435 | } | |
3436 | ||
3437 | /* allow to reduce meta or sys integrity only if force set */ | |
3438 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3439 | BTRFS_BLOCK_GROUP_RAID10 | |
3440 | BTRFS_BLOCK_GROUP_RAID5 | | |
3441 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3442 | do { |
3443 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3444 | ||
3445 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3446 | (fs_info->avail_system_alloc_bits & allowed) && | |
3447 | !(bctl->sys.target & allowed)) || | |
3448 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3449 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3450 | !(bctl->meta.target & allowed))) { | |
3451 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
efe120a0 | 3452 | btrfs_info(fs_info, "force reducing metadata integrity"); |
de98ced9 | 3453 | } else { |
efe120a0 FH |
3454 | btrfs_err(fs_info, "balance will reduce metadata " |
3455 | "integrity, use force if you want this"); | |
de98ced9 MX |
3456 | ret = -EINVAL; |
3457 | goto out; | |
3458 | } | |
e4d8ec0f | 3459 | } |
de98ced9 | 3460 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3461 | |
5af3e8cc SB |
3462 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3463 | int num_tolerated_disk_barrier_failures; | |
3464 | u64 target = bctl->sys.target; | |
3465 | ||
3466 | num_tolerated_disk_barrier_failures = | |
3467 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3468 | if (num_tolerated_disk_barrier_failures > 0 && | |
3469 | (target & | |
3470 | (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | | |
3471 | BTRFS_AVAIL_ALLOC_BIT_SINGLE))) | |
3472 | num_tolerated_disk_barrier_failures = 0; | |
3473 | else if (num_tolerated_disk_barrier_failures > 1 && | |
3474 | (target & | |
3475 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))) | |
3476 | num_tolerated_disk_barrier_failures = 1; | |
3477 | ||
3478 | fs_info->num_tolerated_disk_barrier_failures = | |
3479 | num_tolerated_disk_barrier_failures; | |
3480 | } | |
3481 | ||
0940ebf6 | 3482 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3483 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3484 | goto out; |
3485 | ||
59641015 ID |
3486 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3487 | BUG_ON(ret == -EEXIST); | |
3488 | set_balance_control(bctl); | |
3489 | } else { | |
3490 | BUG_ON(ret != -EEXIST); | |
3491 | spin_lock(&fs_info->balance_lock); | |
3492 | update_balance_args(bctl); | |
3493 | spin_unlock(&fs_info->balance_lock); | |
3494 | } | |
c9e9f97b | 3495 | |
837d5b6e | 3496 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3497 | mutex_unlock(&fs_info->balance_mutex); |
3498 | ||
3499 | ret = __btrfs_balance(fs_info); | |
3500 | ||
3501 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3502 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3503 | |
bf023ecf ID |
3504 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3505 | fs_info->num_tolerated_disk_barrier_failures = | |
3506 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3507 | } | |
3508 | ||
c9e9f97b ID |
3509 | if (bargs) { |
3510 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3511 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3512 | } |
3513 | ||
3a01aa7a ID |
3514 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3515 | balance_need_close(fs_info)) { | |
3516 | __cancel_balance(fs_info); | |
3517 | } | |
3518 | ||
837d5b6e | 3519 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3520 | |
3521 | return ret; | |
3522 | out: | |
59641015 ID |
3523 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3524 | __cancel_balance(fs_info); | |
ed0fb78f | 3525 | else { |
59641015 | 3526 | kfree(bctl); |
ed0fb78f ID |
3527 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3528 | } | |
59641015 ID |
3529 | return ret; |
3530 | } | |
3531 | ||
3532 | static int balance_kthread(void *data) | |
3533 | { | |
2b6ba629 | 3534 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3535 | int ret = 0; |
59641015 ID |
3536 | |
3537 | mutex_lock(&fs_info->volume_mutex); | |
3538 | mutex_lock(&fs_info->balance_mutex); | |
3539 | ||
2b6ba629 | 3540 | if (fs_info->balance_ctl) { |
efe120a0 | 3541 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3542 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3543 | } |
59641015 ID |
3544 | |
3545 | mutex_unlock(&fs_info->balance_mutex); | |
3546 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3547 | |
59641015 ID |
3548 | return ret; |
3549 | } | |
3550 | ||
2b6ba629 ID |
3551 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3552 | { | |
3553 | struct task_struct *tsk; | |
3554 | ||
3555 | spin_lock(&fs_info->balance_lock); | |
3556 | if (!fs_info->balance_ctl) { | |
3557 | spin_unlock(&fs_info->balance_lock); | |
3558 | return 0; | |
3559 | } | |
3560 | spin_unlock(&fs_info->balance_lock); | |
3561 | ||
3562 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
efe120a0 | 3563 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3564 | return 0; |
3565 | } | |
3566 | ||
3567 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3568 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3569 | } |
3570 | ||
68310a5e | 3571 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3572 | { |
59641015 ID |
3573 | struct btrfs_balance_control *bctl; |
3574 | struct btrfs_balance_item *item; | |
3575 | struct btrfs_disk_balance_args disk_bargs; | |
3576 | struct btrfs_path *path; | |
3577 | struct extent_buffer *leaf; | |
3578 | struct btrfs_key key; | |
3579 | int ret; | |
3580 | ||
3581 | path = btrfs_alloc_path(); | |
3582 | if (!path) | |
3583 | return -ENOMEM; | |
3584 | ||
59641015 ID |
3585 | key.objectid = BTRFS_BALANCE_OBJECTID; |
3586 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3587 | key.offset = 0; | |
3588 | ||
68310a5e | 3589 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3590 | if (ret < 0) |
68310a5e | 3591 | goto out; |
59641015 ID |
3592 | if (ret > 0) { /* ret = -ENOENT; */ |
3593 | ret = 0; | |
68310a5e ID |
3594 | goto out; |
3595 | } | |
3596 | ||
3597 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3598 | if (!bctl) { | |
3599 | ret = -ENOMEM; | |
3600 | goto out; | |
59641015 ID |
3601 | } |
3602 | ||
3603 | leaf = path->nodes[0]; | |
3604 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3605 | ||
68310a5e ID |
3606 | bctl->fs_info = fs_info; |
3607 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3608 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3609 | |
3610 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3611 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3612 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3613 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3614 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3615 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3616 | ||
ed0fb78f ID |
3617 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3618 | ||
68310a5e ID |
3619 | mutex_lock(&fs_info->volume_mutex); |
3620 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3621 | |
68310a5e ID |
3622 | set_balance_control(bctl); |
3623 | ||
3624 | mutex_unlock(&fs_info->balance_mutex); | |
3625 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3626 | out: |
3627 | btrfs_free_path(path); | |
ec44a35c CM |
3628 | return ret; |
3629 | } | |
3630 | ||
837d5b6e ID |
3631 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3632 | { | |
3633 | int ret = 0; | |
3634 | ||
3635 | mutex_lock(&fs_info->balance_mutex); | |
3636 | if (!fs_info->balance_ctl) { | |
3637 | mutex_unlock(&fs_info->balance_mutex); | |
3638 | return -ENOTCONN; | |
3639 | } | |
3640 | ||
3641 | if (atomic_read(&fs_info->balance_running)) { | |
3642 | atomic_inc(&fs_info->balance_pause_req); | |
3643 | mutex_unlock(&fs_info->balance_mutex); | |
3644 | ||
3645 | wait_event(fs_info->balance_wait_q, | |
3646 | atomic_read(&fs_info->balance_running) == 0); | |
3647 | ||
3648 | mutex_lock(&fs_info->balance_mutex); | |
3649 | /* we are good with balance_ctl ripped off from under us */ | |
3650 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3651 | atomic_dec(&fs_info->balance_pause_req); | |
3652 | } else { | |
3653 | ret = -ENOTCONN; | |
3654 | } | |
3655 | ||
3656 | mutex_unlock(&fs_info->balance_mutex); | |
3657 | return ret; | |
3658 | } | |
3659 | ||
a7e99c69 ID |
3660 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3661 | { | |
e649e587 ID |
3662 | if (fs_info->sb->s_flags & MS_RDONLY) |
3663 | return -EROFS; | |
3664 | ||
a7e99c69 ID |
3665 | mutex_lock(&fs_info->balance_mutex); |
3666 | if (!fs_info->balance_ctl) { | |
3667 | mutex_unlock(&fs_info->balance_mutex); | |
3668 | return -ENOTCONN; | |
3669 | } | |
3670 | ||
3671 | atomic_inc(&fs_info->balance_cancel_req); | |
3672 | /* | |
3673 | * if we are running just wait and return, balance item is | |
3674 | * deleted in btrfs_balance in this case | |
3675 | */ | |
3676 | if (atomic_read(&fs_info->balance_running)) { | |
3677 | mutex_unlock(&fs_info->balance_mutex); | |
3678 | wait_event(fs_info->balance_wait_q, | |
3679 | atomic_read(&fs_info->balance_running) == 0); | |
3680 | mutex_lock(&fs_info->balance_mutex); | |
3681 | } else { | |
3682 | /* __cancel_balance needs volume_mutex */ | |
3683 | mutex_unlock(&fs_info->balance_mutex); | |
3684 | mutex_lock(&fs_info->volume_mutex); | |
3685 | mutex_lock(&fs_info->balance_mutex); | |
3686 | ||
3687 | if (fs_info->balance_ctl) | |
3688 | __cancel_balance(fs_info); | |
3689 | ||
3690 | mutex_unlock(&fs_info->volume_mutex); | |
3691 | } | |
3692 | ||
3693 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3694 | atomic_dec(&fs_info->balance_cancel_req); | |
3695 | mutex_unlock(&fs_info->balance_mutex); | |
3696 | return 0; | |
3697 | } | |
3698 | ||
803b2f54 SB |
3699 | static int btrfs_uuid_scan_kthread(void *data) |
3700 | { | |
3701 | struct btrfs_fs_info *fs_info = data; | |
3702 | struct btrfs_root *root = fs_info->tree_root; | |
3703 | struct btrfs_key key; | |
3704 | struct btrfs_key max_key; | |
3705 | struct btrfs_path *path = NULL; | |
3706 | int ret = 0; | |
3707 | struct extent_buffer *eb; | |
3708 | int slot; | |
3709 | struct btrfs_root_item root_item; | |
3710 | u32 item_size; | |
f45388f3 | 3711 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
3712 | |
3713 | path = btrfs_alloc_path(); | |
3714 | if (!path) { | |
3715 | ret = -ENOMEM; | |
3716 | goto out; | |
3717 | } | |
3718 | ||
3719 | key.objectid = 0; | |
3720 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3721 | key.offset = 0; | |
3722 | ||
3723 | max_key.objectid = (u64)-1; | |
3724 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
3725 | max_key.offset = (u64)-1; | |
3726 | ||
803b2f54 | 3727 | while (1) { |
6174d3cb | 3728 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
3729 | if (ret) { |
3730 | if (ret > 0) | |
3731 | ret = 0; | |
3732 | break; | |
3733 | } | |
3734 | ||
3735 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
3736 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
3737 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
3738 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
3739 | goto skip; | |
3740 | ||
3741 | eb = path->nodes[0]; | |
3742 | slot = path->slots[0]; | |
3743 | item_size = btrfs_item_size_nr(eb, slot); | |
3744 | if (item_size < sizeof(root_item)) | |
3745 | goto skip; | |
3746 | ||
803b2f54 SB |
3747 | read_extent_buffer(eb, &root_item, |
3748 | btrfs_item_ptr_offset(eb, slot), | |
3749 | (int)sizeof(root_item)); | |
3750 | if (btrfs_root_refs(&root_item) == 0) | |
3751 | goto skip; | |
f45388f3 FDBM |
3752 | |
3753 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
3754 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
3755 | if (trans) | |
3756 | goto update_tree; | |
3757 | ||
3758 | btrfs_release_path(path); | |
803b2f54 SB |
3759 | /* |
3760 | * 1 - subvol uuid item | |
3761 | * 1 - received_subvol uuid item | |
3762 | */ | |
3763 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
3764 | if (IS_ERR(trans)) { | |
3765 | ret = PTR_ERR(trans); | |
3766 | break; | |
3767 | } | |
f45388f3 FDBM |
3768 | continue; |
3769 | } else { | |
3770 | goto skip; | |
3771 | } | |
3772 | update_tree: | |
3773 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
803b2f54 SB |
3774 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3775 | root_item.uuid, | |
3776 | BTRFS_UUID_KEY_SUBVOL, | |
3777 | key.objectid); | |
3778 | if (ret < 0) { | |
efe120a0 | 3779 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3780 | ret); |
803b2f54 SB |
3781 | break; |
3782 | } | |
3783 | } | |
3784 | ||
3785 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
803b2f54 SB |
3786 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3787 | root_item.received_uuid, | |
3788 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
3789 | key.objectid); | |
3790 | if (ret < 0) { | |
efe120a0 | 3791 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3792 | ret); |
803b2f54 SB |
3793 | break; |
3794 | } | |
3795 | } | |
3796 | ||
f45388f3 | 3797 | skip: |
803b2f54 SB |
3798 | if (trans) { |
3799 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 3800 | trans = NULL; |
803b2f54 SB |
3801 | if (ret) |
3802 | break; | |
3803 | } | |
3804 | ||
803b2f54 SB |
3805 | btrfs_release_path(path); |
3806 | if (key.offset < (u64)-1) { | |
3807 | key.offset++; | |
3808 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
3809 | key.offset = 0; | |
3810 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3811 | } else if (key.objectid < (u64)-1) { | |
3812 | key.offset = 0; | |
3813 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3814 | key.objectid++; | |
3815 | } else { | |
3816 | break; | |
3817 | } | |
3818 | cond_resched(); | |
3819 | } | |
3820 | ||
3821 | out: | |
3822 | btrfs_free_path(path); | |
f45388f3 FDBM |
3823 | if (trans && !IS_ERR(trans)) |
3824 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 3825 | if (ret) |
efe120a0 | 3826 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 SB |
3827 | else |
3828 | fs_info->update_uuid_tree_gen = 1; | |
803b2f54 SB |
3829 | up(&fs_info->uuid_tree_rescan_sem); |
3830 | return 0; | |
3831 | } | |
3832 | ||
70f80175 SB |
3833 | /* |
3834 | * Callback for btrfs_uuid_tree_iterate(). | |
3835 | * returns: | |
3836 | * 0 check succeeded, the entry is not outdated. | |
3837 | * < 0 if an error occured. | |
3838 | * > 0 if the check failed, which means the caller shall remove the entry. | |
3839 | */ | |
3840 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
3841 | u8 *uuid, u8 type, u64 subid) | |
3842 | { | |
3843 | struct btrfs_key key; | |
3844 | int ret = 0; | |
3845 | struct btrfs_root *subvol_root; | |
3846 | ||
3847 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
3848 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
3849 | goto out; | |
3850 | ||
3851 | key.objectid = subid; | |
3852 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3853 | key.offset = (u64)-1; | |
3854 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
3855 | if (IS_ERR(subvol_root)) { | |
3856 | ret = PTR_ERR(subvol_root); | |
3857 | if (ret == -ENOENT) | |
3858 | ret = 1; | |
3859 | goto out; | |
3860 | } | |
3861 | ||
3862 | switch (type) { | |
3863 | case BTRFS_UUID_KEY_SUBVOL: | |
3864 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
3865 | ret = 1; | |
3866 | break; | |
3867 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
3868 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
3869 | BTRFS_UUID_SIZE)) | |
3870 | ret = 1; | |
3871 | break; | |
3872 | } | |
3873 | ||
3874 | out: | |
3875 | return ret; | |
3876 | } | |
3877 | ||
