]>
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> |
b765ead5 | 23 | #include <linux/iocontext.h> |
6f88a440 | 24 | #include <linux/capability.h> |
442a4f63 | 25 | #include <linux/ratelimit.h> |
59641015 | 26 | #include <linux/kthread.h> |
53b381b3 | 27 | #include <linux/raid/pq.h> |
803b2f54 | 28 | #include <linux/semaphore.h> |
8da4b8c4 | 29 | #include <linux/uuid.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 | |
af902047 ZL |
45 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
46 | [BTRFS_RAID_RAID10] = { | |
47 | .sub_stripes = 2, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 0, /* 0 == as many as possible */ | |
50 | .devs_min = 4, | |
8789f4fe | 51 | .tolerated_failures = 1, |
af902047 ZL |
52 | .devs_increment = 2, |
53 | .ncopies = 2, | |
54 | }, | |
55 | [BTRFS_RAID_RAID1] = { | |
56 | .sub_stripes = 1, | |
57 | .dev_stripes = 1, | |
58 | .devs_max = 2, | |
59 | .devs_min = 2, | |
8789f4fe | 60 | .tolerated_failures = 1, |
af902047 ZL |
61 | .devs_increment = 2, |
62 | .ncopies = 2, | |
63 | }, | |
64 | [BTRFS_RAID_DUP] = { | |
65 | .sub_stripes = 1, | |
66 | .dev_stripes = 2, | |
67 | .devs_max = 1, | |
68 | .devs_min = 1, | |
8789f4fe | 69 | .tolerated_failures = 0, |
af902047 ZL |
70 | .devs_increment = 1, |
71 | .ncopies = 2, | |
72 | }, | |
73 | [BTRFS_RAID_RAID0] = { | |
74 | .sub_stripes = 1, | |
75 | .dev_stripes = 1, | |
76 | .devs_max = 0, | |
77 | .devs_min = 2, | |
8789f4fe | 78 | .tolerated_failures = 0, |
af902047 ZL |
79 | .devs_increment = 1, |
80 | .ncopies = 1, | |
81 | }, | |
82 | [BTRFS_RAID_SINGLE] = { | |
83 | .sub_stripes = 1, | |
84 | .dev_stripes = 1, | |
85 | .devs_max = 1, | |
86 | .devs_min = 1, | |
8789f4fe | 87 | .tolerated_failures = 0, |
af902047 ZL |
88 | .devs_increment = 1, |
89 | .ncopies = 1, | |
90 | }, | |
91 | [BTRFS_RAID_RAID5] = { | |
92 | .sub_stripes = 1, | |
93 | .dev_stripes = 1, | |
94 | .devs_max = 0, | |
95 | .devs_min = 2, | |
8789f4fe | 96 | .tolerated_failures = 1, |
af902047 ZL |
97 | .devs_increment = 1, |
98 | .ncopies = 2, | |
99 | }, | |
100 | [BTRFS_RAID_RAID6] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 3, | |
8789f4fe | 105 | .tolerated_failures = 2, |
af902047 ZL |
106 | .devs_increment = 1, |
107 | .ncopies = 3, | |
108 | }, | |
109 | }; | |
110 | ||
fb75d857 | 111 | const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { |
af902047 ZL |
112 | [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10, |
113 | [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1, | |
114 | [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP, | |
115 | [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0, | |
116 | [BTRFS_RAID_SINGLE] = 0, | |
117 | [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5, | |
118 | [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6, | |
119 | }; | |
120 | ||
621292ba DS |
121 | /* |
122 | * Table to convert BTRFS_RAID_* to the error code if minimum number of devices | |
123 | * condition is not met. Zero means there's no corresponding | |
124 | * BTRFS_ERROR_DEV_*_NOT_MET value. | |
125 | */ | |
126 | const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = { | |
127 | [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, | |
128 | [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, | |
129 | [BTRFS_RAID_DUP] = 0, | |
130 | [BTRFS_RAID_RAID0] = 0, | |
131 | [BTRFS_RAID_SINGLE] = 0, | |
132 | [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, | |
133 | [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, | |
134 | }; | |
135 | ||
2b82032c | 136 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 137 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 138 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 139 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 140 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 141 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
142 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
143 | enum btrfs_map_op op, | |
144 | u64 logical, u64 *length, | |
145 | struct btrfs_bio **bbio_ret, | |
146 | int mirror_num, int need_raid_map); | |
2b82032c | 147 | |
9c6b1c4d DS |
148 | /* |
149 | * Device locking | |
150 | * ============== | |
151 | * | |
152 | * There are several mutexes that protect manipulation of devices and low-level | |
153 | * structures like chunks but not block groups, extents or files | |
154 | * | |
155 | * uuid_mutex (global lock) | |
156 | * ------------------------ | |
157 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
158 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
159 | * device) or requested by the device= mount option | |
160 | * | |
161 | * the mutex can be very coarse and can cover long-running operations | |
162 | * | |
163 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
164 | * seeding, structure cloning, openning/closing devices at mount/umount time | |
165 | * | |
166 | * global::fs_devs - add, remove, updates to the global list | |
167 | * | |
168 | * does not protect: manipulation of the fs_devices::devices list! | |
169 | * | |
170 | * btrfs_device::name - renames (write side), read is RCU | |
171 | * | |
172 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
173 | * ------------------------------------------------ | |
174 | * protects updates to fs_devices::devices, ie. adding and deleting | |
175 | * | |
176 | * simple list traversal with read-only actions can be done with RCU protection | |
177 | * | |
178 | * may be used to exclude some operations from running concurrently without any | |
179 | * modifications to the list (see write_all_supers) | |
180 | * | |
181 | * volume_mutex | |
182 | * ------------ | |
183 | * coarse lock owned by a mounted filesystem; used to exclude some operations | |
184 | * that cannot run in parallel and affect the higher-level properties of the | |
185 | * filesystem like: device add/deleting/resize/replace, or balance | |
186 | * | |
187 | * balance_mutex | |
188 | * ------------- | |
189 | * protects balance structures (status, state) and context accessed from | |
190 | * several places (internally, ioctl) | |
191 | * | |
192 | * chunk_mutex | |
193 | * ----------- | |
194 | * protects chunks, adding or removing during allocation, trim or when a new | |
195 | * device is added/removed | |
196 | * | |
197 | * cleaner_mutex | |
198 | * ------------- | |
199 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
200 | * cleaning together with relocation or delayed iputs | |
201 | * | |
202 | * | |
203 | * Lock nesting | |
204 | * ============ | |
205 | * | |
206 | * uuid_mutex | |
207 | * volume_mutex | |
208 | * device_list_mutex | |
209 | * chunk_mutex | |
210 | * balance_mutex | |
211 | */ | |
212 | ||
67a2c45e | 213 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 214 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
215 | struct list_head *btrfs_get_fs_uuids(void) |
216 | { | |
217 | return &fs_uuids; | |
218 | } | |
8a4b83cc | 219 | |
2dfeca9b DS |
220 | /* |
221 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
222 | * @fsid: if not NULL, copy the uuid to fs_devices::fsid | |
223 | * | |
224 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
225 | * The returned struct is not linked onto any lists and can be destroyed with | |
226 | * kfree() right away. | |
227 | */ | |
228 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
2208a378 ID |
229 | { |
230 | struct btrfs_fs_devices *fs_devs; | |
231 | ||
78f2c9e6 | 232 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
233 | if (!fs_devs) |
234 | return ERR_PTR(-ENOMEM); | |
235 | ||
236 | mutex_init(&fs_devs->device_list_mutex); | |
237 | ||
238 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 239 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
240 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
241 | INIT_LIST_HEAD(&fs_devs->list); | |
2208a378 ID |
242 | if (fsid) |
243 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 ID |
244 | |
245 | return fs_devs; | |
246 | } | |
247 | ||
48dae9cf DS |
248 | static void free_device(struct btrfs_device *device) |
249 | { | |
250 | rcu_string_free(device->name); | |
251 | bio_put(device->flush_bio); | |
252 | kfree(device); | |
253 | } | |
254 | ||
e4404d6e YZ |
255 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
256 | { | |
257 | struct btrfs_device *device; | |
258 | WARN_ON(fs_devices->opened); | |
259 | while (!list_empty(&fs_devices->devices)) { | |
260 | device = list_entry(fs_devices->devices.next, | |
261 | struct btrfs_device, dev_list); | |
262 | list_del(&device->dev_list); | |
55de4803 | 263 | free_device(device); |
e4404d6e YZ |
264 | } |
265 | kfree(fs_devices); | |
266 | } | |
267 | ||
b8b8ff59 LC |
268 | static void btrfs_kobject_uevent(struct block_device *bdev, |
269 | enum kobject_action action) | |
270 | { | |
271 | int ret; | |
272 | ||
273 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
274 | if (ret) | |
efe120a0 | 275 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
276 | action, |
277 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
278 | &disk_to_dev(bdev->bd_disk)->kobj); | |
279 | } | |
280 | ||
143bede5 | 281 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
282 | { |
283 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 284 | |
2b82032c YZ |
285 | while (!list_empty(&fs_uuids)) { |
286 | fs_devices = list_entry(fs_uuids.next, | |
287 | struct btrfs_fs_devices, list); | |
288 | list_del(&fs_devices->list); | |
e4404d6e | 289 | free_fs_devices(fs_devices); |
8a4b83cc | 290 | } |
8a4b83cc CM |
291 | } |
292 | ||
48dae9cf DS |
293 | /* |
294 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
295 | * Returned struct is not linked onto any lists and must be destroyed using | |
296 | * free_device. | |
297 | */ | |
12bd2fc0 ID |
298 | static struct btrfs_device *__alloc_device(void) |
299 | { | |
300 | struct btrfs_device *dev; | |
301 | ||
78f2c9e6 | 302 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
303 | if (!dev) |
304 | return ERR_PTR(-ENOMEM); | |
305 | ||
e0ae9994 DS |
306 | /* |
307 | * Preallocate a bio that's always going to be used for flushing device | |
308 | * barriers and matches the device lifespan | |
309 | */ | |
310 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
311 | if (!dev->flush_bio) { | |
312 | kfree(dev); | |
313 | return ERR_PTR(-ENOMEM); | |
314 | } | |
e0ae9994 | 315 | |
12bd2fc0 ID |
316 | INIT_LIST_HEAD(&dev->dev_list); |
317 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 318 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
319 | |
320 | spin_lock_init(&dev->io_lock); | |
321 | ||
12bd2fc0 | 322 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 323 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 324 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 325 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 326 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
327 | |
328 | return dev; | |
329 | } | |
330 | ||
35c70103 DS |
331 | /* |
332 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
333 | * return NULL. | |
334 | * | |
335 | * If devid and uuid are both specified, the match must be exact, otherwise | |
336 | * only devid is used. | |
337 | */ | |
338 | static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices, | |
339 | u64 devid, const u8 *uuid) | |
8a4b83cc | 340 | { |
35c70103 | 341 | struct list_head *head = &fs_devices->devices; |
8a4b83cc | 342 | struct btrfs_device *dev; |
8a4b83cc | 343 | |
c6e30871 | 344 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 345 | if (dev->devid == devid && |
8f18cf13 | 346 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 347 | return dev; |
a443755f | 348 | } |
8a4b83cc CM |
349 | } |
350 | return NULL; | |
351 | } | |
352 | ||
a1b32a59 | 353 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 354 | { |
8a4b83cc CM |
355 | struct btrfs_fs_devices *fs_devices; |
356 | ||
c6e30871 | 357 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
358 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
359 | return fs_devices; | |
360 | } | |
361 | return NULL; | |
362 | } | |
363 | ||
beaf8ab3 SB |
364 | static int |
365 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
366 | int flush, struct block_device **bdev, | |
367 | struct buffer_head **bh) | |
368 | { | |
369 | int ret; | |
370 | ||
371 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
372 | ||
373 | if (IS_ERR(*bdev)) { | |
374 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
375 | goto error; |
376 | } | |
377 | ||
378 | if (flush) | |
379 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 380 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
381 | if (ret) { |
382 | blkdev_put(*bdev, flags); | |
383 | goto error; | |
384 | } | |
385 | invalidate_bdev(*bdev); | |
386 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
387 | if (IS_ERR(*bh)) { |
388 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
389 | blkdev_put(*bdev, flags); |
390 | goto error; | |
391 | } | |
392 | ||
393 | return 0; | |
394 | ||
395 | error: | |
396 | *bdev = NULL; | |
397 | *bh = NULL; | |
398 | return ret; | |
399 | } | |
400 | ||
ffbd517d CM |
401 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
402 | struct bio *head, struct bio *tail) | |
403 | { | |
404 | ||
405 | struct bio *old_head; | |
406 | ||
407 | old_head = pending_bios->head; | |
408 | pending_bios->head = head; | |
409 | if (pending_bios->tail) | |
410 | tail->bi_next = old_head; | |
411 | else | |
412 | pending_bios->tail = tail; | |
413 | } | |
414 | ||
8b712842 CM |
415 | /* |
416 | * we try to collect pending bios for a device so we don't get a large | |
417 | * number of procs sending bios down to the same device. This greatly | |
418 | * improves the schedulers ability to collect and merge the bios. | |
419 | * | |
420 | * But, it also turns into a long list of bios to process and that is sure | |
421 | * to eventually make the worker thread block. The solution here is to | |
422 | * make some progress and then put this work struct back at the end of | |
423 | * the list if the block device is congested. This way, multiple devices | |
424 | * can make progress from a single worker thread. | |
425 | */ | |
143bede5 | 426 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 427 | { |
0b246afa | 428 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
429 | struct bio *pending; |
430 | struct backing_dev_info *bdi; | |
ffbd517d | 431 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
432 | struct bio *tail; |
433 | struct bio *cur; | |
434 | int again = 0; | |
ffbd517d | 435 | unsigned long num_run; |
d644d8a1 | 436 | unsigned long batch_run = 0; |
b765ead5 | 437 | unsigned long last_waited = 0; |
d84275c9 | 438 | int force_reg = 0; |
0e588859 | 439 | int sync_pending = 0; |
211588ad CM |
440 | struct blk_plug plug; |
441 | ||
442 | /* | |
443 | * this function runs all the bios we've collected for | |
444 | * a particular device. We don't want to wander off to | |
445 | * another device without first sending all of these down. | |
446 | * So, setup a plug here and finish it off before we return | |
447 | */ | |
448 | blk_start_plug(&plug); | |
8b712842 | 449 | |
efa7c9f9 | 450 | bdi = device->bdev->bd_bdi; |
b64a2851 | 451 | |
8b712842 CM |
452 | loop: |
453 | spin_lock(&device->io_lock); | |
454 | ||
a6837051 | 455 | loop_lock: |
d84275c9 | 456 | num_run = 0; |
ffbd517d | 457 | |
8b712842 CM |
458 | /* take all the bios off the list at once and process them |
459 | * later on (without the lock held). But, remember the | |
460 | * tail and other pointers so the bios can be properly reinserted | |
461 | * into the list if we hit congestion | |
462 | */ | |
d84275c9 | 463 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 464 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
465 | force_reg = 1; |
466 | } else { | |
ffbd517d | 467 | pending_bios = &device->pending_bios; |
d84275c9 CM |
468 | force_reg = 0; |
469 | } | |
ffbd517d CM |
470 | |
471 | pending = pending_bios->head; | |
472 | tail = pending_bios->tail; | |
8b712842 | 473 | WARN_ON(pending && !tail); |
8b712842 CM |
474 | |
475 | /* | |
476 | * if pending was null this time around, no bios need processing | |
477 | * at all and we can stop. Otherwise it'll loop back up again | |
478 | * and do an additional check so no bios are missed. | |
479 | * | |
480 | * device->running_pending is used to synchronize with the | |
481 | * schedule_bio code. | |
482 | */ | |
ffbd517d CM |
483 | if (device->pending_sync_bios.head == NULL && |
484 | device->pending_bios.head == NULL) { | |
8b712842 CM |
485 | again = 0; |
486 | device->running_pending = 0; | |
ffbd517d CM |
487 | } else { |
488 | again = 1; | |
489 | device->running_pending = 1; | |
8b712842 | 490 | } |
ffbd517d CM |
491 | |
492 | pending_bios->head = NULL; | |
493 | pending_bios->tail = NULL; | |
494 | ||
8b712842 CM |
495 | spin_unlock(&device->io_lock); |
496 | ||
d397712b | 497 | while (pending) { |
ffbd517d CM |
498 | |
499 | rmb(); | |
d84275c9 CM |
500 | /* we want to work on both lists, but do more bios on the |
501 | * sync list than the regular list | |
502 | */ | |
503 | if ((num_run > 32 && | |
504 | pending_bios != &device->pending_sync_bios && | |
505 | device->pending_sync_bios.head) || | |
506 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
507 | device->pending_bios.head)) { | |
ffbd517d CM |
508 | spin_lock(&device->io_lock); |
509 | requeue_list(pending_bios, pending, tail); | |
510 | goto loop_lock; | |
511 | } | |
512 | ||
8b712842 CM |
513 | cur = pending; |
514 | pending = pending->bi_next; | |
515 | cur->bi_next = NULL; | |
b64a2851 | 516 | |
dac56212 | 517 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 518 | |
2ab1ba68 CM |
519 | /* |
520 | * if we're doing the sync list, record that our | |
521 | * plug has some sync requests on it | |
522 | * | |
523 | * If we're doing the regular list and there are | |
524 | * sync requests sitting around, unplug before | |
525 | * we add more | |
526 | */ | |
527 | if (pending_bios == &device->pending_sync_bios) { | |
528 | sync_pending = 1; | |
529 | } else if (sync_pending) { | |
530 | blk_finish_plug(&plug); | |
531 | blk_start_plug(&plug); | |
532 | sync_pending = 0; | |
533 | } | |
534 | ||
4e49ea4a | 535 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
536 | num_run++; |
537 | batch_run++; | |
853d8ec4 DS |
538 | |
539 | cond_resched(); | |
8b712842 CM |
540 | |
541 | /* | |
542 | * we made progress, there is more work to do and the bdi | |
543 | * is now congested. Back off and let other work structs | |
544 | * run instead | |
545 | */ | |
57fd5a5f | 546 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 547 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 548 | struct io_context *ioc; |
8b712842 | 549 | |
b765ead5 CM |
550 | ioc = current->io_context; |
551 | ||
552 | /* | |
553 | * the main goal here is that we don't want to | |
554 | * block if we're going to be able to submit | |
555 | * more requests without blocking. | |
556 | * | |
557 | * This code does two great things, it pokes into | |
558 | * the elevator code from a filesystem _and_ | |
559 | * it makes assumptions about how batching works. | |
560 | */ | |
561 | if (ioc && ioc->nr_batch_requests > 0 && | |
562 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
563 | (last_waited == 0 || | |
564 | ioc->last_waited == last_waited)) { | |
565 | /* | |
566 | * we want to go through our batch of | |
567 | * requests and stop. So, we copy out | |
568 | * the ioc->last_waited time and test | |
569 | * against it before looping | |
570 | */ | |
571 | last_waited = ioc->last_waited; | |
853d8ec4 | 572 | cond_resched(); |
b765ead5 CM |
573 | continue; |
574 | } | |
8b712842 | 575 | spin_lock(&device->io_lock); |
ffbd517d | 576 | requeue_list(pending_bios, pending, tail); |
a6837051 | 577 | device->running_pending = 1; |
8b712842 CM |
578 | |
579 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
580 | btrfs_queue_work(fs_info->submit_workers, |
581 | &device->work); | |
8b712842 CM |
582 | goto done; |
583 | } | |
584 | } | |
ffbd517d | 585 | |
51684082 CM |
586 | cond_resched(); |
587 | if (again) | |
588 | goto loop; | |
589 | ||
590 | spin_lock(&device->io_lock); | |
591 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
592 | goto loop_lock; | |
593 | spin_unlock(&device->io_lock); | |
594 | ||
8b712842 | 595 | done: |
211588ad | 596 | blk_finish_plug(&plug); |
8b712842 CM |
597 | } |
598 | ||
b2950863 | 599 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
600 | { |
601 | struct btrfs_device *device; | |
602 | ||
603 | device = container_of(work, struct btrfs_device, work); | |
604 | run_scheduled_bios(device); | |
605 | } | |
606 | ||
d8367db3 AJ |
607 | /* |
608 | * Search and remove all stale (devices which are not mounted) devices. | |
609 | * When both inputs are NULL, it will search and release all stale devices. | |
610 | * path: Optional. When provided will it release all unmounted devices | |
611 | * matching this path only. | |
612 | * skip_dev: Optional. Will skip this device when searching for the stale | |
613 | * devices. | |
614 | */ | |
615 | static void btrfs_free_stale_devices(const char *path, | |
616 | struct btrfs_device *skip_dev) | |
4fde46f0 | 617 | { |
38cf665d AJ |
618 | struct btrfs_fs_devices *fs_devs, *tmp_fs_devs; |
619 | struct btrfs_device *dev, *tmp_dev; | |
4fde46f0 | 620 | |
38cf665d | 621 | list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, list) { |
4fde46f0 AJ |
622 | |
623 | if (fs_devs->opened) | |
624 | continue; | |
4fde46f0 | 625 | |
38cf665d AJ |
626 | list_for_each_entry_safe(dev, tmp_dev, |
627 | &fs_devs->devices, dev_list) { | |
522f1b45 | 628 | int not_found = 0; |
4fde46f0 | 629 | |
d8367db3 AJ |
630 | if (skip_dev && skip_dev == dev) |
631 | continue; | |
632 | if (path && !dev->name) | |
4fde46f0 AJ |
633 | continue; |
634 | ||
4fde46f0 | 635 | rcu_read_lock(); |
d8367db3 | 636 | if (path) |
522f1b45 | 637 | not_found = strcmp(rcu_str_deref(dev->name), |
d8367db3 | 638 | path); |
4fde46f0 | 639 | rcu_read_unlock(); |
38cf665d AJ |
640 | if (not_found) |
641 | continue; | |
4fde46f0 | 642 | |
4fde46f0 AJ |
643 | /* delete the stale device */ |
644 | if (fs_devs->num_devices == 1) { | |
645 | btrfs_sysfs_remove_fsid(fs_devs); | |
646 | list_del(&fs_devs->list); | |
647 | free_fs_devices(fs_devs); | |
648 | } else { | |
649 | fs_devs->num_devices--; | |
650 | list_del(&dev->dev_list); | |
55de4803 | 651 | free_device(dev); |
4fde46f0 | 652 | } |
4fde46f0 AJ |
653 | } |
654 | } | |
655 | } | |
656 | ||
0fb08bcc AJ |
657 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
658 | struct btrfs_device *device, fmode_t flags, | |
659 | void *holder) | |
660 | { | |
661 | struct request_queue *q; | |
662 | struct block_device *bdev; | |
663 | struct buffer_head *bh; | |
664 | struct btrfs_super_block *disk_super; | |
665 | u64 devid; | |
666 | int ret; | |
667 | ||
668 | if (device->bdev) | |
669 | return -EINVAL; | |
670 | if (!device->name) | |
671 | return -EINVAL; | |
672 | ||
673 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
674 | &bdev, &bh); | |
675 | if (ret) | |
676 | return ret; | |
677 | ||
678 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
679 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
680 | if (devid != device->devid) | |
681 | goto error_brelse; | |
682 | ||
683 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
684 | goto error_brelse; | |
685 | ||
686 | device->generation = btrfs_super_generation(disk_super); | |
687 | ||
688 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
ebbede42 | 689 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
690 | fs_devices->seeding = 1; |
691 | } else { | |
ebbede42 AJ |
692 | if (bdev_read_only(bdev)) |
693 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
694 | else | |
695 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
696 | } |
697 | ||
698 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
699 | if (!blk_queue_nonrot(q)) |
700 | fs_devices->rotating = 1; | |
701 | ||
702 | device->bdev = bdev; | |
e12c9621 | 703 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
704 | device->mode = flags; |
705 | ||
706 | fs_devices->open_devices++; | |
ebbede42 AJ |
707 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
708 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc AJ |
709 | fs_devices->rw_devices++; |
710 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
711 | } | |
712 | brelse(bh); | |
713 | ||
714 | return 0; | |
715 | ||
716 | error_brelse: | |
717 | brelse(bh); | |
718 | blkdev_put(bdev, flags); | |
719 | ||
720 | return -EINVAL; | |
721 | } | |
722 | ||
60999ca4 DS |
723 | /* |
724 | * Add new device to list of registered devices | |
725 | * | |
726 | * Returns: | |
727 | * 1 - first time device is seen | |
728 | * 0 - device already known | |
729 | * < 0 - error | |
730 | */ | |
a1b32a59 | 731 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
732 | struct btrfs_super_block *disk_super, |
733 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
734 | { | |
735 | struct btrfs_device *device; | |
736 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 737 | struct rcu_string *name; |
60999ca4 | 738 | int ret = 0; |
8a4b83cc CM |
739 | u64 found_transid = btrfs_super_generation(disk_super); |
740 | ||
741 | fs_devices = find_fsid(disk_super->fsid); | |
742 | if (!fs_devices) { | |
2208a378 ID |
743 | fs_devices = alloc_fs_devices(disk_super->fsid); |
744 | if (IS_ERR(fs_devices)) | |
745 | return PTR_ERR(fs_devices); | |
746 | ||
8a4b83cc | 747 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 748 | |
8a4b83cc CM |
749 | device = NULL; |
750 | } else { | |
35c70103 DS |
751 | device = find_device(fs_devices, devid, |
752 | disk_super->dev_item.uuid); | |
8a4b83cc | 753 | } |
443f24fe | 754 | |
8a4b83cc | 755 | if (!device) { |
2b82032c YZ |
756 | if (fs_devices->opened) |
757 | return -EBUSY; | |
758 | ||
12bd2fc0 ID |
759 | device = btrfs_alloc_device(NULL, &devid, |
760 | disk_super->dev_item.uuid); | |
761 | if (IS_ERR(device)) { | |
8a4b83cc | 762 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 763 | return PTR_ERR(device); |
8a4b83cc | 764 | } |
606686ee JB |
765 | |
766 | name = rcu_string_strdup(path, GFP_NOFS); | |
767 | if (!name) { | |
55de4803 | 768 | free_device(device); |
8a4b83cc CM |
769 | return -ENOMEM; |
770 | } | |
606686ee | 771 | rcu_assign_pointer(device->name, name); |
90519d66 | 772 | |
e5e9a520 | 773 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 774 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 775 | fs_devices->num_devices++; |
e5e9a520 CM |
776 | mutex_unlock(&fs_devices->device_list_mutex); |
777 | ||
60999ca4 | 778 | ret = 1; |
2b82032c | 779 | device->fs_devices = fs_devices; |
d8367db3 | 780 | btrfs_free_stale_devices(path, device); |
606686ee | 781 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
782 | /* |
783 | * When FS is already mounted. | |
784 | * 1. If you are here and if the device->name is NULL that | |
785 | * means this device was missing at time of FS mount. | |
786 | * 2. If you are here and if the device->name is different | |
787 | * from 'path' that means either | |
788 | * a. The same device disappeared and reappeared with | |
789 | * different name. or | |
790 | * b. The missing-disk-which-was-replaced, has | |
791 | * reappeared now. | |
792 | * | |
793 | * We must allow 1 and 2a above. But 2b would be a spurious | |
794 | * and unintentional. | |
795 | * | |
796 | * Further in case of 1 and 2a above, the disk at 'path' | |
797 | * would have missed some transaction when it was away and | |
798 | * in case of 2a the stale bdev has to be updated as well. | |
799 | * 2b must not be allowed at all time. | |
800 | */ | |
801 | ||
802 | /* | |
0f23ae74 CM |
803 | * For now, we do allow update to btrfs_fs_device through the |
804 | * btrfs dev scan cli after FS has been mounted. We're still | |
805 | * tracking a problem where systems fail mount by subvolume id | |
806 | * when we reject replacement on a mounted FS. | |
b96de000 | 807 | */ |
0f23ae74 | 808 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
809 | /* |
810 | * That is if the FS is _not_ mounted and if you | |
811 | * are here, that means there is more than one | |
812 | * disk with same uuid and devid.We keep the one | |
813 | * with larger generation number or the last-in if | |
814 | * generation are equal. | |
815 | */ | |
0f23ae74 | 816 | return -EEXIST; |
77bdae4d | 817 | } |
b96de000 | 818 | |
606686ee | 819 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
820 | if (!name) |
821 | return -ENOMEM; | |
606686ee JB |
822 | rcu_string_free(device->name); |
823 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 824 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 825 | fs_devices->missing_devices--; |
e6e674bd | 826 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 827 | } |
8a4b83cc CM |
828 | } |
829 | ||
77bdae4d AJ |
830 | /* |
831 | * Unmount does not free the btrfs_device struct but would zero | |
832 | * generation along with most of the other members. So just update | |
833 | * it back. We need it to pick the disk with largest generation | |
834 | * (as above). | |
835 | */ | |
836 | if (!fs_devices->opened) | |
837 | device->generation = found_transid; | |
838 | ||
8a4b83cc | 839 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
840 | |
841 | return ret; | |
8a4b83cc CM |
842 | } |
843 | ||
e4404d6e YZ |
844 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
845 | { | |
846 | struct btrfs_fs_devices *fs_devices; | |
847 | struct btrfs_device *device; | |
848 | struct btrfs_device *orig_dev; | |
849 | ||
2208a378 ID |
850 | fs_devices = alloc_fs_devices(orig->fsid); |
851 | if (IS_ERR(fs_devices)) | |
852 | return fs_devices; | |
e4404d6e | 853 | |
adbbb863 | 854 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 855 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 856 | |
46224705 | 857 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 858 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
859 | struct rcu_string *name; |
860 | ||
12bd2fc0 ID |
861 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
862 | orig_dev->uuid); | |
863 | if (IS_ERR(device)) | |
e4404d6e YZ |
864 | goto error; |
865 | ||
606686ee JB |
866 | /* |
867 | * This is ok to do without rcu read locked because we hold the | |
868 | * uuid mutex so nothing we touch in here is going to disappear. | |
869 | */ | |
e755f780 | 870 | if (orig_dev->name) { |
78f2c9e6 DS |
871 | name = rcu_string_strdup(orig_dev->name->str, |
872 | GFP_KERNEL); | |
e755f780 | 873 | if (!name) { |
55de4803 | 874 | free_device(device); |
e755f780 AJ |
875 | goto error; |
876 | } | |
877 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 878 | } |
e4404d6e | 879 | |
e4404d6e YZ |
880 | list_add(&device->dev_list, &fs_devices->devices); |
881 | device->fs_devices = fs_devices; | |
882 | fs_devices->num_devices++; | |
883 | } | |
adbbb863 | 884 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
885 | return fs_devices; |
886 | error: | |
adbbb863 | 887 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
888 | free_fs_devices(fs_devices); |
889 | return ERR_PTR(-ENOMEM); | |
890 | } | |
891 | ||
9eaed21e | 892 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 893 | { |
c6e30871 | 894 | struct btrfs_device *device, *next; |
443f24fe | 895 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 896 | |
dfe25020 CM |
897 | mutex_lock(&uuid_mutex); |
898 | again: | |
46224705 | 899 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 900 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
901 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
902 | &device->dev_state)) { | |
401e29c1 AJ |
903 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
904 | &device->dev_state) && | |
905 | (!latest_dev || | |
906 | device->generation > latest_dev->generation)) { | |
443f24fe | 907 | latest_dev = device; |
a6b0d5c8 | 908 | } |
2b82032c | 909 | continue; |
a6b0d5c8 | 910 | } |
2b82032c | 911 | |
8dabb742 SB |
912 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
913 | /* | |
914 | * In the first step, keep the device which has | |
915 | * the correct fsid and the devid that is used | |
916 | * for the dev_replace procedure. | |
917 | * In the second step, the dev_replace state is | |
918 | * read from the device tree and it is known | |
919 | * whether the procedure is really active or | |
920 | * not, which means whether this device is | |
921 | * used or whether it should be removed. | |
922 | */ | |
401e29c1 AJ |
923 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
924 | &device->dev_state)) { | |
8dabb742 SB |
925 | continue; |
926 | } | |
927 | } | |
2b82032c | 928 | if (device->bdev) { |
d4d77629 | 929 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
930 | device->bdev = NULL; |
931 | fs_devices->open_devices--; | |
932 | } | |