3878 | static int btrfs_uuid_rescan_kthread(void *data) | |
3879 | { | |
3880 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
3881 | int ret; | |
3882 | ||
3883 | /* | |
3884 | * 1st step is to iterate through the existing UUID tree and | |
3885 | * to delete all entries that contain outdated data. | |
3886 | * 2nd step is to add all missing entries to the UUID tree. | |
3887 | */ | |
3888 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
3889 | if (ret < 0) { | |
efe120a0 | 3890 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
3891 | up(&fs_info->uuid_tree_rescan_sem); |
3892 | return ret; | |
3893 | } | |
3894 | return btrfs_uuid_scan_kthread(data); | |
3895 | } | |
3896 | ||
f7a81ea4 SB |
3897 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
3898 | { | |
3899 | struct btrfs_trans_handle *trans; | |
3900 | struct btrfs_root *tree_root = fs_info->tree_root; | |
3901 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
3902 | struct task_struct *task; |
3903 | int ret; | |
f7a81ea4 SB |
3904 | |
3905 | /* | |
3906 | * 1 - root node | |
3907 | * 1 - root item | |
3908 | */ | |
3909 | trans = btrfs_start_transaction(tree_root, 2); | |
3910 | if (IS_ERR(trans)) | |
3911 | return PTR_ERR(trans); | |
3912 | ||
3913 | uuid_root = btrfs_create_tree(trans, fs_info, | |
3914 | BTRFS_UUID_TREE_OBJECTID); | |
3915 | if (IS_ERR(uuid_root)) { | |
3916 | btrfs_abort_transaction(trans, tree_root, | |
3917 | PTR_ERR(uuid_root)); | |
3918 | return PTR_ERR(uuid_root); | |
3919 | } | |
3920 | ||
3921 | fs_info->uuid_root = uuid_root; | |
3922 | ||
803b2f54 SB |
3923 | ret = btrfs_commit_transaction(trans, tree_root); |
3924 | if (ret) | |
3925 | return ret; | |
3926 | ||
3927 | down(&fs_info->uuid_tree_rescan_sem); | |
3928 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
3929 | if (IS_ERR(task)) { | |
70f80175 | 3930 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 3931 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
3932 | up(&fs_info->uuid_tree_rescan_sem); |
3933 | return PTR_ERR(task); | |
3934 | } | |
3935 | ||
3936 | return 0; | |
f7a81ea4 | 3937 | } |
803b2f54 | 3938 | |
70f80175 SB |
3939 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
3940 | { | |
3941 | struct task_struct *task; | |
3942 | ||
3943 | down(&fs_info->uuid_tree_rescan_sem); | |
3944 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
3945 | if (IS_ERR(task)) { | |
3946 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 3947 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
3948 | up(&fs_info->uuid_tree_rescan_sem); |
3949 | return PTR_ERR(task); | |
3950 | } | |
3951 | ||
3952 | return 0; | |
3953 | } | |
3954 | ||
8f18cf13 CM |
3955 | /* |
3956 | * shrinking a device means finding all of the device extents past | |
3957 | * the new size, and then following the back refs to the chunks. | |
3958 | * The chunk relocation code actually frees the device extent | |
3959 | */ | |
3960 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
3961 | { | |
3962 | struct btrfs_trans_handle *trans; | |
3963 | struct btrfs_root *root = device->dev_root; | |
3964 | struct btrfs_dev_extent *dev_extent = NULL; | |
3965 | struct btrfs_path *path; | |
3966 | u64 length; | |
8f18cf13 CM |
3967 | u64 chunk_objectid; |
3968 | u64 chunk_offset; | |
3969 | int ret; | |
3970 | int slot; | |
ba1bf481 JB |
3971 | int failed = 0; |
3972 | bool retried = false; | |
8f18cf13 CM |
3973 | struct extent_buffer *l; |
3974 | struct btrfs_key key; | |
6c41761f | 3975 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 3976 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
3977 | u64 old_size = btrfs_device_get_total_bytes(device); |
3978 | u64 diff = old_size - new_size; | |
8f18cf13 | 3979 | |
63a212ab SB |
3980 | if (device->is_tgtdev_for_dev_replace) |
3981 | return -EINVAL; | |
3982 | ||
8f18cf13 CM |
3983 | path = btrfs_alloc_path(); |
3984 | if (!path) | |
3985 | return -ENOMEM; | |
3986 | ||
8f18cf13 CM |
3987 | path->reada = 2; |
3988 | ||
7d9eb12c CM |
3989 | lock_chunks(root); |
3990 | ||
7cc8e58d | 3991 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 3992 | if (device->writeable) { |
2b82032c | 3993 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
3994 | spin_lock(&root->fs_info->free_chunk_lock); |
3995 | root->fs_info->free_chunk_space -= diff; | |
3996 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3997 | } | |
7d9eb12c | 3998 | unlock_chunks(root); |
8f18cf13 | 3999 | |
ba1bf481 | 4000 | again: |
8f18cf13 CM |
4001 | key.objectid = device->devid; |
4002 | key.offset = (u64)-1; | |
4003 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4004 | ||
213e64da | 4005 | do { |
8f18cf13 CM |
4006 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4007 | if (ret < 0) | |
4008 | goto done; | |
4009 | ||
4010 | ret = btrfs_previous_item(root, path, 0, key.type); | |
4011 | if (ret < 0) | |
4012 | goto done; | |
4013 | if (ret) { | |
4014 | ret = 0; | |
b3b4aa74 | 4015 | btrfs_release_path(path); |
bf1fb512 | 4016 | break; |
8f18cf13 CM |
4017 | } |
4018 | ||
4019 | l = path->nodes[0]; | |
4020 | slot = path->slots[0]; | |
4021 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4022 | ||
ba1bf481 | 4023 | if (key.objectid != device->devid) { |
b3b4aa74 | 4024 | btrfs_release_path(path); |
bf1fb512 | 4025 | break; |
ba1bf481 | 4026 | } |
8f18cf13 CM |
4027 | |
4028 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4029 | length = btrfs_dev_extent_length(l, dev_extent); | |
4030 | ||
ba1bf481 | 4031 | if (key.offset + length <= new_size) { |
b3b4aa74 | 4032 | btrfs_release_path(path); |
d6397bae | 4033 | break; |
ba1bf481 | 4034 | } |
8f18cf13 | 4035 | |
8f18cf13 CM |
4036 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); |
4037 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 4038 | btrfs_release_path(path); |
8f18cf13 | 4039 | |
a688a04a | 4040 | ret = btrfs_relocate_chunk(root, chunk_objectid, chunk_offset); |
ba1bf481 | 4041 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4042 | goto done; |
ba1bf481 JB |
4043 | if (ret == -ENOSPC) |
4044 | failed++; | |
213e64da | 4045 | } while (key.offset-- > 0); |
ba1bf481 JB |
4046 | |
4047 | if (failed && !retried) { | |
4048 | failed = 0; | |
4049 | retried = true; | |
4050 | goto again; | |
4051 | } else if (failed && retried) { | |
4052 | ret = -ENOSPC; | |
4053 | lock_chunks(root); | |
4054 | ||
7cc8e58d | 4055 | btrfs_device_set_total_bytes(device, old_size); |
ba1bf481 JB |
4056 | if (device->writeable) |
4057 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
4058 | spin_lock(&root->fs_info->free_chunk_lock); |
4059 | root->fs_info->free_chunk_space += diff; | |
4060 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
4061 | unlock_chunks(root); |
4062 | goto done; | |
8f18cf13 CM |
4063 | } |
4064 | ||
d6397bae | 4065 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4066 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4067 | if (IS_ERR(trans)) { |
4068 | ret = PTR_ERR(trans); | |
4069 | goto done; | |
4070 | } | |
4071 | ||
d6397bae | 4072 | lock_chunks(root); |
7cc8e58d | 4073 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4074 | if (list_empty(&device->resized_list)) |
4075 | list_add_tail(&device->resized_list, | |
4076 | &root->fs_info->fs_devices->resized_devices); | |
d6397bae | 4077 | |
d6397bae CB |
4078 | WARN_ON(diff > old_total); |
4079 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4080 | unlock_chunks(root); | |
2196d6e8 MX |
4081 | |
4082 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4083 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4084 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4085 | done: |
4086 | btrfs_free_path(path); | |
4087 | return ret; | |
4088 | } | |
4089 | ||
125ccb0a | 4090 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4091 | struct btrfs_key *key, |
4092 | struct btrfs_chunk *chunk, int item_size) | |
4093 | { | |
6c41761f | 4094 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
4095 | struct btrfs_disk_key disk_key; |
4096 | u32 array_size; | |
4097 | u8 *ptr; | |
4098 | ||
fe48a5c0 | 4099 | lock_chunks(root); |
0b86a832 | 4100 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4101 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 MX |
4102 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
4103 | unlock_chunks(root); | |
0b86a832 | 4104 | return -EFBIG; |
fe48a5c0 | 4105 | } |
0b86a832 CM |
4106 | |
4107 | ptr = super_copy->sys_chunk_array + array_size; | |
4108 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4109 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4110 | ptr += sizeof(disk_key); | |
4111 | memcpy(ptr, chunk, item_size); | |
4112 | item_size += sizeof(disk_key); | |
4113 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 MX |
4114 | unlock_chunks(root); |
4115 | ||
0b86a832 CM |
4116 | return 0; |
4117 | } | |
4118 | ||
73c5de00 AJ |
4119 | /* |
4120 | * sort the devices in descending order by max_avail, total_avail | |
4121 | */ | |
4122 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4123 | { |
73c5de00 AJ |
4124 | const struct btrfs_device_info *di_a = a; |
4125 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4126 | |
73c5de00 | 4127 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4128 | return -1; |
73c5de00 | 4129 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4130 | return 1; |
73c5de00 AJ |
4131 | if (di_a->total_avail > di_b->total_avail) |
4132 | return -1; | |
4133 | if (di_a->total_avail < di_b->total_avail) | |
4134 | return 1; | |
4135 | return 0; | |
b2117a39 | 4136 | } |
0b86a832 | 4137 | |
e8c9f186 | 4138 | static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
e6ec716f MX |
4139 | [BTRFS_RAID_RAID10] = { |
4140 | .sub_stripes = 2, | |
4141 | .dev_stripes = 1, | |
4142 | .devs_max = 0, /* 0 == as many as possible */ | |
4143 | .devs_min = 4, | |
4144 | .devs_increment = 2, | |
4145 | .ncopies = 2, | |
4146 | }, | |
4147 | [BTRFS_RAID_RAID1] = { | |
4148 | .sub_stripes = 1, | |
4149 | .dev_stripes = 1, | |
4150 | .devs_max = 2, | |
4151 | .devs_min = 2, | |
4152 | .devs_increment = 2, | |
4153 | .ncopies = 2, | |
4154 | }, | |
4155 | [BTRFS_RAID_DUP] = { | |
4156 | .sub_stripes = 1, | |
4157 | .dev_stripes = 2, | |
4158 | .devs_max = 1, | |
4159 | .devs_min = 1, | |
4160 | .devs_increment = 1, | |
4161 | .ncopies = 2, | |
4162 | }, | |
4163 | [BTRFS_RAID_RAID0] = { | |
4164 | .sub_stripes = 1, | |
4165 | .dev_stripes = 1, | |
4166 | .devs_max = 0, | |
4167 | .devs_min = 2, | |
4168 | .devs_increment = 1, | |
4169 | .ncopies = 1, | |
4170 | }, | |
4171 | [BTRFS_RAID_SINGLE] = { | |
4172 | .sub_stripes = 1, | |
4173 | .dev_stripes = 1, | |
4174 | .devs_max = 1, | |
4175 | .devs_min = 1, | |
4176 | .devs_increment = 1, | |
4177 | .ncopies = 1, | |
4178 | }, | |
e942f883 CM |
4179 | [BTRFS_RAID_RAID5] = { |
4180 | .sub_stripes = 1, | |
4181 | .dev_stripes = 1, | |
4182 | .devs_max = 0, | |
4183 | .devs_min = 2, | |
4184 | .devs_increment = 1, | |
4185 | .ncopies = 2, | |
4186 | }, | |
4187 | [BTRFS_RAID_RAID6] = { | |
4188 | .sub_stripes = 1, | |
4189 | .dev_stripes = 1, | |
4190 | .devs_max = 0, | |
4191 | .devs_min = 3, | |
4192 | .devs_increment = 1, | |
4193 | .ncopies = 3, | |
4194 | }, | |
31e50229 LB |
4195 | }; |
4196 | ||
53b381b3 DW |
4197 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4198 | { | |
4199 | /* TODO allow them to set a preferred stripe size */ | |
4200 | return 64 * 1024; | |
4201 | } | |
4202 | ||
4203 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4204 | { | |
ffe2d203 | 4205 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4206 | return; |
4207 | ||
ceda0864 | 4208 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4209 | } |
4210 | ||
23f8f9b7 GH |
4211 | #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \ |
4212 | - sizeof(struct btrfs_item) \ | |
4213 | - sizeof(struct btrfs_chunk)) \ | |
4214 | / sizeof(struct btrfs_stripe) + 1) | |
4215 | ||
4216 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4217 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4218 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4219 | / sizeof(struct btrfs_stripe) + 1) | |
4220 | ||
73c5de00 | 4221 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4222 | struct btrfs_root *extent_root, u64 start, |
4223 | u64 type) | |
b2117a39 | 4224 | { |
73c5de00 AJ |
4225 | struct btrfs_fs_info *info = extent_root->fs_info; |
4226 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4227 | struct list_head *cur; | |
4228 | struct map_lookup *map = NULL; | |
4229 | struct extent_map_tree *em_tree; | |
4230 | struct extent_map *em; | |
4231 | struct btrfs_device_info *devices_info = NULL; | |
4232 | u64 total_avail; | |
4233 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4234 | int data_stripes; /* number of stripes that count for |
4235 | block group size */ | |
73c5de00 AJ |
4236 | int sub_stripes; /* sub_stripes info for map */ |
4237 | int dev_stripes; /* stripes per dev */ | |
4238 | int devs_max; /* max devs to use */ | |
4239 | int devs_min; /* min devs needed */ | |
4240 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4241 | int ncopies; /* how many copies to data has */ | |
4242 | int ret; | |
4243 | u64 max_stripe_size; | |
4244 | u64 max_chunk_size; | |
4245 | u64 stripe_size; | |
4246 | u64 num_bytes; | |
53b381b3 | 4247 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4248 | int ndevs; |
4249 | int i; | |
4250 | int j; | |
31e50229 | 4251 | int index; |
593060d7 | 4252 | |
0c460c0d | 4253 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4254 | |
73c5de00 AJ |
4255 | if (list_empty(&fs_devices->alloc_list)) |
4256 | return -ENOSPC; | |
b2117a39 | 4257 | |
31e50229 | 4258 | index = __get_raid_index(type); |
73c5de00 | 4259 | |
31e50229 LB |
4260 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4261 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4262 | devs_max = btrfs_raid_array[index].devs_max; | |
4263 | devs_min = btrfs_raid_array[index].devs_min; | |
4264 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4265 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4266 | |
9b3f68b9 | 4267 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
4268 | max_stripe_size = 1024 * 1024 * 1024; |
4269 | max_chunk_size = 10 * max_stripe_size; | |
23f8f9b7 GH |
4270 | if (!devs_max) |
4271 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4272 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
4273 | /* for larger filesystems, use larger metadata chunks */ |
4274 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
4275 | max_stripe_size = 1024 * 1024 * 1024; | |
4276 | else | |
4277 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 4278 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4279 | if (!devs_max) |
4280 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4281 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 4282 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 | 4283 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4284 | if (!devs_max) |
4285 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4286 | } else { |
351fd353 | 4287 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4288 | type); |
4289 | BUG_ON(1); | |
9b3f68b9 CM |
4290 | } |
4291 | ||
2b82032c YZ |
4292 | /* we don't want a chunk larger than 10% of writeable space */ |
4293 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4294 | max_chunk_size); | |
9b3f68b9 | 4295 | |
31e818fe | 4296 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4297 | GFP_NOFS); |
4298 | if (!devices_info) | |
4299 | return -ENOMEM; | |
0cad8a11 | 4300 | |