ebbede42 | 933 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 934 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 935 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
936 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
937 | &device->dev_state)) | |
8dabb742 | 938 | fs_devices->rw_devices--; |
2b82032c | 939 | } |
e4404d6e YZ |
940 | list_del_init(&device->dev_list); |
941 | fs_devices->num_devices--; | |
55de4803 | 942 | free_device(device); |
dfe25020 | 943 | } |
2b82032c YZ |
944 | |
945 | if (fs_devices->seed) { | |
946 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
947 | goto again; |
948 | } | |
949 | ||
443f24fe | 950 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 951 | |
dfe25020 | 952 | mutex_unlock(&uuid_mutex); |
dfe25020 | 953 | } |
a0af469b | 954 | |
f06c5965 | 955 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
956 | { |
957 | struct btrfs_device *device; | |
958 | ||
9f5316c1 | 959 | device = container_of(head, struct btrfs_device, rcu); |
55de4803 | 960 | free_device(device); |
1f78160c XG |
961 | } |
962 | ||
14238819 AJ |
963 | static void btrfs_close_bdev(struct btrfs_device *device) |
964 | { | |
08ffcae8 DS |
965 | if (!device->bdev) |
966 | return; | |
967 | ||
ebbede42 | 968 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
969 | sync_blockdev(device->bdev); |
970 | invalidate_bdev(device->bdev); | |
971 | } | |
972 | ||
08ffcae8 | 973 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
974 | } |
975 | ||
0ccd0528 | 976 | static void btrfs_prepare_close_one_device(struct btrfs_device *device) |
f448341a AJ |
977 | { |
978 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
979 | struct btrfs_device *new_device; | |
980 | struct rcu_string *name; | |
981 | ||
982 | if (device->bdev) | |
983 | fs_devices->open_devices--; | |
984 | ||
ebbede42 | 985 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
986 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
987 | list_del_init(&device->dev_alloc_list); | |
988 | fs_devices->rw_devices--; | |
989 | } | |
990 | ||
e6e674bd | 991 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
992 | fs_devices->missing_devices--; |
993 | ||
994 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
995 | device->uuid); | |
996 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
997 | ||
998 | /* Safe because we are under uuid_mutex */ | |
999 | if (device->name) { | |
1000 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1001 | BUG_ON(!name); /* -ENOMEM */ | |
1002 | rcu_assign_pointer(new_device->name, name); | |
1003 | } | |
1004 | ||
1005 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1006 | new_device->fs_devices = device->fs_devices; | |
f448341a AJ |
1007 | } |
1008 | ||
2b82032c | 1009 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1010 | { |
2037a093 | 1011 | struct btrfs_device *device, *tmp; |
0ccd0528 AJ |
1012 | struct list_head pending_put; |
1013 | ||
1014 | INIT_LIST_HEAD(&pending_put); | |
e4404d6e | 1015 | |
2b82032c YZ |
1016 | if (--fs_devices->opened > 0) |
1017 | return 0; | |
8a4b83cc | 1018 | |
c9513edb | 1019 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1020 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
0ccd0528 AJ |
1021 | btrfs_prepare_close_one_device(device); |
1022 | list_add(&device->dev_list, &pending_put); | |
8a4b83cc | 1023 | } |
c9513edb XG |
1024 | mutex_unlock(&fs_devices->device_list_mutex); |
1025 | ||
0ccd0528 AJ |
1026 | /* |
1027 | * btrfs_show_devname() is using the device_list_mutex, | |
1028 | * sometimes call to blkdev_put() leads vfs calling | |
1029 | * into this func. So do put outside of device_list_mutex, | |
1030 | * as of now. | |
1031 | */ | |
1032 | while (!list_empty(&pending_put)) { | |
1033 | device = list_first_entry(&pending_put, | |
1034 | struct btrfs_device, dev_list); | |
1035 | list_del(&device->dev_list); | |
1036 | btrfs_close_bdev(device); | |
f06c5965 | 1037 | call_rcu(&device->rcu, free_device_rcu); |
0ccd0528 AJ |
1038 | } |
1039 | ||
e4404d6e YZ |
1040 | WARN_ON(fs_devices->open_devices); |
1041 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1042 | fs_devices->opened = 0; |
1043 | fs_devices->seeding = 0; | |
2b82032c | 1044 | |
8a4b83cc CM |
1045 | return 0; |
1046 | } | |
1047 | ||
2b82032c YZ |
1048 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1049 | { | |
e4404d6e | 1050 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1051 | int ret; |
1052 | ||
1053 | mutex_lock(&uuid_mutex); | |
1054 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
1055 | if (!fs_devices->opened) { |
1056 | seed_devices = fs_devices->seed; | |
1057 | fs_devices->seed = NULL; | |
1058 | } | |
2b82032c | 1059 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1060 | |
1061 | while (seed_devices) { | |
1062 | fs_devices = seed_devices; | |
1063 | seed_devices = fs_devices->seed; | |
1064 | __btrfs_close_devices(fs_devices); | |
1065 | free_fs_devices(fs_devices); | |
1066 | } | |
2b82032c YZ |
1067 | return ret; |
1068 | } | |
1069 | ||
e4404d6e YZ |
1070 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
1071 | fmode_t flags, void *holder) | |
8a4b83cc | 1072 | { |
8a4b83cc | 1073 | struct list_head *head = &fs_devices->devices; |
8a4b83cc | 1074 | struct btrfs_device *device; |
443f24fe | 1075 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1076 | int ret = 0; |
8a4b83cc | 1077 | |
d4d77629 TH |
1078 | flags |= FMODE_EXCL; |
1079 | ||
c6e30871 | 1080 | list_for_each_entry(device, head, dev_list) { |
f63e0cca | 1081 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1082 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1083 | continue; |
a0af469b | 1084 | |
9f050db4 AJ |
1085 | if (!latest_dev || |
1086 | device->generation > latest_dev->generation) | |
1087 | latest_dev = device; | |
8a4b83cc | 1088 | } |
a0af469b | 1089 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1090 | ret = -EINVAL; |
a0af469b CM |
1091 | goto out; |
1092 | } | |
2b82032c | 1093 | fs_devices->opened = 1; |
443f24fe | 1094 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1095 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1096 | out: |
2b82032c YZ |
1097 | return ret; |
1098 | } | |
1099 | ||
1100 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 1101 | fmode_t flags, void *holder) |
2b82032c YZ |
1102 | { |
1103 | int ret; | |
1104 | ||
1105 | mutex_lock(&uuid_mutex); | |
1106 | if (fs_devices->opened) { | |
e4404d6e YZ |
1107 | fs_devices->opened++; |
1108 | ret = 0; | |
2b82032c | 1109 | } else { |
15916de8 | 1110 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1111 | } |
8a4b83cc | 1112 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1113 | return ret; |
1114 | } | |
1115 | ||
c9162bdf | 1116 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1117 | { |
1118 | kunmap(page); | |
1119 | put_page(page); | |
1120 | } | |
1121 | ||
c9162bdf OS |
1122 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1123 | struct page **page, | |
1124 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1125 | { |
1126 | void *p; | |
1127 | pgoff_t index; | |
1128 | ||
1129 | /* make sure our super fits in the device */ | |
1130 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1131 | return 1; | |
1132 | ||
1133 | /* make sure our super fits in the page */ | |
1134 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1135 | return 1; | |
1136 | ||
1137 | /* make sure our super doesn't straddle pages on disk */ | |
1138 | index = bytenr >> PAGE_SHIFT; | |
1139 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1140 | return 1; | |
1141 | ||
1142 | /* pull in the page with our super */ | |
1143 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1144 | index, GFP_KERNEL); | |
1145 | ||
1146 | if (IS_ERR_OR_NULL(*page)) | |
1147 | return 1; | |
1148 | ||
1149 | p = kmap(*page); | |
1150 | ||
1151 | /* align our pointer to the offset of the super block */ | |
1152 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1153 | ||
1154 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1155 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1156 | btrfs_release_disk_super(*page); | |
1157 | return 1; | |
1158 | } | |
1159 | ||
1160 | if ((*disk_super)->label[0] && | |
1161 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1162 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1163 | ||
1164 | return 0; | |
1165 | } | |
1166 | ||
6f60cbd3 DS |
1167 | /* |
1168 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1169 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1170 | * is read via pagecache | |
1171 | */ | |
97288f2c | 1172 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1173 | struct btrfs_fs_devices **fs_devices_ret) |
1174 | { | |
1175 | struct btrfs_super_block *disk_super; | |
1176 | struct block_device *bdev; | |
6f60cbd3 | 1177 | struct page *page; |
05a5c55d | 1178 | int ret; |
8a4b83cc | 1179 | u64 devid; |
f2984462 | 1180 | u64 transid; |
02db0844 | 1181 | u64 total_devices; |
6f60cbd3 | 1182 | u64 bytenr; |
8a4b83cc | 1183 | |
6f60cbd3 DS |
1184 | /* |
1185 | * we would like to check all the supers, but that would make | |
1186 | * a btrfs mount succeed after a mkfs from a different FS. | |
1187 | * So, we need to add a special mount option to scan for | |
1188 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1189 | */ | |
1190 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1191 | flags |= FMODE_EXCL; |
10f6327b | 1192 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1193 | |
1194 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1195 | if (IS_ERR(bdev)) { |
1196 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1197 | goto error; |
6f60cbd3 DS |
1198 | } |
1199 | ||
05a5c55d AJ |
1200 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
1201 | ret = -EINVAL; | |
6f60cbd3 | 1202 | goto error_bdev_put; |
05a5c55d | 1203 | } |
6f60cbd3 | 1204 | |
a343832f | 1205 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1206 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1207 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1208 | |
8a4b83cc | 1209 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 | 1210 | if (ret > 0) { |
05a5c55d | 1211 | if (disk_super->label[0]) |
62e85577 | 1212 | pr_info("BTRFS: device label %s ", disk_super->label); |
05a5c55d | 1213 | else |
62e85577 | 1214 | pr_info("BTRFS: device fsid %pU ", disk_super->fsid); |
60999ca4 | 1215 | |
62e85577 | 1216 | pr_cont("devid %llu transid %llu %s\n", devid, transid, path); |
60999ca4 DS |
1217 | ret = 0; |
1218 | } | |
02db0844 JB |
1219 | if (!ret && fs_devices_ret) |
1220 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 | 1221 | |
6cf86a00 | 1222 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1223 | |
1224 | error_bdev_put: | |
d4d77629 | 1225 | blkdev_put(bdev, flags); |
8a4b83cc | 1226 | error: |
beaf8ab3 | 1227 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1228 | return ret; |
1229 | } | |
0b86a832 | 1230 | |
6d07bcec MX |
1231 | /* helper to account the used device space in the range */ |
1232 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1233 | u64 end, u64 *length) | |
1234 | { | |
1235 | struct btrfs_key key; | |
fb456252 | 1236 | struct btrfs_root *root = device->fs_info->dev_root; |
6d07bcec MX |
1237 | struct btrfs_dev_extent *dev_extent; |
1238 | struct btrfs_path *path; | |
1239 | u64 extent_end; | |
1240 | int ret; | |
1241 | int slot; | |
1242 | struct extent_buffer *l; | |
1243 | ||
1244 | *length = 0; | |
1245 | ||
401e29c1 AJ |
1246 | if (start >= device->total_bytes || |
1247 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) | |
6d07bcec MX |
1248 | return 0; |
1249 | ||
1250 | path = btrfs_alloc_path(); | |
1251 | if (!path) | |
1252 | return -ENOMEM; | |
e4058b54 | 1253 | path->reada = READA_FORWARD; |
6d07bcec MX |
1254 | |
1255 | key.objectid = device->devid; | |
1256 | key.offset = start; | |
1257 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1258 | ||
1259 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1260 | if (ret < 0) | |
1261 | goto out; | |
1262 | if (ret > 0) { | |
1263 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1264 | if (ret < 0) | |
1265 | goto out; | |
1266 | } | |
1267 | ||
1268 | while (1) { | |
1269 | l = path->nodes[0]; | |
1270 | slot = path->slots[0]; | |
1271 | if (slot >= btrfs_header_nritems(l)) { | |
1272 | ret = btrfs_next_leaf(root, path); | |
1273 | if (ret == 0) | |
1274 | continue; | |
1275 | if (ret < 0) | |
1276 | goto out; | |
1277 | ||
1278 | break; | |
1279 | } | |
1280 | btrfs_item_key_to_cpu(l, &key, slot); | |
1281 | ||
1282 | if (key.objectid < device->devid) | |
1283 | goto next; | |
1284 | ||
1285 | if (key.objectid > device->devid) | |
1286 | break; | |
1287 | ||
962a298f | 1288 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1289 | goto next; |
1290 | ||
1291 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1292 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1293 | dev_extent); | |
1294 | if (key.offset <= start && extent_end > end) { | |
1295 | *length = end - start + 1; | |
1296 | break; | |
1297 | } else if (key.offset <= start && extent_end > start) | |
1298 | *length += extent_end - start; | |
1299 | else if (key.offset > start && extent_end <= end) | |
1300 | *length += extent_end - key.offset; | |
1301 | else if (key.offset > start && key.offset <= end) { | |
1302 | *length += end - key.offset + 1; | |
1303 | break; | |
1304 | } else if (key.offset > end) | |
1305 | break; | |
1306 | ||
1307 | next: | |
1308 | path->slots[0]++; | |
1309 | } | |
1310 | ret = 0; | |
1311 | out: | |
1312 | btrfs_free_path(path); | |
1313 | return ret; | |
1314 | } | |
1315 | ||
499f377f | 1316 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1317 | struct btrfs_device *device, |
1318 | u64 *start, u64 len) | |
1319 | { | |
fb456252 | 1320 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1321 | struct extent_map *em; |
499f377f | 1322 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1323 | int ret = 0; |
1b984508 | 1324 | u64 physical_start = *start; |
6df9a95e | 1325 | |
499f377f JM |
1326 | if (transaction) |
1327 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1328 | again: |
1329 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1330 | struct map_lookup *map; |
1331 | int i; | |
1332 | ||
95617d69 | 1333 | map = em->map_lookup; |
6df9a95e | 1334 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1335 | u64 end; |
1336 | ||
6df9a95e JB |
1337 | if (map->stripes[i].dev != device) |
1338 | continue; | |
1b984508 | 1339 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1340 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1341 | physical_start) |
6df9a95e | 1342 | continue; |
c152b63e FM |
1343 | /* |
1344 | * Make sure that while processing the pinned list we do | |
1345 | * not override our *start with a lower value, because | |
1346 | * we can have pinned chunks that fall within this | |
1347 | * device hole and that have lower physical addresses | |
1348 | * than the pending chunks we processed before. If we | |
1349 | * do not take this special care we can end up getting | |
1350 | * 2 pending chunks that start at the same physical | |
1351 | * device offsets because the end offset of a pinned | |
1352 | * chunk can be equal to the start offset of some | |
1353 | * pending chunk. | |
1354 | */ | |
1355 | end = map->stripes[i].physical + em->orig_block_len; | |
1356 | if (end > *start) { | |
1357 | *start = end; | |
1358 | ret = 1; | |
1359 | } | |
6df9a95e JB |
1360 | } |
1361 | } | |
499f377f JM |
1362 | if (search_list != &fs_info->pinned_chunks) { |
1363 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1364 | goto again; |
1365 | } | |
6df9a95e JB |
1366 | |
1367 | return ret; | |
1368 | } | |
1369 | ||
1370 | ||
0b86a832 | 1371 | /* |
499f377f JM |
1372 | * find_free_dev_extent_start - find free space in the specified device |
1373 | * @device: the device which we search the free space in | |
1374 | * @num_bytes: the size of the free space that we need | |
1375 | * @search_start: the position from which to begin the search | |
1376 | * @start: store the start of the free space. | |
1377 | * @len: the size of the free space. that we find, or the size | |
1378 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1379 | * |
0b86a832 CM |
1380 | * this uses a pretty simple search, the expectation is that it is |
1381 | * called very infrequently and that a given device has a small number | |
1382 | * of extents | |
7bfc837d MX |
1383 | * |
1384 | * @start is used to store the start of the free space if we find. But if we | |
1385 | * don't find suitable free space, it will be used to store the start position | |
1386 | * of the max free space. | |
1387 | * | |
1388 | * @len is used to store the size of the free space that we find. | |
1389 | * But if we don't find suitable free space, it is used to store the size of | |
1390 | * the max free space. | |
0b86a832 | 1391 | */ |
499f377f JM |
1392 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1393 | struct btrfs_device *device, u64 num_bytes, | |
1394 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1395 | { |
0b246afa JM |
1396 | struct btrfs_fs_info *fs_info = device->fs_info; |
1397 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1398 | struct btrfs_key key; |
7bfc837d | 1399 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1400 | struct btrfs_path *path; |
7bfc837d MX |
1401 | u64 hole_size; |
1402 | u64 max_hole_start; | |
1403 | u64 max_hole_size; | |
1404 | u64 extent_end; | |
0b86a832 CM |
1405 | u64 search_end = device->total_bytes; |
1406 | int ret; | |
7bfc837d | 1407 | int slot; |
0b86a832 | 1408 | struct extent_buffer *l; |
8cdc7c5b FM |
1409 | |
1410 | /* | |
1411 | * We don't want to overwrite the superblock on the drive nor any area | |
1412 | * used by the boot loader (grub for example), so we make sure to start | |
1413 | * at an offset of at least 1MB. | |
1414 | */ | |
0d0c71b3 | 1415 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1416 | |
6df9a95e JB |
1417 | path = btrfs_alloc_path(); |
1418 | if (!path) | |
1419 | return -ENOMEM; | |
f2ab7618 | 1420 | |
7bfc837d MX |
1421 | max_hole_start = search_start; |
1422 | max_hole_size = 0; | |
1423 | ||
f2ab7618 | 1424 | again: |
401e29c1 AJ |
1425 | if (search_start >= search_end || |
1426 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1427 | ret = -ENOSPC; |
6df9a95e | 1428 | goto out; |
7bfc837d MX |
1429 | } |
1430 | ||
e4058b54 | 1431 | path->reada = READA_FORWARD; |
6df9a95e JB |
1432 | path->search_commit_root = 1; |
1433 | path->skip_locking = 1; | |
7bfc837d | 1434 | |
0b86a832 CM |
1435 | key.objectid = device->devid; |
1436 | key.offset = search_start; | |
1437 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1438 | |
125ccb0a | 1439 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1440 | if (ret < 0) |
7bfc837d | 1441 | goto out; |
1fcbac58 YZ |
1442 | if (ret > 0) { |
1443 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1444 | if (ret < 0) | |
7bfc837d | 1445 | goto out; |
1fcbac58 | 1446 | } |
7bfc837d | 1447 | |
0b86a832 CM |
1448 | while (1) { |
1449 | l = path->nodes[0]; | |
1450 | slot = path->slots[0]; | |
1451 | if (slot >= btrfs_header_nritems(l)) { | |
1452 | ret = btrfs_next_leaf(root, path); | |
1453 | if (ret == 0) | |
1454 | continue; | |
1455 | if (ret < 0) | |
7bfc837d MX |
1456 | goto out; |
1457 | ||
1458 | break; | |
0b86a832 CM |
1459 | } |
1460 | btrfs_item_key_to_cpu(l, &key, slot); | |
1461 | ||
1462 | if (key.objectid < device->devid) | |
1463 | goto next; | |
1464 | ||
1465 | if (key.objectid > device->devid) | |
7bfc837d | 1466 | break; |
0b86a832 | 1467 | |
962a298f | 1468 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1469 | goto next; |
9779b72f | 1470 | |
7bfc837d MX |
1471 | if (key.offset > search_start) { |
1472 | hole_size = key.offset - search_start; | |
9779b72f | 1473 | |
6df9a95e JB |
1474 | /* |
1475 | * Have to check before we set max_hole_start, otherwise | |
1476 | * we could end up sending back this offset anyway. | |
1477 | */ | |
499f377f | 1478 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1479 | &search_start, |
1b984508 FL |
1480 | hole_size)) { |
1481 | if (key.offset >= search_start) { | |
1482 | hole_size = key.offset - search_start; | |
1483 | } else { | |
1484 | WARN_ON_ONCE(1); | |
1485 | hole_size = 0; | |
1486 | } | |
1487 | } | |
6df9a95e | 1488 | |
7bfc837d MX |
1489 | if (hole_size > max_hole_size) { |
1490 | max_hole_start = search_start; | |
1491 | max_hole_size = hole_size; | |
1492 | } | |
9779b72f | 1493 | |
7bfc837d MX |
1494 | /* |
1495 | * If this free space is greater than which we need, | |
1496 | * it must be the max free space that we have found | |
1497 | * until now, so max_hole_start must point to the start | |
1498 | * of this free space and the length of this free space | |
1499 | * is stored in max_hole_size. Thus, we return | |
1500 | * max_hole_start and max_hole_size and go back to the | |
1501 | * caller. | |
1502 | */ | |
1503 | if (hole_size >= num_bytes) { | |
1504 | ret = 0; | |
1505 | goto out; | |
0b86a832 CM |
1506 | } |
1507 | } | |
0b86a832 | 1508 | |
0b86a832 | 1509 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1510 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1511 | dev_extent); | |
1512 | if (extent_end > search_start) | |
1513 | search_start = extent_end; | |
0b86a832 CM |
1514 | next: |
1515 | path->slots[0]++; | |
1516 | cond_resched(); | |
1517 | } | |
0b86a832 | 1518 | |
38c01b96 | 1519 | /* |
1520 | * At this point, search_start should be the end of | |
1521 | * allocated dev extents, and when shrinking the device, | |
1522 | * search_end may be smaller than search_start. | |
1523 | */ | |
f2ab7618 | 1524 | if (search_end > search_start) { |
38c01b96 | 1525 | hole_size = search_end - search_start; |
1526 | ||
499f377f | 1527 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1528 | hole_size)) { |
1529 | btrfs_release_path(path); | |
1530 | goto again; | |
1531 | } | |
0b86a832 | 1532 | |
f2ab7618 ZL |
1533 | if (hole_size > max_hole_size) { |
1534 | max_hole_start = search_start; | |
1535 | max_hole_size = hole_size; | |
1536 | } | |
6df9a95e JB |
1537 | } |
1538 | ||
7bfc837d | 1539 | /* See above. */ |
f2ab7618 | 1540 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1541 | ret = -ENOSPC; |
1542 | else | |
1543 | ret = 0; | |
1544 | ||
1545 | out: | |
2b82032c | 1546 | btrfs_free_path(path); |
7bfc837d | 1547 | *start = max_hole_start; |
b2117a39 | 1548 | if (len) |
7bfc837d | 1549 | *len = max_hole_size; |
0b86a832 CM |
1550 | return ret; |
1551 | } | |
1552 | ||
499f377f JM |
1553 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1554 | struct btrfs_device *device, u64 num_bytes, | |
1555 | u64 *start, u64 *len) | |
1556 | { | |
499f377f | 1557 | /* FIXME use last free of some kind */ |
499f377f | 1558 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1559 | num_bytes, 0, start, len); |
499f377f JM |
1560 | } |
1561 | ||
b2950863 | 1562 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1563 | struct btrfs_device *device, |
2196d6e8 | 1564 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1565 | { |
0b246afa JM |
1566 | struct btrfs_fs_info *fs_info = device->fs_info; |
1567 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1568 | int ret; |
1569 | struct btrfs_path *path; | |
8f18cf13 | 1570 | struct btrfs_key key; |
a061fc8d CM |
1571 | struct btrfs_key found_key; |
1572 | struct extent_buffer *leaf = NULL; | |
1573 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1574 | |
1575 | path = btrfs_alloc_path(); | |
1576 | if (!path) | |
1577 | return -ENOMEM; | |
1578 | ||
1579 | key.objectid = device->devid; | |
1580 | key.offset = start; | |
1581 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1582 | again: |
8f18cf13 | 1583 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1584 | if (ret > 0) { |
1585 | ret = btrfs_previous_item(root, path, key.objectid, | |
1586 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1587 | if (ret) |
1588 | goto out; | |
a061fc8d CM |
1589 | leaf = path->nodes[0]; |
1590 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1591 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1592 | struct btrfs_dev_extent); | |
1593 | BUG_ON(found_key.offset > start || found_key.offset + | |
1594 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1595 | key = found_key; |
1596 | btrfs_release_path(path); | |
1597 | goto again; | |
a061fc8d CM |
1598 | } else if (ret == 0) { |
1599 | leaf = path->nodes[0]; | |
1600 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1601 | struct btrfs_dev_extent); | |
79787eaa | 1602 | } else { |
0b246afa | 1603 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1604 | goto out; |
a061fc8d | 1605 | } |
8f18cf13 | 1606 | |
2196d6e8 MX |
1607 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1608 | ||
8f18cf13 | 1609 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1610 | if (ret) { |
0b246afa JM |
1611 | btrfs_handle_fs_error(fs_info, ret, |
1612 | "Failed to remove dev extent item"); | |
13212b54 | 1613 | } else { |
3204d33c | 1614 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1615 | } |
b0b802d7 | 1616 | out: |
8f18cf13 CM |
1617 | btrfs_free_path(path); |
1618 | return ret; | |
1619 | } | |
1620 | ||
48a3b636 ES |
1621 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1622 | struct btrfs_device *device, | |
48a3b636 | 1623 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1624 | { |
1625 | int ret; | |
1626 | struct btrfs_path *path; | |
0b246afa JM |
1627 | struct btrfs_fs_info *fs_info = device->fs_info; |
1628 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1629 | struct btrfs_dev_extent *extent; |
1630 | struct extent_buffer *leaf; | |
1631 | struct btrfs_key key; | |
1632 | ||
e12c9621 | 1633 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1634 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1635 | path = btrfs_alloc_path(); |
1636 | if (!path) | |
1637 | return -ENOMEM; | |
1638 | ||
0b86a832 | 1639 | key.objectid = device->devid; |
2b82032c | 1640 | key.offset = start; |
0b86a832 CM |
1641 | key.type = BTRFS_DEV_EXTENT_KEY; |
1642 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1643 | sizeof(*extent)); | |
2cdcecbc MF |
1644 | if (ret) |
1645 | goto out; | |
0b86a832 CM |
1646 | |
1647 | leaf = path->nodes[0]; | |
1648 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1649 | struct btrfs_dev_extent); | |
b5d9071c NB |
1650 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1651 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1652 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1653 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1654 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1655 | ||
0b86a832 CM |
1656 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1657 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1658 | out: |
0b86a832 CM |
1659 | btrfs_free_path(path); |
1660 | return ret; | |
1661 | } | |
1662 | ||
6df9a95e | 1663 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1664 | { |
6df9a95e JB |
1665 | struct extent_map_tree *em_tree; |
1666 | struct extent_map *em; | |
1667 | struct rb_node *n; | |
1668 | u64 ret = 0; | |
0b86a832 | 1669 | |
6df9a95e JB |
1670 | em_tree = &fs_info->mapping_tree.map_tree; |
1671 | read_lock(&em_tree->lock); | |
1672 | n = rb_last(&em_tree->map); | |
1673 | if (n) { | |
1674 | em = rb_entry(n, struct extent_map, rb_node); | |
1675 | ret = em->start + em->len; | |
0b86a832 | 1676 | } |
6df9a95e JB |
1677 | read_unlock(&em_tree->lock); |
1678 | ||
0b86a832 CM |
1679 | return ret; |
1680 | } | |
1681 | ||
53f10659 ID |
1682 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1683 | u64 *devid_ret) | |
0b86a832 CM |
1684 | { |
1685 | int ret; | |
1686 | struct btrfs_key key; | |
1687 | struct btrfs_key found_key; | |
2b82032c YZ |
1688 | struct btrfs_path *path; |
1689 | ||
2b82032c YZ |
1690 | path = btrfs_alloc_path(); |
1691 | if (!path) | |
1692 | return -ENOMEM; | |
0b86a832 CM |
1693 | |
1694 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1695 | key.type = BTRFS_DEV_ITEM_KEY; | |
1696 | key.offset = (u64)-1; | |
1697 | ||
53f10659 | 1698 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1699 | if (ret < 0) |
1700 | goto error; | |
1701 | ||
79787eaa | 1702 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1703 | |
53f10659 ID |
1704 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1705 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1706 | BTRFS_DEV_ITEM_KEY); |
1707 | if (ret) { | |
53f10659 | 1708 | *devid_ret = 1; |
0b86a832 CM |
1709 | } else { |
1710 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1711 | path->slots[0]); | |
53f10659 | 1712 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1713 | } |
1714 | ret = 0; | |
1715 | error: | |
2b82032c | 1716 | btrfs_free_path(path); |
0b86a832 CM |
1717 | return ret; |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * the device information is stored in the chunk root | |
1722 | * the btrfs_device struct should be fully filled in | |
1723 | */ | |
c74a0b02 | 1724 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
5b4aacef | 1725 | struct btrfs_fs_info *fs_info, |
48a3b636 | 1726 | struct btrfs_device *device) |
0b86a832 | 1727 | { |
5b4aacef | 1728 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
1729 | int ret; |
1730 | struct btrfs_path *path; | |
1731 | struct btrfs_dev_item *dev_item; | |
1732 | struct extent_buffer *leaf; | |
1733 | struct btrfs_key key; | |
1734 | unsigned long ptr; | |
0b86a832 | 1735 | |
0b86a832 CM |
1736 | path = btrfs_alloc_path(); |
1737 | if (!path) | |
1738 | return -ENOMEM; | |
1739 | ||
0b86a832 CM |
1740 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1741 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1742 | key.offset = device->devid; |
0b86a832 CM |
1743 | |
1744 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1745 | sizeof(*dev_item)); |
0b86a832 CM |
1746 | if (ret) |
1747 | goto out; | |
1748 | ||
1749 | leaf = path->nodes[0]; | |
1750 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1751 | ||
1752 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1753 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1754 | btrfs_set_device_type(leaf, dev_item, device->type); |
1755 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1756 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1757 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1758 | btrfs_set_device_total_bytes(leaf, dev_item, |
1759 | btrfs_device_get_disk_total_bytes(device)); | |
1760 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1761 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1762 | btrfs_set_device_group(leaf, dev_item, 0); |
1763 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1764 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1765 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1766 | |
410ba3a2 | 1767 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1768 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1769 | ptr = btrfs_device_fsid(dev_item); |
44880fdc | 1770 | write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE); |
0b86a832 | 1771 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1772 | |
2b82032c | 1773 | ret = 0; |
0b86a832 CM |
1774 | out: |
1775 | btrfs_free_path(path); | |
1776 | return ret; | |
1777 | } | |
8f18cf13 | 1778 | |
5a1972bd QW |
1779 | /* |
1780 | * Function to update ctime/mtime for a given device path. | |
1781 | * Mainly used for ctime/mtime based probe like libblkid. | |
1782 | */ | |
da353f6b | 1783 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1784 | { |
1785 | struct file *filp; | |
1786 | ||
1787 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1788 | if (IS_ERR(filp)) |
5a1972bd QW |
1789 | return; |
1790 | file_update_time(filp); | |
1791 | filp_close(filp, NULL); | |
5a1972bd QW |
1792 | } |
1793 | ||
5b4aacef | 1794 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1795 | struct btrfs_device *device) |
1796 | { | |
5b4aacef | 1797 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1798 | int ret; |
1799 | struct btrfs_path *path; | |
a061fc8d | 1800 | struct btrfs_key key; |
a061fc8d CM |
1801 | struct btrfs_trans_handle *trans; |
1802 | ||
a061fc8d CM |
1803 | path = btrfs_alloc_path(); |
1804 | if (!path) | |
1805 | return -ENOMEM; | |
1806 | ||
a22285a6 | 1807 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1808 | if (IS_ERR(trans)) { |
1809 | btrfs_free_path(path); | |
1810 | return PTR_ERR(trans); | |
1811 | } | |
a061fc8d CM |
1812 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1813 | key.type = BTRFS_DEV_ITEM_KEY; | |
1814 | key.offset = device->devid; | |
1815 | ||
1816 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1817 | if (ret) { |
1818 | if (ret > 0) | |
1819 | ret = -ENOENT; | |
1820 | btrfs_abort_transaction(trans, ret); | |
1821 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1822 | goto out; |
1823 | } | |
1824 | ||
1825 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1826 | if (ret) { |
1827 | btrfs_abort_transaction(trans, ret); | |
1828 | btrfs_end_transaction(trans); | |
1829 | } | |
1830 | ||
a061fc8d CM |
1831 | out: |
1832 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1833 | if (!ret) |
1834 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1835 | return ret; |
1836 | } | |
1837 | ||
3cc31a0d DS |
1838 | /* |
1839 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1840 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1841 | * replace. | |
1842 | */ | |
1843 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1844 | u64 num_devices) | |
a061fc8d | 1845 | { |
a061fc8d | 1846 | u64 all_avail; |
de98ced9 | 1847 | unsigned seq; |
418775a2 | 1848 | int i; |
a061fc8d | 1849 | |
de98ced9 | 1850 | do { |
bd45ffbc | 1851 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1852 | |
bd45ffbc AJ |
1853 | all_avail = fs_info->avail_data_alloc_bits | |
1854 | fs_info->avail_system_alloc_bits | | |
1855 | fs_info->avail_metadata_alloc_bits; | |
1856 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1857 | |
418775a2 DS |
1858 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1859 | if (!(all_avail & btrfs_raid_group[i])) | |
1860 | continue; | |
a061fc8d | 1861 | |
418775a2 DS |
1862 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1863 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1864 | |
418775a2 DS |
1865 | if (ret) |
1866 | return ret; | |
1867 | } | |
53b381b3 DW |
1868 | } |
1869 | ||
bd45ffbc | 1870 | return 0; |
f1fa7f26 AJ |
1871 | } |
1872 | ||
c9162bdf OS |
1873 | static struct btrfs_device * btrfs_find_next_active_device( |
1874 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1875 | { |
2b82032c | 1876 | struct btrfs_device *next_device; |
88acff64 AJ |
1877 | |
1878 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1879 | if (next_device != device && | |
e6e674bd AJ |
1880 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1881 | && next_device->bdev) | |
88acff64 AJ |
1882 | return next_device; |
1883 | } | |
1884 | ||
1885 | return NULL; | |
1886 | } | |
1887 | ||
1888 | /* | |
1889 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1890 | * and replace it with the provided or the next active device, in the context | |
1891 | * where this function called, there should be always be another device (or | |
1892 | * this_dev) which is active. | |
1893 | */ | |
1894 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1895 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1896 | { | |
1897 | struct btrfs_device *next_device; | |
1898 | ||
1899 | if (this_dev) | |
1900 | next_device = this_dev; | |
1901 | else | |
1902 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1903 | device); | |
1904 | ASSERT(next_device); | |
1905 | ||
1906 | if (fs_info->sb->s_bdev && | |
1907 | (fs_info->sb->s_bdev == device->bdev)) | |
1908 | fs_info->sb->s_bdev = next_device->bdev; | |
1909 | ||
1910 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1911 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1912 | } | |
1913 | ||
da353f6b DS |
1914 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1915 | u64 devid) | |
f1fa7f26 AJ |
1916 | { |
1917 | struct btrfs_device *device; | |
1f78160c | 1918 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1919 | u64 num_devices; |
a061fc8d CM |
1920 | int ret = 0; |
1921 | ||
2c997384 | 1922 | mutex_lock(&fs_info->volume_mutex); |
a061fc8d CM |
1923 | mutex_lock(&uuid_mutex); |
1924 | ||
0b246afa JM |
1925 | num_devices = fs_info->fs_devices->num_devices; |
1926 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
1927 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
8dabb742 SB |
1928 | WARN_ON(num_devices < 1); |
1929 | num_devices--; | |
1930 | } | |
0b246afa | 1931 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
8dabb742 | 1932 | |
0b246afa | 1933 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1934 | if (ret) |
a061fc8d | 1935 | goto out; |
a061fc8d | 1936 | |
2ff7e61e JM |
1937 | ret = btrfs_find_device_by_devspec(fs_info, devid, device_path, |
1938 | &device); | |
24fc572f | 1939 | if (ret) |
53b381b3 | 1940 | goto out; |
dfe25020 | 1941 | |
401e29c1 | 1942 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1943 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1944 | goto out; |
63a212ab SB |
1945 | } |
1946 | ||
ebbede42 AJ |
1947 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1948 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1949 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1950 | goto out; |
2b82032c YZ |
1951 | } |
1952 | ||
ebbede42 | 1953 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1954 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1955 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1956 | device->fs_devices->rw_devices--; |
34441361 | 1957 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 1958 | } |
a061fc8d | 1959 | |
d7901554 | 1960 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1961 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1962 | mutex_lock(&uuid_mutex); |
a061fc8d | 1963 | if (ret) |
9b3517e9 | 1964 | goto error_undo; |
a061fc8d | 1965 | |
63a212ab SB |
1966 | /* |
1967 | * TODO: the superblock still includes this device in its num_devices | |
1968 | * counter although write_all_supers() is not locked out. This | |
1969 | * could give a filesystem state which requires a degraded mount. | |
1970 | */ | |
0b246afa | 1971 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1972 | if (ret) |
9b3517e9 | 1973 | goto error_undo; |
a061fc8d | 1974 | |
e12c9621 | 1975 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 1976 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1977 | |
1978 | /* | |
1979 | * the device list mutex makes sure that we don't change | |
1980 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1981 | * the device supers. Whoever is writing all supers, should |
1982 | * lock the device list mutex before getting the number of | |
1983 | * devices in the super block (super_copy). Conversely, | |
1984 | * whoever updates the number of devices in the super block | |
1985 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1986 | */ |
1f78160c XG |
1987 | |
1988 | cur_devices = device->fs_devices; | |
0b246afa | 1989 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c | 1990 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1991 | |
e4404d6e | 1992 | device->fs_devices->num_devices--; |
02db0844 | 1993 | device->fs_devices->total_devices--; |
2b82032c | 1994 | |
e6e674bd | 1995 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
3a7d55c8 | 1996 | device->fs_devices->missing_devices--; |
cd02dca5 | 1997 | |
0b246afa | 1998 | btrfs_assign_next_active_device(fs_info, device, NULL); |
2b82032c | 1999 | |
0bfaa9c5 | 2000 | if (device->bdev) { |
e4404d6e | 2001 | device->fs_devices->open_devices--; |
0bfaa9c5 | 2002 | /* remove sysfs entry */ |
0b246afa | 2003 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
0bfaa9c5 | 2004 | } |
99994cde | 2005 | |
0b246afa JM |
2006 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2007 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
2008 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2009 | |
cea67ab9 JM |
2010 | /* |
2011 | * at this point, the device is zero sized and detached from | |
2012 | * the devices list. All that's left is to zero out the old | |
2013 | * supers and free the device. | |
2014 | */ | |
ebbede42 | 2015 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2016 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2017 | ||
2018 | btrfs_close_bdev(device); | |
f06c5965 | 2019 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2020 | |
1f78160c | 2021 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2022 | struct btrfs_fs_devices *fs_devices; |
0b246afa | 2023 | fs_devices = fs_info->fs_devices; |
e4404d6e | 2024 | while (fs_devices) { |
8321cf25 RS |
2025 | if (fs_devices->seed == cur_devices) { |
2026 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2027 | break; |
8321cf25 | 2028 | } |
e4404d6e | 2029 | fs_devices = fs_devices->seed; |
2b82032c | 2030 | } |
1f78160c | 2031 | cur_devices->seed = NULL; |
1f78160c | 2032 | __btrfs_close_devices(cur_devices); |
1f78160c | 2033 | free_fs_devices(cur_devices); |
2b82032c YZ |
2034 | } |
2035 | ||
a061fc8d CM |
2036 | out: |
2037 | mutex_unlock(&uuid_mutex); | |
2c997384 | 2038 | mutex_unlock(&fs_info->volume_mutex); |
a061fc8d | 2039 | return ret; |
24fc572f | 2040 | |
9b3517e9 | 2041 | error_undo: |
ebbede42 | 2042 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2043 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2044 | list_add(&device->dev_alloc_list, |
0b246afa | 2045 | &fs_info->fs_devices->alloc_list); |
c3929c36 | 2046 | device->fs_devices->rw_devices++; |
34441361 | 2047 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2048 | } |
24fc572f | 2049 | goto out; |
a061fc8d CM |
2050 | } |
2051 | ||
084b6e7c QW |
2052 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
2053 | struct btrfs_device *srcdev) | |
e93c89c1 | 2054 | { |
d51908ce AJ |
2055 | struct btrfs_fs_devices *fs_devices; |
2056 | ||
e93c89c1 | 2057 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 2058 | |
25e8e911 AJ |
2059 | /* |
2060 | * in case of fs with no seed, srcdev->fs_devices will point | |
2061 | * to fs_devices of fs_info. However when the dev being replaced is | |
2062 | * a seed dev it will point to the seed's local fs_devices. In short | |
2063 | * srcdev will have its correct fs_devices in both the cases. | |
2064 | */ | |
2065 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2066 | |
e93c89c1 | 2067 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2068 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2069 | fs_devices->num_devices--; |
e6e674bd | 2070 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2071 | fs_devices->missing_devices--; |
e93c89c1 | 2072 | |
ebbede42 | 2073 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2074 | fs_devices->rw_devices--; |
1357272f | 2075 | |
82372bc8 | 2076 | if (srcdev->bdev) |
d51908ce | 2077 | fs_devices->open_devices--; |
084b6e7c QW |
2078 | } |
2079 | ||
2080 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2081 | struct btrfs_device *srcdev) | |
2082 | { | |
2083 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2084 | |
ebbede42 | 2085 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2086 | /* zero out the old super if it is writable */ |
2087 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2088 | } | |
14238819 AJ |
2089 | |
2090 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2091 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2092 | |
94d5f0c2 AJ |
2093 | /* if this is no devs we rather delete the fs_devices */ |
2094 | if (!fs_devices->num_devices) { | |
2095 | struct btrfs_fs_devices *tmp_fs_devices; | |
2096 | ||
6dd38f81 AJ |
2097 | /* |
2098 | * On a mounted FS, num_devices can't be zero unless it's a | |
2099 | * seed. In case of a seed device being replaced, the replace | |
2100 | * target added to the sprout FS, so there will be no more | |
2101 | * device left under the seed FS. | |
2102 | */ | |
2103 | ASSERT(fs_devices->seeding); | |
2104 | ||
94d5f0c2 AJ |
2105 | tmp_fs_devices = fs_info->fs_devices; |
2106 | while (tmp_fs_devices) { | |
2107 | if (tmp_fs_devices->seed == fs_devices) { | |
2108 | tmp_fs_devices->seed = fs_devices->seed; | |
2109 | break; | |
2110 | } | |
2111 | tmp_fs_devices = tmp_fs_devices->seed; | |
2112 | } | |
2113 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2114 | __btrfs_close_devices(fs_devices); |
2115 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2116 | } |
e93c89c1 SB |
2117 | } |
2118 | ||
2119 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2120 | struct btrfs_device *tgtdev) | |
2121 | { | |
67a2c45e | 2122 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2123 | WARN_ON(!tgtdev); |
2124 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2125 | |
32576040 | 2126 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2127 | |
779bf3fe | 2128 | if (tgtdev->bdev) |
e93c89c1 | 2129 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2130 | |
e93c89c1 | 2131 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2132 | |
88acff64 | 2133 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2134 | |
e93c89c1 | 2135 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2136 | |
2137 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2138 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2139 | |
2140 | /* | |
2141 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2142 | * may lead to a call to btrfs_show_devname() which will try | |
2143 | * to hold device_list_mutex. And here this device | |
2144 | * is already out of device list, so we don't have to hold | |
2145 | * the device_list_mutex lock. | |
2146 | */ | |
2147 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2148 | |
2149 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2150 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2151 | } |
2152 | ||
2ff7e61e | 2153 | static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2154 | const char *device_path, |
48a3b636 | 2155 | struct btrfs_device **device) |
7ba15b7d SB |
2156 | { |
2157 | int ret = 0; | |
2158 | struct btrfs_super_block *disk_super; | |
2159 | u64 devid; | |
2160 | u8 *dev_uuid; | |
2161 | struct block_device *bdev; | |
2162 | struct buffer_head *bh; | |
2163 | ||
2164 | *device = NULL; | |
2165 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2166 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2167 | if (ret) |
2168 | return ret; | |
2169 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2170 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2171 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2172 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2173 | brelse(bh); |
2174 | if (!*device) | |
2175 | ret = -ENOENT; | |
2176 | blkdev_put(bdev, FMODE_READ); | |
2177 | return ret; | |
2178 | } | |
2179 | ||
2ff7e61e | 2180 | int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2181 | const char *device_path, |
7ba15b7d SB |
2182 | struct btrfs_device **device) |
2183 | { | |
2184 | *device = NULL; | |
2185 | if (strcmp(device_path, "missing") == 0) { | |
2186 | struct list_head *devices; | |
2187 | struct btrfs_device *tmp; | |
2188 | ||
0b246afa | 2189 | devices = &fs_info->fs_devices->devices; |
7ba15b7d SB |
2190 | /* |
2191 | * It is safe to read the devices since the volume_mutex | |
2192 | * is held by the caller. | |
2193 | */ | |
2194 | list_for_each_entry(tmp, devices, dev_list) { | |
e12c9621 AJ |
2195 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2196 | &tmp->dev_state) && !tmp->bdev) { | |
7ba15b7d SB |
2197 | *device = tmp; |
2198 | break; | |
2199 | } | |
2200 | } | |
2201 | ||
d74a6259 AJ |
2202 | if (!*device) |
2203 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2204 | |
2205 | return 0; | |
2206 | } else { | |
2ff7e61e | 2207 | return btrfs_find_device_by_path(fs_info, device_path, device); |
7ba15b7d SB |
2208 | } |
2209 | } | |
2210 | ||
5c5c0df0 DS |
2211 | /* |
2212 | * Lookup a device given by device id, or the path if the id is 0. | |
2213 | */ | |
2ff7e61e | 2214 | int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid, |
da353f6b DS |
2215 | const char *devpath, |
2216 | struct btrfs_device **device) | |
24e0474b AJ |
2217 | { |
2218 | int ret; | |
2219 | ||
5c5c0df0 | 2220 | if (devid) { |
24e0474b | 2221 | ret = 0; |
0b246afa | 2222 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2223 | if (!*device) |
2224 | ret = -ENOENT; | |
2225 | } else { | |
5c5c0df0 | 2226 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2227 | return -EINVAL; |
2228 | ||
2ff7e61e | 2229 | ret = btrfs_find_device_missing_or_by_path(fs_info, devpath, |
24e0474b AJ |
2230 | device); |
2231 | } | |
2232 | return ret; | |
2233 | } | |
2234 | ||
2b82032c YZ |
2235 | /* |
2236 | * does all the dirty work required for changing file system's UUID. | |
2237 | */ | |
2ff7e61e | 2238 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2239 | { |
0b246afa | 2240 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2241 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2242 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2243 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2244 | struct btrfs_device *device; |
2245 | u64 super_flags; | |
2246 | ||
2247 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2248 | if (!fs_devices->seeding) |
2b82032c YZ |
2249 | return -EINVAL; |
2250 | ||
2dfeca9b | 2251 | seed_devices = alloc_fs_devices(NULL); |
2208a378 ID |
2252 | if (IS_ERR(seed_devices)) |
2253 | return PTR_ERR(seed_devices); | |
2b82032c | 2254 | |
e4404d6e YZ |
2255 | old_devices = clone_fs_devices(fs_devices); |
2256 | if (IS_ERR(old_devices)) { | |
2257 | kfree(seed_devices); | |
2258 | return PTR_ERR(old_devices); | |
2b82032c | 2259 | } |
e4404d6e | 2260 | |
2b82032c YZ |
2261 | list_add(&old_devices->list, &fs_uuids); |
2262 | ||
e4404d6e YZ |
2263 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2264 | seed_devices->opened = 1; | |
2265 | INIT_LIST_HEAD(&seed_devices->devices); | |
2266 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2267 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2268 | |
0b246afa | 2269 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c XG |
2270 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2271 | synchronize_rcu); | |
2196d6e8 MX |
2272 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2273 | device->fs_devices = seed_devices; | |
c9513edb | 2274 | |
34441361 | 2275 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2276 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2277 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2278 | |
2b82032c YZ |
2279 | fs_devices->seeding = 0; |
2280 | fs_devices->num_devices = 0; | |
2281 | fs_devices->open_devices = 0; | |
69611ac8 | 2282 | fs_devices->missing_devices = 0; |
69611ac8 | 2283 | fs_devices->rotating = 0; |
e4404d6e | 2284 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2285 | |
2286 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2287 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2288 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
0b246afa | 2289 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
f7171750 | 2290 | |
2b82032c YZ |
2291 | super_flags = btrfs_super_flags(disk_super) & |
2292 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2293 | btrfs_set_super_flags(disk_super, super_flags); | |
2294 | ||
2295 | return 0; | |
2296 | } | |
2297 | ||
2298 | /* | |
01327610 | 2299 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2300 | */ |
2301 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2302 | struct btrfs_fs_info *fs_info) |
2b82032c | 2303 | { |
5b4aacef | 2304 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2305 | struct btrfs_path *path; |
2306 | struct extent_buffer *leaf; | |
2307 | struct btrfs_dev_item *dev_item; | |
2308 | struct btrfs_device *device; | |
2309 | struct btrfs_key key; | |
44880fdc | 2310 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2311 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2312 | u64 devid; | |
2313 | int ret; | |
2314 | ||
2315 | path = btrfs_alloc_path(); | |
2316 | if (!path) | |
2317 | return -ENOMEM; | |
2318 | ||
2b82032c YZ |
2319 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2320 | key.offset = 0; | |
2321 | key.type = BTRFS_DEV_ITEM_KEY; | |
2322 | ||
2323 | while (1) { | |
2324 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2325 | if (ret < 0) | |
2326 | goto error; | |
2327 | ||
2328 | leaf = path->nodes[0]; | |
2329 | next_slot: | |
2330 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2331 | ret = btrfs_next_leaf(root, path); | |
2332 | if (ret > 0) | |
2333 | break; | |
2334 | if (ret < 0) | |
2335 | goto error; | |
2336 | leaf = path->nodes[0]; | |
2337 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2338 | btrfs_release_path(path); |
2b82032c YZ |
2339 | continue; |
2340 | } | |
2341 | ||
2342 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2343 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2344 | key.type != BTRFS_DEV_ITEM_KEY) | |
2345 | break; | |
2346 | ||
2347 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2348 | struct btrfs_dev_item); | |
2349 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2350 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2351 | BTRFS_UUID_SIZE); |
1473b24e | 2352 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2353 | BTRFS_FSID_SIZE); |
0b246afa | 2354 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2355 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2356 | |
2357 | if (device->fs_devices->seeding) { | |
2358 | btrfs_set_device_generation(leaf, dev_item, | |
2359 | device->generation); | |
2360 | btrfs_mark_buffer_dirty(leaf); | |
2361 | } | |
2362 | ||
2363 | path->slots[0]++; | |
2364 | goto next_slot; | |
2365 | } | |
2366 | ret = 0; | |
2367 | error: | |
2368 | btrfs_free_path(path); | |
2369 | return ret; | |
2370 | } | |
2371 | ||
da353f6b | 2372 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2373 | { |
5112febb | 2374 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2375 | struct request_queue *q; |
788f20eb CM |
2376 | struct btrfs_trans_handle *trans; |
2377 | struct btrfs_device *device; | |
2378 | struct block_device *bdev; | |
788f20eb | 2379 | struct list_head *devices; |
0b246afa | 2380 | struct super_block *sb = fs_info->sb; |
606686ee | 2381 | struct rcu_string *name; |
3c1dbdf5 | 2382 | u64 tmp; |
2b82032c | 2383 | int seeding_dev = 0; |
788f20eb | 2384 | int ret = 0; |
7132a262 | 2385 | bool unlocked = false; |
788f20eb | 2386 | |
bc98a42c | 2387 | if (sb_rdonly(sb) && !fs_info->fs_devices->seeding) |
f8c5d0b4 | 2388 | return -EROFS; |
788f20eb | 2389 | |
a5d16333 | 2390 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2391 | fs_info->bdev_holder); |
7f59203a JB |
2392 | if (IS_ERR(bdev)) |
2393 | return PTR_ERR(bdev); | |
a2135011 | 2394 | |
0b246afa | 2395 | if (fs_info->fs_devices->seeding) { |
2b82032c YZ |
2396 | seeding_dev = 1; |
2397 | down_write(&sb->s_umount); | |
2398 | mutex_lock(&uuid_mutex); | |
2399 | } | |
2400 | ||
8c8bee1d | 2401 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2402 | |
0b246afa | 2403 | devices = &fs_info->fs_devices->devices; |
d25628bd | 2404 | |
0b246afa | 2405 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
c6e30871 | 2406 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2407 | if (device->bdev == bdev) { |
2408 | ret = -EEXIST; | |
d25628bd | 2409 | mutex_unlock( |
0b246afa | 2410 | &fs_info->fs_devices->device_list_mutex); |
2b82032c | 2411 | goto error; |
788f20eb CM |
2412 | } |
2413 | } | |
0b246afa | 2414 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2415 | |
0b246afa | 2416 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2417 | if (IS_ERR(device)) { |
788f20eb | 2418 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2419 | ret = PTR_ERR(device); |
2b82032c | 2420 | goto error; |
788f20eb CM |
2421 | } |
2422 | ||
78f2c9e6 | 2423 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2424 | if (!name) { |
2b82032c | 2425 | ret = -ENOMEM; |
5c4cf6c9 | 2426 | goto error_free_device; |
788f20eb | 2427 | } |
606686ee | 2428 | rcu_assign_pointer(device->name, name); |
2b82032c | 2429 | |
a22285a6 | 2430 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2431 | if (IS_ERR(trans)) { |
98d5dc13 | 2432 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2433 | goto error_free_device; |
98d5dc13 TI |
2434 | } |
2435 | ||
d5e2003c | 2436 | q = bdev_get_queue(bdev); |
ebbede42 | 2437 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2438 | device->generation = trans->transid; |
0b246afa JM |
2439 | device->io_width = fs_info->sectorsize; |
2440 | device->io_align = fs_info->sectorsize; | |
2441 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2442 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2443 | fs_info->sectorsize); | |
2cc3c559 | 2444 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2445 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2446 | device->fs_info = fs_info; |
788f20eb | 2447 | device->bdev = bdev; |
e12c9621 | 2448 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2449 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2450 | device->mode = FMODE_EXCL; |
27087f37 | 2451 | device->dev_stats_valid = 1; |
9f6d2510 | 2452 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2453 | |
2b82032c | 2454 | if (seeding_dev) { |
1751e8a6 | 2455 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2456 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2457 | if (ret) { |
2458 | btrfs_abort_transaction(trans, ret); | |
2459 | goto error_trans; | |
2460 | } | |
2b82032c | 2461 | } |
788f20eb | 2462 | |
0b246afa | 2463 | device->fs_devices = fs_info->fs_devices; |
e5e9a520 | 2464 | |
0b246afa | 2465 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
34441361 | 2466 | mutex_lock(&fs_info->chunk_mutex); |
0b246afa | 2467 | list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices); |
2b82032c | 2468 | list_add(&device->dev_alloc_list, |
0b246afa JM |
2469 | &fs_info->fs_devices->alloc_list); |
2470 | fs_info->fs_devices->num_devices++; | |
2471 | fs_info->fs_devices->open_devices++; | |
2472 | fs_info->fs_devices->rw_devices++; | |
2473 | fs_info->fs_devices->total_devices++; | |
2474 | fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2475 | |
a5ed45f8 | 2476 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2477 | |
e884f4f0 | 2478 | if (!blk_queue_nonrot(q)) |
0b246afa | 2479 | fs_info->fs_devices->rotating = 1; |
c289811c | 2480 | |
0b246afa JM |
2481 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2482 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
7dfb8be1 | 2483 | round_down(tmp + device->total_bytes, fs_info->sectorsize)); |
788f20eb | 2484 | |
0b246afa JM |
2485 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2486 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2487 | |
2488 | /* add sysfs device entry */ | |
0b246afa | 2489 | btrfs_sysfs_add_device_link(fs_info->fs_devices, device); |
0d39376a | 2490 | |
2196d6e8 MX |
2491 | /* |
2492 | * we've got more storage, clear any full flags on the space | |
2493 | * infos | |
2494 | */ | |
0b246afa | 2495 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2496 | |
34441361 | 2497 | mutex_unlock(&fs_info->chunk_mutex); |
0b246afa | 2498 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2499 | |
2b82032c | 2500 | if (seeding_dev) { |
34441361 | 2501 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2502 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2503 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2504 | if (ret) { |
66642832 | 2505 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2506 | goto error_sysfs; |
005d6427 | 2507 | } |
2196d6e8 MX |
2508 | } |
2509 | ||
c74a0b02 | 2510 | ret = btrfs_add_dev_item(trans, fs_info, device); |
2196d6e8 | 2511 | if (ret) { |
66642832 | 2512 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2513 | goto error_sysfs; |
2196d6e8 MX |
2514 | } |
2515 | ||
2516 | if (seeding_dev) { | |
2517 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2518 | ||
0b246afa | 2519 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2520 | if (ret) { |
66642832 | 2521 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2522 | goto error_sysfs; |
005d6427 | 2523 | } |
b2373f25 AJ |
2524 | |
2525 | /* Sprouting would change fsid of the mounted root, | |
2526 | * so rename the fsid on the sysfs | |
2527 | */ | |
2528 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa JM |
2529 | fs_info->fsid); |
2530 | if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf)) | |
2531 | btrfs_warn(fs_info, | |
2532 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2533 | } |
2534 | ||
3a45bb20 | 2535 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2536 | |
2b82032c YZ |
2537 | if (seeding_dev) { |
2538 | mutex_unlock(&uuid_mutex); | |
2539 | up_write(&sb->s_umount); | |
7132a262 | 2540 | unlocked = true; |
788f20eb | 2541 | |
79787eaa JM |
2542 | if (ret) /* transaction commit */ |
2543 | return ret; | |
2544 | ||
2ff7e61e | 2545 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2546 | if (ret < 0) |
0b246afa | 2547 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2548 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2549 | trans = btrfs_attach_transaction(root); |
2550 | if (IS_ERR(trans)) { | |
2551 | if (PTR_ERR(trans) == -ENOENT) | |
2552 | return 0; | |
7132a262 AJ |
2553 | ret = PTR_ERR(trans); |
2554 | trans = NULL; | |
2555 | goto error_sysfs; | |
671415b7 | 2556 | } |
3a45bb20 | 2557 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2558 | } |
c9e9f97b | 2559 | |
5a1972bd QW |
2560 | /* Update ctime/mtime for libblkid */ |
2561 | update_dev_time(device_path); | |
2b82032c | 2562 | return ret; |
79787eaa | 2563 | |
d31c32f6 AJ |
2564 | error_sysfs: |
2565 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); | |
79787eaa | 2566 | error_trans: |
0af2c4bf | 2567 | if (seeding_dev) |
1751e8a6 | 2568 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2569 | if (trans) |
2570 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2571 | error_free_device: |
55de4803 | 2572 | free_device(device); |
2b82032c | 2573 | error: |
e525fd89 | 2574 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2575 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2576 | mutex_unlock(&uuid_mutex); |
2577 | up_write(&sb->s_umount); | |
2578 | } | |
c9e9f97b | 2579 | return ret; |
788f20eb CM |
2580 | } |
2581 | ||
2ff7e61e | 2582 | int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, |
da353f6b | 2583 | const char *device_path, |
1c43366d | 2584 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2585 | struct btrfs_device **device_out) |
2586 | { | |
e93c89c1 SB |
2587 | struct btrfs_device *device; |
2588 | struct block_device *bdev; | |
e93c89c1 SB |
2589 | struct list_head *devices; |
2590 | struct rcu_string *name; | |
12bd2fc0 | 2591 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2592 | int ret = 0; |
2593 | ||
2594 | *device_out = NULL; | |
1c43366d MX |
2595 | if (fs_info->fs_devices->seeding) { |
2596 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2597 | return -EINVAL; |
1c43366d | 2598 | } |
e93c89c1 SB |
2599 | |
2600 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2601 | fs_info->bdev_holder); | |
1c43366d MX |
2602 | if (IS_ERR(bdev)) { |
2603 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2604 | return PTR_ERR(bdev); |
1c43366d | 2605 | } |
e93c89c1 SB |
2606 | |
2607 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2608 | ||
2609 | devices = &fs_info->fs_devices->devices; | |
2610 | list_for_each_entry(device, devices, dev_list) { | |
2611 | if (device->bdev == bdev) { | |
5d163e0e JM |
2612 | btrfs_err(fs_info, |
2613 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2614 | ret = -EEXIST; |
2615 | goto error; | |
2616 | } | |
2617 | } | |
2618 | ||
1c43366d | 2619 | |
7cc8e58d MX |
2620 | if (i_size_read(bdev->bd_inode) < |
2621 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2622 | btrfs_err(fs_info, |
2623 | "target device is smaller than source device!"); | |
1c43366d MX |
2624 | ret = -EINVAL; |
2625 | goto error; | |
2626 | } | |
2627 | ||
2628 | ||
12bd2fc0 ID |
2629 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2630 | if (IS_ERR(device)) { | |
2631 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2632 | goto error; |
2633 | } | |
2634 | ||
6165572c | 2635 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
e93c89c1 | 2636 | if (!name) { |
55de4803 | 2637 | free_device(device); |
e93c89c1 SB |
2638 | ret = -ENOMEM; |
2639 | goto error; | |
2640 | } | |
2641 | rcu_assign_pointer(device->name, name); | |
2642 | ||
0b246afa | 2643 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
ebbede42 | 2644 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
e93c89c1 | 2645 | device->generation = 0; |
0b246afa JM |
2646 | device->io_width = fs_info->sectorsize; |
2647 | device->io_align = fs_info->sectorsize; | |
2648 | device->sector_size = fs_info->sectorsize; | |
7cc8e58d MX |
2649 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2650 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2651 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2652 | ASSERT(list_empty(&srcdev->resized_list)); |
2653 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2654 | device->commit_bytes_used = device->bytes_used; |
fb456252 | 2655 | device->fs_info = fs_info; |
e93c89c1 | 2656 | device->bdev = bdev; |
e12c9621 | 2657 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2658 | set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
e93c89c1 | 2659 | device->mode = FMODE_EXCL; |
27087f37 | 2660 | device->dev_stats_valid = 1; |
9f6d2510 | 2661 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
e93c89c1 SB |