73c5de00 | 4301 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4302 | |
9f680ce0 | 4303 | /* |
73c5de00 AJ |
4304 | * in the first pass through the devices list, we gather information |
4305 | * about the available holes on each device. | |
9f680ce0 | 4306 | */ |
73c5de00 AJ |
4307 | ndevs = 0; |
4308 | while (cur != &fs_devices->alloc_list) { | |
4309 | struct btrfs_device *device; | |
4310 | u64 max_avail; | |
4311 | u64 dev_offset; | |
b2117a39 | 4312 | |
73c5de00 | 4313 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4314 | |
73c5de00 | 4315 | cur = cur->next; |
b2117a39 | 4316 | |
73c5de00 | 4317 | if (!device->writeable) { |
31b1a2bd | 4318 | WARN(1, KERN_ERR |
efe120a0 | 4319 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4320 | continue; |
4321 | } | |
b2117a39 | 4322 | |
63a212ab SB |
4323 | if (!device->in_fs_metadata || |
4324 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4325 | continue; |
b2117a39 | 4326 | |
73c5de00 AJ |
4327 | if (device->total_bytes > device->bytes_used) |
4328 | total_avail = device->total_bytes - device->bytes_used; | |
4329 | else | |
4330 | total_avail = 0; | |
38c01b96 | 4331 | |
4332 | /* If there is no space on this device, skip it. */ | |
4333 | if (total_avail == 0) | |
4334 | continue; | |
b2117a39 | 4335 | |
6df9a95e | 4336 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4337 | max_stripe_size * dev_stripes, |
4338 | &dev_offset, &max_avail); | |
4339 | if (ret && ret != -ENOSPC) | |
4340 | goto error; | |
b2117a39 | 4341 | |
73c5de00 AJ |
4342 | if (ret == 0) |
4343 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4344 | |
73c5de00 AJ |
4345 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4346 | continue; | |
b2117a39 | 4347 | |
063d006f ES |
4348 | if (ndevs == fs_devices->rw_devices) { |
4349 | WARN(1, "%s: found more than %llu devices\n", | |
4350 | __func__, fs_devices->rw_devices); | |
4351 | break; | |
4352 | } | |
73c5de00 AJ |
4353 | devices_info[ndevs].dev_offset = dev_offset; |
4354 | devices_info[ndevs].max_avail = max_avail; | |
4355 | devices_info[ndevs].total_avail = total_avail; | |
4356 | devices_info[ndevs].dev = device; | |
4357 | ++ndevs; | |
4358 | } | |
b2117a39 | 4359 | |
73c5de00 AJ |
4360 | /* |
4361 | * now sort the devices by hole size / available space | |
4362 | */ | |
4363 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4364 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4365 | |
73c5de00 AJ |
4366 | /* round down to number of usable stripes */ |
4367 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4368 | |
73c5de00 AJ |
4369 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4370 | ret = -ENOSPC; | |
4371 | goto error; | |
b2117a39 | 4372 | } |
9f680ce0 | 4373 | |
73c5de00 AJ |
4374 | if (devs_max && ndevs > devs_max) |
4375 | ndevs = devs_max; | |
4376 | /* | |
4377 | * the primary goal is to maximize the number of stripes, so use as many | |
4378 | * devices as possible, even if the stripes are not maximum sized. | |
4379 | */ | |
4380 | stripe_size = devices_info[ndevs-1].max_avail; | |
4381 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4382 | |
53b381b3 DW |
4383 | /* |
4384 | * this will have to be fixed for RAID1 and RAID10 over | |
4385 | * more drives | |
4386 | */ | |
4387 | data_stripes = num_stripes / ncopies; | |
4388 | ||
53b381b3 DW |
4389 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4390 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4391 | btrfs_super_stripesize(info->super_copy)); | |
4392 | data_stripes = num_stripes - 1; | |
4393 | } | |
4394 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4395 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4396 | btrfs_super_stripesize(info->super_copy)); | |
4397 | data_stripes = num_stripes - 2; | |
4398 | } | |
86db2578 CM |
4399 | |
4400 | /* | |
4401 | * Use the number of data stripes to figure out how big this chunk | |
4402 | * is really going to be in terms of logical address space, | |
4403 | * and compare that answer with the max chunk size | |
4404 | */ | |
4405 | if (stripe_size * data_stripes > max_chunk_size) { | |
4406 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4407 | |
4408 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4409 | |
4410 | /* bump the answer up to a 16MB boundary */ | |
4411 | stripe_size = (stripe_size + mask) & ~mask; | |
4412 | ||
4413 | /* but don't go higher than the limits we found | |
4414 | * while searching for free extents | |
4415 | */ | |
4416 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4417 | stripe_size = devices_info[ndevs-1].max_avail; | |
4418 | } | |
4419 | ||
b8b93add | 4420 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4421 | |
4422 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4423 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4424 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4425 | |
4426 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4427 | if (!map) { | |
4428 | ret = -ENOMEM; | |
4429 | goto error; | |
4430 | } | |
4431 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4432 | |
73c5de00 AJ |
4433 | for (i = 0; i < ndevs; ++i) { |
4434 | for (j = 0; j < dev_stripes; ++j) { | |
4435 | int s = i * dev_stripes + j; | |
4436 | map->stripes[s].dev = devices_info[i].dev; | |
4437 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4438 | j * stripe_size; | |
6324fbf3 | 4439 | } |
6324fbf3 | 4440 | } |
2b82032c | 4441 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4442 | map->stripe_len = raid_stripe_len; |
4443 | map->io_align = raid_stripe_len; | |
4444 | map->io_width = raid_stripe_len; | |
2b82032c | 4445 | map->type = type; |
2b82032c | 4446 | map->sub_stripes = sub_stripes; |
0b86a832 | 4447 | |
53b381b3 | 4448 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4449 | |
73c5de00 | 4450 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4451 | |
172ddd60 | 4452 | em = alloc_extent_map(); |
2b82032c | 4453 | if (!em) { |
298a8f9c | 4454 | kfree(map); |
b2117a39 MX |
4455 | ret = -ENOMEM; |
4456 | goto error; | |
593060d7 | 4457 | } |
298a8f9c | 4458 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
2b82032c YZ |
4459 | em->bdev = (struct block_device *)map; |
4460 | em->start = start; | |
73c5de00 | 4461 | em->len = num_bytes; |
2b82032c YZ |
4462 | em->block_start = 0; |
4463 | em->block_len = em->len; | |
6df9a95e | 4464 | em->orig_block_len = stripe_size; |
593060d7 | 4465 | |
2b82032c | 4466 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4467 | write_lock(&em_tree->lock); |
09a2a8f9 | 4468 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4469 | if (!ret) { |
4470 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4471 | atomic_inc(&em->refs); | |
4472 | } | |
890871be | 4473 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4474 | if (ret) { |
4475 | free_extent_map(em); | |
1dd4602f | 4476 | goto error; |
0f5d42b2 | 4477 | } |
0b86a832 | 4478 | |
04487488 JB |
4479 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4480 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4481 | start, num_bytes); | |
6df9a95e JB |
4482 | if (ret) |
4483 | goto error_del_extent; | |
2b82032c | 4484 | |
7cc8e58d MX |
4485 | for (i = 0; i < map->num_stripes; i++) { |
4486 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4487 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4488 | } | |
43530c46 | 4489 | |
1c116187 MX |
4490 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4491 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4492 | map->num_stripes); | |
4493 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4494 | ||
0f5d42b2 | 4495 | free_extent_map(em); |
53b381b3 DW |
4496 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4497 | ||
b2117a39 | 4498 | kfree(devices_info); |
2b82032c | 4499 | return 0; |
b2117a39 | 4500 | |
6df9a95e | 4501 | error_del_extent: |
0f5d42b2 JB |
4502 | write_lock(&em_tree->lock); |
4503 | remove_extent_mapping(em_tree, em); | |
4504 | write_unlock(&em_tree->lock); | |
4505 | ||
4506 | /* One for our allocation */ | |
4507 | free_extent_map(em); | |
4508 | /* One for the tree reference */ | |
4509 | free_extent_map(em); | |
495e64f4 FM |
4510 | /* One for the pending_chunks list reference */ |
4511 | free_extent_map(em); | |
b2117a39 | 4512 | error: |
b2117a39 MX |
4513 | kfree(devices_info); |
4514 | return ret; | |
2b82032c YZ |
4515 | } |
4516 | ||
6df9a95e | 4517 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4518 | struct btrfs_root *extent_root, |
6df9a95e | 4519 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4520 | { |
2b82032c YZ |
4521 | struct btrfs_key key; |
4522 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4523 | struct btrfs_device *device; | |
4524 | struct btrfs_chunk *chunk; | |
4525 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4526 | struct extent_map_tree *em_tree; |
4527 | struct extent_map *em; | |
4528 | struct map_lookup *map; | |
4529 | size_t item_size; | |
4530 | u64 dev_offset; | |
4531 | u64 stripe_size; | |
4532 | int i = 0; | |
2b82032c YZ |
4533 | int ret; |
4534 | ||
6df9a95e JB |
4535 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4536 | read_lock(&em_tree->lock); | |
4537 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4538 | read_unlock(&em_tree->lock); | |
4539 | ||
4540 | if (!em) { | |
4541 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4542 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4543 | return -EINVAL; | |
4544 | } | |
4545 | ||
4546 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4547 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
351fd353 | 4548 | " %Lu-%Lu, found %Lu-%Lu", chunk_offset, |
6df9a95e JB |
4549 | chunk_size, em->start, em->len); |
4550 | free_extent_map(em); | |
4551 | return -EINVAL; | |
4552 | } | |
4553 | ||
4554 | map = (struct map_lookup *)em->bdev; | |
4555 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
4556 | stripe_size = em->orig_block_len; | |
4557 | ||
2b82032c | 4558 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4559 | if (!chunk) { |
4560 | ret = -ENOMEM; | |
4561 | goto out; | |
4562 | } | |
4563 | ||
4564 | for (i = 0; i < map->num_stripes; i++) { | |
4565 | device = map->stripes[i].dev; | |
4566 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4567 | |
0b86a832 | 4568 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4569 | if (ret) |
6df9a95e JB |
4570 | goto out; |
4571 | ret = btrfs_alloc_dev_extent(trans, device, | |
4572 | chunk_root->root_key.objectid, | |
4573 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4574 | chunk_offset, dev_offset, | |
4575 | stripe_size); | |
4576 | if (ret) | |
4577 | goto out; | |
2b82032c YZ |
4578 | } |
4579 | ||
2b82032c | 4580 | stripe = &chunk->stripe; |
6df9a95e JB |
4581 | for (i = 0; i < map->num_stripes; i++) { |
4582 | device = map->stripes[i].dev; | |
4583 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4584 | |
e17cade2 CM |
4585 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4586 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4587 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4588 | stripe++; |
0b86a832 CM |
4589 | } |
4590 | ||
2b82032c | 4591 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4592 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4593 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4594 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4595 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4596 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4597 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4598 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4599 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4600 | |
2b82032c YZ |
4601 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4602 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4603 | key.offset = chunk_offset; | |
0b86a832 | 4604 | |
2b82032c | 4605 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4606 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4607 | /* | |
4608 | * TODO: Cleanup of inserted chunk root in case of | |
4609 | * failure. | |
4610 | */ | |
125ccb0a | 4611 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4612 | item_size); |
8f18cf13 | 4613 | } |
1abe9b8a | 4614 | |
6df9a95e | 4615 | out: |
0b86a832 | 4616 | kfree(chunk); |
6df9a95e | 4617 | free_extent_map(em); |
4ed1d16e | 4618 | return ret; |
2b82032c | 4619 | } |
0b86a832 | 4620 | |
2b82032c YZ |
4621 | /* |
4622 | * Chunk allocation falls into two parts. The first part does works | |
4623 | * that make the new allocated chunk useable, but not do any operation | |
4624 | * that modifies the chunk tree. The second part does the works that | |
4625 | * require modifying the chunk tree. This division is important for the | |
4626 | * bootstrap process of adding storage to a seed btrfs. | |
4627 | */ | |
4628 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4629 | struct btrfs_root *extent_root, u64 type) | |
4630 | { | |
4631 | u64 chunk_offset; | |
2b82032c | 4632 | |
a9629596 | 4633 | ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex)); |
6df9a95e JB |
4634 | chunk_offset = find_next_chunk(extent_root->fs_info); |
4635 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
4636 | } |
4637 | ||
d397712b | 4638 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4639 | struct btrfs_root *root, |
4640 | struct btrfs_device *device) | |
4641 | { | |
4642 | u64 chunk_offset; | |
4643 | u64 sys_chunk_offset; | |
2b82032c | 4644 | u64 alloc_profile; |
2b82032c YZ |
4645 | struct btrfs_fs_info *fs_info = root->fs_info; |
4646 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4647 | int ret; | |
4648 | ||
6df9a95e | 4649 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 4650 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
4651 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
4652 | alloc_profile); | |
79787eaa JM |
4653 | if (ret) |
4654 | return ret; | |
2b82032c | 4655 | |
6df9a95e | 4656 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 4657 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
4658 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
4659 | alloc_profile); | |
79787eaa | 4660 | return ret; |
2b82032c YZ |
4661 | } |
4662 | ||
d20983b4 MX |
4663 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
4664 | { | |
4665 | int max_errors; | |
4666 | ||
4667 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4668 | BTRFS_BLOCK_GROUP_RAID10 | | |
4669 | BTRFS_BLOCK_GROUP_RAID5 | | |
4670 | BTRFS_BLOCK_GROUP_DUP)) { | |
4671 | max_errors = 1; | |
4672 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4673 | max_errors = 2; | |
4674 | } else { | |
4675 | max_errors = 0; | |
005d6427 | 4676 | } |
2b82032c | 4677 | |
d20983b4 | 4678 | return max_errors; |
2b82032c YZ |
4679 | } |
4680 | ||
4681 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4682 | { | |
4683 | struct extent_map *em; | |
4684 | struct map_lookup *map; | |
4685 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4686 | int readonly = 0; | |
d20983b4 | 4687 | int miss_ndevs = 0; |
2b82032c YZ |
4688 | int i; |
4689 | ||
890871be | 4690 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4691 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4692 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4693 | if (!em) |
4694 | return 1; | |
4695 | ||
4696 | map = (struct map_lookup *)em->bdev; | |
4697 | for (i = 0; i < map->num_stripes; i++) { | |
d20983b4 MX |
4698 | if (map->stripes[i].dev->missing) { |
4699 | miss_ndevs++; | |
4700 | continue; | |
4701 | } | |
4702 | ||
2b82032c YZ |
4703 | if (!map->stripes[i].dev->writeable) { |