2662 | device->fs_devices = fs_info->fs_devices; |
2663 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2664 | fs_info->fs_devices->num_devices++; | |
2665 | fs_info->fs_devices->open_devices++; | |
0b246afa | 2666 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 SB |
2667 | |
2668 | *device_out = device; | |
2669 | return ret; | |
2670 | ||
2671 | error: | |
2672 | blkdev_put(bdev, FMODE_EXCL); | |
2673 | return ret; | |
2674 | } | |
2675 | ||
2676 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2677 | struct btrfs_device *tgtdev) | |
2678 | { | |
da17066c JM |
2679 | u32 sectorsize = fs_info->sectorsize; |
2680 | ||
e93c89c1 | 2681 | WARN_ON(fs_info->fs_devices->rw_devices == 0); |
da17066c JM |
2682 | tgtdev->io_width = sectorsize; |
2683 | tgtdev->io_align = sectorsize; | |
2684 | tgtdev->sector_size = sectorsize; | |
fb456252 | 2685 | tgtdev->fs_info = fs_info; |
e12c9621 | 2686 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &tgtdev->dev_state); |
e93c89c1 SB |
2687 | } |
2688 | ||
d397712b CM |
2689 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2690 | struct btrfs_device *device) | |
0b86a832 CM |
2691 | { |
2692 | int ret; | |
2693 | struct btrfs_path *path; | |
0b246afa | 2694 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2695 | struct btrfs_dev_item *dev_item; |
2696 | struct extent_buffer *leaf; | |
2697 | struct btrfs_key key; | |
2698 | ||
0b86a832 CM |
2699 | path = btrfs_alloc_path(); |
2700 | if (!path) | |
2701 | return -ENOMEM; | |
2702 | ||
2703 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2704 | key.type = BTRFS_DEV_ITEM_KEY; | |
2705 | key.offset = device->devid; | |
2706 | ||
2707 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2708 | if (ret < 0) | |
2709 | goto out; | |
2710 | ||
2711 | if (ret > 0) { | |
2712 | ret = -ENOENT; | |
2713 | goto out; | |
2714 | } | |
2715 | ||
2716 | leaf = path->nodes[0]; | |
2717 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2718 | ||
2719 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2720 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2721 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2722 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2723 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2724 | btrfs_set_device_total_bytes(leaf, dev_item, |
2725 | btrfs_device_get_disk_total_bytes(device)); | |
2726 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2727 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2728 | btrfs_mark_buffer_dirty(leaf); |
2729 | ||
2730 | out: | |
2731 | btrfs_free_path(path); | |
2732 | return ret; | |
2733 | } | |
2734 | ||
2196d6e8 | 2735 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2736 | struct btrfs_device *device, u64 new_size) |
2737 | { | |
0b246afa JM |
2738 | struct btrfs_fs_info *fs_info = device->fs_info; |
2739 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2740 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2741 | u64 old_total; |
2742 | u64 diff; | |
8f18cf13 | 2743 | |
ebbede42 | 2744 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2745 | return -EACCES; |
2196d6e8 | 2746 | |
7dfb8be1 NB |
2747 | new_size = round_down(new_size, fs_info->sectorsize); |
2748 | ||
34441361 | 2749 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2750 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2751 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2752 | |
63a212ab | 2753 | if (new_size <= device->total_bytes || |
401e29c1 | 2754 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2755 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2756 | return -EINVAL; |
2196d6e8 | 2757 | } |
2b82032c | 2758 | |
0b246afa | 2759 | fs_devices = fs_info->fs_devices; |
2b82032c | 2760 | |
7dfb8be1 NB |
2761 | btrfs_set_super_total_bytes(super_copy, |
2762 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2763 | device->fs_devices->total_rw_bytes += diff; |
2764 | ||
7cc8e58d MX |
2765 | btrfs_device_set_total_bytes(device, new_size); |
2766 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2767 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2768 | if (list_empty(&device->resized_list)) |
2769 | list_add_tail(&device->resized_list, | |
2770 | &fs_devices->resized_devices); | |
34441361 | 2771 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2772 | |
8f18cf13 CM |
2773 | return btrfs_update_device(trans, device); |
2774 | } | |
2775 | ||
2776 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
408fbf19 | 2777 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2778 | { |
5b4aacef | 2779 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2780 | int ret; |
2781 | struct btrfs_path *path; | |
2782 | struct btrfs_key key; | |
2783 | ||
8f18cf13 CM |
2784 | path = btrfs_alloc_path(); |
2785 | if (!path) | |
2786 | return -ENOMEM; | |
2787 | ||
408fbf19 | 2788 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2789 | key.offset = chunk_offset; |
2790 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2791 | ||
2792 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2793 | if (ret < 0) |
2794 | goto out; | |
2795 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2796 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2797 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2798 | ret = -ENOENT; |
2799 | goto out; | |
2800 | } | |
8f18cf13 CM |
2801 | |
2802 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2803 | if (ret < 0) |
0b246afa JM |
2804 | btrfs_handle_fs_error(fs_info, ret, |
2805 | "Failed to delete chunk item."); | |
79787eaa | 2806 | out: |
8f18cf13 | 2807 | btrfs_free_path(path); |
65a246c5 | 2808 | return ret; |
8f18cf13 CM |
2809 | } |
2810 | ||
408fbf19 | 2811 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2812 | { |
0b246afa | 2813 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2814 | struct btrfs_disk_key *disk_key; |
2815 | struct btrfs_chunk *chunk; | |
2816 | u8 *ptr; | |
2817 | int ret = 0; | |
2818 | u32 num_stripes; | |
2819 | u32 array_size; | |
2820 | u32 len = 0; | |
2821 | u32 cur; | |
2822 | struct btrfs_key key; | |
2823 | ||
34441361 | 2824 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2825 | array_size = btrfs_super_sys_array_size(super_copy); |
2826 | ||
2827 | ptr = super_copy->sys_chunk_array; | |
2828 | cur = 0; | |
2829 | ||
2830 | while (cur < array_size) { | |
2831 | disk_key = (struct btrfs_disk_key *)ptr; | |
2832 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2833 | ||
2834 | len = sizeof(*disk_key); | |
2835 | ||
2836 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2837 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2838 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2839 | len += btrfs_chunk_item_size(num_stripes); | |
2840 | } else { | |
2841 | ret = -EIO; | |
2842 | break; | |
2843 | } | |
408fbf19 | 2844 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2845 | key.offset == chunk_offset) { |
2846 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2847 | array_size -= len; | |
2848 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2849 | } else { | |
2850 | ptr += len; | |
2851 | cur += len; | |
2852 | } | |
2853 | } | |
34441361 | 2854 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2855 | return ret; |
2856 | } | |
2857 | ||
592d92ee LB |
2858 | static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info, |
2859 | u64 logical, u64 length) | |
2860 | { | |
2861 | struct extent_map_tree *em_tree; | |
2862 | struct extent_map *em; | |
2863 | ||
2864 | em_tree = &fs_info->mapping_tree.map_tree; | |
2865 | read_lock(&em_tree->lock); | |
2866 | em = lookup_extent_mapping(em_tree, logical, length); | |
2867 | read_unlock(&em_tree->lock); | |
2868 | ||
2869 | if (!em) { | |
2870 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2871 | logical, length); | |
2872 | return ERR_PTR(-EINVAL); | |
2873 | } | |
2874 | ||
2875 | if (em->start > logical || em->start + em->len < logical) { | |
2876 | btrfs_crit(fs_info, | |
2877 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2878 | logical, length, em->start, em->start + em->len); | |
2879 | free_extent_map(em); | |
2880 | return ERR_PTR(-EINVAL); | |
2881 | } | |
2882 | ||
2883 | /* callers are responsible for dropping em's ref. */ | |
2884 | return em; | |
2885 | } | |
2886 | ||
47ab2a6c | 2887 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
5b4aacef | 2888 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2889 | { |
8f18cf13 CM |
2890 | struct extent_map *em; |
2891 | struct map_lookup *map; | |
2196d6e8 | 2892 | u64 dev_extent_len = 0; |
47ab2a6c | 2893 | int i, ret = 0; |
0b246afa | 2894 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2895 | |
592d92ee LB |
2896 | em = get_chunk_map(fs_info, chunk_offset, 1); |
2897 | if (IS_ERR(em)) { | |
47ab2a6c JB |
2898 | /* |
2899 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2900 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2901 | * do anything we still error out. |
2902 | */ | |
2903 | ASSERT(0); | |
592d92ee | 2904 | return PTR_ERR(em); |
47ab2a6c | 2905 | } |
95617d69 | 2906 | map = em->map_lookup; |
34441361 | 2907 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 2908 | check_system_chunk(trans, fs_info, map->type); |
34441361 | 2909 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2910 | |
57ba4cb8 FM |
2911 | /* |
2912 | * Take the device list mutex to prevent races with the final phase of | |
2913 | * a device replace operation that replaces the device object associated | |
2914 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2915 | */ | |
2916 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2917 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2918 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2919 | ret = btrfs_free_dev_extent(trans, device, |
2920 | map->stripes[i].physical, | |
2921 | &dev_extent_len); | |
47ab2a6c | 2922 | if (ret) { |
57ba4cb8 | 2923 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2924 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2925 | goto out; |
2926 | } | |
a061fc8d | 2927 | |
2196d6e8 | 2928 | if (device->bytes_used > 0) { |
34441361 | 2929 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2930 | btrfs_device_set_bytes_used(device, |
2931 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2932 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2933 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2934 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2935 | } |
a061fc8d | 2936 | |
dfe25020 CM |
2937 | if (map->stripes[i].dev) { |
2938 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2939 | if (ret) { |
57ba4cb8 | 2940 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2941 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2942 | goto out; |
2943 | } | |
dfe25020 | 2944 | } |
8f18cf13 | 2945 | } |
57ba4cb8 FM |
2946 | mutex_unlock(&fs_devices->device_list_mutex); |
2947 | ||
408fbf19 | 2948 | ret = btrfs_free_chunk(trans, fs_info, chunk_offset); |
47ab2a6c | 2949 | if (ret) { |
66642832 | 2950 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2951 | goto out; |
2952 | } | |
8f18cf13 | 2953 | |
6bccf3ab | 2954 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2955 | |
8f18cf13 | 2956 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2957 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2958 | if (ret) { |
66642832 | 2959 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2960 | goto out; |
2961 | } | |
8f18cf13 CM |
2962 | } |
2963 | ||
6bccf3ab | 2964 | ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em); |
47ab2a6c | 2965 | if (ret) { |
66642832 | 2966 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2967 | goto out; |
2968 | } | |
2b82032c | 2969 | |
47ab2a6c | 2970 | out: |
2b82032c YZ |
2971 | /* once for us */ |
2972 | free_extent_map(em); | |
47ab2a6c JB |
2973 | return ret; |
2974 | } | |
2b82032c | 2975 | |
5b4aacef | 2976 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2977 | { |
5b4aacef | 2978 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2979 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2980 | int ret; |
2b82032c | 2981 | |
67c5e7d4 FM |
2982 | /* |
2983 | * Prevent races with automatic removal of unused block groups. | |
2984 | * After we relocate and before we remove the chunk with offset | |
2985 | * chunk_offset, automatic removal of the block group can kick in, | |
2986 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2987 | * | |
2988 | * Make sure to acquire this mutex before doing a tree search (dev | |
2989 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2990 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2991 | * we release the path used to search the chunk/dev tree and before | |
2992 | * the current task acquires this mutex and calls us. | |
2993 | */ | |
0b246afa | 2994 | ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex)); |
67c5e7d4 | 2995 | |
0b246afa | 2996 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2997 | if (ret) |
2998 | return -ENOSPC; | |
2999 | ||
3000 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 3001 | btrfs_scrub_pause(fs_info); |
0b246afa | 3002 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3003 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3004 | if (ret) |
3005 | return ret; | |
3006 | ||
19c4d2f9 CM |
3007 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3008 | chunk_offset); | |
3009 | if (IS_ERR(trans)) { | |
3010 | ret = PTR_ERR(trans); | |
3011 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3012 | return ret; | |
3013 | } | |
3014 | ||
47ab2a6c | 3015 | /* |
19c4d2f9 CM |
3016 | * step two, delete the device extents and the |
3017 | * chunk tree entries | |
47ab2a6c | 3018 | */ |
5b4aacef | 3019 | ret = btrfs_remove_chunk(trans, fs_info, chunk_offset); |
3a45bb20 | 3020 | btrfs_end_transaction(trans); |
19c4d2f9 | 3021 | return ret; |
2b82032c YZ |
3022 | } |
3023 | ||
2ff7e61e | 3024 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3025 | { |
0b246afa | 3026 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3027 | struct btrfs_path *path; |
3028 | struct extent_buffer *leaf; | |
3029 | struct btrfs_chunk *chunk; | |
3030 | struct btrfs_key key; | |
3031 | struct btrfs_key found_key; | |
2b82032c | 3032 | u64 chunk_type; |
ba1bf481 JB |
3033 | bool retried = false; |
3034 | int failed = 0; | |
2b82032c YZ |
3035 | int ret; |
3036 | ||
3037 | path = btrfs_alloc_path(); | |
3038 | if (!path) | |
3039 | return -ENOMEM; | |
3040 | ||
ba1bf481 | 3041 | again: |
2b82032c YZ |
3042 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3043 | key.offset = (u64)-1; | |
3044 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3045 | ||
3046 | while (1) { | |
0b246afa | 3047 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3048 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3049 | if (ret < 0) { |
0b246afa | 3050 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3051 | goto error; |
67c5e7d4 | 3052 | } |
79787eaa | 3053 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3054 | |
3055 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3056 | key.type); | |
67c5e7d4 | 3057 | if (ret) |
0b246afa | 3058 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3059 | if (ret < 0) |
3060 | goto error; | |
3061 | if (ret > 0) | |
3062 | break; | |
1a40e23b | 3063 | |
2b82032c YZ |
3064 | leaf = path->nodes[0]; |
3065 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3066 | |
2b82032c YZ |
3067 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3068 | struct btrfs_chunk); | |
3069 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3070 | btrfs_release_path(path); |
8f18cf13 | 3071 | |
2b82032c | 3072 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3073 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3074 | if (ret == -ENOSPC) |
3075 | failed++; | |
14586651 HS |
3076 | else |
3077 | BUG_ON(ret); | |
2b82032c | 3078 | } |
0b246afa | 3079 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3080 | |
2b82032c YZ |
3081 | if (found_key.offset == 0) |
3082 | break; | |
3083 | key.offset = found_key.offset - 1; | |
3084 | } | |
3085 | ret = 0; | |
ba1bf481 JB |
3086 | if (failed && !retried) { |
3087 | failed = 0; | |
3088 | retried = true; | |
3089 | goto again; | |
fae7f21c | 3090 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3091 | ret = -ENOSPC; |
3092 | } | |
2b82032c YZ |
3093 | error: |
3094 | btrfs_free_path(path); | |
3095 | return ret; | |
8f18cf13 CM |
3096 | } |
3097 | ||
a6f93c71 LB |
3098 | /* |
3099 | * return 1 : allocate a data chunk successfully, | |
3100 | * return <0: errors during allocating a data chunk, | |
3101 | * return 0 : no need to allocate a data chunk. | |
3102 | */ | |
3103 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3104 | u64 chunk_offset) | |
3105 | { | |
3106 | struct btrfs_block_group_cache *cache; | |
3107 | u64 bytes_used; | |
3108 | u64 chunk_type; | |
3109 | ||
3110 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3111 | ASSERT(cache); | |
3112 | chunk_type = cache->flags; | |
3113 | btrfs_put_block_group(cache); | |
3114 | ||
3115 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3116 | spin_lock(&fs_info->data_sinfo->lock); | |
3117 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3118 | spin_unlock(&fs_info->data_sinfo->lock); | |
3119 | ||
3120 | if (!bytes_used) { | |
3121 | struct btrfs_trans_handle *trans; | |
3122 | int ret; | |
3123 | ||
3124 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3125 | if (IS_ERR(trans)) | |
3126 | return PTR_ERR(trans); | |
3127 | ||
3128 | ret = btrfs_force_chunk_alloc(trans, fs_info, | |
3129 | BTRFS_BLOCK_GROUP_DATA); | |
3130 | btrfs_end_transaction(trans); | |
3131 | if (ret < 0) | |
3132 | return ret; | |
3133 | ||
3134 | return 1; | |
3135 | } | |
3136 | } | |
3137 | return 0; | |
3138 | } | |
3139 | ||
6bccf3ab | 3140 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3141 | struct btrfs_balance_control *bctl) |
3142 | { | |
6bccf3ab | 3143 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3144 | struct btrfs_trans_handle *trans; |
3145 | struct btrfs_balance_item *item; | |
3146 | struct btrfs_disk_balance_args disk_bargs; | |
3147 | struct btrfs_path *path; | |
3148 | struct extent_buffer *leaf; | |
3149 | struct btrfs_key key; | |
3150 | int ret, err; | |
3151 | ||
3152 | path = btrfs_alloc_path(); | |
3153 | if (!path) | |
3154 | return -ENOMEM; | |
3155 | ||
3156 | trans = btrfs_start_transaction(root, 0); | |
3157 | if (IS_ERR(trans)) { | |
3158 | btrfs_free_path(path); | |
3159 | return PTR_ERR(trans); | |
3160 | } | |
3161 | ||
3162 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3163 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3164 | key.offset = 0; |
3165 | ||
3166 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3167 | sizeof(*item)); | |
3168 | if (ret) | |
3169 | goto out; | |
3170 | ||
3171 | leaf = path->nodes[0]; | |
3172 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3173 | ||
b159fa28 | 3174 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3175 | |
3176 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3177 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3178 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3179 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3180 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3181 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3182 | ||
3183 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3184 | ||
3185 | btrfs_mark_buffer_dirty(leaf); | |
3186 | out: | |
3187 | btrfs_free_path(path); | |
3a45bb20 | 3188 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3189 | if (err && !ret) |
3190 | ret = err; | |
3191 | return ret; | |
3192 | } | |
3193 | ||
6bccf3ab | 3194 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3195 | { |
6bccf3ab | 3196 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3197 | struct btrfs_trans_handle *trans; |
3198 | struct btrfs_path *path; | |
3199 | struct btrfs_key key; | |
3200 | int ret, err; | |
3201 | ||
3202 | path = btrfs_alloc_path(); | |
3203 | if (!path) | |
3204 | return -ENOMEM; | |
3205 | ||
3206 | trans = btrfs_start_transaction(root, 0); | |
3207 | if (IS_ERR(trans)) { | |
3208 | btrfs_free_path(path); | |
3209 | return PTR_ERR(trans); | |
3210 | } | |
3211 | ||
3212 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3213 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3214 | key.offset = 0; |
3215 | ||
3216 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3217 | if (ret < 0) | |
3218 | goto out; | |
3219 | if (ret > 0) { | |
3220 | ret = -ENOENT; | |
3221 | goto out; | |
3222 | } | |
3223 | ||
3224 | ret = btrfs_del_item(trans, root, path); | |
3225 | out: | |
3226 | btrfs_free_path(path); | |
3a45bb20 | 3227 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3228 | if (err && !ret) |
3229 | ret = err; | |
3230 | return ret; | |
3231 | } | |
3232 | ||
59641015 ID |
3233 | /* |
3234 | * This is a heuristic used to reduce the number of chunks balanced on | |
3235 | * resume after balance was interrupted. | |
3236 | */ | |
3237 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3238 | { | |
3239 | /* | |
3240 | * Turn on soft mode for chunk types that were being converted. | |
3241 | */ | |
3242 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3243 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3244 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3245 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3246 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3247 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3248 | ||
3249 | /* | |
3250 | * Turn on usage filter if is not already used. The idea is | |
3251 | * that chunks that we have already balanced should be | |
3252 | * reasonably full. Don't do it for chunks that are being | |
3253 | * converted - that will keep us from relocating unconverted | |
3254 | * (albeit full) chunks. | |
3255 | */ | |
3256 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3257 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3258 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3259 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3260 | bctl->data.usage = 90; | |
3261 | } | |
3262 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3263 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3264 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3265 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3266 | bctl->sys.usage = 90; | |
3267 | } | |
3268 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3269 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3270 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3271 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3272 | bctl->meta.usage = 90; | |
3273 | } | |
3274 | } | |
3275 | ||
c9e9f97b ID |
3276 | /* |
3277 | * Should be called with both balance and volume mutexes held to | |
3278 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3279 | * restriper. Same goes for unset_balance_control. | |
3280 | */ | |
3281 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3282 | { | |
3283 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3284 | ||
3285 | BUG_ON(fs_info->balance_ctl); | |
3286 | ||
3287 | spin_lock(&fs_info->balance_lock); | |
3288 | fs_info->balance_ctl = bctl; | |
3289 | spin_unlock(&fs_info->balance_lock); | |
3290 | } | |
3291 | ||
3292 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3293 | { | |
3294 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3295 | ||
3296 | BUG_ON(!fs_info->balance_ctl); | |
3297 | ||
3298 | spin_lock(&fs_info->balance_lock); | |
3299 | fs_info->balance_ctl = NULL; | |
3300 | spin_unlock(&fs_info->balance_lock); | |
3301 | ||
3302 | kfree(bctl); | |
3303 | } | |
3304 | ||
ed25e9b2 ID |
3305 | /* |
3306 | * Balance filters. Return 1 if chunk should be filtered out | |
3307 | * (should not be balanced). | |
3308 | */ | |
899c81ea | 3309 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3310 | struct btrfs_balance_args *bargs) |
3311 | { | |
899c81ea ID |
3312 | chunk_type = chunk_to_extended(chunk_type) & |
3313 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3314 | |
899c81ea | 3315 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3316 | return 0; |
3317 | ||
3318 | return 1; | |
3319 | } | |
3320 | ||
dba72cb3 | 3321 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3322 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3323 | { |
3324 | struct btrfs_block_group_cache *cache; | |
3325 | u64 chunk_used; | |
3326 | u64 user_thresh_min; | |
3327 | u64 user_thresh_max; | |
3328 | int ret = 1; | |
3329 | ||
3330 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3331 | chunk_used = btrfs_block_group_used(&cache->item); | |
3332 | ||
3333 | if (bargs->usage_min == 0) | |
3334 | user_thresh_min = 0; | |
3335 | else | |
3336 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3337 | bargs->usage_min); | |
3338 | ||
3339 | if (bargs->usage_max == 0) | |
3340 | user_thresh_max = 1; | |
3341 | else if (bargs->usage_max > 100) | |
3342 | user_thresh_max = cache->key.offset; | |
3343 | else | |
3344 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3345 | bargs->usage_max); | |
3346 | ||
3347 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3348 | ret = 0; | |
3349 | ||
3350 | btrfs_put_block_group(cache); | |
3351 | return ret; | |
3352 | } | |
3353 | ||
dba72cb3 | 3354 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3355 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3356 | { |
3357 | struct btrfs_block_group_cache *cache; | |
3358 | u64 chunk_used, user_thresh; | |
3359 | int ret = 1; | |
3360 | ||
3361 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3362 | chunk_used = btrfs_block_group_used(&cache->item); | |
3363 | ||
bc309467 | 3364 | if (bargs->usage_min == 0) |
3e39cea6 | 3365 | user_thresh = 1; |
a105bb88 ID |
3366 | else if (bargs->usage > 100) |
3367 | user_thresh = cache->key.offset; | |
3368 | else | |
3369 | user_thresh = div_factor_fine(cache->key.offset, | |
3370 | bargs->usage); | |
3371 | ||
5ce5b3c0 ID |
3372 | if (chunk_used < user_thresh) |
3373 | ret = 0; | |
3374 | ||
3375 | btrfs_put_block_group(cache); | |
3376 | return ret; | |
3377 | } | |
3378 | ||
409d404b ID |
3379 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3380 | struct btrfs_chunk *chunk, | |
3381 | struct btrfs_balance_args *bargs) | |
3382 | { | |
3383 | struct btrfs_stripe *stripe; | |
3384 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3385 | int i; | |
3386 | ||
3387 | for (i = 0; i < num_stripes; i++) { | |
3388 | stripe = btrfs_stripe_nr(chunk, i); | |
3389 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3390 | return 0; | |
3391 | } | |
3392 | ||
3393 | return 1; | |
3394 | } | |
3395 | ||
94e60d5a ID |
3396 | /* [pstart, pend) */ |
3397 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3398 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3399 | struct btrfs_balance_args *bargs) |
3400 | { | |
3401 | struct btrfs_stripe *stripe; | |
3402 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3403 | u64 stripe_offset; | |
3404 | u64 stripe_length; | |
3405 | int factor; | |
3406 | int i; | |
3407 | ||
3408 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3409 | return 0; | |
3410 | ||
3411 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3412 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3413 | factor = num_stripes / 2; | |
3414 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3415 | factor = num_stripes - 1; | |
3416 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3417 | factor = num_stripes - 2; | |
3418 | } else { | |
3419 | factor = num_stripes; | |
3420 | } | |
94e60d5a ID |
3421 | |
3422 | for (i = 0; i < num_stripes; i++) { | |
3423 | stripe = btrfs_stripe_nr(chunk, i); | |
3424 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3425 | continue; | |
3426 | ||
3427 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3428 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3429 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3430 | |
3431 | if (stripe_offset < bargs->pend && | |
3432 | stripe_offset + stripe_length > bargs->pstart) | |
3433 | return 0; | |
3434 | } | |
3435 | ||
3436 | return 1; | |
3437 | } | |
3438 | ||
ea67176a ID |
3439 | /* [vstart, vend) */ |
3440 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3441 | struct btrfs_chunk *chunk, | |
3442 | u64 chunk_offset, | |
3443 | struct btrfs_balance_args *bargs) | |
3444 | { | |
3445 | if (chunk_offset < bargs->vend && | |
3446 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3447 | /* at least part of the chunk is inside this vrange */ | |
3448 | return 0; | |
3449 | ||
3450 | return 1; | |
3451 | } | |
3452 | ||
dee32d0a GAP |
3453 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3454 | struct btrfs_chunk *chunk, | |
3455 | struct btrfs_balance_args *bargs) | |
3456 | { | |
3457 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3458 | ||
3459 | if (bargs->stripes_min <= num_stripes | |
3460 | && num_stripes <= bargs->stripes_max) | |
3461 | return 0; | |
3462 | ||
3463 | return 1; | |
3464 | } | |
3465 | ||
899c81ea | 3466 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3467 | struct btrfs_balance_args *bargs) |
3468 | { | |
3469 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3470 | return 0; | |
3471 | ||
899c81ea ID |
3472 | chunk_type = chunk_to_extended(chunk_type) & |
3473 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3474 | |
899c81ea | 3475 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3476 | return 1; |
3477 | ||
3478 | return 0; | |
3479 | } | |
3480 | ||
2ff7e61e | 3481 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3482 | struct extent_buffer *leaf, |
3483 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3484 | { | |
0b246afa | 3485 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3486 | struct btrfs_balance_args *bargs = NULL; |
3487 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3488 | ||
3489 | /* type filter */ | |
3490 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3491 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3492 | return 0; | |
3493 | } | |
3494 | ||
3495 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3496 | bargs = &bctl->data; | |
3497 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3498 | bargs = &bctl->sys; | |
3499 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3500 | bargs = &bctl->meta; | |
3501 | ||
ed25e9b2 ID |
3502 | /* profiles filter */ |
3503 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3504 | chunk_profiles_filter(chunk_type, bargs)) { | |
3505 | return 0; | |
5ce5b3c0 ID |
3506 | } |
3507 | ||
3508 | /* usage filter */ | |
3509 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3510 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3511 | return 0; |
bc309467 | 3512 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3513 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3514 | return 0; |
409d404b ID |
3515 | } |
3516 | ||
3517 | /* devid filter */ | |
3518 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3519 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3520 | return 0; | |
94e60d5a ID |
3521 | } |
3522 | ||
3523 | /* drange filter, makes sense only with devid filter */ | |
3524 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3525 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3526 | return 0; |
ea67176a ID |
3527 | } |
3528 | ||
3529 | /* vrange filter */ | |
3530 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3531 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3532 | return 0; | |
ed25e9b2 ID |
3533 | } |
3534 | ||
dee32d0a GAP |
3535 | /* stripes filter */ |
3536 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3537 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3538 | return 0; | |
3539 | } | |
3540 | ||
cfa4c961 ID |
3541 | /* soft profile changing mode */ |
3542 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3543 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3544 | return 0; | |
3545 | } | |
3546 | ||
7d824b6f DS |
3547 | /* |
3548 | * limited by count, must be the last filter | |
3549 | */ | |
3550 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3551 | if (bargs->limit == 0) | |
3552 | return 0; | |
3553 | else | |
3554 | bargs->limit--; | |
12907fc7 DS |
3555 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3556 | /* | |
3557 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3558 | * determined here because we do not have the global information |
12907fc7 DS |
3559 | * about the count of all chunks that satisfy the filters. |
3560 | */ | |
3561 | if (bargs->limit_max == 0) | |
3562 | return 0; | |
3563 | else | |
3564 | bargs->limit_max--; | |
7d824b6f DS |
3565 | } |
3566 | ||
f43ffb60 ID |
3567 | return 1; |
3568 | } | |
3569 | ||
c9e9f97b | 3570 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3571 | { |