4704 | readonly = 1; | |
d20983b4 | 4705 | goto end; |
2b82032c YZ |
4706 | } |
4707 | } | |
d20983b4 MX |
4708 | |
4709 | /* | |
4710 | * If the number of missing devices is larger than max errors, | |
4711 | * we can not write the data into that chunk successfully, so | |
4712 | * set it readonly. | |
4713 | */ | |
4714 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
4715 | readonly = 1; | |
4716 | end: | |
0b86a832 | 4717 | free_extent_map(em); |
2b82032c | 4718 | return readonly; |
0b86a832 CM |
4719 | } |
4720 | ||
4721 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4722 | { | |
a8067e02 | 4723 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
4724 | } |
4725 | ||
4726 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4727 | { | |
4728 | struct extent_map *em; | |
4729 | ||
d397712b | 4730 | while (1) { |
890871be | 4731 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
4732 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
4733 | if (em) | |
4734 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 4735 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
4736 | if (!em) |
4737 | break; | |
0b86a832 CM |
4738 | /* once for us */ |
4739 | free_extent_map(em); | |
4740 | /* once for the tree */ | |
4741 | free_extent_map(em); | |
4742 | } | |
4743 | } | |
4744 | ||
5d964051 | 4745 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 4746 | { |
5d964051 | 4747 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
4748 | struct extent_map *em; |
4749 | struct map_lookup *map; | |
4750 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4751 | int ret; | |
4752 | ||
890871be | 4753 | read_lock(&em_tree->lock); |
f188591e | 4754 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 4755 | read_unlock(&em_tree->lock); |
f188591e | 4756 | |
fb7669b5 JB |
4757 | /* |
4758 | * We could return errors for these cases, but that could get ugly and | |
4759 | * we'd probably do the same thing which is just not do anything else | |
4760 | * and exit, so return 1 so the callers don't try to use other copies. | |
4761 | */ | |
4762 | if (!em) { | |
351fd353 | 4763 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
4764 | logical+len); |
4765 | return 1; | |
4766 | } | |
4767 | ||
4768 | if (em->start > logical || em->start + em->len < logical) { | |
ccf39f92 | 4769 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " |
351fd353 | 4770 | "%Lu-%Lu", logical, logical+len, em->start, |
fb7669b5 | 4771 | em->start + em->len); |
7d3d1744 | 4772 | free_extent_map(em); |
fb7669b5 JB |
4773 | return 1; |
4774 | } | |
4775 | ||
f188591e CM |
4776 | map = (struct map_lookup *)em->bdev; |
4777 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
4778 | ret = map->num_stripes; | |
321aecc6 CM |
4779 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
4780 | ret = map->sub_stripes; | |
53b381b3 DW |
4781 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
4782 | ret = 2; | |
4783 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4784 | ret = 3; | |
f188591e CM |
4785 | else |
4786 | ret = 1; | |
4787 | free_extent_map(em); | |
ad6d620e SB |
4788 | |
4789 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
4790 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
4791 | ret++; | |
4792 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
4793 | ||
f188591e CM |
4794 | return ret; |
4795 | } | |
4796 | ||
53b381b3 DW |
4797 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
4798 | struct btrfs_mapping_tree *map_tree, | |
4799 | u64 logical) | |
4800 | { | |
4801 | struct extent_map *em; | |
4802 | struct map_lookup *map; | |
4803 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4804 | unsigned long len = root->sectorsize; | |
4805 | ||
4806 | read_lock(&em_tree->lock); | |
4807 | em = lookup_extent_mapping(em_tree, logical, len); | |
4808 | read_unlock(&em_tree->lock); | |
4809 | BUG_ON(!em); | |
4810 | ||
4811 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4812 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 4813 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 4814 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
4815 | free_extent_map(em); |
4816 | return len; | |
4817 | } | |
4818 | ||
4819 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
4820 | u64 logical, u64 len, int mirror_num) | |
4821 | { | |
4822 | struct extent_map *em; | |
4823 | struct map_lookup *map; | |
4824 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4825 | int ret = 0; | |
4826 | ||
4827 | read_lock(&em_tree->lock); | |
4828 | em = lookup_extent_mapping(em_tree, logical, len); | |
4829 | read_unlock(&em_tree->lock); | |
4830 | BUG_ON(!em); | |
4831 | ||
4832 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4833 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 4834 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
4835 | ret = 1; |
4836 | free_extent_map(em); | |
4837 | return ret; | |
4838 | } | |
4839 | ||
30d9861f SB |
4840 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
4841 | struct map_lookup *map, int first, int num, | |
4842 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
4843 | { |
4844 | int i; | |
30d9861f SB |
4845 | int tolerance; |
4846 | struct btrfs_device *srcdev; | |
4847 | ||
4848 | if (dev_replace_is_ongoing && | |
4849 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
4850 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
4851 | srcdev = fs_info->dev_replace.srcdev; | |
4852 | else | |
4853 | srcdev = NULL; | |
4854 | ||
4855 | /* | |
4856 | * try to avoid the drive that is the source drive for a | |
4857 | * dev-replace procedure, only choose it if no other non-missing | |
4858 | * mirror is available | |
4859 | */ | |
4860 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
4861 | if (map->stripes[optimal].dev->bdev && | |
4862 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
4863 | return optimal; | |
4864 | for (i = first; i < first + num; i++) { | |
4865 | if (map->stripes[i].dev->bdev && | |
4866 | (tolerance || map->stripes[i].dev != srcdev)) | |
4867 | return i; | |
4868 | } | |
dfe25020 | 4869 | } |
30d9861f | 4870 | |
dfe25020 CM |
4871 | /* we couldn't find one that doesn't fail. Just return something |
4872 | * and the io error handling code will clean up eventually | |
4873 | */ | |
4874 | return optimal; | |
4875 | } | |
4876 | ||
53b381b3 DW |
4877 | static inline int parity_smaller(u64 a, u64 b) |
4878 | { | |
4879 | return a > b; | |
4880 | } | |
4881 | ||
4882 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 4883 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
4884 | { |
4885 | struct btrfs_bio_stripe s; | |
4886 | int i; | |
4887 | u64 l; | |
4888 | int again = 1; | |
4889 | ||
4890 | while (again) { | |
4891 | again = 0; | |
cc7539ed | 4892 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
4893 | if (parity_smaller(bbio->raid_map[i], |
4894 | bbio->raid_map[i+1])) { | |
53b381b3 | 4895 | s = bbio->stripes[i]; |
8e5cfb55 | 4896 | l = bbio->raid_map[i]; |
53b381b3 | 4897 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 4898 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 4899 | bbio->stripes[i+1] = s; |
8e5cfb55 | 4900 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 4901 | |
53b381b3 DW |
4902 | again = 1; |
4903 | } | |
4904 | } | |
4905 | } | |
4906 | } | |
4907 | ||
6e9606d2 ZL |
4908 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
4909 | { | |
4910 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 4911 | /* the size of the btrfs_bio */ |
6e9606d2 | 4912 | sizeof(struct btrfs_bio) + |
e57cf21e | 4913 | /* plus the variable array for the stripes */ |
6e9606d2 | 4914 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 4915 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 4916 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
4917 | /* |
4918 | * plus the raid_map, which includes both the tgt dev | |
4919 | * and the stripes | |
4920 | */ | |
4921 | sizeof(u64) * (total_stripes), | |
6e9606d2 ZL |
4922 | GFP_NOFS); |
4923 | if (!bbio) | |
4924 | return NULL; | |
4925 | ||
4926 | atomic_set(&bbio->error, 0); | |
4927 | atomic_set(&bbio->refs, 1); | |
4928 | ||
4929 | return bbio; | |
4930 | } | |
4931 | ||
4932 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
4933 | { | |
4934 | WARN_ON(!atomic_read(&bbio->refs)); | |
4935 | atomic_inc(&bbio->refs); | |
4936 | } | |
4937 | ||
4938 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
4939 | { | |
4940 | if (!bbio) | |
4941 | return; | |
4942 | if (atomic_dec_and_test(&bbio->refs)) | |
4943 | kfree(bbio); | |
4944 | } | |
4945 | ||
3ec706c8 | 4946 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 4947 | u64 logical, u64 *length, |
a1d3c478 | 4948 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 4949 | int mirror_num, int need_raid_map) |
0b86a832 CM |
4950 | { |
4951 | struct extent_map *em; | |
4952 | struct map_lookup *map; | |
3ec706c8 | 4953 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
4954 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
4955 | u64 offset; | |
593060d7 | 4956 | u64 stripe_offset; |
fce3bb9a | 4957 | u64 stripe_end_offset; |
593060d7 | 4958 | u64 stripe_nr; |
fce3bb9a LD |
4959 | u64 stripe_nr_orig; |
4960 | u64 stripe_nr_end; | |
53b381b3 | 4961 | u64 stripe_len; |
9d644a62 | 4962 | u32 stripe_index; |
cea9e445 | 4963 | int i; |
de11cc12 | 4964 | int ret = 0; |
f2d8d74d | 4965 | int num_stripes; |
a236aed1 | 4966 | int max_errors = 0; |
2c8cdd6e | 4967 | int tgtdev_indexes = 0; |
a1d3c478 | 4968 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
4969 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
4970 | int dev_replace_is_ongoing = 0; | |
4971 | int num_alloc_stripes; | |
ad6d620e SB |
4972 | int patch_the_first_stripe_for_dev_replace = 0; |
4973 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 4974 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 4975 | |
890871be | 4976 | read_lock(&em_tree->lock); |
0b86a832 | 4977 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 4978 | read_unlock(&em_tree->lock); |
f2d8d74d | 4979 | |
3b951516 | 4980 | if (!em) { |
c2cf52eb | 4981 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 4982 | logical, *length); |
9bb91873 JB |
4983 | return -EINVAL; |
4984 | } | |
4985 | ||
4986 | if (em->start > logical || em->start + em->len < logical) { | |
4987 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
351fd353 | 4988 | "found %Lu-%Lu", logical, em->start, |
9bb91873 | 4989 | em->start + em->len); |
7d3d1744 | 4990 | free_extent_map(em); |
9bb91873 | 4991 | return -EINVAL; |
3b951516 | 4992 | } |
0b86a832 | 4993 | |
0b86a832 CM |
4994 | map = (struct map_lookup *)em->bdev; |
4995 | offset = logical - em->start; | |
593060d7 | 4996 | |
53b381b3 | 4997 | stripe_len = map->stripe_len; |
593060d7 CM |
4998 | stripe_nr = offset; |
4999 | /* | |
5000 | * stripe_nr counts the total number of stripes we have to stride | |
5001 | * to get to this block | |
5002 | */ | |
47c5713f | 5003 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5004 | |
53b381b3 | 5005 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
5006 | BUG_ON(offset < stripe_offset); |
5007 | ||
5008 | /* stripe_offset is the offset of this block in its stripe*/ | |
5009 | stripe_offset = offset - stripe_offset; | |
5010 | ||
53b381b3 | 5011 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5012 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5013 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5014 | raid56_full_stripe_start = offset; | |
5015 | ||
5016 | /* allow a write of a full stripe, but make sure we don't | |
5017 | * allow straddling of stripes | |
5018 | */ | |
47c5713f DS |
5019 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5020 | full_stripe_len); | |
53b381b3 DW |
5021 | raid56_full_stripe_start *= full_stripe_len; |
5022 | } | |
5023 | ||
5024 | if (rw & REQ_DISCARD) { | |
5025 | /* we don't discard raid56 yet */ | |
ffe2d203 | 5026 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5027 | ret = -EOPNOTSUPP; |
5028 | goto out; | |
5029 | } | |
fce3bb9a | 5030 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
5031 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
5032 | u64 max_len; | |
5033 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5034 | For other RAID types and for RAID[56] reads, just allow a single | |
5035 | stripe (on a single disk). */ | |
ffe2d203 | 5036 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
53b381b3 DW |
5037 | (rw & REQ_WRITE)) { |
5038 | max_len = stripe_len * nr_data_stripes(map) - | |
5039 | (offset - raid56_full_stripe_start); | |
5040 | } else { | |
5041 | /* we limit the length of each bio to what fits in a stripe */ | |
5042 | max_len = stripe_len - stripe_offset; | |
5043 | } | |
5044 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5045 | } else { |
5046 | *length = em->len - offset; | |
5047 | } | |
f2d8d74d | 5048 | |
53b381b3 DW |
5049 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5050 | it cares about is the length */ | |
a1d3c478 | 5051 | if (!bbio_ret) |
cea9e445 CM |
5052 | goto out; |
5053 | ||
472262f3 SB |
5054 | btrfs_dev_replace_lock(dev_replace); |
5055 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
5056 | if (!dev_replace_is_ongoing) | |
5057 | btrfs_dev_replace_unlock(dev_replace); | |
5058 | ||
ad6d620e SB |
5059 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
5060 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
5061 | dev_replace->tgtdev != NULL) { | |
5062 | /* | |
5063 | * in dev-replace case, for repair case (that's the only | |
5064 | * case where the mirror is selected explicitly when | |
5065 | * calling btrfs_map_block), blocks left of the left cursor | |
5066 | * can also be read from the target drive. | |
5067 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5068 | * the last one to the array of stripes. For READ, it also | |
5069 | * needs to be supported using the same mirror number. | |
5070 | * If the requested block is not left of the left cursor, | |
5071 | * EIO is returned. This can happen because btrfs_num_copies() | |
5072 | * returns one more in the dev-replace case. | |
5073 | */ | |
5074 | u64 tmp_length = *length; | |
5075 | struct btrfs_bio *tmp_bbio = NULL; | |
5076 | int tmp_num_stripes; | |
5077 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5078 | int index_srcdev = 0; | |
5079 | int found = 0; | |
5080 | u64 physical_of_found = 0; | |
5081 | ||
5082 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
8e5cfb55 | 5083 | logical, &tmp_length, &tmp_bbio, 0, 0); |
ad6d620e SB |
5084 | if (ret) { |
5085 | WARN_ON(tmp_bbio != NULL); | |
5086 | goto out; | |
5087 | } | |
5088 | ||
5089 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5090 | if (mirror_num > tmp_num_stripes) { | |
5091 | /* | |
5092 | * REQ_GET_READ_MIRRORS does not contain this | |
5093 | * mirror, that means that the requested area | |
5094 | * is not left of the left cursor | |
5095 | */ | |
5096 | ret = -EIO; | |
6e9606d2 | 5097 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5098 | goto out; |
5099 | } | |
5100 | ||
5101 | /* | |
5102 | * process the rest of the function using the mirror_num | |
5103 | * of the source drive. Therefore look it up first. | |
5104 | * At the end, patch the device pointer to the one of the | |
5105 | * target drive. | |
5106 | */ | |
5107 | for (i = 0; i < tmp_num_stripes; i++) { | |
5108 | if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) { | |
5109 | /* | |
5110 | * In case of DUP, in order to keep it | |
5111 | * simple, only add the mirror with the | |
5112 | * lowest physical address | |
5113 | */ | |
5114 | if (found && | |
5115 | physical_of_found <= | |
5116 | tmp_bbio->stripes[i].physical) | |
5117 | continue; | |
5118 | index_srcdev = i; | |
5119 | found = 1; | |
5120 | physical_of_found = | |
5121 | tmp_bbio->stripes[i].physical; | |
5122 | } | |
5123 | } | |
5124 | ||
5125 | if (found) { | |
5126 | mirror_num = index_srcdev + 1; | |
5127 | patch_the_first_stripe_for_dev_replace = 1; | |
5128 | physical_to_patch_in_first_stripe = physical_of_found; | |
5129 | } else { | |
5130 | WARN_ON(1); | |
5131 | ret = -EIO; | |
6e9606d2 | 5132 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5133 | goto out; |
5134 | } | |
5135 | ||
6e9606d2 | 5136 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5137 | } else if (mirror_num > map->num_stripes) { |
5138 | mirror_num = 0; | |
5139 | } | |
5140 | ||
f2d8d74d | 5141 | num_stripes = 1; |
cea9e445 | 5142 | stripe_index = 0; |
fce3bb9a | 5143 | stripe_nr_orig = stripe_nr; |
fda2832f | 5144 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
b8b93add | 5145 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); |
fce3bb9a LD |
5146 | stripe_end_offset = stripe_nr_end * map->stripe_len - |
5147 | (offset + *length); | |
53b381b3 | 5148 | |
fce3bb9a LD |
5149 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5150 | if (rw & REQ_DISCARD) | |
5151 | num_stripes = min_t(u64, map->num_stripes, | |
5152 | stripe_nr_end - stripe_nr_orig); | |
47c5713f DS |
5153 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5154 | &stripe_index); | |
28e1cc7d MX |
5155 | if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))) |
5156 | mirror_num = 1; | |
fce3bb9a | 5157 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
29a8d9a0 | 5158 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5159 | num_stripes = map->num_stripes; |
2fff734f | 5160 | else if (mirror_num) |
f188591e | 5161 | stripe_index = mirror_num - 1; |
dfe25020 | 5162 | else { |
30d9861f | 5163 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5164 | map->num_stripes, |
30d9861f SB |
5165 | current->pid % map->num_stripes, |
5166 | dev_replace_is_ongoing); | |
a1d3c478 | 5167 | mirror_num = stripe_index + 1; |
dfe25020 | 5168 | } |
2fff734f | 5169 | |
611f0e00 | 5170 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 5171 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 5172 | num_stripes = map->num_stripes; |
a1d3c478 | 5173 | } else if (mirror_num) { |
f188591e | 5174 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5175 | } else { |
5176 | mirror_num = 1; | |
5177 | } | |
2fff734f | 5178 | |
321aecc6 | 5179 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5180 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5181 | |
47c5713f | 5182 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5183 | stripe_index *= map->sub_stripes; |
5184 | ||
29a8d9a0 | 5185 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5186 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
5187 | else if (rw & REQ_DISCARD) |
5188 | num_stripes = min_t(u64, map->sub_stripes * | |
5189 | (stripe_nr_end - stripe_nr_orig), | |
5190 | map->num_stripes); | |
321aecc6 CM |
5191 | else if (mirror_num) |
5192 | stripe_index += mirror_num - 1; | |
dfe25020 | 5193 | else { |
3e74317a | 5194 | int old_stripe_index = stripe_index; |
30d9861f SB |
5195 | stripe_index = find_live_mirror(fs_info, map, |
5196 | stripe_index, | |
dfe25020 | 5197 | map->sub_stripes, stripe_index + |
30d9861f SB |
5198 | current->pid % map->sub_stripes, |
5199 | dev_replace_is_ongoing); | |
3e74317a | 5200 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5201 | } |
53b381b3 | 5202 | |
ffe2d203 | 5203 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5204 | if (need_raid_map && |
af8e2d1d MX |
5205 | ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5206 | mirror_num > 1)) { | |
53b381b3 | 5207 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5208 | stripe_nr = div_u64(raid56_full_stripe_start, |
5209 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5210 | |
5211 | /* RAID[56] write or recovery. Return all stripes */ | |
5212 | num_stripes = map->num_stripes; | |
5213 | max_errors = nr_parity_stripes(map); | |
5214 | ||
53b381b3 DW |
5215 | *length = map->stripe_len; |
5216 | stripe_index = 0; | |
5217 | stripe_offset = 0; | |
5218 | } else { | |
5219 | /* | |
5220 | * Mirror #0 or #1 means the original data block. | |
5221 | * Mirror #2 is RAID5 parity block. | |
5222 | * Mirror #3 is RAID6 Q block. | |
5223 | */ | |
47c5713f DS |
5224 | stripe_nr = div_u64_rem(stripe_nr, |
5225 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5226 | if (mirror_num > 1) |
5227 | stripe_index = nr_data_stripes(map) + | |
5228 | mirror_num - 2; | |
5229 | ||
5230 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5231 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5232 | &stripe_index); | |
28e1cc7d MX |
5233 | if (!(rw & (REQ_WRITE | REQ_DISCARD | |
5234 | REQ_GET_READ_MIRRORS)) && mirror_num <= 1) | |
5235 | mirror_num = 1; | |
53b381b3 | 5236 | } |
8790d502 CM |
5237 | } else { |
5238 | /* | |
47c5713f DS |
5239 | * after this, stripe_nr is the number of stripes on this |
5240 | * device we have to walk to find the data, and stripe_index is | |
5241 | * the number of our device in the stripe array | |
8790d502 | 5242 | */ |
47c5713f DS |
5243 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5244 | &stripe_index); | |
a1d3c478 | 5245 | mirror_num = stripe_index + 1; |
8790d502 | 5246 | } |
593060d7 | 5247 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 5248 | |
472262f3 | 5249 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
5250 | if (dev_replace_is_ongoing) { |
5251 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
5252 | num_alloc_stripes <<= 1; | |
5253 | if (rw & REQ_GET_READ_MIRRORS) | |
5254 | num_alloc_stripes++; | |
2c8cdd6e | 5255 | tgtdev_indexes = num_stripes; |
ad6d620e | 5256 | } |
2c8cdd6e | 5257 | |
6e9606d2 | 5258 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5259 | if (!bbio) { |
5260 | ret = -ENOMEM; | |
5261 | goto out; | |
5262 | } | |
2c8cdd6e MX |
5263 | if (dev_replace_is_ongoing) |
5264 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
de11cc12 | 5265 | |
8e5cfb55 | 5266 | /* build raid_map */ |
ffe2d203 | 5267 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
8e5cfb55 ZL |
5268 | need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5269 | mirror_num > 1)) { | |
5270 | u64 tmp; | |
9d644a62 | 5271 | unsigned rot; |
8e5cfb55 ZL |
5272 | |
5273 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5274 | sizeof(struct btrfs_bio_stripe) * | |
5275 | num_alloc_stripes + | |
5276 | sizeof(int) * tgtdev_indexes); | |
5277 | ||
5278 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5279 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5280 | |
5281 | /* Fill in the logical address of each stripe */ | |
5282 | tmp = stripe_nr * nr_data_stripes(map); | |
5283 | for (i = 0; i < nr_data_stripes(map); i++) | |
5284 | bbio->raid_map[(i+rot) % num_stripes] = | |
5285 | em->start + (tmp + i) * map->stripe_len; | |
5286 | ||
5287 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5288 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5289 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5290 | RAID6_Q_STRIPE; | |
5291 | } | |
5292 | ||
fce3bb9a | 5293 | if (rw & REQ_DISCARD) { |
9d644a62 DS |
5294 | u32 factor = 0; |
5295 | u32 sub_stripes = 0; | |
ec9ef7a1 LZ |
5296 | u64 stripes_per_dev = 0; |
5297 | u32 remaining_stripes = 0; | |
b89203f7 | 5298 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5299 | |
5300 | if (map->type & | |
5301 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5302 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5303 | sub_stripes = 1; | |
5304 | else | |
5305 | sub_stripes = map->sub_stripes; | |
5306 | ||
5307 | factor = map->num_stripes / sub_stripes; | |
5308 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5309 | stripe_nr_orig, | |
5310 | factor, | |
5311 | &remaining_stripes); | |
b89203f7 LB |
5312 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5313 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5314 | } |
5315 | ||
fce3bb9a | 5316 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5317 | bbio->stripes[i].physical = |
f2d8d74d CM |
5318 | map->stripes[stripe_index].physical + |
5319 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5320 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5321 | |
ec9ef7a1 LZ |
5322 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5323 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5324 | bbio->stripes[i].length = stripes_per_dev * | |
5325 | map->stripe_len; | |
b89203f7 | 5326 | |
ec9ef7a1 LZ |
5327 | if (i / sub_stripes < remaining_stripes) |
5328 | bbio->stripes[i].length += | |
5329 | map->stripe_len; | |
b89203f7 LB |
5330 | |
5331 | /* | |
5332 | * Special for the first stripe and | |
5333 | * the last stripe: | |
5334 | * | |
5335 | * |-------|...|-------| | |
5336 | * |----------| | |
5337 | * off end_off | |
5338 | */ | |
ec9ef7a1 | 5339 | if (i < sub_stripes) |
a1d3c478 | 5340 | bbio->stripes[i].length -= |
fce3bb9a | 5341 | stripe_offset; |
b89203f7 LB |
5342 | |
5343 | if (stripe_index >= last_stripe && | |
5344 | stripe_index <= (last_stripe + | |
5345 | sub_stripes - 1)) | |
a1d3c478 | 5346 | bbio->stripes[i].length -= |
fce3bb9a | 5347 | stripe_end_offset; |
b89203f7 | 5348 | |
ec9ef7a1 LZ |
5349 | if (i == sub_stripes - 1) |
5350 | stripe_offset = 0; | |
fce3bb9a | 5351 | } else |
a1d3c478 | 5352 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5353 | |
5354 | stripe_index++; | |
5355 | if (stripe_index == map->num_stripes) { | |
5356 | /* This could only happen for RAID0/10 */ | |
5357 | stripe_index = 0; | |
5358 | stripe_nr++; | |
5359 | } | |
5360 | } | |
5361 | } else { | |
5362 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5363 | bbio->stripes[i].physical = |
212a17ab LT |
5364 | map->stripes[stripe_index].physical + |
5365 | stripe_offset + | |
5366 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5367 | bbio->stripes[i].dev = |
212a17ab | 5368 | map->stripes[stripe_index].dev; |
fce3bb9a | 5369 | stripe_index++; |
f2d8d74d | 5370 | } |
593060d7 | 5371 | } |
de11cc12 | 5372 | |
d20983b4 MX |
5373 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
5374 | max_errors = btrfs_chunk_max_errors(map); | |
de11cc12 | 5375 | |
8e5cfb55 ZL |
5376 | if (bbio->raid_map) |
5377 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5378 | |
2c8cdd6e | 5379 | tgtdev_indexes = 0; |
472262f3 SB |
5380 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
5381 | dev_replace->tgtdev != NULL) { | |
5382 | int index_where_to_add; | |
5383 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5384 | ||
5385 | /* | |
5386 | * duplicate the write operations while the dev replace | |
5387 | * procedure is running. Since the copying of the old disk | |
5388 | * to the new disk takes place at run time while the | |
5389 | * filesystem is mounted writable, the regular write | |
5390 | * operations to the old disk have to be duplicated to go | |
5391 | * to the new disk as well. | |
5392 | * Note that device->missing is handled by the caller, and | |
5393 | * that the write to the old disk is already set up in the | |
5394 | * stripes array. | |
5395 | */ | |
5396 | index_where_to_add = num_stripes; | |
5397 | for (i = 0; i < num_stripes; i++) { | |
5398 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5399 | /* write to new disk, too */ | |
5400 | struct btrfs_bio_stripe *new = | |
5401 | bbio->stripes + index_where_to_add; | |
5402 | struct btrfs_bio_stripe *old = | |
5403 | bbio->stripes + i; | |
5404 | ||
5405 | new->physical = old->physical; | |
5406 | new->length = old->length; | |
5407 | new->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5408 | bbio->tgtdev_map[i] = index_where_to_add; |
472262f3 SB |
5409 | index_where_to_add++; |
5410 | max_errors++; | |
2c8cdd6e | 5411 | tgtdev_indexes++; |
472262f3 SB |
5412 | } |
5413 | } | |
5414 | num_stripes = index_where_to_add; | |
ad6d620e SB |
5415 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
5416 | dev_replace->tgtdev != NULL) { | |
5417 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5418 | int index_srcdev = 0; | |
5419 | int found = 0; | |
5420 | u64 physical_of_found = 0; | |
5421 | ||
5422 | /* | |
5423 | * During the dev-replace procedure, the target drive can | |
5424 | * also be used to read data in case it is needed to repair | |
5425 | * a corrupt block elsewhere. This is possible if the | |
5426 | * requested area is left of the left cursor. In this area, | |
5427 | * the target drive is a full copy of the source drive. | |
5428 | */ | |
5429 | for (i = 0; i < num_stripes; i++) { | |
5430 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5431 | /* | |
5432 | * In case of DUP, in order to keep it | |
5433 | * simple, only add the mirror with the | |
5434 | * lowest physical address | |
5435 | */ | |
5436 | if (found && | |
5437 | physical_of_found <= | |
5438 | bbio->stripes[i].physical) | |
5439 | continue; | |
5440 | index_srcdev = i; | |
5441 | found = 1; | |
5442 | physical_of_found = bbio->stripes[i].physical; | |
5443 | } | |
5444 | } | |
5445 | if (found) { | |
258ece02 | 5446 | if (physical_of_found + map->stripe_len <= |
ad6d620e SB |
5447 | dev_replace->cursor_left) { |
5448 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5449 | bbio->stripes + num_stripes; | |
5450 | ||
5451 | tgtdev_stripe->physical = physical_of_found; | |
5452 | tgtdev_stripe->length = | |
5453 | bbio->stripes[index_srcdev].length; | |
5454 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5455 | bbio->tgtdev_map[index_srcdev] = num_stripes; |
ad6d620e | 5456 | |
2c8cdd6e | 5457 | tgtdev_indexes++; |
ad6d620e SB |
5458 | num_stripes++; |
5459 | } | |
5460 | } | |
472262f3 SB |
5461 | } |
5462 | ||
de11cc12 | 5463 | *bbio_ret = bbio; |
10f11900 | 5464 | bbio->map_type = map->type; |
de11cc12 LZ |
5465 | bbio->num_stripes = num_stripes; |
5466 | bbio->max_errors = max_errors; | |
5467 | bbio->mirror_num = mirror_num; | |
2c8cdd6e | 5468 | bbio->num_tgtdevs = tgtdev_indexes; |
ad6d620e SB |
5469 | |
5470 | /* | |
5471 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5472 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5473 | * available as a mirror | |
5474 | */ | |
5475 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5476 | WARN_ON(num_stripes > 1); | |
5477 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5478 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5479 | bbio->mirror_num = map->num_stripes + 1; | |
5480 | } | |
cea9e445 | 5481 | out: |
472262f3 SB |
5482 | if (dev_replace_is_ongoing) |
5483 | btrfs_dev_replace_unlock(dev_replace); | |
0b86a832 | 5484 | free_extent_map(em); |
de11cc12 | 5485 | return ret; |
0b86a832 CM |
5486 | } |
5487 | ||
3ec706c8 | 5488 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5489 | u64 logical, u64 *length, |
a1d3c478 | 5490 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5491 | { |
3ec706c8 | 5492 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
8e5cfb55 | 5493 | mirror_num, 0); |
f2d8d74d CM |
5494 | } |
5495 | ||
af8e2d1d MX |
5496 | /* For Scrub/replace */ |
5497 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw, | |
5498 | u64 logical, u64 *length, | |
5499 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5500 | int need_raid_map) |
af8e2d1d MX |
5501 | { |
5502 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, | |
8e5cfb55 | 5503 | mirror_num, need_raid_map); |