19a39dce | 3572 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3573 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3574 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3575 | struct list_head *devices; | |
ec44a35c CM |
3576 | struct btrfs_device *device; |
3577 | u64 old_size; | |
3578 | u64 size_to_free; | |
12907fc7 | 3579 | u64 chunk_type; |
f43ffb60 | 3580 | struct btrfs_chunk *chunk; |
5a488b9d | 3581 | struct btrfs_path *path = NULL; |
ec44a35c | 3582 | struct btrfs_key key; |
ec44a35c | 3583 | struct btrfs_key found_key; |
c9e9f97b | 3584 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3585 | struct extent_buffer *leaf; |
3586 | int slot; | |
c9e9f97b ID |
3587 | int ret; |
3588 | int enospc_errors = 0; | |
19a39dce | 3589 | bool counting = true; |
12907fc7 | 3590 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3591 | u64 limit_data = bctl->data.limit; |
3592 | u64 limit_meta = bctl->meta.limit; | |
3593 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3594 | u32 count_data = 0; |
3595 | u32 count_meta = 0; | |
3596 | u32 count_sys = 0; | |
2c9fe835 | 3597 | int chunk_reserved = 0; |
ec44a35c | 3598 | |
ec44a35c | 3599 | /* step one make some room on all the devices */ |
c9e9f97b | 3600 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3601 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3602 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3603 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3604 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3605 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3606 | btrfs_device_get_total_bytes(device) - |
3607 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3608 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3609 | continue; |
3610 | ||
3611 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3612 | if (ret == -ENOSPC) |
3613 | break; | |
5a488b9d LB |
3614 | if (ret) { |
3615 | /* btrfs_shrink_device never returns ret > 0 */ | |
3616 | WARN_ON(ret > 0); | |
3617 | goto error; | |
3618 | } | |
ec44a35c | 3619 | |
a22285a6 | 3620 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3621 | if (IS_ERR(trans)) { |
3622 | ret = PTR_ERR(trans); | |
3623 | btrfs_info_in_rcu(fs_info, | |
3624 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3625 | rcu_str_deref(device->name), ret, | |
3626 | old_size, old_size - size_to_free); | |
3627 | goto error; | |
3628 | } | |
ec44a35c CM |
3629 | |
3630 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3631 | if (ret) { |
3a45bb20 | 3632 | btrfs_end_transaction(trans); |
5a488b9d LB |
3633 | /* btrfs_grow_device never returns ret > 0 */ |
3634 | WARN_ON(ret > 0); | |
3635 | btrfs_info_in_rcu(fs_info, | |
3636 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3637 | rcu_str_deref(device->name), ret, | |
3638 | old_size, old_size - size_to_free); | |
3639 | goto error; | |
3640 | } | |
ec44a35c | 3641 | |
3a45bb20 | 3642 | btrfs_end_transaction(trans); |
ec44a35c CM |
3643 | } |
3644 | ||
3645 | /* step two, relocate all the chunks */ | |
3646 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3647 | if (!path) { |
3648 | ret = -ENOMEM; | |
3649 | goto error; | |
3650 | } | |
19a39dce ID |
3651 | |
3652 | /* zero out stat counters */ | |
3653 | spin_lock(&fs_info->balance_lock); | |
3654 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3655 | spin_unlock(&fs_info->balance_lock); | |
3656 | again: | |
7d824b6f | 3657 | if (!counting) { |
12907fc7 DS |
3658 | /* |
3659 | * The single value limit and min/max limits use the same bytes | |
3660 | * in the | |
3661 | */ | |
7d824b6f DS |
3662 | bctl->data.limit = limit_data; |
3663 | bctl->meta.limit = limit_meta; | |
3664 | bctl->sys.limit = limit_sys; | |
3665 | } | |
ec44a35c CM |
3666 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3667 | key.offset = (u64)-1; | |
3668 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3669 | ||
d397712b | 3670 | while (1) { |
19a39dce | 3671 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3672 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3673 | ret = -ECANCELED; |
3674 | goto error; | |
3675 | } | |
3676 | ||
67c5e7d4 | 3677 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3678 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3679 | if (ret < 0) { |
3680 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3681 | goto error; |
67c5e7d4 | 3682 | } |
ec44a35c CM |
3683 | |
3684 | /* | |
3685 | * this shouldn't happen, it means the last relocate | |
3686 | * failed | |
3687 | */ | |
3688 | if (ret == 0) | |
c9e9f97b | 3689 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3690 | |
3691 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3692 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3693 | if (ret) { |
67c5e7d4 | 3694 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3695 | ret = 0; |
ec44a35c | 3696 | break; |
c9e9f97b | 3697 | } |
7d9eb12c | 3698 | |
f43ffb60 ID |
3699 | leaf = path->nodes[0]; |
3700 | slot = path->slots[0]; | |
3701 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3702 | |
67c5e7d4 FM |
3703 | if (found_key.objectid != key.objectid) { |
3704 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3705 | break; |
67c5e7d4 | 3706 | } |
7d9eb12c | 3707 | |
f43ffb60 | 3708 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3709 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3710 | |
19a39dce ID |
3711 | if (!counting) { |
3712 | spin_lock(&fs_info->balance_lock); | |
3713 | bctl->stat.considered++; | |
3714 | spin_unlock(&fs_info->balance_lock); | |
3715 | } | |
3716 | ||
2ff7e61e | 3717 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3718 | found_key.offset); |
2c9fe835 | 3719 | |
b3b4aa74 | 3720 | btrfs_release_path(path); |
67c5e7d4 FM |
3721 | if (!ret) { |
3722 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3723 | goto loop; |
67c5e7d4 | 3724 | } |
f43ffb60 | 3725 | |
19a39dce | 3726 | if (counting) { |
67c5e7d4 | 3727 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3728 | spin_lock(&fs_info->balance_lock); |
3729 | bctl->stat.expected++; | |
3730 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3731 | |
3732 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3733 | count_data++; | |
3734 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3735 | count_sys++; | |
3736 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3737 | count_meta++; | |
3738 | ||
3739 | goto loop; | |
3740 | } | |
3741 | ||
3742 | /* | |
3743 | * Apply limit_min filter, no need to check if the LIMITS | |
3744 | * filter is used, limit_min is 0 by default | |
3745 | */ | |
3746 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3747 | count_data < bctl->data.limit_min) | |
3748 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3749 | count_meta < bctl->meta.limit_min) | |
3750 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3751 | count_sys < bctl->sys.limit_min)) { | |
3752 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3753 | goto loop; |
3754 | } | |
3755 | ||
a6f93c71 LB |
3756 | if (!chunk_reserved) { |
3757 | /* | |
3758 | * We may be relocating the only data chunk we have, | |
3759 | * which could potentially end up with losing data's | |
3760 | * raid profile, so lets allocate an empty one in | |
3761 | * advance. | |
3762 | */ | |
3763 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3764 | found_key.offset); | |
2c9fe835 ZL |
3765 | if (ret < 0) { |
3766 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3767 | goto error; | |
a6f93c71 LB |
3768 | } else if (ret == 1) { |
3769 | chunk_reserved = 1; | |
2c9fe835 | 3770 | } |
2c9fe835 ZL |
3771 | } |
3772 | ||
5b4aacef | 3773 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3774 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3775 | if (ret && ret != -ENOSPC) |
3776 | goto error; | |
19a39dce | 3777 | if (ret == -ENOSPC) { |
c9e9f97b | 3778 | enospc_errors++; |
19a39dce ID |
3779 | } else { |
3780 | spin_lock(&fs_info->balance_lock); | |
3781 | bctl->stat.completed++; | |
3782 | spin_unlock(&fs_info->balance_lock); | |
3783 | } | |
f43ffb60 | 3784 | loop: |
795a3321 ID |
3785 | if (found_key.offset == 0) |
3786 | break; | |
ba1bf481 | 3787 | key.offset = found_key.offset - 1; |
ec44a35c | 3788 | } |
c9e9f97b | 3789 | |
19a39dce ID |
3790 | if (counting) { |
3791 | btrfs_release_path(path); | |
3792 | counting = false; | |
3793 | goto again; | |
3794 | } | |
ec44a35c CM |
3795 | error: |
3796 | btrfs_free_path(path); | |
c9e9f97b | 3797 | if (enospc_errors) { |
efe120a0 | 3798 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3799 | enospc_errors); |
c9e9f97b ID |
3800 | if (!ret) |
3801 | ret = -ENOSPC; | |
3802 | } | |
3803 | ||
ec44a35c CM |
3804 | return ret; |
3805 | } | |
3806 | ||
0c460c0d ID |
3807 | /** |
3808 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3809 | * @flags: profile to validate | |
3810 | * @extended: if true @flags is treated as an extended profile | |
3811 | */ | |
3812 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3813 | { | |
3814 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3815 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3816 | ||
3817 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3818 | ||
3819 | /* 1) check that all other bits are zeroed */ | |
3820 | if (flags & ~mask) | |
3821 | return 0; | |
3822 | ||
3823 | /* 2) see if profile is reduced */ | |
3824 | if (flags == 0) | |
3825 | return !extended; /* "0" is valid for usual profiles */ | |
3826 | ||
3827 | /* true if exactly one bit set */ | |
3828 | return (flags & (flags - 1)) == 0; | |
3829 | } | |
3830 | ||
837d5b6e ID |
3831 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3832 | { | |
a7e99c69 ID |
3833 | /* cancel requested || normal exit path */ |
3834 | return atomic_read(&fs_info->balance_cancel_req) || | |
3835 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3836 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3837 | } |
3838 | ||
c9e9f97b ID |
3839 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3840 | { | |
0940ebf6 ID |
3841 | int ret; |
3842 | ||
c9e9f97b | 3843 | unset_balance_control(fs_info); |
6bccf3ab | 3844 | ret = del_balance_item(fs_info); |
0f788c58 | 3845 | if (ret) |
34d97007 | 3846 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f | 3847 | |
171938e5 | 3848 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
c9e9f97b ID |
3849 | } |
3850 | ||
bdcd3c97 AM |
3851 | /* Non-zero return value signifies invalidity */ |
3852 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3853 | u64 allowed) | |
3854 | { | |
3855 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3856 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3857 | (bctl_arg->target & ~allowed))); | |
3858 | } | |
3859 | ||
c9e9f97b ID |
3860 | /* |
3861 | * Should be called with both balance and volume mutexes held | |
3862 | */ | |
3863 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3864 | struct btrfs_ioctl_balance_args *bargs) | |
3865 | { | |
3866 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
14506127 | 3867 | u64 meta_target, data_target; |
f43ffb60 | 3868 | u64 allowed; |
e4837f8f | 3869 | int mixed = 0; |
c9e9f97b | 3870 | int ret; |
8dabb742 | 3871 | u64 num_devices; |
de98ced9 | 3872 | unsigned seq; |
c9e9f97b | 3873 | |
837d5b6e | 3874 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3875 | atomic_read(&fs_info->balance_pause_req) || |
3876 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3877 | ret = -EINVAL; |
3878 | goto out; | |
3879 | } | |
3880 | ||
e4837f8f ID |
3881 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3882 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3883 | mixed = 1; | |
3884 | ||
f43ffb60 ID |
3885 | /* |
3886 | * In case of mixed groups both data and meta should be picked, | |
3887 | * and identical options should be given for both of them. | |
3888 | */ | |
e4837f8f ID |
3889 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3890 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3891 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3892 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3893 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3894 | btrfs_err(fs_info, |
3895 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3896 | ret = -EINVAL; |
3897 | goto out; | |
3898 | } | |
3899 | } | |
3900 | ||
8dabb742 | 3901 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3902 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3903 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3904 | BUG_ON(num_devices < 1); | |
3905 | num_devices--; | |
3906 | } | |
73beece9 | 3907 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
88be159c AH |
3908 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3909 | if (num_devices > 1) | |
e4d8ec0f | 3910 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3911 | if (num_devices > 2) |
3912 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3913 | if (num_devices > 3) | |
3914 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3915 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3916 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3917 | btrfs_err(fs_info, |
3918 | "unable to start balance with target data profile %llu", | |
3919 | bctl->data.target); | |
e4d8ec0f ID |
3920 | ret = -EINVAL; |
3921 | goto out; | |
3922 | } | |
bdcd3c97 | 3923 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3924 | btrfs_err(fs_info, |
5d163e0e JM |
3925 | "unable to start balance with target metadata profile %llu", |
3926 | bctl->meta.target); | |
e4d8ec0f ID |
3927 | ret = -EINVAL; |
3928 | goto out; | |
3929 | } | |
bdcd3c97 | 3930 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3931 | btrfs_err(fs_info, |
5d163e0e JM |
3932 | "unable to start balance with target system profile %llu", |
3933 | bctl->sys.target); | |
e4d8ec0f ID |
3934 | ret = -EINVAL; |
3935 | goto out; | |
3936 | } | |
3937 | ||
e4d8ec0f ID |
3938 | /* allow to reduce meta or sys integrity only if force set */ |
3939 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3940 | BTRFS_BLOCK_GROUP_RAID10 | |
3941 | BTRFS_BLOCK_GROUP_RAID5 | | |
3942 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3943 | do { |
3944 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3945 | ||
3946 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3947 | (fs_info->avail_system_alloc_bits & allowed) && | |
3948 | !(bctl->sys.target & allowed)) || | |
3949 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3950 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3951 | !(bctl->meta.target & allowed))) { | |
3952 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3953 | btrfs_info(fs_info, |
3954 | "force reducing metadata integrity"); | |
de98ced9 | 3955 | } else { |
5d163e0e JM |
3956 | btrfs_err(fs_info, |
3957 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3958 | ret = -EINVAL; |
3959 | goto out; | |
3960 | } | |
e4d8ec0f | 3961 | } |
de98ced9 | 3962 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3963 | |
14506127 AB |
3964 | /* if we're not converting, the target field is uninitialized */ |
3965 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3966 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3967 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3968 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3969 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3970 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 3971 | btrfs_warn(fs_info, |
5d163e0e | 3972 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
14506127 | 3973 | meta_target, data_target); |
ee592d07 ST |
3974 | } |
3975 | ||
6bccf3ab | 3976 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3977 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3978 | goto out; |
3979 | ||
59641015 ID |
3980 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3981 | BUG_ON(ret == -EEXIST); | |
3982 | set_balance_control(bctl); | |
3983 | } else { | |
3984 | BUG_ON(ret != -EEXIST); | |
3985 | spin_lock(&fs_info->balance_lock); | |
3986 | update_balance_args(bctl); | |
3987 | spin_unlock(&fs_info->balance_lock); | |
3988 | } | |
c9e9f97b | 3989 | |
837d5b6e | 3990 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3991 | mutex_unlock(&fs_info->balance_mutex); |
3992 | ||
3993 | ret = __btrfs_balance(fs_info); | |
3994 | ||
3995 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3996 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
3997 | |
3998 | if (bargs) { | |
3999 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 4000 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
4001 | } |
4002 | ||
3a01aa7a ID |
4003 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4004 | balance_need_close(fs_info)) { | |
4005 | __cancel_balance(fs_info); | |
4006 | } | |
4007 | ||
837d5b6e | 4008 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4009 | |
4010 | return ret; | |
4011 | out: | |
59641015 ID |
4012 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
4013 | __cancel_balance(fs_info); | |
ed0fb78f | 4014 | else { |
59641015 | 4015 | kfree(bctl); |
171938e5 | 4016 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
ed0fb78f | 4017 | } |
59641015 ID |
4018 | return ret; |
4019 | } | |
4020 | ||
4021 | static int balance_kthread(void *data) | |
4022 | { | |
2b6ba629 | 4023 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4024 | int ret = 0; |
59641015 ID |
4025 | |
4026 | mutex_lock(&fs_info->volume_mutex); | |
4027 | mutex_lock(&fs_info->balance_mutex); | |
4028 | ||
2b6ba629 | 4029 | if (fs_info->balance_ctl) { |
efe120a0 | 4030 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 4031 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 4032 | } |
59641015 ID |
4033 | |
4034 | mutex_unlock(&fs_info->balance_mutex); | |
4035 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 4036 | |
59641015 ID |
4037 | return ret; |
4038 | } | |
4039 | ||
2b6ba629 ID |
4040 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4041 | { | |
4042 | struct task_struct *tsk; | |
4043 | ||
4044 | spin_lock(&fs_info->balance_lock); | |
4045 | if (!fs_info->balance_ctl) { | |
4046 | spin_unlock(&fs_info->balance_lock); | |
4047 | return 0; | |
4048 | } | |
4049 | spin_unlock(&fs_info->balance_lock); | |
4050 | ||
3cdde224 | 4051 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 4052 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
4053 | return 0; |
4054 | } | |
4055 | ||
4056 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 4057 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4058 | } |
4059 | ||
68310a5e | 4060 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4061 | { |
59641015 ID |
4062 | struct btrfs_balance_control *bctl; |
4063 | struct btrfs_balance_item *item; | |
4064 | struct btrfs_disk_balance_args disk_bargs; | |
4065 | struct btrfs_path *path; | |
4066 | struct extent_buffer *leaf; | |
4067 | struct btrfs_key key; | |
4068 | int ret; | |
4069 | ||
4070 | path = btrfs_alloc_path(); | |
4071 | if (!path) | |
4072 | return -ENOMEM; | |
4073 | ||
59641015 | 4074 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4075 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4076 | key.offset = 0; |
4077 | ||
68310a5e | 4078 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4079 | if (ret < 0) |
68310a5e | 4080 | goto out; |
59641015 ID |
4081 | if (ret > 0) { /* ret = -ENOENT; */ |
4082 | ret = 0; | |
68310a5e ID |
4083 | goto out; |
4084 | } | |
4085 | ||
4086 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4087 | if (!bctl) { | |
4088 | ret = -ENOMEM; | |
4089 | goto out; | |
59641015 ID |
4090 | } |
4091 | ||
4092 | leaf = path->nodes[0]; | |
4093 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4094 | ||
68310a5e ID |
4095 | bctl->fs_info = fs_info; |
4096 | bctl->flags = btrfs_balance_flags(leaf, item); | |
4097 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4098 | |
4099 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4100 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4101 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4102 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4103 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4104 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4105 | ||
171938e5 | 4106 | WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)); |
ed0fb78f | 4107 | |
68310a5e ID |
4108 | mutex_lock(&fs_info->volume_mutex); |
4109 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 4110 | |
68310a5e ID |
4111 | set_balance_control(bctl); |
4112 | ||
4113 | mutex_unlock(&fs_info->balance_mutex); | |
4114 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4115 | out: |
4116 | btrfs_free_path(path); | |
ec44a35c CM |
4117 | return ret; |
4118 | } | |
4119 | ||
837d5b6e ID |
4120 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4121 | { | |
4122 | int ret = 0; | |
4123 | ||
4124 | mutex_lock(&fs_info->balance_mutex); | |
4125 | if (!fs_info->balance_ctl) { | |
4126 | mutex_unlock(&fs_info->balance_mutex); | |
4127 | return -ENOTCONN; | |
4128 | } | |
4129 | ||
4130 | if (atomic_read(&fs_info->balance_running)) { | |
4131 | atomic_inc(&fs_info->balance_pause_req); | |
4132 | mutex_unlock(&fs_info->balance_mutex); | |
4133 | ||
4134 | wait_event(fs_info->balance_wait_q, | |
4135 | atomic_read(&fs_info->balance_running) == 0); | |
4136 | ||
4137 | mutex_lock(&fs_info->balance_mutex); | |
4138 | /* we are good with balance_ctl ripped off from under us */ | |
4139 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4140 | atomic_dec(&fs_info->balance_pause_req); | |
4141 | } else { | |
4142 | ret = -ENOTCONN; | |
4143 | } | |
4144 | ||
4145 | mutex_unlock(&fs_info->balance_mutex); | |
4146 | return ret; | |
4147 | } | |
4148 | ||
a7e99c69 ID |
4149 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4150 | { | |
bc98a42c | 4151 | if (sb_rdonly(fs_info->sb)) |
e649e587 ID |
4152 | return -EROFS; |
4153 | ||
a7e99c69 ID |
4154 | mutex_lock(&fs_info->balance_mutex); |
4155 | if (!fs_info->balance_ctl) { | |
4156 | mutex_unlock(&fs_info->balance_mutex); | |
4157 | return -ENOTCONN; | |
4158 | } | |
4159 | ||
4160 | atomic_inc(&fs_info->balance_cancel_req); | |
4161 | /* | |
4162 | * if we are running just wait and return, balance item is | |
4163 | * deleted in btrfs_balance in this case | |
4164 | */ | |
4165 | if (atomic_read(&fs_info->balance_running)) { | |
4166 | mutex_unlock(&fs_info->balance_mutex); | |
4167 | wait_event(fs_info->balance_wait_q, | |
4168 | atomic_read(&fs_info->balance_running) == 0); | |
4169 | mutex_lock(&fs_info->balance_mutex); | |
4170 | } else { | |
4171 | /* __cancel_balance needs volume_mutex */ | |
4172 | mutex_unlock(&fs_info->balance_mutex); | |
4173 | mutex_lock(&fs_info->volume_mutex); | |
4174 | mutex_lock(&fs_info->balance_mutex); | |
4175 | ||
4176 | if (fs_info->balance_ctl) | |
4177 | __cancel_balance(fs_info); | |
4178 | ||
4179 | mutex_unlock(&fs_info->volume_mutex); | |
4180 | } | |
4181 | ||
4182 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4183 | atomic_dec(&fs_info->balance_cancel_req); | |
4184 | mutex_unlock(&fs_info->balance_mutex); | |
4185 | return 0; | |
4186 | } | |
4187 | ||
803b2f54 SB |
4188 | static int btrfs_uuid_scan_kthread(void *data) |
4189 | { | |
4190 | struct btrfs_fs_info *fs_info = data; | |
4191 | struct btrfs_root *root = fs_info->tree_root; | |
4192 | struct btrfs_key key; | |
803b2f54 SB |
4193 | struct btrfs_path *path = NULL; |
4194 | int ret = 0; | |
4195 | struct extent_buffer *eb; | |
4196 | int slot; | |
4197 | struct btrfs_root_item root_item; | |
4198 | u32 item_size; | |
f45388f3 | 4199 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4200 | |
4201 | path = btrfs_alloc_path(); | |
4202 | if (!path) { | |
4203 | ret = -ENOMEM; | |
4204 | goto out; | |
4205 | } | |
4206 | ||
4207 | key.objectid = 0; | |
4208 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4209 | key.offset = 0; | |
4210 | ||
803b2f54 | 4211 | while (1) { |
6174d3cb | 4212 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4213 | if (ret) { |
4214 | if (ret > 0) | |
4215 | ret = 0; | |
4216 | break; | |
4217 | } | |
4218 | ||
4219 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4220 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4221 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4222 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4223 | goto skip; | |
4224 | ||
4225 | eb = path->nodes[0]; | |
4226 | slot = path->slots[0]; | |
4227 | item_size = btrfs_item_size_nr(eb, slot); | |
4228 | if (item_size < sizeof(root_item)) | |
4229 | goto skip; | |
4230 | ||
803b2f54 SB |
4231 | read_extent_buffer(eb, &root_item, |
4232 | btrfs_item_ptr_offset(eb, slot), | |
4233 | (int)sizeof(root_item)); | |
4234 | if (btrfs_root_refs(&root_item) == 0) | |
4235 | goto skip; | |
f45388f3 FDBM |
4236 | |
4237 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4238 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4239 | if (trans) | |
4240 | goto update_tree; | |
4241 | ||
4242 | btrfs_release_path(path); | |
803b2f54 SB |
4243 | /* |
4244 | * 1 - subvol uuid item | |
4245 | * 1 - received_subvol uuid item | |
4246 | */ | |
4247 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4248 | if (IS_ERR(trans)) { | |
4249 | ret = PTR_ERR(trans); | |
4250 | break; | |
4251 | } | |
f45388f3 FDBM |
4252 | continue; |
4253 | } else { | |
4254 | goto skip; | |
4255 | } | |
4256 | update_tree: | |
4257 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
6bccf3ab | 4258 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4259 | root_item.uuid, |
4260 | BTRFS_UUID_KEY_SUBVOL, | |
4261 | key.objectid); | |
4262 | if (ret < 0) { | |
efe120a0 | 4263 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4264 | ret); |
803b2f54 SB |
4265 | break; |
4266 | } | |
4267 | } | |
4268 | ||
4269 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
6bccf3ab | 4270 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4271 | root_item.received_uuid, |
4272 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4273 | key.objectid); | |
4274 | if (ret < 0) { | |
efe120a0 | 4275 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4276 | ret); |
803b2f54 SB |
4277 | break; |
4278 | } | |
4279 | } | |
4280 | ||
f45388f3 | 4281 | skip: |
803b2f54 | 4282 | if (trans) { |
3a45bb20 | 4283 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4284 | trans = NULL; |
803b2f54 SB |
4285 | if (ret) |
4286 | break; | |
4287 | } | |
4288 | ||
803b2f54 SB |
4289 | btrfs_release_path(path); |
4290 | if (key.offset < (u64)-1) { | |
4291 | key.offset++; | |
4292 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4293 | key.offset = 0; | |
4294 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4295 | } else if (key.objectid < (u64)-1) { | |
4296 | key.offset = 0; | |
4297 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4298 | key.objectid++; | |
4299 | } else { | |
4300 | break; | |
4301 | } | |
4302 | cond_resched(); | |
4303 | } | |
4304 | ||
4305 | out: | |
4306 | btrfs_free_path(path); | |
f45388f3 | 4307 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4308 | btrfs_end_transaction(trans); |
803b2f54 | 4309 | if (ret) |
efe120a0 | 4310 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4311 | else |
afcdd129 | 4312 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4313 | up(&fs_info->uuid_tree_rescan_sem); |
4314 | return 0; | |
4315 | } | |
4316 | ||
70f80175 SB |
4317 | /* |
4318 | * Callback for btrfs_uuid_tree_iterate(). | |
4319 | * returns: | |
4320 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4321 | * < 0 if an error occurred. |
70f80175 SB |
4322 | * > 0 if the check failed, which means the caller shall remove the entry. |
4323 | */ | |
4324 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4325 | u8 *uuid, u8 type, u64 subid) | |
4326 | { | |
4327 | struct btrfs_key key; | |
4328 | int ret = 0; | |
4329 | struct btrfs_root *subvol_root; | |
4330 | ||
4331 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4332 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4333 | goto out; | |
4334 | ||
4335 | key.objectid = subid; | |
4336 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4337 | key.offset = (u64)-1; | |
4338 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4339 | if (IS_ERR(subvol_root)) { | |
4340 | ret = PTR_ERR(subvol_root); | |
4341 | if (ret == -ENOENT) | |
4342 | ret = 1; | |
4343 | goto out; | |
4344 | } | |
4345 | ||
4346 | switch (type) { | |
4347 | case BTRFS_UUID_KEY_SUBVOL: | |
4348 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4349 | ret = 1; | |
4350 | break; | |
4351 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4352 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4353 | BTRFS_UUID_SIZE)) | |
4354 | ret = 1; | |
4355 | break; | |
4356 | } | |
4357 | ||
4358 | out: | |
4359 | return ret; | |
4360 | } | |
4361 | ||
4362 | static int btrfs_uuid_rescan_kthread(void *data) | |
4363 | { | |
4364 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4365 | int ret; | |
4366 | ||
4367 | /* | |
4368 | * 1st step is to iterate through the existing UUID tree and | |
4369 | * to delete all entries that contain outdated data. | |
4370 | * 2nd step is to add all missing entries to the UUID tree. | |
4371 | */ | |
4372 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4373 | if (ret < 0) { | |
efe120a0 | 4374 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4375 | up(&fs_info->uuid_tree_rescan_sem); |
4376 | return ret; | |
4377 | } | |
4378 | return btrfs_uuid_scan_kthread(data); | |
4379 | } | |
4380 | ||
f7a81ea4 SB |
4381 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4382 | { | |
4383 | struct btrfs_trans_handle *trans; | |
4384 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4385 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4386 | struct task_struct *task; |
4387 | int ret; | |
f7a81ea4 SB |
4388 | |
4389 | /* | |
4390 | * 1 - root node | |
4391 | * 1 - root item | |
4392 | */ | |
4393 | trans = btrfs_start_transaction(tree_root, 2); | |
4394 | if (IS_ERR(trans)) | |
4395 | return PTR_ERR(trans); | |
4396 | ||
4397 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4398 | BTRFS_UUID_TREE_OBJECTID); | |
4399 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4400 | ret = PTR_ERR(uuid_root); |
66642832 | 4401 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4402 | btrfs_end_transaction(trans); |
6d13f549 | 4403 | return ret; |
f7a81ea4 SB |
4404 | } |
4405 | ||
4406 | fs_info->uuid_root = uuid_root; | |
4407 | ||
3a45bb20 | 4408 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4409 | if (ret) |
4410 | return ret; | |
4411 | ||
4412 | down(&fs_info->uuid_tree_rescan_sem); | |
4413 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4414 | if (IS_ERR(task)) { | |
70f80175 | 4415 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4416 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4417 | up(&fs_info->uuid_tree_rescan_sem); |
4418 | return PTR_ERR(task); | |
4419 | } | |
4420 | ||
4421 | return 0; | |
f7a81ea4 | 4422 | } |
803b2f54 | 4423 | |
70f80175 SB |
4424 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4425 | { | |
4426 | struct task_struct *task; | |
4427 | ||
4428 | down(&fs_info->uuid_tree_rescan_sem); | |
4429 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4430 | if (IS_ERR(task)) { | |
4431 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4432 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4433 | up(&fs_info->uuid_tree_rescan_sem); |
4434 | return PTR_ERR(task); | |
4435 | } | |
4436 | ||
4437 | return 0; | |
4438 | } | |
4439 | ||
8f18cf13 CM |
4440 | /* |
4441 | * shrinking a device means finding all of the device extents past | |
4442 | * the new size, and then following the back refs to the chunks. | |
4443 | * The chunk relocation code actually frees the device extent | |
4444 | */ | |
4445 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4446 | { | |
0b246afa JM |
4447 | struct btrfs_fs_info *fs_info = device->fs_info; |
4448 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4449 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4450 | struct btrfs_dev_extent *dev_extent = NULL; |
4451 | struct btrfs_path *path; | |
4452 | u64 length; | |
8f18cf13 CM |
4453 | u64 chunk_offset; |
4454 | int ret; | |
4455 | int slot; | |
ba1bf481 JB |
4456 | int failed = 0; |
4457 | bool retried = false; | |
53e489bc | 4458 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4459 | struct extent_buffer *l; |
4460 | struct btrfs_key key; | |
0b246afa | 4461 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4462 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4463 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4464 | u64 diff; |
4465 | ||
4466 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4467 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4468 | |
401e29c1 | 4469 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4470 | return -EINVAL; |
4471 | ||
8f18cf13 CM |
4472 | path = btrfs_alloc_path(); |
4473 | if (!path) | |
4474 | return -ENOMEM; | |
4475 | ||
e4058b54 | 4476 | path->reada = READA_FORWARD; |
8f18cf13 | 4477 | |
34441361 | 4478 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4479 | |
7cc8e58d | 4480 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4481 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4482 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4483 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4484 | } |
34441361 | 4485 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4486 | |
ba1bf481 | 4487 | again: |
8f18cf13 CM |
4488 | key.objectid = device->devid; |
4489 | key.offset = (u64)-1; | |
4490 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4491 | ||
213e64da | 4492 | do { |
0b246afa | 4493 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4494 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4495 | if (ret < 0) { |
0b246afa | 4496 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4497 | goto done; |
67c5e7d4 | 4498 | } |
8f18cf13 CM |
4499 | |
4500 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4501 | if (ret) |
0b246afa | 4502 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4503 | if (ret < 0) |
4504 | goto done; | |
4505 | if (ret) { | |
4506 | ret = 0; | |
b3b4aa74 | 4507 | btrfs_release_path(path); |
bf1fb512 | 4508 | break; |
8f18cf13 CM |
4509 | } |
4510 | ||
4511 | l = path->nodes[0]; | |
4512 | slot = path->slots[0]; | |
4513 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4514 | ||
ba1bf481 | 4515 | if (key.objectid != device->devid) { |
0b246afa | 4516 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4517 | btrfs_release_path(path); |
bf1fb512 | 4518 | break; |
ba1bf481 | 4519 | } |
8f18cf13 CM |
4520 | |
4521 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4522 | length = btrfs_dev_extent_length(l, dev_extent); | |
4523 | ||
ba1bf481 | 4524 | if (key.offset + length <= new_size) { |
0b246afa | 4525 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4526 | btrfs_release_path(path); |
d6397bae | 4527 | break; |
ba1bf481 | 4528 | } |
8f18cf13 | 4529 | |
8f18cf13 | 4530 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4531 | btrfs_release_path(path); |
8f18cf13 | 4532 | |
a6f93c71 LB |
4533 | /* |
4534 | * We may be relocating the only data chunk we have, | |
4535 | * which could potentially end up with losing data's | |
4536 | * raid profile, so lets allocate an empty one in | |
4537 | * advance. | |
4538 | */ | |
4539 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4540 | if (ret < 0) { | |
4541 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4542 | goto done; | |
4543 | } | |
4544 | ||
0b246afa JM |
4545 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4546 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4547 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4548 | goto done; |
ba1bf481 JB |
4549 | if (ret == -ENOSPC) |
4550 | failed++; | |
213e64da | 4551 | } while (key.offset-- > 0); |
ba1bf481 JB |
4552 | |
4553 | if (failed && !retried) { | |
4554 | failed = 0; | |
4555 | retried = true; | |
4556 | goto again; | |
4557 | } else if (failed && retried) { | |
4558 | ret = -ENOSPC; | |
ba1bf481 | 4559 | goto done; |
8f18cf13 CM |
4560 | } |
4561 | ||
d6397bae | 4562 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4563 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4564 | if (IS_ERR(trans)) { |
4565 | ret = PTR_ERR(trans); | |
4566 | goto done; | |
4567 | } | |
4568 | ||
34441361 | 4569 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4570 | |
4571 | /* | |
4572 | * We checked in the above loop all device extents that were already in | |
4573 | * the device tree. However before we have updated the device's | |
4574 | * total_bytes to the new size, we might have had chunk allocations that | |
4575 | * have not complete yet (new block groups attached to transaction | |
4576 | * handles), and therefore their device extents were not yet in the | |
4577 | * device tree and we missed them in the loop above. So if we have any | |
4578 | * pending chunk using a device extent that overlaps the device range | |
4579 | * that we can not use anymore, commit the current transaction and | |
4580 | * repeat the search on the device tree - this way we guarantee we will | |
4581 | * not have chunks using device extents that end beyond 'new_size'. | |
4582 | */ | |
4583 | if (!checked_pending_chunks) { | |
4584 | u64 start = new_size; | |
4585 | u64 len = old_size - new_size; | |
4586 | ||
499f377f JM |
4587 | if (contains_pending_extent(trans->transaction, device, |
4588 | &start, len)) { | |
34441361 | 4589 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4590 | checked_pending_chunks = true; |
4591 | failed = 0; | |
4592 | retried = false; | |
3a45bb20 | 4593 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4594 | if (ret) |
4595 | goto done; | |
4596 | goto again; | |
4597 | } | |
4598 | } | |
4599 | ||
7cc8e58d | 4600 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4601 | if (list_empty(&device->resized_list)) |
4602 | list_add_tail(&device->resized_list, | |
0b246afa | 4603 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4604 | |
d6397bae | 4605 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4606 | btrfs_set_super_total_bytes(super_copy, |
4607 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4608 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4609 | |
4610 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4611 | ret = btrfs_update_device(trans, device); | |
3a45bb20 | 4612 | btrfs_end_transaction(trans); |
8f18cf13 CM |
4613 | done: |
4614 | btrfs_free_path(path); | |
53e489bc | 4615 | if (ret) { |
34441361 | 4616 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4617 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4618 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4619 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4620 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4621 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4622 | } |
8f18cf13 CM |
4623 | return ret; |
4624 | } | |
4625 | ||
2ff7e61e | 4626 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4627 | struct btrfs_key *key, |
4628 | struct btrfs_chunk *chunk, int item_size) | |
4629 | { | |
0b246afa | 4630 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4631 | struct btrfs_disk_key disk_key; |
4632 | u32 array_size; | |
4633 | u8 *ptr; | |
4634 | ||
34441361 | 4635 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4636 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4637 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4638 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4639 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4640 | return -EFBIG; |
fe48a5c0 | 4641 | } |
0b86a832 CM |
4642 | |
4643 | ptr = super_copy->sys_chunk_array + array_size; | |
4644 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4645 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4646 | ptr += sizeof(disk_key); | |
4647 | memcpy(ptr, chunk, item_size); | |
4648 | item_size += sizeof(disk_key); | |
4649 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4650 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4651 | |
0b86a832 CM |
4652 | return 0; |
4653 | } | |
4654 | ||
73c5de00 AJ |
4655 | /* |
4656 | * sort the devices in descending order by max_avail, total_avail | |
4657 | */ | |
4658 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4659 | { |
73c5de00 AJ |
4660 | const struct btrfs_device_info *di_a = a; |
4661 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4662 | |
73c5de00 | 4663 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4664 | return -1; |
73c5de00 | 4665 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4666 | return 1; |
73c5de00 AJ |
4667 | if (di_a->total_avail > di_b->total_avail) |
4668 | return -1; | |
4669 | if (di_a->total_avail < di_b->total_avail) | |
4670 | return 1; | |
4671 | return 0; | |
b2117a39 | 4672 | } |
0b86a832 | 4673 | |
53b381b3 DW |
4674 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4675 | { | |
ffe2d203 | 4676 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4677 | return; |
4678 | ||
ceda0864 | 4679 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4680 | } |
4681 | ||
da17066c | 4682 | #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \ |
23f8f9b7 GH |
4683 | - sizeof(struct btrfs_chunk)) \ |
4684 | / sizeof(struct btrfs_stripe) + 1) | |
4685 | ||
4686 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4687 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4688 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4689 | / sizeof(struct btrfs_stripe) + 1) | |
4690 | ||
73c5de00 | 4691 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4692 | u64 start, u64 type) |
b2117a39 | 4693 | { |
2ff7e61e | 4694 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4695 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4696 | struct btrfs_device *device; |
73c5de00 AJ |
4697 | struct map_lookup *map = NULL; |
4698 | struct extent_map_tree *em_tree; | |
4699 | struct extent_map *em; | |
4700 | struct btrfs_device_info *devices_info = NULL; | |
4701 | u64 total_avail; | |
4702 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4703 | int data_stripes; /* number of stripes that count for |
4704 | block group size */ | |
73c5de00 AJ |
4705 | int sub_stripes; /* sub_stripes info for map */ |
4706 | int dev_stripes; /* stripes per dev */ | |
4707 | int devs_max; /* max devs to use */ | |
4708 | int devs_min; /* min devs needed */ | |
4709 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4710 | int ncopies; /* how many copies to data has */ | |
4711 | int ret; | |
4712 | u64 max_stripe_size; | |
4713 | u64 max_chunk_size; | |
4714 | u64 stripe_size; | |
4715 | u64 num_bytes; | |
4716 | int ndevs; | |
4717 | int i; | |
4718 | int j; | |
31e50229 | 4719 | int index; |
593060d7 | 4720 | |
0c460c0d | 4721 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4722 | |
73c5de00 AJ |
4723 | if (list_empty(&fs_devices->alloc_list)) |
4724 | return -ENOSPC; | |
b2117a39 | 4725 | |
31e50229 | 4726 | index = __get_raid_index(type); |
73c5de00 | 4727 | |
31e50229 LB |
4728 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4729 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4730 | devs_max = btrfs_raid_array[index].devs_max; | |
4731 | devs_min = btrfs_raid_array[index].devs_min; | |
4732 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4733 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4734 | |
9b3f68b9 | 4735 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4736 | max_stripe_size = SZ_1G; |
73c5de00 | 4737 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 GH |
4738 | if (!devs_max) |
4739 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4740 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4741 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4742 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4743 | max_stripe_size = SZ_1G; | |
1100373f | 4744 | else |
ee22184b | 4745 | max_stripe_size = SZ_256M; |
73c5de00 | 4746 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4747 | if (!devs_max) |
4748 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4749 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4750 | max_stripe_size = SZ_32M; |
73c5de00 | 4751 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4752 | if (!devs_max) |
4753 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4754 | } else { |
351fd353 | 4755 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4756 | type); |
4757 | BUG_ON(1); | |
9b3f68b9 CM |
4758 | } |
4759 | ||
2b82032c YZ |
4760 | /* we don't want a chunk larger than 10% of writeable space */ |
4761 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4762 | max_chunk_size); | |
9b3f68b9 | 4763 | |
31e818fe | 4764 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4765 | GFP_NOFS); |
4766 | if (!devices_info) | |
4767 | return -ENOMEM; | |
0cad8a11 | 4768 | |
9f680ce0 | 4769 | /* |
73c5de00 AJ |
4770 | * in the first pass through the devices list, we gather information |
4771 | * about the available holes on each device. | |
9f680ce0 | 4772 | */ |
73c5de00 | 4773 | ndevs = 0; |
ebcc9301 | 4774 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4775 | u64 max_avail; |
4776 | u64 dev_offset; | |
b2117a39 | 4777 | |
ebbede42 | 4778 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4779 | WARN(1, KERN_ERR |
efe120a0 | 4780 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4781 | continue; |
4782 | } | |
b2117a39 | 4783 | |
e12c9621 AJ |
4784 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4785 | &device->dev_state) || | |
401e29c1 | 4786 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4787 | continue; |
b2117a39 | 4788 | |
73c5de00 AJ |
4789 | if (device->total_bytes > device->bytes_used) |
4790 | total_avail = device->total_bytes - device->bytes_used; | |
4791 | else | |
4792 | total_avail = 0; | |
38c01b96 | 4793 | |
4794 | /* If there is no space on this device, skip it. */ | |
4795 | if (total_avail == 0) | |
4796 | continue; | |
b2117a39 | 4797 | |
6df9a95e | 4798 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4799 | max_stripe_size * dev_stripes, |
4800 | &dev_offset, &max_avail); | |
4801 | if (ret && ret != -ENOSPC) | |
4802 | goto error; | |
b2117a39 | 4803 | |
73c5de00 AJ |
4804 | if (ret == 0) |
4805 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4806 | |
73c5de00 AJ |
4807 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4808 | continue; | |
b2117a39 | 4809 | |
063d006f ES |
4810 | if (ndevs == fs_devices->rw_devices) { |
4811 | WARN(1, "%s: found more than %llu devices\n", | |
4812 | __func__, fs_devices->rw_devices); | |
4813 | break; | |
4814 | } | |
73c5de00 AJ |
4815 | devices_info[ndevs].dev_offset = dev_offset; |
4816 | devices_info[ndevs].max_avail = max_avail; | |
4817 | devices_info[ndevs].total_avail = total_avail; | |
4818 | devices_info[ndevs].dev = device; | |
4819 | ++ndevs; | |
4820 | } | |
b2117a39 | 4821 | |
73c5de00 AJ |
4822 | /* |
4823 | * now sort the devices by hole size / available space | |
4824 | */ | |
4825 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4826 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4827 | |
73c5de00 | 4828 | /* round down to number of usable stripes */ |
e5600fd6 | 4829 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4830 | |
73c5de00 AJ |
4831 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4832 | ret = -ENOSPC; | |
4833 | goto error; | |
b2117a39 | 4834 | } |
9f680ce0 | 4835 | |
f148ef4d NB |
4836 | ndevs = min(ndevs, devs_max); |
4837 | ||
73c5de00 AJ |
4838 | /* |
4839 | * the primary goal is to maximize the number of stripes, so use as many | |
4840 | * devices as possible, even if the stripes are not maximum sized. | |
4841 | */ | |
4842 | stripe_size = devices_info[ndevs-1].max_avail; | |
4843 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4844 | |
53b381b3 DW |
4845 | /* |
4846 | * this will have to be fixed for RAID1 and RAID10 over | |
4847 | * more drives | |
4848 | */ | |
4849 | data_stripes = num_stripes / ncopies; | |
4850 | ||
500ceed8 | 4851 | if (type & BTRFS_BLOCK_GROUP_RAID5) |
53b381b3 | 4852 | data_stripes = num_stripes - 1; |
500ceed8 NB |
4853 | |
4854 | if (type & BTRFS_BLOCK_GROUP_RAID6) | |
53b381b3 | 4855 | data_stripes = num_stripes - 2; |
86db2578 CM |
4856 | |
4857 | /* | |
4858 | * Use the number of data stripes to figure out how big this chunk | |
4859 | * is really going to be in terms of logical address space, | |
4860 | * and compare that answer with the max chunk size | |
4861 | */ | |
4862 | if (stripe_size * data_stripes > max_chunk_size) { | |
4863 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4864 | |
4865 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4866 | |
4867 | /* bump the answer up to a 16MB boundary */ | |
4868 | stripe_size = (stripe_size + mask) & ~mask; | |
4869 | ||
4870 | /* but don't go higher than the limits we found | |
4871 | * while searching for free extents | |
4872 | */ | |
4873 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4874 | stripe_size = devices_info[ndevs-1].max_avail; | |
4875 | } | |
4876 | ||
b8b93add | 4877 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4878 | |
4879 | /* align to BTRFS_STRIPE_LEN */ | |
500ceed8 | 4880 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4881 | |
4882 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4883 | if (!map) { | |
4884 | ret = -ENOMEM; | |
4885 | goto error; | |
4886 | } | |
4887 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4888 | |
73c5de00 AJ |
4889 | for (i = 0; i < ndevs; ++i) { |
4890 | for (j = 0; j < dev_stripes; ++j) { | |
4891 | int s = i * dev_stripes + j; | |
4892 | map->stripes[s].dev = devices_info[i].dev; | |
4893 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4894 | j * stripe_size; | |
6324fbf3 | 4895 | } |
6324fbf3 | 4896 | } |
500ceed8 NB |
4897 | map->stripe_len = BTRFS_STRIPE_LEN; |
4898 | map->io_align = BTRFS_STRIPE_LEN; | |
4899 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4900 | map->type = type; |
2b82032c | 4901 | map->sub_stripes = sub_stripes; |
0b86a832 | 4902 | |
53b381b3 | 4903 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4904 | |
6bccf3ab | 4905 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4906 | |
172ddd60 | 4907 | em = alloc_extent_map(); |
2b82032c | 4908 | if (!em) { |
298a8f9c | 4909 | kfree(map); |
b2117a39 MX |
4910 | ret = -ENOMEM; |
4911 | goto error; | |
593060d7 | 4912 | } |
298a8f9c | 4913 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4914 | em->map_lookup = map; |
2b82032c | 4915 | em->start = start; |
73c5de00 | 4916 | em->len = num_bytes; |
2b82032c YZ |
4917 | em->block_start = 0; |
4918 | em->block_len = em->len; | |
6df9a95e | 4919 | em->orig_block_len = stripe_size; |
593060d7 | 4920 | |
0b246afa | 4921 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4922 | write_lock(&em_tree->lock); |
09a2a8f9 | 4923 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 4924 | if (ret) { |
1efb72a3 | 4925 | write_unlock(&em_tree->lock); |
0f5d42b2 | 4926 | free_extent_map(em); |
1dd4602f | 4927 | goto error; |
0f5d42b2 | 4928 | } |
0b86a832 | 4929 | |
1efb72a3 NB |
4930 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
4931 | refcount_inc(&em->refs); | |
4932 | write_unlock(&em_tree->lock); | |
4933 | ||
0174484d | 4934 | ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes); |
6df9a95e JB |
4935 | if (ret) |
4936 | goto error_del_extent; | |
2b82032c | 4937 | |
7cc8e58d MX |
4938 | for (i = 0; i < map->num_stripes; i++) { |
4939 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4940 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4941 | } | |
43530c46 | 4942 | |
a5ed45f8 | 4943 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4944 | |
0f5d42b2 | 4945 | free_extent_map(em); |
0b246afa | 4946 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4947 | |
b2117a39 | 4948 | kfree(devices_info); |
2b82032c | 4949 | return 0; |
b2117a39 | 4950 | |
6df9a95e | 4951 | error_del_extent: |
0f5d42b2 JB |
4952 | write_lock(&em_tree->lock); |
4953 | remove_extent_mapping(em_tree, em); | |
4954 | write_unlock(&em_tree->lock); | |
4955 | ||
4956 | /* One for our allocation */ | |
4957 | free_extent_map(em); | |
4958 | /* One for the tree reference */ | |
4959 | free_extent_map(em); | |
495e64f4 FM |
4960 | /* One for the pending_chunks list reference */ |
4961 | free_extent_map(em); | |
b2117a39 | 4962 | error: |
b2117a39 MX |
4963 | kfree(devices_info); |
4964 | return ret; | |
2b82032c YZ |
4965 | } |
4966 | ||
6df9a95e | 4967 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
6bccf3ab | 4968 | struct btrfs_fs_info *fs_info, |
6df9a95e | 4969 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4970 | { |
6bccf3ab JM |
4971 | struct btrfs_root *extent_root = fs_info->extent_root; |
4972 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4973 | struct btrfs_key key; |
2b82032c YZ |
4974 | struct btrfs_device *device; |
4975 | struct btrfs_chunk *chunk; | |
4976 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4977 | struct extent_map *em; |
4978 | struct map_lookup *map; | |
4979 | size_t item_size; | |
4980 | u64 dev_offset; | |
4981 | u64 stripe_size; | |
4982 | int i = 0; | |
140e639f | 4983 | int ret = 0; |
2b82032c | 4984 | |
592d92ee LB |
4985 | em = get_chunk_map(fs_info, chunk_offset, chunk_size); |
4986 | if (IS_ERR(em)) | |
4987 | return PTR_ERR(em); | |
6df9a95e | 4988 | |
95617d69 | 4989 | map = em->map_lookup; |
6df9a95e JB |
4990 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4991 | stripe_size = em->orig_block_len; | |
4992 | ||
2b82032c | 4993 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4994 | if (!chunk) { |
4995 | ret = -ENOMEM; | |
4996 | goto out; | |
4997 | } | |
4998 | ||
50460e37 FM |
4999 | /* |
5000 | * Take the device list mutex to prevent races with the final phase of | |
5001 | * a device replace operation that replaces the device object associated | |
5002 | * with the map's stripes, because the device object's id can change | |
5003 | * at any time during that final phase of the device replace operation | |
5004 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5005 | */ | |
0b246afa | 5006 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5007 | for (i = 0; i < map->num_stripes; i++) { |
5008 | device = map->stripes[i].dev; | |
5009 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5010 | |
0b86a832 | 5011 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5012 | if (ret) |
50460e37 | 5013 | break; |
b5d9071c NB |
5014 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5015 | dev_offset, stripe_size); | |
6df9a95e | 5016 | if (ret) |
50460e37 FM |
5017 | break; |
5018 | } | |
5019 | if (ret) { | |
0b246afa | 5020 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5021 | goto out; |
2b82032c YZ |
5022 | } |
5023 | ||
2b82032c | 5024 | stripe = &chunk->stripe; |
6df9a95e JB |
5025 | for (i = 0; i < map->num_stripes; i++) { |
5026 | device = map->stripes[i].dev; | |
5027 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5028 | |
e17cade2 CM |
5029 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5030 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5031 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5032 | stripe++; |
0b86a832 | 5033 | } |
0b246afa | 5034 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5035 | |
2b82032c | 5036 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5037 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5038 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5039 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5040 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5041 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5042 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5043 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5044 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5045 | |
2b82032c YZ |
5046 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5047 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5048 | key.offset = chunk_offset; | |
0b86a832 | 5049 | |
2b82032c | 5050 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5051 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5052 | /* | |
5053 | * TODO: Cleanup of inserted chunk root in case of | |
5054 | * failure. | |
5055 | */ | |
2ff7e61e | 5056 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5057 | } |
1abe9b8a | 5058 | |
6df9a95e | 5059 | out: |
0b86a832 | 5060 | kfree(chunk); |
6df9a95e | 5061 | free_extent_map(em); |
4ed1d16e | 5062 | return ret; |
2b82032c | 5063 | } |
0b86a832 | 5064 | |
2b82032c YZ |
5065 | /* |
5066 | * Chunk allocation falls into two parts. The first part does works | |
5067 | * that make the new allocated chunk useable, but not do any operation | |
5068 | * that modifies the chunk tree. The second part does the works that | |
5069 | * require modifying the chunk tree. This division is important for the | |
5070 | * bootstrap process of adding storage to a seed btrfs. | |
5071 | */ | |
5072 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 5073 | struct btrfs_fs_info *fs_info, u64 type) |
2b82032c YZ |
5074 | { |
5075 | u64 chunk_offset; | |
2b82032c | 5076 | |
0b246afa JM |
5077 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
5078 | chunk_offset = find_next_chunk(fs_info); | |
72b468c8 | 5079 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5080 | } |
5081 | ||
d397712b | 5082 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5083 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5084 | { |
5085 | u64 chunk_offset; | |
5086 | u64 sys_chunk_offset; | |
2b82032c | 5087 | u64 alloc_profile; |
2b82032c YZ |
5088 | int ret; |
5089 | ||
6df9a95e | 5090 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5091 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5092 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5093 | if (ret) |
5094 | return ret; | |
2b82032c | 5095 | |
0b246afa | 5096 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5097 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5098 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5099 | return ret; |
2b82032c YZ |
5100 | } |
5101 | ||
d20983b4 MX |
5102 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5103 | { | |
5104 | int max_errors; | |
5105 | ||
5106 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5107 | BTRFS_BLOCK_GROUP_RAID10 | | |
5108 | BTRFS_BLOCK_GROUP_RAID5 | | |
5109 | BTRFS_BLOCK_GROUP_DUP)) { | |
5110 | max_errors = 1; | |
5111 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5112 | max_errors = 2; | |
5113 | } else { | |
5114 | max_errors = 0; | |
005d6427 | 5115 | } |
2b82032c | 5116 | |
d20983b4 | 5117 | return max_errors; |
2b82032c YZ |
5118 | } |
5119 | ||
2ff7e61e | 5120 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5121 | { |
5122 | struct extent_map *em; | |
5123 | struct map_lookup *map; | |
2b82032c | 5124 | int readonly = 0; |
d20983b4 | 5125 | int miss_ndevs = 0; |
2b82032c YZ |
5126 | int i; |
5127 | ||
592d92ee LB |
5128 | em = get_chunk_map(fs_info, chunk_offset, 1); |
5129 | if (IS_ERR(em)) | |
2b82032c YZ |
5130 | return 1; |
5131 | ||
95617d69 | 5132 | map = em->map_lookup; |
2b82032c | 5133 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5134 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5135 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5136 | miss_ndevs++; |
5137 | continue; | |
5138 | } | |
ebbede42 AJ |
5139 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5140 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5141 | readonly = 1; |
d20983b4 | 5142 | goto end; |
2b82032c YZ |
5143 | } |
5144 | } | |
d20983b4 MX |
5145 | |
5146 | /* | |
5147 | * If the number of missing devices is larger than max errors, | |
5148 | * we can not write the data into that chunk successfully, so | |
5149 | * set it readonly. | |
5150 | */ | |
5151 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5152 | readonly = 1; | |
5153 | end: | |
0b86a832 | 5154 | free_extent_map(em); |
2b82032c | 5155 | return readonly; |
0b86a832 CM |
5156 | } |
5157 | ||
5158 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5159 | { | |
a8067e02 | 5160 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5161 | } |
5162 | ||
5163 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5164 | { | |
5165 | struct extent_map *em; | |
5166 | ||
d397712b | 5167 | while (1) { |
890871be | 5168 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5169 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5170 | if (em) | |
5171 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5172 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5173 | if (!em) |
5174 | break; | |
0b86a832 CM |
5175 | /* once for us */ |
5176 | free_extent_map(em); | |
5177 | /* once for the tree */ | |
5178 | free_extent_map(em); | |
5179 | } | |
5180 | } | |
5181 | ||
5d964051 | 5182 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5183 | { |
5184 | struct extent_map *em; | |
5185 | struct map_lookup *map; | |
f188591e CM |
5186 | int ret; |
5187 | ||
592d92ee LB |
5188 | em = get_chunk_map(fs_info, logical, len); |
5189 | if (IS_ERR(em)) | |
5190 | /* | |
5191 | * We could return errors for these cases, but that could get | |
5192 | * ugly and we'd probably do the same thing which is just not do | |
5193 | * anything else and exit, so return 1 so the callers don't try | |
5194 | * to use other copies. | |
5195 | */ | |
fb7669b5 | 5196 | return 1; |
fb7669b5 | 5197 | |
95617d69 | 5198 | map = em->map_lookup; |
f188591e CM |
5199 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5200 | ret = map->num_stripes; | |
321aecc6 CM |
5201 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5202 | ret = map->sub_stripes; | |
53b381b3 DW |
5203 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5204 | ret = 2; | |
5205 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5206 | /* |
5207 | * There could be two corrupted data stripes, we need | |
5208 | * to loop retry in order to rebuild the correct data. | |
5209 | * | |
5210 | * Fail a stripe at a time on every retry except the | |
5211 | * stripe under reconstruction. | |
5212 | */ | |
5213 | ret = map->num_stripes; | |
f188591e CM |
5214 | else |
5215 | ret = 1; | |
5216 | free_extent_map(em); | |
ad6d620e | 5217 | |
73beece9 | 5218 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
6fad823f LB |
5219 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5220 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5221 | ret++; |
73beece9 | 5222 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5223 | |
f188591e CM |
5224 | return ret; |
5225 | } | |
5226 | ||
2ff7e61e | 5227 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5228 | u64 logical) |
5229 | { | |
5230 | struct extent_map *em; | |
5231 | struct map_lookup *map; | |
0b246afa | 5232 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5233 | |
592d92ee | 5234 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5235 | |
69f03f13 NB |
5236 | if (!WARN_ON(IS_ERR(em))) { |
5237 | map = em->map_lookup; | |
5238 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5239 | len = map->stripe_len * nr_data_stripes(map); | |
5240 | free_extent_map(em); | |
5241 | } | |
53b381b3 DW |
5242 | return len; |
5243 | } | |
5244 | ||
e4ff5fb5 | 5245 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5246 | { |
5247 | struct extent_map *em; | |
5248 | struct map_lookup *map; | |
53b381b3 DW |
5249 | int ret = 0; |
5250 | ||
592d92ee | 5251 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5252 | |
69f03f13 NB |
5253 | if(!WARN_ON(IS_ERR(em))) { |
5254 | map = em->map_lookup; | |
5255 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5256 | ret = 1; | |
5257 | free_extent_map(em); | |
5258 | } | |
53b381b3 DW |
5259 | return ret; |
5260 | } | |
5261 | ||
30d9861f SB |
5262 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5263 | struct map_lookup *map, int first, int num, | |
5264 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5265 | { |
5266 | int i; | |
30d9861f SB |
5267 | int tolerance; |
5268 | struct btrfs_device *srcdev; | |
5269 | ||
5270 | if (dev_replace_is_ongoing && | |
5271 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5272 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5273 | srcdev = fs_info->dev_replace.srcdev; | |
5274 | else | |
5275 | srcdev = NULL; | |
5276 | ||
5277 | /* | |
5278 | * try to avoid the drive that is the source drive for a | |
5279 | * dev-replace procedure, only choose it if no other non-missing | |
5280 | * mirror is available | |
5281 | */ | |
5282 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5283 | if (map->stripes[optimal].dev->bdev && | |
5284 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5285 | return optimal; | |
5286 | for (i = first; i < first + num; i++) { | |
5287 | if (map->stripes[i].dev->bdev && | |
5288 | (tolerance || map->stripes[i].dev != srcdev)) | |
5289 | return i; | |
5290 | } | |
dfe25020 | 5291 | } |
30d9861f | 5292 | |
dfe25020 CM |
5293 | /* we couldn't find one that doesn't fail. Just return something |
5294 | * and the io error handling code will clean up eventually | |
5295 | */ | |
5296 | return optimal; | |
5297 | } | |
5298 | ||
53b381b3 DW |
5299 | static inline int parity_smaller(u64 a, u64 b) |
5300 | { | |
5301 | return a > b; | |
5302 | } | |
5303 | ||
5304 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5305 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5306 | { |
5307 | struct btrfs_bio_stripe s; | |
5308 | int i; | |
5309 | u64 l; | |
5310 | int again = 1; | |
5311 | ||
5312 | while (again) { | |
5313 | again = 0; | |
cc7539ed | 5314 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5315 | if (parity_smaller(bbio->raid_map[i], |
5316 | bbio->raid_map[i+1])) { | |
53b381b3 | 5317 | s = bbio->stripes[i]; |
8e5cfb55 | 5318 | l = bbio->raid_map[i]; |
53b381b3 | 5319 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5320 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5321 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5322 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5323 | |