af8e2d1d MX |
5504 | } |
5505 | ||
a512bbf8 YZ |
5506 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
5507 | u64 chunk_start, u64 physical, u64 devid, | |
5508 | u64 **logical, int *naddrs, int *stripe_len) | |
5509 | { | |
5510 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5511 | struct extent_map *em; | |
5512 | struct map_lookup *map; | |
5513 | u64 *buf; | |
5514 | u64 bytenr; | |
5515 | u64 length; | |
5516 | u64 stripe_nr; | |
53b381b3 | 5517 | u64 rmap_len; |
a512bbf8 YZ |
5518 | int i, j, nr = 0; |
5519 | ||
890871be | 5520 | read_lock(&em_tree->lock); |
a512bbf8 | 5521 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5522 | read_unlock(&em_tree->lock); |
a512bbf8 | 5523 | |
835d974f | 5524 | if (!em) { |
efe120a0 | 5525 | printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n", |
835d974f JB |
5526 | chunk_start); |
5527 | return -EIO; | |
5528 | } | |
5529 | ||
5530 | if (em->start != chunk_start) { | |
efe120a0 | 5531 | printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n", |
835d974f JB |
5532 | em->start, chunk_start); |
5533 | free_extent_map(em); | |
5534 | return -EIO; | |
5535 | } | |
a512bbf8 YZ |
5536 | map = (struct map_lookup *)em->bdev; |
5537 | ||
5538 | length = em->len; | |
53b381b3 DW |
5539 | rmap_len = map->stripe_len; |
5540 | ||
a512bbf8 | 5541 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5542 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5543 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5544 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5545 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5546 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5547 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5548 | } | |
a512bbf8 | 5549 | |
31e818fe | 5550 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5551 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5552 | |
5553 | for (i = 0; i < map->num_stripes; i++) { | |
5554 | if (devid && map->stripes[i].dev->devid != devid) | |
5555 | continue; | |
5556 | if (map->stripes[i].physical > physical || | |
5557 | map->stripes[i].physical + length <= physical) | |
5558 | continue; | |
5559 | ||
5560 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 5561 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5562 | |
5563 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5564 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5565 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5566 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5567 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5568 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5569 | * Alternatively, just use rmap_len below instead of | |
5570 | * map->stripe_len */ | |
5571 | ||
5572 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5573 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5574 | for (j = 0; j < nr; j++) { |
5575 | if (buf[j] == bytenr) | |
5576 | break; | |
5577 | } | |
934d375b CM |
5578 | if (j == nr) { |
5579 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5580 | buf[nr++] = bytenr; |
934d375b | 5581 | } |
a512bbf8 YZ |
5582 | } |
5583 | ||
a512bbf8 YZ |
5584 | *logical = buf; |
5585 | *naddrs = nr; | |
53b381b3 | 5586 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5587 | |
5588 | free_extent_map(em); | |
5589 | return 0; | |
f2d8d74d CM |
5590 | } |
5591 | ||
8408c716 MX |
5592 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err) |
5593 | { | |
5594 | if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED)) | |
5595 | bio_endio_nodec(bio, err); | |
5596 | else | |
5597 | bio_endio(bio, err); | |
6e9606d2 | 5598 | btrfs_put_bbio(bbio); |
8408c716 MX |
5599 | } |
5600 | ||
a1d3c478 | 5601 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 5602 | { |
9be3395b | 5603 | struct btrfs_bio *bbio = bio->bi_private; |
c404e0dc | 5604 | struct btrfs_device *dev = bbio->stripes[0].dev; |
7d2b4daa | 5605 | int is_orig_bio = 0; |
8790d502 | 5606 | |
442a4f63 | 5607 | if (err) { |
a1d3c478 | 5608 | atomic_inc(&bbio->error); |
442a4f63 SB |
5609 | if (err == -EIO || err == -EREMOTEIO) { |
5610 | unsigned int stripe_index = | |
9be3395b | 5611 | btrfs_io_bio(bio)->stripe_index; |
442a4f63 SB |
5612 | |
5613 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5614 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5615 | if (dev->bdev) { |
5616 | if (bio->bi_rw & WRITE) | |
5617 | btrfs_dev_stat_inc(dev, | |
5618 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5619 | else | |
5620 | btrfs_dev_stat_inc(dev, | |
5621 | BTRFS_DEV_STAT_READ_ERRS); | |
5622 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5623 | btrfs_dev_stat_inc(dev, | |
5624 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5625 | btrfs_dev_stat_print_on_error(dev); | |
5626 | } | |
442a4f63 SB |
5627 | } |
5628 | } | |
8790d502 | 5629 | |
a1d3c478 | 5630 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5631 | is_orig_bio = 1; |
5632 | ||
c404e0dc MX |
5633 | btrfs_bio_counter_dec(bbio->fs_info); |
5634 | ||
a1d3c478 | 5635 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5636 | if (!is_orig_bio) { |
5637 | bio_put(bio); | |
a1d3c478 | 5638 | bio = bbio->orig_bio; |
7d2b4daa | 5639 | } |
c7b22bb1 | 5640 | |
a1d3c478 JS |
5641 | bio->bi_private = bbio->private; |
5642 | bio->bi_end_io = bbio->end_io; | |
9be3395b | 5643 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 5644 | /* only send an error to the higher layers if it is |
53b381b3 | 5645 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5646 | */ |
a1d3c478 | 5647 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 5648 | err = -EIO; |
5dbc8fca | 5649 | } else { |
1259ab75 CM |
5650 | /* |
5651 | * this bio is actually up to date, we didn't | |
5652 | * go over the max number of errors | |
5653 | */ | |
5654 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 5655 | err = 0; |
1259ab75 | 5656 | } |
c55f1396 | 5657 | |
8408c716 | 5658 | btrfs_end_bbio(bbio, bio, err); |
7d2b4daa | 5659 | } else if (!is_orig_bio) { |
8790d502 CM |
5660 | bio_put(bio); |
5661 | } | |
8790d502 CM |
5662 | } |
5663 | ||
8b712842 CM |
5664 | /* |
5665 | * see run_scheduled_bios for a description of why bios are collected for | |
5666 | * async submit. | |
5667 | * | |
5668 | * This will add one bio to the pending list for a device and make sure | |
5669 | * the work struct is scheduled. | |
5670 | */ | |
48a3b636 ES |
5671 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
5672 | struct btrfs_device *device, | |
5673 | int rw, struct bio *bio) | |
8b712842 CM |
5674 | { |
5675 | int should_queue = 1; | |
ffbd517d | 5676 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5677 | |
53b381b3 DW |
5678 | if (device->missing || !device->bdev) { |
5679 | bio_endio(bio, -EIO); | |
5680 | return; | |
5681 | } | |
5682 | ||
8b712842 | 5683 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5684 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5685 | bio_get(bio); |
21adbd5c | 5686 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5687 | bio_put(bio); |
143bede5 | 5688 | return; |
8b712842 CM |
5689 | } |
5690 | ||
5691 | /* | |
0986fe9e | 5692 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5693 | * higher layers. Otherwise, the async bio makes it appear we have |
5694 | * made progress against dirty pages when we've really just put it | |
5695 | * on a queue for later | |
5696 | */ | |
0986fe9e | 5697 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5698 | WARN_ON(bio->bi_next); |
8b712842 CM |
5699 | bio->bi_next = NULL; |
5700 | bio->bi_rw |= rw; | |
5701 | ||
5702 | spin_lock(&device->io_lock); | |
7b6d91da | 5703 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5704 | pending_bios = &device->pending_sync_bios; |
5705 | else | |
5706 | pending_bios = &device->pending_bios; | |
8b712842 | 5707 | |
ffbd517d CM |
5708 | if (pending_bios->tail) |
5709 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5710 | |
ffbd517d CM |
5711 | pending_bios->tail = bio; |
5712 | if (!pending_bios->head) | |
5713 | pending_bios->head = bio; | |
8b712842 CM |
5714 | if (device->running_pending) |
5715 | should_queue = 0; | |
5716 | ||
5717 | spin_unlock(&device->io_lock); | |
5718 | ||
5719 | if (should_queue) | |
a8c93d4e QW |
5720 | btrfs_queue_work(root->fs_info->submit_workers, |
5721 | &device->work); | |
8b712842 CM |
5722 | } |
5723 | ||
de1ee92a JB |
5724 | static int bio_size_ok(struct block_device *bdev, struct bio *bio, |
5725 | sector_t sector) | |
5726 | { | |
5727 | struct bio_vec *prev; | |
5728 | struct request_queue *q = bdev_get_queue(bdev); | |
475bf36f | 5729 | unsigned int max_sectors = queue_max_sectors(q); |
de1ee92a JB |
5730 | struct bvec_merge_data bvm = { |
5731 | .bi_bdev = bdev, | |
5732 | .bi_sector = sector, | |
5733 | .bi_rw = bio->bi_rw, | |
5734 | }; | |
5735 | ||
fae7f21c | 5736 | if (WARN_ON(bio->bi_vcnt == 0)) |
de1ee92a | 5737 | return 1; |
de1ee92a JB |
5738 | |
5739 | prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
aa8b57aa | 5740 | if (bio_sectors(bio) > max_sectors) |
de1ee92a JB |
5741 | return 0; |
5742 | ||
5743 | if (!q->merge_bvec_fn) | |
5744 | return 1; | |
5745 | ||
4f024f37 | 5746 | bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len; |
de1ee92a JB |
5747 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) |
5748 | return 0; | |
5749 | return 1; | |
5750 | } | |
5751 | ||
5752 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5753 | struct bio *bio, u64 physical, int dev_nr, | |
5754 | int rw, int async) | |
5755 | { | |
5756 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5757 | ||
5758 | bio->bi_private = bbio; | |
9be3395b | 5759 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 5760 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 5761 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
5762 | #ifdef DEBUG |
5763 | { | |
5764 | struct rcu_string *name; | |
5765 | ||
5766 | rcu_read_lock(); | |
5767 | name = rcu_dereference(dev->name); | |
d1423248 | 5768 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a | 5769 | "(%s id %llu), size=%u\n", rw, |
1b6e4469 FF |
5770 | (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev, |
5771 | name->str, dev->devid, bio->bi_iter.bi_size); | |
de1ee92a JB |
5772 | rcu_read_unlock(); |
5773 | } | |
5774 | #endif | |
5775 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
5776 | |
5777 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
5778 | ||
de1ee92a | 5779 | if (async) |
53b381b3 | 5780 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
5781 | else |
5782 | btrfsic_submit_bio(rw, bio); | |
5783 | } | |
5784 | ||
5785 | static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5786 | struct bio *first_bio, struct btrfs_device *dev, | |
5787 | int dev_nr, int rw, int async) | |
5788 | { | |
5789 | struct bio_vec *bvec = first_bio->bi_io_vec; | |
5790 | struct bio *bio; | |
5791 | int nr_vecs = bio_get_nr_vecs(dev->bdev); | |
5792 | u64 physical = bbio->stripes[dev_nr].physical; | |
5793 | ||
5794 | again: | |
5795 | bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS); | |
5796 | if (!bio) | |
5797 | return -ENOMEM; | |
5798 | ||
5799 | while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) { | |
5800 | if (bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5801 | bvec->bv_offset) < bvec->bv_len) { | |
4f024f37 | 5802 | u64 len = bio->bi_iter.bi_size; |
de1ee92a JB |
5803 | |
5804 | atomic_inc(&bbio->stripes_pending); | |
5805 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, | |
5806 | rw, async); | |
5807 | physical += len; | |
5808 | goto again; | |
5809 | } | |
5810 | bvec++; | |
5811 | } | |
5812 | ||
5813 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async); | |
5814 | return 0; | |
5815 | } | |
5816 | ||
5817 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
5818 | { | |
5819 | atomic_inc(&bbio->error); | |
5820 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
8408c716 MX |
5821 | /* Shoud be the original bio. */ |
5822 | WARN_ON(bio != bbio->orig_bio); | |
5823 | ||
de1ee92a JB |
5824 | bio->bi_private = bbio->private; |
5825 | bio->bi_end_io = bbio->end_io; | |
9be3395b | 5826 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 5827 | bio->bi_iter.bi_sector = logical >> 9; |
8408c716 MX |
5828 | |
5829 | btrfs_end_bbio(bbio, bio, -EIO); | |
de1ee92a JB |
5830 | } |
5831 | } | |
5832 | ||
f188591e | 5833 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 5834 | int mirror_num, int async_submit) |
0b86a832 | 5835 | { |
0b86a832 | 5836 | struct btrfs_device *dev; |
8790d502 | 5837 | struct bio *first_bio = bio; |
4f024f37 | 5838 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
5839 | u64 length = 0; |
5840 | u64 map_length; | |
0b86a832 | 5841 | int ret; |
08da757d ZL |
5842 | int dev_nr; |
5843 | int total_devs; | |
a1d3c478 | 5844 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 5845 | |
4f024f37 | 5846 | length = bio->bi_iter.bi_size; |
0b86a832 | 5847 | map_length = length; |
cea9e445 | 5848 | |
c404e0dc | 5849 | btrfs_bio_counter_inc_blocked(root->fs_info); |
53b381b3 | 5850 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
8e5cfb55 | 5851 | mirror_num, 1); |
c404e0dc MX |
5852 | if (ret) { |
5853 | btrfs_bio_counter_dec(root->fs_info); | |
79787eaa | 5854 | return ret; |
c404e0dc | 5855 | } |
cea9e445 | 5856 | |
a1d3c478 | 5857 | total_devs = bbio->num_stripes; |
53b381b3 DW |
5858 | bbio->orig_bio = first_bio; |
5859 | bbio->private = first_bio->bi_private; | |
5860 | bbio->end_io = first_bio->bi_end_io; | |
c404e0dc | 5861 | bbio->fs_info = root->fs_info; |
53b381b3 DW |
5862 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
5863 | ||
8e5cfb55 | 5864 | if (bbio->raid_map) { |
53b381b3 DW |
5865 | /* In this case, map_length has been set to the length of |
5866 | a single stripe; not the whole write */ | |
5867 | if (rw & WRITE) { | |
8e5cfb55 | 5868 | ret = raid56_parity_write(root, bio, bbio, map_length); |
53b381b3 | 5869 | } else { |
8e5cfb55 | 5870 | ret = raid56_parity_recover(root, bio, bbio, map_length, |
4245215d | 5871 | mirror_num, 1); |
53b381b3 | 5872 | } |
4245215d | 5873 | |
c404e0dc MX |
5874 | btrfs_bio_counter_dec(root->fs_info); |
5875 | return ret; | |
53b381b3 DW |
5876 | } |
5877 | ||
cea9e445 | 5878 | if (map_length < length) { |
c2cf52eb | 5879 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", |
c1c9ff7c | 5880 | logical, length, map_length); |
cea9e445 CM |
5881 | BUG(); |
5882 | } | |
a1d3c478 | 5883 | |
08da757d | 5884 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a JB |
5885 | dev = bbio->stripes[dev_nr].dev; |
5886 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
5887 | bbio_error(bbio, first_bio, logical); | |
de1ee92a JB |
5888 | continue; |
5889 | } | |
5890 | ||
5891 | /* | |
5892 | * Check and see if we're ok with this bio based on it's size | |
5893 | * and offset with the given device. | |
5894 | */ | |
5895 | if (!bio_size_ok(dev->bdev, first_bio, | |
5896 | bbio->stripes[dev_nr].physical >> 9)) { | |
5897 | ret = breakup_stripe_bio(root, bbio, first_bio, dev, | |
5898 | dev_nr, rw, async_submit); | |
5899 | BUG_ON(ret); | |
de1ee92a JB |
5900 | continue; |
5901 | } | |
5902 | ||
a1d3c478 | 5903 | if (dev_nr < total_devs - 1) { |
9be3395b | 5904 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 5905 | BUG_ON(!bio); /* -ENOMEM */ |
a1d3c478 JS |
5906 | } else { |
5907 | bio = first_bio; | |
c55f1396 | 5908 | bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED; |
8790d502 | 5909 | } |
de1ee92a JB |
5910 | |
5911 | submit_stripe_bio(root, bbio, bio, | |
5912 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
5913 | async_submit); | |
8790d502 | 5914 | } |
c404e0dc | 5915 | btrfs_bio_counter_dec(root->fs_info); |
0b86a832 CM |
5916 | return 0; |
5917 | } | |
5918 | ||
aa1b8cd4 | 5919 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 5920 | u8 *uuid, u8 *fsid) |
0b86a832 | 5921 | { |
2b82032c YZ |
5922 | struct btrfs_device *device; |
5923 | struct btrfs_fs_devices *cur_devices; | |
5924 | ||
aa1b8cd4 | 5925 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
5926 | while (cur_devices) { |
5927 | if (!fsid || | |
5928 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
5929 | device = __find_device(&cur_devices->devices, | |
5930 | devid, uuid); | |
5931 | if (device) | |
5932 | return device; | |
5933 | } | |
5934 | cur_devices = cur_devices->seed; | |
5935 | } | |
5936 | return NULL; | |
0b86a832 CM |
5937 | } |
5938 | ||
dfe25020 | 5939 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 5940 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
5941 | u64 devid, u8 *dev_uuid) |
5942 | { | |
5943 | struct btrfs_device *device; | |
dfe25020 | 5944 | |
12bd2fc0 ID |
5945 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
5946 | if (IS_ERR(device)) | |
7cbd8a83 | 5947 | return NULL; |
12bd2fc0 ID |
5948 | |
5949 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 5950 | device->fs_devices = fs_devices; |
dfe25020 | 5951 | fs_devices->num_devices++; |
12bd2fc0 ID |
5952 | |
5953 | device->missing = 1; | |
cd02dca5 | 5954 | fs_devices->missing_devices++; |
12bd2fc0 | 5955 | |
dfe25020 CM |
5956 | return device; |
5957 | } | |
5958 | ||
12bd2fc0 ID |
5959 | /** |
5960 | * btrfs_alloc_device - allocate struct btrfs_device | |
5961 | * @fs_info: used only for generating a new devid, can be NULL if | |
5962 | * devid is provided (i.e. @devid != NULL). | |
5963 | * @devid: a pointer to devid for this device. If NULL a new devid | |
5964 | * is generated. | |
5965 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
5966 | * is generated. | |
5967 | * | |
5968 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
5969 | * on error. Returned struct is not linked onto any lists and can be | |
5970 | * destroyed with kfree() right away. | |
5971 | */ | |
5972 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
5973 | const u64 *devid, | |
5974 | const u8 *uuid) | |
5975 | { | |
5976 | struct btrfs_device *dev; | |
5977 | u64 tmp; | |
5978 | ||
fae7f21c | 5979 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 5980 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
5981 | |
5982 | dev = __alloc_device(); | |
5983 | if (IS_ERR(dev)) | |
5984 | return dev; | |
5985 | ||
5986 | if (devid) | |
5987 | tmp = *devid; | |
5988 | else { | |
5989 | int ret; | |
5990 | ||
5991 | ret = find_next_devid(fs_info, &tmp); | |
5992 | if (ret) { | |
5993 | kfree(dev); | |
5994 | return ERR_PTR(ret); | |
5995 | } | |
5996 | } | |
5997 | dev->devid = tmp; | |
5998 | ||
5999 | if (uuid) | |
6000 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6001 | else | |
6002 | generate_random_uuid(dev->uuid); | |
6003 | ||
9e0af237 LB |
6004 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6005 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6006 | |
6007 | return dev; | |
6008 | } | |
6009 | ||
0b86a832 CM |
6010 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
6011 | struct extent_buffer *leaf, | |
6012 | struct btrfs_chunk *chunk) | |
6013 | { | |
6014 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6015 | struct map_lookup *map; | |
6016 | struct extent_map *em; | |
6017 | u64 logical; | |
6018 | u64 length; | |
6019 | u64 devid; | |
a443755f | 6020 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 6021 | int num_stripes; |
0b86a832 | 6022 | int ret; |
593060d7 | 6023 | int i; |
0b86a832 | 6024 | |
e17cade2 CM |
6025 | logical = key->offset; |
6026 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 6027 | |
890871be | 6028 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6029 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6030 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6031 | |
6032 | /* already mapped? */ | |
6033 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6034 | free_extent_map(em); | |
0b86a832 CM |
6035 | return 0; |
6036 | } else if (em) { | |
6037 | free_extent_map(em); | |
6038 | } | |
0b86a832 | 6039 | |
172ddd60 | 6040 | em = alloc_extent_map(); |
0b86a832 CM |
6041 | if (!em) |
6042 | return -ENOMEM; | |
593060d7 CM |
6043 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6044 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
6045 | if (!map) { |
6046 | free_extent_map(em); | |
6047 | return -ENOMEM; | |
6048 | } | |
6049 | ||
298a8f9c | 6050 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
0b86a832 CM |
6051 | em->bdev = (struct block_device *)map; |
6052 | em->start = logical; | |
6053 | em->len = length; | |
70c8a91c | 6054 | em->orig_start = 0; |
0b86a832 | 6055 | em->block_start = 0; |
c8b97818 | 6056 | em->block_len = em->len; |
0b86a832 | 6057 | |
593060d7 CM |
6058 | map->num_stripes = num_stripes; |
6059 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6060 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6061 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6062 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6063 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6064 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6065 | for (i = 0; i < num_stripes; i++) { |
6066 | map->stripes[i].physical = | |
6067 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6068 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6069 | read_extent_buffer(leaf, uuid, (unsigned long) |
6070 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6071 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
6072 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
6073 | uuid, NULL); | |
dfe25020 | 6074 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
6075 | free_extent_map(em); |
6076 | return -EIO; | |
6077 | } | |
dfe25020 CM |
6078 | if (!map->stripes[i].dev) { |
6079 | map->stripes[i].dev = | |
5f375835 MX |
6080 | add_missing_dev(root, root->fs_info->fs_devices, |
6081 | devid, uuid); | |
dfe25020 | 6082 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6083 | free_extent_map(em); |
6084 | return -EIO; | |
6085 | } | |
6086 | } | |
6087 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6088 | } |
6089 | ||
890871be | 6090 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6091 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6092 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6093 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6094 | free_extent_map(em); |
6095 | ||
6096 | return 0; | |
6097 | } | |
6098 | ||
143bede5 | 6099 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6100 | struct btrfs_dev_item *dev_item, |
6101 | struct btrfs_device *device) | |
6102 | { | |
6103 | unsigned long ptr; | |
0b86a832 CM |
6104 | |
6105 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6106 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6107 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6108 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6109 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6110 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6111 | device->type = btrfs_device_type(leaf, dev_item); |
6112 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6113 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6114 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6115 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6116 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6117 | |
410ba3a2 | 6118 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6119 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6120 | } |
6121 | ||
5f375835 MX |
6122 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
6123 | u8 *fsid) | |
2b82032c YZ |
6124 | { |
6125 | struct btrfs_fs_devices *fs_devices; | |
6126 | int ret; | |
6127 | ||
b367e47f | 6128 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
6129 | |
6130 | fs_devices = root->fs_info->fs_devices->seed; | |
6131 | while (fs_devices) { | |
5f375835 MX |
6132 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6133 | return fs_devices; | |
6134 | ||
2b82032c YZ |
6135 | fs_devices = fs_devices->seed; |
6136 | } | |
6137 | ||
6138 | fs_devices = find_fsid(fsid); | |
6139 | if (!fs_devices) { | |
5f375835 MX |
6140 | if (!btrfs_test_opt(root, DEGRADED)) |
6141 | return ERR_PTR(-ENOENT); | |
6142 | ||
6143 | fs_devices = alloc_fs_devices(fsid); | |
6144 | if (IS_ERR(fs_devices)) | |
6145 | return fs_devices; | |
6146 | ||
6147 | fs_devices->seeding = 1; | |
6148 | fs_devices->opened = 1; | |
6149 | return fs_devices; | |
2b82032c | 6150 | } |
e4404d6e YZ |
6151 | |
6152 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6153 | if (IS_ERR(fs_devices)) |
6154 | return fs_devices; | |
2b82032c | 6155 | |
97288f2c | 6156 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 6157 | root->fs_info->bdev_holder); |
48d28232 JL |
6158 | if (ret) { |
6159 | free_fs_devices(fs_devices); | |
5f375835 | 6160 | fs_devices = ERR_PTR(ret); |
2b82032c | 6161 | goto out; |
48d28232 | 6162 | } |
2b82032c YZ |
6163 | |
6164 | if (!fs_devices->seeding) { | |
6165 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6166 | free_fs_devices(fs_devices); |
5f375835 | 6167 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6168 | goto out; |
6169 | } | |
6170 | ||
6171 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6172 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6173 | out: |
5f375835 | 6174 | return fs_devices; |
2b82032c YZ |
6175 | } |
6176 | ||
0d81ba5d | 6177 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6178 | struct extent_buffer *leaf, |
6179 | struct btrfs_dev_item *dev_item) | |
6180 | { | |
5f375835 | 6181 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
0b86a832 CM |
6182 | struct btrfs_device *device; |
6183 | u64 devid; | |
6184 | int ret; | |
2b82032c | 6185 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6186 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6187 | ||
0b86a832 | 6188 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6189 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6190 | BTRFS_UUID_SIZE); |
1473b24e | 6191 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6192 | BTRFS_UUID_SIZE); |
6193 | ||
6194 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5f375835 MX |
6195 | fs_devices = open_seed_devices(root, fs_uuid); |
6196 | if (IS_ERR(fs_devices)) | |
6197 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6198 | } |
6199 | ||
aa1b8cd4 | 6200 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6201 | if (!device) { |
e4404d6e | 6202 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
6203 | return -EIO; |
6204 | ||
5f375835 MX |
6205 | btrfs_warn(root->fs_info, "devid %llu missing", devid); |
6206 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); | |
6207 | if (!device) | |
6208 | return -ENOMEM; | |
6209 | } else { | |
6210 | if (!device->bdev && !btrfs_test_opt(root, DEGRADED)) | |
6211 | return -EIO; | |
6212 | ||
6213 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6214 | /* |
6215 | * this happens when a device that was properly setup | |
6216 | * in the device info lists suddenly goes bad. | |
6217 | * device->bdev is NULL, and so we have to set | |
6218 | * device->missing to one here | |
6219 | */ | |
5f375835 | 6220 | device->fs_devices->missing_devices++; |
cd02dca5 | 6221 | device->missing = 1; |
2b82032c | 6222 | } |
5f375835 MX |
6223 | |
6224 | /* Move the device to its own fs_devices */ | |
6225 | if (device->fs_devices != fs_devices) { | |
6226 | ASSERT(device->missing); | |
6227 | ||
6228 | list_move(&device->dev_list, &fs_devices->devices); | |
6229 | device->fs_devices->num_devices--; | |
6230 | fs_devices->num_devices++; | |
6231 | ||
6232 | device->fs_devices->missing_devices--; | |
6233 | fs_devices->missing_devices++; | |
6234 | ||
6235 | device->fs_devices = fs_devices; | |
6236 | } | |
2b82032c YZ |
6237 | } |
6238 | ||
6239 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6240 | BUG_ON(device->writeable); | |
6241 | if (device->generation != | |
6242 | btrfs_device_generation(leaf, dev_item)) | |
6243 | return -EINVAL; | |
6324fbf3 | 6244 | } |
0b86a832 CM |
6245 | |
6246 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6247 | device->in_fs_metadata = 1; |
63a212ab | 6248 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6249 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
6250 | spin_lock(&root->fs_info->free_chunk_lock); |
6251 | root->fs_info->free_chunk_space += device->total_bytes - | |
6252 | device->bytes_used; | |
6253 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6254 | } | |
0b86a832 | 6255 | ret = 0; |
0b86a832 CM |
6256 | return ret; |
6257 | } | |
6258 | ||
e4404d6e | 6259 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 6260 | { |
6c41761f | 6261 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 6262 | struct extent_buffer *sb; |
0b86a832 | 6263 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6264 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6265 | u8 *array_ptr; |
6266 | unsigned long sb_array_offset; | |
84eed90f | 6267 | int ret = 0; |
0b86a832 CM |
6268 | u32 num_stripes; |
6269 | u32 array_size; | |
6270 | u32 len = 0; | |
1ffb22cf | 6271 | u32 cur_offset; |
84eed90f | 6272 | struct btrfs_key key; |
0b86a832 | 6273 | |
a83fffb7 DS |
6274 | ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize); |
6275 | /* | |
6276 | * This will create extent buffer of nodesize, superblock size is | |
6277 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6278 | * overallocate but we can keep it as-is, only the first page is used. | |
6279 | */ | |
6280 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
a061fc8d CM |
6281 | if (!sb) |
6282 | return -ENOMEM; | |
6283 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 6284 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
6285 | /* |
6286 | * The sb extent buffer is artifical and just used to read the system array. | |
6287 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
6288 | * pages up-to-date when the page is larger: extent does not cover the | |
6289 | * whole page and consequently check_page_uptodate does not find all | |
6290 | * the page's extents up-to-date (the hole beyond sb), | |
6291 | * write_extent_buffer then triggers a WARN_ON. | |
6292 | * | |
6293 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6294 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6295 | * to silence the warning eg. on PowerPC 64. | |
6296 | */ | |
6297 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 6298 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6299 | |
a061fc8d | 6300 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6301 | array_size = btrfs_super_sys_array_size(super_copy); |
6302 | ||
1ffb22cf DS |
6303 | array_ptr = super_copy->sys_chunk_array; |
6304 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6305 | cur_offset = 0; | |
0b86a832 | 6306 | |
1ffb22cf DS |
6307 | while (cur_offset < array_size) { |
6308 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6309 | len = sizeof(*disk_key); |
6310 | if (cur_offset + len > array_size) | |
6311 | goto out_short_read; | |
6312 | ||
0b86a832 CM |
6313 | btrfs_disk_key_to_cpu(&key, disk_key); |
6314 | ||
1ffb22cf DS |
6315 | array_ptr += len; |
6316 | sb_array_offset += len; | |
6317 | cur_offset += len; | |
0b86a832 | 6318 | |