53b381b3 DW |
5324 | again = 1; |
5325 | } | |
5326 | } | |
5327 | } | |
5328 | } | |
5329 | ||
6e9606d2 ZL |
5330 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5331 | { | |
5332 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5333 | /* the size of the btrfs_bio */ |
6e9606d2 | 5334 | sizeof(struct btrfs_bio) + |
e57cf21e | 5335 | /* plus the variable array for the stripes */ |
6e9606d2 | 5336 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5337 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5338 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5339 | /* |
5340 | * plus the raid_map, which includes both the tgt dev | |
5341 | * and the stripes | |
5342 | */ | |
5343 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5344 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5345 | |
5346 | atomic_set(&bbio->error, 0); | |
140475ae | 5347 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5348 | |
5349 | return bbio; | |
5350 | } | |
5351 | ||
5352 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5353 | { | |
140475ae ER |
5354 | WARN_ON(!refcount_read(&bbio->refs)); |
5355 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5356 | } |
5357 | ||
5358 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5359 | { | |
5360 | if (!bbio) | |
5361 | return; | |
140475ae | 5362 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5363 | kfree(bbio); |
5364 | } | |
5365 | ||
0b3d4cd3 LB |
5366 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5367 | /* | |
5368 | * Please note that, discard won't be sent to target device of device | |
5369 | * replace. | |
5370 | */ | |
5371 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5372 | u64 logical, u64 length, | |
5373 | struct btrfs_bio **bbio_ret) | |
5374 | { | |
5375 | struct extent_map *em; | |
5376 | struct map_lookup *map; | |
5377 | struct btrfs_bio *bbio; | |
5378 | u64 offset; | |
5379 | u64 stripe_nr; | |
5380 | u64 stripe_nr_end; | |
5381 | u64 stripe_end_offset; | |
5382 | u64 stripe_cnt; | |
5383 | u64 stripe_len; | |
5384 | u64 stripe_offset; | |
5385 | u64 num_stripes; | |
5386 | u32 stripe_index; | |
5387 | u32 factor = 0; | |
5388 | u32 sub_stripes = 0; | |
5389 | u64 stripes_per_dev = 0; | |
5390 | u32 remaining_stripes = 0; | |
5391 | u32 last_stripe = 0; | |
5392 | int ret = 0; | |
5393 | int i; | |
5394 | ||
5395 | /* discard always return a bbio */ | |
5396 | ASSERT(bbio_ret); | |
5397 | ||
5398 | em = get_chunk_map(fs_info, logical, length); | |
5399 | if (IS_ERR(em)) | |
5400 | return PTR_ERR(em); | |
5401 | ||
5402 | map = em->map_lookup; | |
5403 | /* we don't discard raid56 yet */ | |
5404 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5405 | ret = -EOPNOTSUPP; | |
5406 | goto out; | |
5407 | } | |
5408 | ||
5409 | offset = logical - em->start; | |
5410 | length = min_t(u64, em->len - offset, length); | |
5411 | ||
5412 | stripe_len = map->stripe_len; | |
5413 | /* | |
5414 | * stripe_nr counts the total number of stripes we have to stride | |
5415 | * to get to this block | |
5416 | */ | |
5417 | stripe_nr = div64_u64(offset, stripe_len); | |
5418 | ||
5419 | /* stripe_offset is the offset of this block in its stripe */ | |
5420 | stripe_offset = offset - stripe_nr * stripe_len; | |
5421 | ||
5422 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5423 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5424 | stripe_cnt = stripe_nr_end - stripe_nr; |
5425 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5426 | (offset + length); | |
5427 | /* | |
5428 | * after this, stripe_nr is the number of stripes on this | |
5429 | * device we have to walk to find the data, and stripe_index is | |
5430 | * the number of our device in the stripe array | |
5431 | */ | |
5432 | num_stripes = 1; | |
5433 | stripe_index = 0; | |
5434 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5435 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5436 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5437 | sub_stripes = 1; | |
5438 | else | |
5439 | sub_stripes = map->sub_stripes; | |
5440 | ||
5441 | factor = map->num_stripes / sub_stripes; | |
5442 | num_stripes = min_t(u64, map->num_stripes, | |
5443 | sub_stripes * stripe_cnt); | |
5444 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5445 | stripe_index *= sub_stripes; | |
5446 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5447 | &remaining_stripes); | |
5448 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5449 | last_stripe *= sub_stripes; | |
5450 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5451 | BTRFS_BLOCK_GROUP_DUP)) { | |
5452 | num_stripes = map->num_stripes; | |
5453 | } else { | |
5454 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5455 | &stripe_index); | |
5456 | } | |
5457 | ||
5458 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5459 | if (!bbio) { | |
5460 | ret = -ENOMEM; | |
5461 | goto out; | |
5462 | } | |
5463 | ||
5464 | for (i = 0; i < num_stripes; i++) { | |
5465 | bbio->stripes[i].physical = | |
5466 | map->stripes[stripe_index].physical + | |
5467 | stripe_offset + stripe_nr * map->stripe_len; | |
5468 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5469 | ||
5470 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5471 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5472 | bbio->stripes[i].length = stripes_per_dev * | |
5473 | map->stripe_len; | |
5474 | ||
5475 | if (i / sub_stripes < remaining_stripes) | |
5476 | bbio->stripes[i].length += | |
5477 | map->stripe_len; | |
5478 | ||
5479 | /* | |
5480 | * Special for the first stripe and | |
5481 | * the last stripe: | |
5482 | * | |
5483 | * |-------|...|-------| | |
5484 | * |----------| | |
5485 | * off end_off | |
5486 | */ | |
5487 | if (i < sub_stripes) | |
5488 | bbio->stripes[i].length -= | |
5489 | stripe_offset; | |
5490 | ||
5491 | if (stripe_index >= last_stripe && | |
5492 | stripe_index <= (last_stripe + | |
5493 | sub_stripes - 1)) | |
5494 | bbio->stripes[i].length -= | |
5495 | stripe_end_offset; | |
5496 | ||
5497 | if (i == sub_stripes - 1) | |
5498 | stripe_offset = 0; | |
5499 | } else { | |
5500 | bbio->stripes[i].length = length; | |
5501 | } | |
5502 | ||
5503 | stripe_index++; | |
5504 | if (stripe_index == map->num_stripes) { | |
5505 | stripe_index = 0; | |
5506 | stripe_nr++; | |
5507 | } | |
5508 | } | |
5509 | ||
5510 | *bbio_ret = bbio; | |
5511 | bbio->map_type = map->type; | |
5512 | bbio->num_stripes = num_stripes; | |
5513 | out: | |
5514 | free_extent_map(em); | |
5515 | return ret; | |
5516 | } | |
5517 | ||
5ab56090 LB |
5518 | /* |
5519 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5520 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5521 | * left cursor can also be read from the target drive. | |
5522 | * | |
5523 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5524 | * array of stripes. | |
5525 | * For READ, it also needs to be supported using the same mirror number. | |
5526 | * | |
5527 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5528 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5529 | * case. | |
5530 | */ | |
5531 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5532 | u64 logical, u64 length, | |
5533 | u64 srcdev_devid, int *mirror_num, | |
5534 | u64 *physical) | |
5535 | { | |
5536 | struct btrfs_bio *bbio = NULL; | |
5537 | int num_stripes; | |
5538 | int index_srcdev = 0; | |
5539 | int found = 0; | |
5540 | u64 physical_of_found = 0; | |
5541 | int i; | |
5542 | int ret = 0; | |
5543 | ||
5544 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5545 | logical, &length, &bbio, 0, 0); | |
5546 | if (ret) { | |
5547 | ASSERT(bbio == NULL); | |
5548 | return ret; | |
5549 | } | |
5550 | ||
5551 | num_stripes = bbio->num_stripes; | |
5552 | if (*mirror_num > num_stripes) { | |
5553 | /* | |
5554 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5555 | * that means that the requested area is not left of the left | |
5556 | * cursor | |
5557 | */ | |
5558 | btrfs_put_bbio(bbio); | |
5559 | return -EIO; | |
5560 | } | |
5561 | ||
5562 | /* | |
5563 | * process the rest of the function using the mirror_num of the source | |
5564 | * drive. Therefore look it up first. At the end, patch the device | |
5565 | * pointer to the one of the target drive. | |
5566 | */ | |
5567 | for (i = 0; i < num_stripes; i++) { | |
5568 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5569 | continue; | |
5570 | ||
5571 | /* | |
5572 | * In case of DUP, in order to keep it simple, only add the | |
5573 | * mirror with the lowest physical address | |
5574 | */ | |
5575 | if (found && | |
5576 | physical_of_found <= bbio->stripes[i].physical) | |
5577 | continue; | |
5578 | ||
5579 | index_srcdev = i; | |
5580 | found = 1; | |
5581 | physical_of_found = bbio->stripes[i].physical; | |
5582 | } | |
5583 | ||
5584 | btrfs_put_bbio(bbio); | |
5585 | ||
5586 | ASSERT(found); | |
5587 | if (!found) | |
5588 | return -EIO; | |
5589 | ||
5590 | *mirror_num = index_srcdev + 1; | |
5591 | *physical = physical_of_found; | |
5592 | return ret; | |
5593 | } | |
5594 | ||
73c0f228 LB |
5595 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5596 | struct btrfs_bio **bbio_ret, | |
5597 | struct btrfs_dev_replace *dev_replace, | |
5598 | int *num_stripes_ret, int *max_errors_ret) | |
5599 | { | |
5600 | struct btrfs_bio *bbio = *bbio_ret; | |
5601 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5602 | int tgtdev_indexes = 0; | |
5603 | int num_stripes = *num_stripes_ret; | |
5604 | int max_errors = *max_errors_ret; | |
5605 | int i; | |
5606 | ||
5607 | if (op == BTRFS_MAP_WRITE) { | |
5608 | int index_where_to_add; | |
5609 | ||
5610 | /* | |
5611 | * duplicate the write operations while the dev replace | |
5612 | * procedure is running. Since the copying of the old disk to | |
5613 | * the new disk takes place at run time while the filesystem is | |
5614 | * mounted writable, the regular write operations to the old | |
5615 | * disk have to be duplicated to go to the new disk as well. | |
5616 | * | |
5617 | * Note that device->missing is handled by the caller, and that | |
5618 | * the write to the old disk is already set up in the stripes | |
5619 | * array. | |
5620 | */ | |
5621 | index_where_to_add = num_stripes; | |
5622 | for (i = 0; i < num_stripes; i++) { | |
5623 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5624 | /* write to new disk, too */ | |
5625 | struct btrfs_bio_stripe *new = | |
5626 | bbio->stripes + index_where_to_add; | |
5627 | struct btrfs_bio_stripe *old = | |
5628 | bbio->stripes + i; | |
5629 | ||
5630 | new->physical = old->physical; | |
5631 | new->length = old->length; | |
5632 | new->dev = dev_replace->tgtdev; | |
5633 | bbio->tgtdev_map[i] = index_where_to_add; | |
5634 | index_where_to_add++; | |
5635 | max_errors++; | |
5636 | tgtdev_indexes++; | |
5637 | } | |
5638 | } | |
5639 | num_stripes = index_where_to_add; | |
5640 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5641 | int index_srcdev = 0; | |
5642 | int found = 0; | |
5643 | u64 physical_of_found = 0; | |
5644 | ||
5645 | /* | |
5646 | * During the dev-replace procedure, the target drive can also | |
5647 | * be used to read data in case it is needed to repair a corrupt | |
5648 | * block elsewhere. This is possible if the requested area is | |
5649 | * left of the left cursor. In this area, the target drive is a | |
5650 | * full copy of the source drive. | |
5651 | */ | |
5652 | for (i = 0; i < num_stripes; i++) { | |
5653 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5654 | /* | |
5655 | * In case of DUP, in order to keep it simple, | |
5656 | * only add the mirror with the lowest physical | |
5657 | * address | |
5658 | */ | |
5659 | if (found && | |
5660 | physical_of_found <= | |
5661 | bbio->stripes[i].physical) | |
5662 | continue; | |
5663 | index_srcdev = i; | |
5664 | found = 1; | |
5665 | physical_of_found = bbio->stripes[i].physical; | |
5666 | } | |
5667 | } | |
5668 | if (found) { | |
5669 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5670 | bbio->stripes + num_stripes; | |
5671 | ||
5672 | tgtdev_stripe->physical = physical_of_found; | |
5673 | tgtdev_stripe->length = | |
5674 | bbio->stripes[index_srcdev].length; | |
5675 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5676 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5677 | ||
5678 | tgtdev_indexes++; | |
5679 | num_stripes++; | |
5680 | } | |
5681 | } | |
5682 | ||
5683 | *num_stripes_ret = num_stripes; | |
5684 | *max_errors_ret = max_errors; | |
5685 | bbio->num_tgtdevs = tgtdev_indexes; | |
5686 | *bbio_ret = bbio; | |
5687 | } | |
5688 | ||
2b19a1fe LB |
5689 | static bool need_full_stripe(enum btrfs_map_op op) |
5690 | { | |
5691 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5692 | } | |
5693 | ||
cf8cddd3 CH |
5694 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5695 | enum btrfs_map_op op, | |
f2d8d74d | 5696 | u64 logical, u64 *length, |
a1d3c478 | 5697 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5698 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5699 | { |
5700 | struct extent_map *em; | |
5701 | struct map_lookup *map; | |
0b86a832 | 5702 | u64 offset; |
593060d7 CM |
5703 | u64 stripe_offset; |
5704 | u64 stripe_nr; | |
53b381b3 | 5705 | u64 stripe_len; |
9d644a62 | 5706 | u32 stripe_index; |
cea9e445 | 5707 | int i; |
de11cc12 | 5708 | int ret = 0; |
f2d8d74d | 5709 | int num_stripes; |
a236aed1 | 5710 | int max_errors = 0; |
2c8cdd6e | 5711 | int tgtdev_indexes = 0; |
a1d3c478 | 5712 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5713 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5714 | int dev_replace_is_ongoing = 0; | |
5715 | int num_alloc_stripes; | |
ad6d620e SB |
5716 | int patch_the_first_stripe_for_dev_replace = 0; |
5717 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5718 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5719 | |
0b3d4cd3 LB |
5720 | if (op == BTRFS_MAP_DISCARD) |
5721 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5722 | *length, bbio_ret); | |
5723 | ||
592d92ee LB |
5724 | em = get_chunk_map(fs_info, logical, *length); |
5725 | if (IS_ERR(em)) | |
5726 | return PTR_ERR(em); | |
0b86a832 | 5727 | |
95617d69 | 5728 | map = em->map_lookup; |
0b86a832 | 5729 | offset = logical - em->start; |
593060d7 | 5730 | |
53b381b3 | 5731 | stripe_len = map->stripe_len; |
593060d7 CM |
5732 | stripe_nr = offset; |
5733 | /* | |
5734 | * stripe_nr counts the total number of stripes we have to stride | |
5735 | * to get to this block | |
5736 | */ | |
47c5713f | 5737 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5738 | |
53b381b3 | 5739 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5740 | if (offset < stripe_offset) { |
5d163e0e JM |
5741 | btrfs_crit(fs_info, |
5742 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5743 | stripe_offset, offset, em->start, logical, |
5744 | stripe_len); | |
5745 | free_extent_map(em); | |
5746 | return -EINVAL; | |
5747 | } | |
593060d7 CM |
5748 | |
5749 | /* stripe_offset is the offset of this block in its stripe*/ | |
5750 | stripe_offset = offset - stripe_offset; | |
5751 | ||
53b381b3 | 5752 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5753 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5754 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5755 | raid56_full_stripe_start = offset; | |
5756 | ||
5757 | /* allow a write of a full stripe, but make sure we don't | |
5758 | * allow straddling of stripes | |
5759 | */ | |
47c5713f DS |
5760 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5761 | full_stripe_len); | |
53b381b3 DW |
5762 | raid56_full_stripe_start *= full_stripe_len; |
5763 | } | |
5764 | ||
0b3d4cd3 | 5765 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5766 | u64 max_len; |
5767 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5768 | For other RAID types and for RAID[56] reads, just allow a single | |
5769 | stripe (on a single disk). */ | |
ffe2d203 | 5770 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5771 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5772 | max_len = stripe_len * nr_data_stripes(map) - |
5773 | (offset - raid56_full_stripe_start); | |
5774 | } else { | |
5775 | /* we limit the length of each bio to what fits in a stripe */ | |
5776 | max_len = stripe_len - stripe_offset; | |
5777 | } | |
5778 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5779 | } else { |
5780 | *length = em->len - offset; | |
5781 | } | |
f2d8d74d | 5782 | |
53b381b3 DW |
5783 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5784 | it cares about is the length */ | |
a1d3c478 | 5785 | if (!bbio_ret) |
cea9e445 CM |
5786 | goto out; |
5787 | ||
73beece9 | 5788 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5789 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5790 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5791 | btrfs_dev_replace_unlock(dev_replace, 0); |
5792 | else | |
5793 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5794 | |
ad6d620e | 5795 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5796 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5797 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5798 | dev_replace->srcdev->devid, | |
5799 | &mirror_num, | |
5800 | &physical_to_patch_in_first_stripe); | |
5801 | if (ret) | |
ad6d620e | 5802 | goto out; |
5ab56090 LB |
5803 | else |
5804 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5805 | } else if (mirror_num > map->num_stripes) { |
5806 | mirror_num = 0; | |
5807 | } | |
5808 | ||
f2d8d74d | 5809 | num_stripes = 1; |
cea9e445 | 5810 | stripe_index = 0; |
fce3bb9a | 5811 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5812 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5813 | &stripe_index); | |
de483734 | 5814 | if (!need_full_stripe(op)) |
28e1cc7d | 5815 | mirror_num = 1; |
fce3bb9a | 5816 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 5817 | if (need_full_stripe(op)) |
f2d8d74d | 5818 | num_stripes = map->num_stripes; |
2fff734f | 5819 | else if (mirror_num) |
f188591e | 5820 | stripe_index = mirror_num - 1; |
dfe25020 | 5821 | else { |
30d9861f | 5822 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5823 | map->num_stripes, |
30d9861f SB |
5824 | current->pid % map->num_stripes, |
5825 | dev_replace_is_ongoing); | |
a1d3c478 | 5826 | mirror_num = stripe_index + 1; |
dfe25020 | 5827 | } |
2fff734f | 5828 | |
611f0e00 | 5829 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5830 | if (need_full_stripe(op)) { |
f2d8d74d | 5831 | num_stripes = map->num_stripes; |
a1d3c478 | 5832 | } else if (mirror_num) { |
f188591e | 5833 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5834 | } else { |
5835 | mirror_num = 1; | |
5836 | } | |
2fff734f | 5837 | |
321aecc6 | 5838 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5839 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5840 | |
47c5713f | 5841 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5842 | stripe_index *= map->sub_stripes; |
5843 | ||
de483734 | 5844 | if (need_full_stripe(op)) |
f2d8d74d | 5845 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5846 | else if (mirror_num) |
5847 | stripe_index += mirror_num - 1; | |
dfe25020 | 5848 | else { |
3e74317a | 5849 | int old_stripe_index = stripe_index; |
30d9861f SB |
5850 | stripe_index = find_live_mirror(fs_info, map, |
5851 | stripe_index, | |
dfe25020 | 5852 | map->sub_stripes, stripe_index + |
30d9861f SB |
5853 | current->pid % map->sub_stripes, |
5854 | dev_replace_is_ongoing); | |
3e74317a | 5855 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5856 | } |
53b381b3 | 5857 | |
ffe2d203 | 5858 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5859 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5860 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5861 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5862 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5863 | |
5864 | /* RAID[56] write or recovery. Return all stripes */ | |
5865 | num_stripes = map->num_stripes; | |
5866 | max_errors = nr_parity_stripes(map); | |
5867 | ||
53b381b3 DW |
5868 | *length = map->stripe_len; |
5869 | stripe_index = 0; | |
5870 | stripe_offset = 0; | |
5871 | } else { | |
5872 | /* | |
5873 | * Mirror #0 or #1 means the original data block. | |
5874 | * Mirror #2 is RAID5 parity block. | |
5875 | * Mirror #3 is RAID6 Q block. | |
5876 | */ | |
47c5713f DS |
5877 | stripe_nr = div_u64_rem(stripe_nr, |
5878 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5879 | if (mirror_num > 1) |
5880 | stripe_index = nr_data_stripes(map) + | |
5881 | mirror_num - 2; | |
5882 | ||
5883 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5884 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5885 | &stripe_index); | |
de483734 | 5886 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 5887 | mirror_num = 1; |
53b381b3 | 5888 | } |
8790d502 CM |
5889 | } else { |
5890 | /* | |
47c5713f DS |
5891 | * after this, stripe_nr is the number of stripes on this |
5892 | * device we have to walk to find the data, and stripe_index is | |
5893 | * the number of our device in the stripe array | |
8790d502 | 5894 | */ |
47c5713f DS |
5895 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5896 | &stripe_index); | |
a1d3c478 | 5897 | mirror_num = stripe_index + 1; |
8790d502 | 5898 | } |
e042d1ec | 5899 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5900 | btrfs_crit(fs_info, |
5901 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5902 | stripe_index, map->num_stripes); |
5903 | ret = -EINVAL; | |
5904 | goto out; | |
5905 | } | |
cea9e445 | 5906 | |
472262f3 | 5907 | num_alloc_stripes = num_stripes; |
6fad823f | 5908 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5909 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5910 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5911 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5912 | num_alloc_stripes++; |
2c8cdd6e | 5913 | tgtdev_indexes = num_stripes; |
ad6d620e | 5914 | } |
2c8cdd6e | 5915 | |
6e9606d2 | 5916 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5917 | if (!bbio) { |
5918 | ret = -ENOMEM; | |
5919 | goto out; | |
5920 | } | |
6fad823f | 5921 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5922 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5923 | |
8e5cfb55 | 5924 | /* build raid_map */ |
2b19a1fe LB |
5925 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5926 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5927 | u64 tmp; |
9d644a62 | 5928 | unsigned rot; |
8e5cfb55 ZL |
5929 | |
5930 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5931 | sizeof(struct btrfs_bio_stripe) * | |
5932 | num_alloc_stripes + | |
5933 | sizeof(int) * tgtdev_indexes); | |
5934 | ||
5935 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5936 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5937 | |
5938 | /* Fill in the logical address of each stripe */ | |
5939 | tmp = stripe_nr * nr_data_stripes(map); | |
5940 | for (i = 0; i < nr_data_stripes(map); i++) | |
5941 | bbio->raid_map[(i+rot) % num_stripes] = | |
5942 | em->start + (tmp + i) * map->stripe_len; | |
5943 | ||
5944 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5945 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5946 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5947 | RAID6_Q_STRIPE; | |
5948 | } | |
5949 | ||
b89203f7 | 5950 | |
0b3d4cd3 LB |
5951 | for (i = 0; i < num_stripes; i++) { |
5952 | bbio->stripes[i].physical = | |
5953 | map->stripes[stripe_index].physical + | |
5954 | stripe_offset + | |
5955 | stripe_nr * map->stripe_len; | |
5956 | bbio->stripes[i].dev = | |
5957 | map->stripes[stripe_index].dev; | |
5958 | stripe_index++; | |
593060d7 | 5959 | } |
de11cc12 | 5960 | |
2b19a1fe | 5961 | if (need_full_stripe(op)) |
d20983b4 | 5962 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5963 | |
8e5cfb55 ZL |
5964 | if (bbio->raid_map) |
5965 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5966 | |
73c0f228 | 5967 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5968 | need_full_stripe(op)) { |
73c0f228 LB |
5969 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5970 | &max_errors); | |
472262f3 SB |
5971 | } |
5972 | ||
de11cc12 | 5973 | *bbio_ret = bbio; |
10f11900 | 5974 | bbio->map_type = map->type; |
de11cc12 LZ |
5975 | bbio->num_stripes = num_stripes; |
5976 | bbio->max_errors = max_errors; | |
5977 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5978 | |
5979 | /* | |
5980 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5981 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5982 | * available as a mirror | |
5983 | */ | |
5984 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5985 | WARN_ON(num_stripes > 1); | |
5986 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5987 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5988 | bbio->mirror_num = map->num_stripes + 1; | |
5989 | } | |
cea9e445 | 5990 | out: |
73beece9 LB |
5991 | if (dev_replace_is_ongoing) { |
5992 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5993 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5994 | } | |
0b86a832 | 5995 | free_extent_map(em); |
de11cc12 | 5996 | return ret; |
0b86a832 CM |
5997 | } |
5998 | ||
cf8cddd3 | 5999 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6000 | u64 logical, u64 *length, |
a1d3c478 | 6001 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6002 | { |
b3d3fa51 | 6003 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6004 | mirror_num, 0); |
f2d8d74d CM |
6005 | } |
6006 | ||
af8e2d1d | 6007 | /* For Scrub/replace */ |
cf8cddd3 | 6008 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6009 | u64 logical, u64 *length, |
825ad4c9 | 6010 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6011 | { |
825ad4c9 | 6012 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6013 | } |
6014 | ||
ab8d0fc4 | 6015 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
6016 | u64 chunk_start, u64 physical, u64 devid, |
6017 | u64 **logical, int *naddrs, int *stripe_len) | |
6018 | { | |
a512bbf8 YZ |
6019 | struct extent_map *em; |
6020 | struct map_lookup *map; | |
6021 | u64 *buf; | |
6022 | u64 bytenr; | |
6023 | u64 length; | |
6024 | u64 stripe_nr; | |
53b381b3 | 6025 | u64 rmap_len; |
a512bbf8 YZ |
6026 | int i, j, nr = 0; |
6027 | ||
592d92ee LB |
6028 | em = get_chunk_map(fs_info, chunk_start, 1); |
6029 | if (IS_ERR(em)) | |
835d974f | 6030 | return -EIO; |
835d974f | 6031 | |
95617d69 | 6032 | map = em->map_lookup; |
a512bbf8 | 6033 | length = em->len; |
53b381b3 DW |
6034 | rmap_len = map->stripe_len; |
6035 | ||
a512bbf8 | 6036 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6037 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6038 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6039 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6040 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6041 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6042 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6043 | } | |
a512bbf8 | 6044 | |
31e818fe | 6045 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6046 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6047 | |
6048 | for (i = 0; i < map->num_stripes; i++) { | |
6049 | if (devid && map->stripes[i].dev->devid != devid) | |
6050 | continue; | |
6051 | if (map->stripes[i].physical > physical || | |
6052 | map->stripes[i].physical + length <= physical) | |
6053 | continue; | |
6054 | ||
6055 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6056 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6057 | |
6058 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6059 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6060 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6061 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6062 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6063 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6064 | * Alternatively, just use rmap_len below instead of | |
6065 | * map->stripe_len */ | |
6066 | ||
6067 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6068 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6069 | for (j = 0; j < nr; j++) { |
6070 | if (buf[j] == bytenr) | |
6071 | break; | |
6072 | } | |
934d375b CM |
6073 | if (j == nr) { |
6074 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6075 | buf[nr++] = bytenr; |
934d375b | 6076 | } |
a512bbf8 YZ |
6077 | } |
6078 | ||
a512bbf8 YZ |
6079 | *logical = buf; |
6080 | *naddrs = nr; | |
53b381b3 | 6081 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6082 | |
6083 | free_extent_map(em); | |
6084 | return 0; | |
f2d8d74d CM |
6085 | } |
6086 | ||
4246a0b6 | 6087 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6088 | { |
326e1dbb MS |
6089 | bio->bi_private = bbio->private; |
6090 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6091 | bio_endio(bio); |
326e1dbb | 6092 | |
6e9606d2 | 6093 | btrfs_put_bbio(bbio); |
8408c716 MX |
6094 | } |
6095 | ||
4246a0b6 | 6096 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6097 | { |
9be3395b | 6098 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6099 | int is_orig_bio = 0; |
8790d502 | 6100 | |
4e4cbee9 | 6101 | if (bio->bi_status) { |
a1d3c478 | 6102 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6103 | if (bio->bi_status == BLK_STS_IOERR || |
6104 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6105 | unsigned int stripe_index = |
9be3395b | 6106 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6107 | struct btrfs_device *dev; |
442a4f63 SB |
6108 | |
6109 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6110 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6111 | if (dev->bdev) { |
37226b21 | 6112 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6113 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6114 | BTRFS_DEV_STAT_WRITE_ERRS); |
6115 | else | |
1cb34c8e | 6116 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6117 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6118 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6119 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6120 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6121 | } |
442a4f63 SB |
6122 | } |
6123 | } | |
8790d502 | 6124 | |
a1d3c478 | 6125 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6126 | is_orig_bio = 1; |
6127 | ||
c404e0dc MX |
6128 | btrfs_bio_counter_dec(bbio->fs_info); |
6129 | ||
a1d3c478 | 6130 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6131 | if (!is_orig_bio) { |
6132 | bio_put(bio); | |
a1d3c478 | 6133 | bio = bbio->orig_bio; |
7d2b4daa | 6134 | } |
c7b22bb1 | 6135 | |
9be3395b | 6136 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6137 | /* only send an error to the higher layers if it is |
53b381b3 | 6138 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6139 | */ |
a1d3c478 | 6140 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6141 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6142 | } else { |
1259ab75 CM |
6143 | /* |
6144 | * this bio is actually up to date, we didn't | |
6145 | * go over the max number of errors | |
6146 | */ | |
2dbe0c77 | 6147 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6148 | } |
c55f1396 | 6149 | |
4246a0b6 | 6150 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6151 | } else if (!is_orig_bio) { |
8790d502 CM |
6152 | bio_put(bio); |
6153 | } | |
8790d502 CM |
6154 | } |
6155 | ||
8b712842 CM |
6156 | /* |
6157 | * see run_scheduled_bios for a description of why bios are collected for | |
6158 | * async submit. | |
6159 | * | |
6160 | * This will add one bio to the pending list for a device and make sure | |
6161 | * the work struct is scheduled. | |
6162 | */ | |
2ff7e61e | 6163 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6164 | struct bio *bio) |
8b712842 | 6165 | { |
0b246afa | 6166 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6167 | int should_queue = 1; |
ffbd517d | 6168 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6169 | |
e6e674bd AJ |
6170 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) || |
6171 | !device->bdev) { | |
4246a0b6 | 6172 | bio_io_error(bio); |
53b381b3 DW |
6173 | return; |
6174 | } | |
6175 | ||
8b712842 | 6176 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6177 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6178 | btrfsic_submit_bio(bio); |
143bede5 | 6179 | return; |
8b712842 CM |
6180 | } |
6181 | ||
492bb6de | 6182 | WARN_ON(bio->bi_next); |
8b712842 | 6183 | bio->bi_next = NULL; |
8b712842 CM |
6184 | |
6185 | spin_lock(&device->io_lock); | |
67f055c7 | 6186 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6187 | pending_bios = &device->pending_sync_bios; |
6188 | else | |
6189 | pending_bios = &device->pending_bios; | |
8b712842 | 6190 | |
ffbd517d CM |
6191 | if (pending_bios->tail) |
6192 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6193 | |
ffbd517d CM |
6194 | pending_bios->tail = bio; |
6195 | if (!pending_bios->head) | |
6196 | pending_bios->head = bio; | |
8b712842 CM |