0d81ba5d | 6319 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6320 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6321 | /* |
6322 | * At least one btrfs_chunk with one stripe must be | |
6323 | * present, exact stripe count check comes afterwards | |
6324 | */ | |
6325 | len = btrfs_chunk_item_size(1); | |
6326 | if (cur_offset + len > array_size) | |
6327 | goto out_short_read; | |
6328 | ||
6329 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
6330 | len = btrfs_chunk_item_size(num_stripes); | |
6331 | if (cur_offset + len > array_size) | |
6332 | goto out_short_read; | |
6333 | ||
0d81ba5d | 6334 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6335 | if (ret) |
6336 | break; | |
0b86a832 | 6337 | } else { |
84eed90f CM |
6338 | ret = -EIO; |
6339 | break; | |
0b86a832 | 6340 | } |
1ffb22cf DS |
6341 | array_ptr += len; |
6342 | sb_array_offset += len; | |
6343 | cur_offset += len; | |
0b86a832 | 6344 | } |
a061fc8d | 6345 | free_extent_buffer(sb); |
84eed90f | 6346 | return ret; |
e3540eab DS |
6347 | |
6348 | out_short_read: | |
6349 | printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n", | |
6350 | len, cur_offset); | |
6351 | free_extent_buffer(sb); | |
6352 | return -EIO; | |
0b86a832 CM |
6353 | } |
6354 | ||
6355 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6356 | { | |
6357 | struct btrfs_path *path; | |
6358 | struct extent_buffer *leaf; | |
6359 | struct btrfs_key key; | |
6360 | struct btrfs_key found_key; | |
6361 | int ret; | |
6362 | int slot; | |
6363 | ||
6364 | root = root->fs_info->chunk_root; | |
6365 | ||
6366 | path = btrfs_alloc_path(); | |
6367 | if (!path) | |
6368 | return -ENOMEM; | |
6369 | ||
b367e47f LZ |
6370 | mutex_lock(&uuid_mutex); |
6371 | lock_chunks(root); | |
6372 | ||
395927a9 FDBM |
6373 | /* |
6374 | * Read all device items, and then all the chunk items. All | |
6375 | * device items are found before any chunk item (their object id | |
6376 | * is smaller than the lowest possible object id for a chunk | |
6377 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6378 | */ |
6379 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6380 | key.offset = 0; | |
6381 | key.type = 0; | |
0b86a832 | 6382 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6383 | if (ret < 0) |
6384 | goto error; | |
d397712b | 6385 | while (1) { |
0b86a832 CM |
6386 | leaf = path->nodes[0]; |
6387 | slot = path->slots[0]; | |
6388 | if (slot >= btrfs_header_nritems(leaf)) { | |
6389 | ret = btrfs_next_leaf(root, path); | |
6390 | if (ret == 0) | |
6391 | continue; | |
6392 | if (ret < 0) | |
6393 | goto error; | |
6394 | break; | |
6395 | } | |
6396 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6397 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6398 | struct btrfs_dev_item *dev_item; | |
6399 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6400 | struct btrfs_dev_item); |
395927a9 FDBM |
6401 | ret = read_one_dev(root, leaf, dev_item); |
6402 | if (ret) | |
6403 | goto error; | |
0b86a832 CM |
6404 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6405 | struct btrfs_chunk *chunk; | |
6406 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6407 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6408 | if (ret) |
6409 | goto error; | |
0b86a832 CM |
6410 | } |
6411 | path->slots[0]++; | |
6412 | } | |
0b86a832 CM |
6413 | ret = 0; |
6414 | error: | |
b367e47f LZ |
6415 | unlock_chunks(root); |
6416 | mutex_unlock(&uuid_mutex); | |
6417 | ||
2b82032c | 6418 | btrfs_free_path(path); |
0b86a832 CM |
6419 | return ret; |
6420 | } | |
442a4f63 | 6421 | |
cb517eab MX |
6422 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6423 | { | |
6424 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6425 | struct btrfs_device *device; | |
6426 | ||
29cc83f6 LB |
6427 | while (fs_devices) { |
6428 | mutex_lock(&fs_devices->device_list_mutex); | |
6429 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6430 | device->dev_root = fs_info->dev_root; | |
6431 | mutex_unlock(&fs_devices->device_list_mutex); | |
6432 | ||
6433 | fs_devices = fs_devices->seed; | |
6434 | } | |
cb517eab MX |
6435 | } |
6436 | ||
733f4fbb SB |
6437 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6438 | { | |
6439 | int i; | |
6440 | ||
6441 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6442 | btrfs_dev_stat_reset(dev, i); | |
6443 | } | |
6444 | ||
6445 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6446 | { | |
6447 | struct btrfs_key key; | |
6448 | struct btrfs_key found_key; | |
6449 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6450 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6451 | struct extent_buffer *eb; | |
6452 | int slot; | |
6453 | int ret = 0; | |
6454 | struct btrfs_device *device; | |
6455 | struct btrfs_path *path = NULL; | |
6456 | int i; | |
6457 | ||
6458 | path = btrfs_alloc_path(); | |
6459 | if (!path) { | |
6460 | ret = -ENOMEM; | |
6461 | goto out; | |
6462 | } | |
6463 | ||
6464 | mutex_lock(&fs_devices->device_list_mutex); | |
6465 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6466 | int item_size; | |
6467 | struct btrfs_dev_stats_item *ptr; | |
6468 | ||
6469 | key.objectid = 0; | |
6470 | key.type = BTRFS_DEV_STATS_KEY; | |
6471 | key.offset = device->devid; | |
6472 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6473 | if (ret) { | |
733f4fbb SB |
6474 | __btrfs_reset_dev_stats(device); |
6475 | device->dev_stats_valid = 1; | |
6476 | btrfs_release_path(path); | |
6477 | continue; | |
6478 | } | |
6479 | slot = path->slots[0]; | |
6480 | eb = path->nodes[0]; | |
6481 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6482 | item_size = btrfs_item_size_nr(eb, slot); | |
6483 | ||
6484 | ptr = btrfs_item_ptr(eb, slot, | |
6485 | struct btrfs_dev_stats_item); | |
6486 | ||
6487 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6488 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6489 | btrfs_dev_stat_set(device, i, | |
6490 | btrfs_dev_stats_value(eb, ptr, i)); | |
6491 | else | |
6492 | btrfs_dev_stat_reset(device, i); | |
6493 | } | |
6494 | ||
6495 | device->dev_stats_valid = 1; | |
6496 | btrfs_dev_stat_print_on_load(device); | |
6497 | btrfs_release_path(path); | |
6498 | } | |
6499 | mutex_unlock(&fs_devices->device_list_mutex); | |
6500 | ||
6501 | out: | |
6502 | btrfs_free_path(path); | |
6503 | return ret < 0 ? ret : 0; | |
6504 | } | |
6505 | ||
6506 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6507 | struct btrfs_root *dev_root, | |
6508 | struct btrfs_device *device) | |
6509 | { | |
6510 | struct btrfs_path *path; | |
6511 | struct btrfs_key key; | |
6512 | struct extent_buffer *eb; | |
6513 | struct btrfs_dev_stats_item *ptr; | |
6514 | int ret; | |
6515 | int i; | |
6516 | ||
6517 | key.objectid = 0; | |
6518 | key.type = BTRFS_DEV_STATS_KEY; | |
6519 | key.offset = device->devid; | |
6520 | ||
6521 | path = btrfs_alloc_path(); | |
6522 | BUG_ON(!path); | |
6523 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6524 | if (ret < 0) { | |
efe120a0 FH |
6525 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6526 | "error %d while searching for dev_stats item for device %s!\n", | |
606686ee | 6527 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6528 | goto out; |
6529 | } | |
6530 | ||
6531 | if (ret == 0 && | |
6532 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6533 | /* need to delete old one and insert a new one */ | |
6534 | ret = btrfs_del_item(trans, dev_root, path); | |
6535 | if (ret != 0) { | |
efe120a0 FH |
6536 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6537 | "delete too small dev_stats item for device %s failed %d!\n", | |
606686ee | 6538 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6539 | goto out; |
6540 | } | |
6541 | ret = 1; | |
6542 | } | |
6543 | ||
6544 | if (ret == 1) { | |
6545 | /* need to insert a new item */ | |
6546 | btrfs_release_path(path); | |
6547 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6548 | &key, sizeof(*ptr)); | |
6549 | if (ret < 0) { | |
efe120a0 FH |
6550 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6551 | "insert dev_stats item for device %s failed %d!\n", | |
606686ee | 6552 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6553 | goto out; |
6554 | } | |
6555 | } | |
6556 | ||
6557 | eb = path->nodes[0]; | |
6558 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6559 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6560 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6561 | btrfs_dev_stat_read(device, i)); | |
6562 | btrfs_mark_buffer_dirty(eb); | |
6563 | ||
6564 | out: | |
6565 | btrfs_free_path(path); | |
6566 | return ret; | |
6567 | } | |
6568 | ||
6569 | /* | |
6570 | * called from commit_transaction. Writes all changed device stats to disk. | |
6571 | */ | |
6572 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6573 | struct btrfs_fs_info *fs_info) | |
6574 | { | |
6575 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6576 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6577 | struct btrfs_device *device; | |
addc3fa7 | 6578 | int stats_cnt; |
733f4fbb SB |
6579 | int ret = 0; |
6580 | ||
6581 | mutex_lock(&fs_devices->device_list_mutex); | |
6582 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 6583 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
6584 | continue; |
6585 | ||
addc3fa7 | 6586 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
733f4fbb SB |
6587 | ret = update_dev_stat_item(trans, dev_root, device); |
6588 | if (!ret) | |
addc3fa7 | 6589 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
6590 | } |
6591 | mutex_unlock(&fs_devices->device_list_mutex); | |
6592 | ||
6593 | return ret; | |
6594 | } | |
6595 | ||
442a4f63 SB |
6596 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
6597 | { | |
6598 | btrfs_dev_stat_inc(dev, index); | |
6599 | btrfs_dev_stat_print_on_error(dev); | |
6600 | } | |
6601 | ||
48a3b636 | 6602 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 6603 | { |
733f4fbb SB |
6604 | if (!dev->dev_stats_valid) |
6605 | return; | |
efe120a0 FH |
6606 | printk_ratelimited_in_rcu(KERN_ERR "BTRFS: " |
6607 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6608 | rcu_str_deref(dev->name), |
442a4f63 SB |
6609 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6610 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6611 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
6612 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
6613 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 6614 | } |
c11d2c23 | 6615 | |
733f4fbb SB |
6616 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
6617 | { | |
a98cdb85 SB |
6618 | int i; |
6619 | ||
6620 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6621 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6622 | break; | |
6623 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6624 | return; /* all values == 0, suppress message */ | |
6625 | ||
efe120a0 FH |
6626 | printk_in_rcu(KERN_INFO "BTRFS: " |
6627 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6628 | rcu_str_deref(dev->name), |
733f4fbb SB |
6629 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6630 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6631 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6632 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6633 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6634 | } | |
6635 | ||
c11d2c23 | 6636 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 6637 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
6638 | { |
6639 | struct btrfs_device *dev; | |
6640 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6641 | int i; | |
6642 | ||
6643 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 6644 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
6645 | mutex_unlock(&fs_devices->device_list_mutex); |
6646 | ||
6647 | if (!dev) { | |
efe120a0 | 6648 | btrfs_warn(root->fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 6649 | return -ENODEV; |
733f4fbb | 6650 | } else if (!dev->dev_stats_valid) { |
efe120a0 | 6651 | btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 6652 | return -ENODEV; |
b27f7c0c | 6653 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
6654 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
6655 | if (stats->nr_items > i) | |
6656 | stats->values[i] = | |
6657 | btrfs_dev_stat_read_and_reset(dev, i); | |
6658 | else | |
6659 | btrfs_dev_stat_reset(dev, i); | |
6660 | } | |
6661 | } else { | |
6662 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6663 | if (stats->nr_items > i) | |
6664 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6665 | } | |
6666 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6667 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6668 | return 0; | |
6669 | } | |
a8a6dab7 SB |
6670 | |
6671 | int btrfs_scratch_superblock(struct btrfs_device *device) | |
6672 | { | |
6673 | struct buffer_head *bh; | |
6674 | struct btrfs_super_block *disk_super; | |
6675 | ||
6676 | bh = btrfs_read_dev_super(device->bdev); | |
6677 | if (!bh) | |
6678 | return -EINVAL; | |
6679 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6680 | ||
6681 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6682 | set_buffer_dirty(bh); | |
6683 | sync_dirty_buffer(bh); | |
6684 | brelse(bh); | |
6685 | ||
6686 | return 0; | |
6687 | } | |
935e5cc9 MX |
6688 | |
6689 | /* | |
6690 | * Update the size of all devices, which is used for writing out the | |
6691 | * super blocks. | |
6692 | */ | |
6693 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
6694 | { | |
6695 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6696 | struct btrfs_device *curr, *next; | |
6697 | ||
6698 | if (list_empty(&fs_devices->resized_devices)) | |
6699 | return; | |
6700 | ||
6701 | mutex_lock(&fs_devices->device_list_mutex); | |
6702 | lock_chunks(fs_info->dev_root); | |
6703 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
6704 | resized_list) { | |
6705 | list_del_init(&curr->resized_list); | |
6706 | curr->commit_total_bytes = curr->disk_total_bytes; | |
6707 | } | |
6708 | unlock_chunks(fs_info->dev_root); | |
6709 | mutex_unlock(&fs_devices->device_list_mutex); | |
6710 | } | |
ce7213c7 MX |
6711 | |
6712 | /* Must be invoked during the transaction commit */ | |
6713 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
6714 | struct btrfs_transaction *transaction) | |
6715 | { | |
6716 | struct extent_map *em; | |
6717 | struct map_lookup *map; | |
6718 | struct btrfs_device *dev; | |
6719 | int i; | |
6720 | ||
6721 | if (list_empty(&transaction->pending_chunks)) | |
6722 | return; | |
6723 | ||
6724 | /* In order to kick the device replace finish process */ | |
6725 | lock_chunks(root); | |
6726 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
6727 | map = (struct map_lookup *)em->bdev; | |
6728 | ||
6729 | for (i = 0; i < map->num_stripes; i++) { | |
6730 | dev = map->stripes[i].dev; | |
6731 | dev->commit_bytes_used = dev->bytes_used; | |
6732 | } | |
6733 | } | |
6734 | unlock_chunks(root); | |
6735 | } | |
5a13f430 AJ |
6736 | |
6737 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6738 | { | |
6739 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6740 | while (fs_devices) { | |
6741 | fs_devices->fs_info = fs_info; | |
6742 | fs_devices = fs_devices->seed; | |
6743 | } | |
6744 | } | |
6745 | ||
6746 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6747 | { | |
6748 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6749 | while (fs_devices) { | |
6750 | fs_devices->fs_info = NULL; | |
6751 | fs_devices = fs_devices->seed; | |
6752 | } | |
6753 | } |