6197 | if (device->running_pending) |
6198 | should_queue = 0; | |
6199 | ||
6200 | spin_unlock(&device->io_lock); | |
6201 | ||
6202 | if (should_queue) | |
0b246afa | 6203 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6204 | } |
6205 | ||
2ff7e61e JM |
6206 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6207 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6208 | { |
6209 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6210 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6211 | |
6212 | bio->bi_private = bbio; | |
9be3395b | 6213 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6214 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6215 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6216 | #ifdef DEBUG |
6217 | { | |
6218 | struct rcu_string *name; | |
6219 | ||
6220 | rcu_read_lock(); | |
6221 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6222 | btrfs_debug(fs_info, |
6223 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6224 | bio_op(bio), bio->bi_opf, | |
6225 | (u64)bio->bi_iter.bi_sector, | |
6226 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6227 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6228 | rcu_read_unlock(); |
6229 | } | |
6230 | #endif | |
74d46992 | 6231 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6232 | |
2ff7e61e | 6233 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6234 | |
de1ee92a | 6235 | if (async) |
2ff7e61e | 6236 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6237 | else |
4e49ea4a | 6238 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6239 | } |
6240 | ||
de1ee92a JB |
6241 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6242 | { | |
6243 | atomic_inc(&bbio->error); | |
6244 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6245 | /* Should be the original bio. */ |
8408c716 MX |
6246 | WARN_ON(bio != bbio->orig_bio); |
6247 | ||
9be3395b | 6248 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6249 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6250 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6251 | bio->bi_status = BLK_STS_IOERR; | |
6252 | else | |
6253 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6254 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6255 | } |
6256 | } | |
6257 | ||
58efbc9f OS |
6258 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6259 | int mirror_num, int async_submit) | |
0b86a832 | 6260 | { |
0b86a832 | 6261 | struct btrfs_device *dev; |
8790d502 | 6262 | struct bio *first_bio = bio; |
4f024f37 | 6263 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6264 | u64 length = 0; |
6265 | u64 map_length; | |
0b86a832 | 6266 | int ret; |
08da757d ZL |
6267 | int dev_nr; |
6268 | int total_devs; | |
a1d3c478 | 6269 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6270 | |
4f024f37 | 6271 | length = bio->bi_iter.bi_size; |
0b86a832 | 6272 | map_length = length; |
cea9e445 | 6273 | |
0b246afa | 6274 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6275 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6276 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6277 | if (ret) { |
0b246afa | 6278 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6279 | return errno_to_blk_status(ret); |
c404e0dc | 6280 | } |
cea9e445 | 6281 | |
a1d3c478 | 6282 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6283 | bbio->orig_bio = first_bio; |
6284 | bbio->private = first_bio->bi_private; | |
6285 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6286 | bbio->fs_info = fs_info; |
53b381b3 DW |
6287 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6288 | ||
ad1ba2a0 | 6289 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6290 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6291 | /* In this case, map_length has been set to the length of |
6292 | a single stripe; not the whole write */ | |
37226b21 | 6293 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6294 | ret = raid56_parity_write(fs_info, bio, bbio, |
6295 | map_length); | |
53b381b3 | 6296 | } else { |
2ff7e61e JM |
6297 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6298 | map_length, mirror_num, 1); | |
53b381b3 | 6299 | } |
4245215d | 6300 | |
0b246afa | 6301 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6302 | return errno_to_blk_status(ret); |
53b381b3 DW |
6303 | } |
6304 | ||
cea9e445 | 6305 | if (map_length < length) { |
0b246afa | 6306 | btrfs_crit(fs_info, |
5d163e0e JM |
6307 | "mapping failed logical %llu bio len %llu len %llu", |
6308 | logical, length, map_length); | |
cea9e445 CM |
6309 | BUG(); |
6310 | } | |
a1d3c478 | 6311 | |
08da757d | 6312 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6313 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6314 | if (!dev || !dev->bdev || |
ebbede42 AJ |
6315 | (bio_op(first_bio) == REQ_OP_WRITE && |
6316 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6317 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6318 | continue; |
6319 | } | |
6320 | ||
3aa8e074 | 6321 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6322 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6323 | else |
a1d3c478 | 6324 | bio = first_bio; |
de1ee92a | 6325 | |
2ff7e61e JM |
6326 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6327 | dev_nr, async_submit); | |
8790d502 | 6328 | } |
0b246afa | 6329 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6330 | return BLK_STS_OK; |
0b86a832 CM |
6331 | } |
6332 | ||
aa1b8cd4 | 6333 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6334 | u8 *uuid, u8 *fsid) |
0b86a832 | 6335 | { |
2b82032c YZ |
6336 | struct btrfs_device *device; |
6337 | struct btrfs_fs_devices *cur_devices; | |
6338 | ||
aa1b8cd4 | 6339 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6340 | while (cur_devices) { |
6341 | if (!fsid || | |
44880fdc | 6342 | !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6343 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6344 | if (device) |
6345 | return device; | |
6346 | } | |
6347 | cur_devices = cur_devices->seed; | |
6348 | } | |
6349 | return NULL; | |
0b86a832 CM |
6350 | } |
6351 | ||
2ff7e61e | 6352 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6353 | u64 devid, u8 *dev_uuid) |
6354 | { | |
6355 | struct btrfs_device *device; | |
dfe25020 | 6356 | |
12bd2fc0 ID |
6357 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6358 | if (IS_ERR(device)) | |
adfb69af | 6359 | return device; |
12bd2fc0 ID |
6360 | |
6361 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6362 | device->fs_devices = fs_devices; |
dfe25020 | 6363 | fs_devices->num_devices++; |
12bd2fc0 | 6364 | |
e6e674bd | 6365 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6366 | fs_devices->missing_devices++; |
12bd2fc0 | 6367 | |
dfe25020 CM |
6368 | return device; |
6369 | } | |
6370 | ||
12bd2fc0 ID |
6371 | /** |
6372 | * btrfs_alloc_device - allocate struct btrfs_device | |
6373 | * @fs_info: used only for generating a new devid, can be NULL if | |
6374 | * devid is provided (i.e. @devid != NULL). | |
6375 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6376 | * is generated. | |
6377 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6378 | * is generated. | |
6379 | * | |
6380 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf DS |
6381 | * on error. Returned struct is not linked onto any lists and must be |
6382 | * destroyed with free_device. | |
12bd2fc0 ID |
6383 | */ |
6384 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6385 | const u64 *devid, | |
6386 | const u8 *uuid) | |
6387 | { | |
6388 | struct btrfs_device *dev; | |
6389 | u64 tmp; | |
6390 | ||
fae7f21c | 6391 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6392 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6393 | |
6394 | dev = __alloc_device(); | |
6395 | if (IS_ERR(dev)) | |
6396 | return dev; | |
6397 | ||
6398 | if (devid) | |
6399 | tmp = *devid; | |
6400 | else { | |
6401 | int ret; | |
6402 | ||
6403 | ret = find_next_devid(fs_info, &tmp); | |
6404 | if (ret) { | |
55de4803 | 6405 | free_device(dev); |
12bd2fc0 ID |
6406 | return ERR_PTR(ret); |
6407 | } | |
6408 | } | |
6409 | dev->devid = tmp; | |
6410 | ||
6411 | if (uuid) | |
6412 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6413 | else | |
6414 | generate_random_uuid(dev->uuid); | |
6415 | ||
9e0af237 LB |
6416 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6417 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6418 | |
6419 | return dev; | |
6420 | } | |
6421 | ||
e06cd3dd | 6422 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6423 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6424 | struct extent_buffer *leaf, |
6425 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6426 | { |
0b86a832 | 6427 | u64 length; |
f04b772b | 6428 | u64 stripe_len; |
e06cd3dd LB |
6429 | u16 num_stripes; |
6430 | u16 sub_stripes; | |
6431 | u64 type; | |
0b86a832 | 6432 | |
e17cade2 | 6433 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6434 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6435 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6436 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6437 | type = btrfs_chunk_type(leaf, chunk); | |
6438 | ||
f04b772b | 6439 | if (!num_stripes) { |
0b246afa | 6440 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6441 | num_stripes); |
6442 | return -EIO; | |
6443 | } | |
0b246afa JM |
6444 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6445 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6446 | return -EIO; |
6447 | } | |
0b246afa JM |
6448 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6449 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6450 | btrfs_chunk_sector_size(leaf, chunk)); |
6451 | return -EIO; | |
6452 | } | |
0b246afa JM |
6453 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6454 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6455 | return -EIO; |
6456 | } | |
3d8da678 | 6457 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6458 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6459 | stripe_len); |
6460 | return -EIO; | |
6461 | } | |
6462 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6463 | type) { |
0b246afa | 6464 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6465 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6466 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6467 | btrfs_chunk_type(leaf, chunk)); | |
6468 | return -EIO; | |
6469 | } | |
e06cd3dd LB |
6470 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6471 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6472 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6473 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6474 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6475 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6476 | num_stripes != 1)) { | |
0b246afa | 6477 | btrfs_err(fs_info, |
e06cd3dd LB |
6478 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6479 | num_stripes, sub_stripes, | |
6480 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6481 | return -EIO; | |
6482 | } | |
6483 | ||
6484 | return 0; | |
6485 | } | |
6486 | ||
5a2b8e60 | 6487 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6488 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6489 | { |
2b902dfc AJ |
6490 | if (error) |
6491 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6492 | devid, uuid); | |
6493 | else | |
6494 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6495 | devid, uuid); | |
5a2b8e60 AJ |
6496 | } |
6497 | ||
2ff7e61e | 6498 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6499 | struct extent_buffer *leaf, |
6500 | struct btrfs_chunk *chunk) | |
6501 | { | |
0b246afa | 6502 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6503 | struct map_lookup *map; |
6504 | struct extent_map *em; | |
6505 | u64 logical; | |
6506 | u64 length; | |
e06cd3dd LB |
6507 | u64 devid; |
6508 | u8 uuid[BTRFS_UUID_SIZE]; | |
6509 | int num_stripes; | |
6510 | int ret; | |
6511 | int i; | |
6512 | ||
6513 | logical = key->offset; | |
6514 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6515 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6516 | ||
2ff7e61e | 6517 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6518 | if (ret) |
6519 | return ret; | |
a061fc8d | 6520 | |
890871be | 6521 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6522 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6523 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6524 | |
6525 | /* already mapped? */ | |
6526 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6527 | free_extent_map(em); | |
0b86a832 CM |
6528 | return 0; |
6529 | } else if (em) { | |
6530 | free_extent_map(em); | |
6531 | } | |
0b86a832 | 6532 | |
172ddd60 | 6533 | em = alloc_extent_map(); |
0b86a832 CM |
6534 | if (!em) |
6535 | return -ENOMEM; | |
593060d7 | 6536 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6537 | if (!map) { |
6538 | free_extent_map(em); | |
6539 | return -ENOMEM; | |
6540 | } | |
6541 | ||
298a8f9c | 6542 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6543 | em->map_lookup = map; |
0b86a832 CM |
6544 | em->start = logical; |
6545 | em->len = length; | |
70c8a91c | 6546 | em->orig_start = 0; |
0b86a832 | 6547 | em->block_start = 0; |
c8b97818 | 6548 | em->block_len = em->len; |
0b86a832 | 6549 | |
593060d7 CM |
6550 | map->num_stripes = num_stripes; |
6551 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6552 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6553 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6554 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6555 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6556 | for (i = 0; i < num_stripes; i++) { |
6557 | map->stripes[i].physical = | |
6558 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6559 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6560 | read_extent_buffer(leaf, uuid, (unsigned long) |
6561 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6562 | BTRFS_UUID_SIZE); | |
0b246afa | 6563 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6564 | uuid, NULL); |
3cdde224 | 6565 | if (!map->stripes[i].dev && |
0b246afa | 6566 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6567 | free_extent_map(em); |
2b902dfc | 6568 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6569 | return -ENOENT; |
593060d7 | 6570 | } |
dfe25020 CM |
6571 | if (!map->stripes[i].dev) { |
6572 | map->stripes[i].dev = | |
2ff7e61e JM |
6573 | add_missing_dev(fs_info->fs_devices, devid, |
6574 | uuid); | |
adfb69af | 6575 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6576 | free_extent_map(em); |
adfb69af AJ |
6577 | btrfs_err(fs_info, |
6578 | "failed to init missing dev %llu: %ld", | |
6579 | devid, PTR_ERR(map->stripes[i].dev)); | |
6580 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6581 | } |
2b902dfc | 6582 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6583 | } |
e12c9621 AJ |
6584 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6585 | &(map->stripes[i].dev->dev_state)); | |
6586 | ||
0b86a832 CM |
6587 | } |
6588 | ||
890871be | 6589 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6590 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6591 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6592 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6593 | free_extent_map(em); |
6594 | ||
6595 | return 0; | |
6596 | } | |
6597 | ||
143bede5 | 6598 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6599 | struct btrfs_dev_item *dev_item, |
6600 | struct btrfs_device *device) | |
6601 | { | |
6602 | unsigned long ptr; | |
0b86a832 CM |
6603 | |
6604 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6605 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6606 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6607 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6608 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6609 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6610 | device->type = btrfs_device_type(leaf, dev_item); |
6611 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6612 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6613 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6614 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6615 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6616 | |
410ba3a2 | 6617 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6618 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6619 | } |
6620 | ||
2ff7e61e | 6621 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6622 | u8 *fsid) |
2b82032c YZ |
6623 | { |
6624 | struct btrfs_fs_devices *fs_devices; | |
6625 | int ret; | |
6626 | ||
b367e47f | 6627 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2dfeca9b | 6628 | ASSERT(fsid); |
2b82032c | 6629 | |
0b246afa | 6630 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6631 | while (fs_devices) { |
44880fdc | 6632 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6633 | return fs_devices; |
6634 | ||
2b82032c YZ |
6635 | fs_devices = fs_devices->seed; |
6636 | } | |
6637 | ||
6638 | fs_devices = find_fsid(fsid); | |
6639 | if (!fs_devices) { | |
0b246afa | 6640 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6641 | return ERR_PTR(-ENOENT); |
6642 | ||
6643 | fs_devices = alloc_fs_devices(fsid); | |
6644 | if (IS_ERR(fs_devices)) | |
6645 | return fs_devices; | |
6646 | ||
6647 | fs_devices->seeding = 1; | |
6648 | fs_devices->opened = 1; | |
6649 | return fs_devices; | |
2b82032c | 6650 | } |
e4404d6e YZ |
6651 | |
6652 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6653 | if (IS_ERR(fs_devices)) |
6654 | return fs_devices; | |
2b82032c | 6655 | |
97288f2c | 6656 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
0b246afa | 6657 | fs_info->bdev_holder); |
48d28232 JL |
6658 | if (ret) { |
6659 | free_fs_devices(fs_devices); | |
5f375835 | 6660 | fs_devices = ERR_PTR(ret); |
2b82032c | 6661 | goto out; |
48d28232 | 6662 | } |
2b82032c YZ |
6663 | |
6664 | if (!fs_devices->seeding) { | |
6665 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6666 | free_fs_devices(fs_devices); |
5f375835 | 6667 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6668 | goto out; |
6669 | } | |
6670 | ||
0b246afa JM |
6671 | fs_devices->seed = fs_info->fs_devices->seed; |
6672 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6673 | out: |
5f375835 | 6674 | return fs_devices; |
2b82032c YZ |
6675 | } |
6676 | ||
2ff7e61e | 6677 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6678 | struct extent_buffer *leaf, |
6679 | struct btrfs_dev_item *dev_item) | |
6680 | { | |
0b246afa | 6681 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6682 | struct btrfs_device *device; |
6683 | u64 devid; | |
6684 | int ret; | |
44880fdc | 6685 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6686 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6687 | ||
0b86a832 | 6688 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6689 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6690 | BTRFS_UUID_SIZE); |
1473b24e | 6691 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6692 | BTRFS_FSID_SIZE); |
2b82032c | 6693 | |
44880fdc | 6694 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6695 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6696 | if (IS_ERR(fs_devices)) |
6697 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6698 | } |
6699 | ||
0b246afa | 6700 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6701 | if (!device) { |
c5502451 | 6702 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6703 | btrfs_report_missing_device(fs_info, devid, |
6704 | dev_uuid, true); | |
45dbdbc9 | 6705 | return -ENOENT; |
c5502451 | 6706 | } |
2b82032c | 6707 | |
2ff7e61e | 6708 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6709 | if (IS_ERR(device)) { |
6710 | btrfs_err(fs_info, | |
6711 | "failed to add missing dev %llu: %ld", | |
6712 | devid, PTR_ERR(device)); | |
6713 | return PTR_ERR(device); | |
6714 | } | |
2b902dfc | 6715 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6716 | } else { |
c5502451 | 6717 | if (!device->bdev) { |
2b902dfc AJ |
6718 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6719 | btrfs_report_missing_device(fs_info, | |
6720 | devid, dev_uuid, true); | |
45dbdbc9 | 6721 | return -ENOENT; |
2b902dfc AJ |
6722 | } |
6723 | btrfs_report_missing_device(fs_info, devid, | |
6724 | dev_uuid, false); | |
c5502451 | 6725 | } |
5f375835 | 6726 | |
e6e674bd AJ |
6727 | if (!device->bdev && |
6728 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6729 | /* |
6730 | * this happens when a device that was properly setup | |
6731 | * in the device info lists suddenly goes bad. | |
6732 | * device->bdev is NULL, and so we have to set | |
6733 | * device->missing to one here | |
6734 | */ | |
5f375835 | 6735 | device->fs_devices->missing_devices++; |
e6e674bd | 6736 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6737 | } |
5f375835 MX |
6738 | |
6739 | /* Move the device to its own fs_devices */ | |
6740 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6741 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6742 | &device->dev_state)); | |
5f375835 MX |
6743 | |
6744 | list_move(&device->dev_list, &fs_devices->devices); | |
6745 | device->fs_devices->num_devices--; | |
6746 | fs_devices->num_devices++; | |
6747 | ||
6748 | device->fs_devices->missing_devices--; | |
6749 | fs_devices->missing_devices++; | |
6750 | ||
6751 | device->fs_devices = fs_devices; | |
6752 | } | |
2b82032c YZ |
6753 | } |
6754 | ||
0b246afa | 6755 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6756 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6757 | if (device->generation != |
6758 | btrfs_device_generation(leaf, dev_item)) | |
6759 | return -EINVAL; | |
6324fbf3 | 6760 | } |
0b86a832 CM |
6761 | |
6762 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6763 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6764 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6765 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6766 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6767 | atomic64_add(device->total_bytes - device->bytes_used, |
6768 | &fs_info->free_chunk_space); | |
2bf64758 | 6769 | } |
0b86a832 | 6770 | ret = 0; |
0b86a832 CM |
6771 | return ret; |
6772 | } | |
6773 | ||
6bccf3ab | 6774 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6775 | { |
6bccf3ab | 6776 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6777 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6778 | struct extent_buffer *sb; |
0b86a832 | 6779 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6780 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6781 | u8 *array_ptr; |
6782 | unsigned long sb_array_offset; | |
84eed90f | 6783 | int ret = 0; |
0b86a832 CM |
6784 | u32 num_stripes; |
6785 | u32 array_size; | |
6786 | u32 len = 0; | |
1ffb22cf | 6787 | u32 cur_offset; |
e06cd3dd | 6788 | u64 type; |
84eed90f | 6789 | struct btrfs_key key; |
0b86a832 | 6790 | |
0b246afa | 6791 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6792 | /* |
6793 | * This will create extent buffer of nodesize, superblock size is | |
6794 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6795 | * overallocate but we can keep it as-is, only the first page is used. | |
6796 | */ | |
2ff7e61e | 6797 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6798 | if (IS_ERR(sb)) |
6799 | return PTR_ERR(sb); | |
4db8c528 | 6800 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6801 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6802 | /* |
01327610 | 6803 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6804 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6805 | * pages up-to-date when the page is larger: extent does not cover the |
6806 | * whole page and consequently check_page_uptodate does not find all | |
6807 | * the page's extents up-to-date (the hole beyond sb), | |
6808 | * write_extent_buffer then triggers a WARN_ON. | |
6809 | * | |
6810 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6811 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6812 | * to silence the warning eg. on PowerPC 64. | |
6813 | */ | |
09cbfeaf | 6814 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6815 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6816 | |
a061fc8d | 6817 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6818 | array_size = btrfs_super_sys_array_size(super_copy); |
6819 | ||
1ffb22cf DS |
6820 | array_ptr = super_copy->sys_chunk_array; |
6821 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6822 | cur_offset = 0; | |
0b86a832 | 6823 | |
1ffb22cf DS |
6824 | while (cur_offset < array_size) { |
6825 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6826 | len = sizeof(*disk_key); |
6827 | if (cur_offset + len > array_size) | |
6828 | goto out_short_read; | |
6829 | ||
0b86a832 CM |
6830 | btrfs_disk_key_to_cpu(&key, disk_key); |
6831 | ||
1ffb22cf DS |
6832 | array_ptr += len; |
6833 | sb_array_offset += len; | |
6834 | cur_offset += len; | |
0b86a832 | 6835 | |
0d81ba5d | 6836 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6837 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6838 | /* |
6839 | * At least one btrfs_chunk with one stripe must be | |
6840 | * present, exact stripe count check comes afterwards | |
6841 | */ | |
6842 | len = btrfs_chunk_item_size(1); | |
6843 | if (cur_offset + len > array_size) | |
6844 | goto out_short_read; | |
6845 | ||
6846 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6847 | if (!num_stripes) { |
ab8d0fc4 JM |
6848 | btrfs_err(fs_info, |
6849 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6850 | num_stripes, cur_offset); |
6851 | ret = -EIO; | |
6852 | break; | |
6853 | } | |
6854 | ||
e06cd3dd LB |
6855 | type = btrfs_chunk_type(sb, chunk); |
6856 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6857 | btrfs_err(fs_info, |
e06cd3dd LB |
6858 | "invalid chunk type %llu in sys_array at offset %u", |
6859 | type, cur_offset); | |
6860 | ret = -EIO; | |
6861 | break; | |
6862 | } | |
6863 | ||
e3540eab DS |
6864 | len = btrfs_chunk_item_size(num_stripes); |
6865 | if (cur_offset + len > array_size) | |
6866 | goto out_short_read; | |
6867 | ||
2ff7e61e | 6868 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6869 | if (ret) |
6870 | break; | |
0b86a832 | 6871 | } else { |
ab8d0fc4 JM |
6872 | btrfs_err(fs_info, |
6873 | "unexpected item type %u in sys_array at offset %u", | |
6874 | (u32)key.type, cur_offset); | |
84eed90f CM |
6875 | ret = -EIO; |
6876 | break; | |
0b86a832 | 6877 | } |
1ffb22cf DS |
6878 | array_ptr += len; |
6879 | sb_array_offset += len; | |
6880 | cur_offset += len; | |
0b86a832 | 6881 | } |
d865177a | 6882 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6883 | free_extent_buffer_stale(sb); |
84eed90f | 6884 | return ret; |
e3540eab DS |
6885 | |
6886 | out_short_read: | |
ab8d0fc4 | 6887 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6888 | len, cur_offset); |
d865177a | 6889 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6890 | free_extent_buffer_stale(sb); |
e3540eab | 6891 | return -EIO; |
0b86a832 CM |
6892 | } |
6893 | ||
21634a19 QW |
6894 | /* |
6895 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6896 | * | |
6528b99d AJ |
6897 | * If the @failing_dev is specified, it's accounted as missing. |
6898 | * | |
21634a19 QW |
6899 | * Return true if all chunks meet the minimal RW mount requirements. |
6900 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6901 | */ | |
6528b99d AJ |
6902 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6903 | struct btrfs_device *failing_dev) | |
21634a19 QW |
6904 | { |
6905 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
6906 | struct extent_map *em; | |
6907 | u64 next_start = 0; | |
6908 | bool ret = true; | |
6909 | ||
6910 | read_lock(&map_tree->map_tree.lock); | |
6911 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
6912 | read_unlock(&map_tree->map_tree.lock); | |
6913 | /* No chunk at all? Return false anyway */ | |
6914 | if (!em) { | |
6915 | ret = false; | |
6916 | goto out; | |
6917 | } | |
6918 | while (em) { | |
6919 | struct map_lookup *map; | |
6920 | int missing = 0; | |
6921 | int max_tolerated; | |
6922 | int i; | |
6923 | ||
6924 | map = em->map_lookup; | |
6925 | max_tolerated = | |
6926 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6927 | map->type); | |
6928 | for (i = 0; i < map->num_stripes; i++) { | |
6929 | struct btrfs_device *dev = map->stripes[i].dev; | |
6930 | ||
e6e674bd AJ |
6931 | if (!dev || !dev->bdev || |
6932 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6933 | dev->last_flush_error) |
6934 | missing++; | |
6528b99d AJ |
6935 | else if (failing_dev && failing_dev == dev) |
6936 | missing++; | |
21634a19 QW |
6937 | } |
6938 | if (missing > max_tolerated) { | |
6528b99d AJ |
6939 | if (!failing_dev) |
6940 | btrfs_warn(fs_info, | |
21634a19 QW |
6941 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
6942 | em->start, missing, max_tolerated); | |
6943 | free_extent_map(em); | |
6944 | ret = false; | |
6945 | goto out; | |
6946 | } | |
6947 | next_start = extent_map_end(em); | |
6948 | free_extent_map(em); | |
6949 | ||
6950 | read_lock(&map_tree->map_tree.lock); | |
6951 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
6952 | (u64)(-1) - next_start); | |
6953 | read_unlock(&map_tree->map_tree.lock); | |
6954 | } | |
6955 | out: | |
6956 | return ret; | |
6957 | } | |
6958 | ||
5b4aacef | 6959 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6960 | { |
5b4aacef | 6961 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6962 | struct btrfs_path *path; |
6963 | struct extent_buffer *leaf; | |
6964 | struct btrfs_key key; | |
6965 | struct btrfs_key found_key; | |
6966 | int ret; | |
6967 | int slot; | |
99e3ecfc | 6968 | u64 total_dev = 0; |
0b86a832 | 6969 | |
0b86a832 CM |
6970 | path = btrfs_alloc_path(); |
6971 | if (!path) | |
6972 | return -ENOMEM; | |
6973 | ||
b367e47f | 6974 | mutex_lock(&uuid_mutex); |
34441361 | 6975 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6976 | |
395927a9 FDBM |
6977 | /* |
6978 | * Read all device items, and then all the chunk items. All | |
6979 | * device items are found before any chunk item (their object id | |
6980 | * is smaller than the lowest possible object id for a chunk | |
6981 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6982 | */ |
6983 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6984 | key.offset = 0; | |
6985 | key.type = 0; | |
0b86a832 | 6986 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6987 | if (ret < 0) |
6988 | goto error; | |
d397712b | 6989 | while (1) { |
0b86a832 CM |
6990 | leaf = path->nodes[0]; |
6991 | slot = path->slots[0]; | |
6992 | if (slot >= btrfs_header_nritems(leaf)) { | |
6993 | ret = btrfs_next_leaf(root, path); | |
6994 | if (ret == 0) | |
6995 | continue; | |
6996 | if (ret < 0) | |
6997 | goto error; | |
6998 | break; | |
6999 | } | |
7000 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7001 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7002 | struct btrfs_dev_item *dev_item; | |
7003 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7004 | struct btrfs_dev_item); |
2ff7e61e | 7005 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
7006 | if (ret) |
7007 | goto error; | |
99e3ecfc | 7008 | total_dev++; |
0b86a832 CM |
7009 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7010 | struct btrfs_chunk *chunk; | |
7011 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7012 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7013 | if (ret) |
7014 | goto error; | |
0b86a832 CM |
7015 | } |
7016 | path->slots[0]++; | |
7017 | } | |
99e3ecfc LB |
7018 | |
7019 | /* | |
7020 | * After loading chunk tree, we've got all device information, | |
7021 | * do another round of validation checks. | |
7022 | */ | |
0b246afa JM |
7023 | if (total_dev != fs_info->fs_devices->total_devices) { |
7024 | btrfs_err(fs_info, | |
99e3ecfc | 7025 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7026 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7027 | total_dev); |
7028 | ret = -EINVAL; | |
7029 | goto error; | |
7030 | } | |
0b246afa JM |
7031 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7032 | fs_info->fs_devices->total_rw_bytes) { | |
7033 | btrfs_err(fs_info, | |
99e3ecfc | 7034 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7035 | btrfs_super_total_bytes(fs_info->super_copy), |
7036 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7037 | ret = -EINVAL; |
7038 | goto error; | |
7039 | } | |
0b86a832 CM |
7040 | ret = 0; |
7041 | error: | |
34441361 | 7042 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7043 | mutex_unlock(&uuid_mutex); |
7044 | ||
2b82032c | 7045 | btrfs_free_path(path); |
0b86a832 CM |
7046 | return ret; |
7047 | } | |
442a4f63 | 7048 | |
cb517eab MX |
7049 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7050 | { | |
7051 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7052 | struct btrfs_device *device; | |
7053 | ||
29cc83f6 LB |
7054 | while (fs_devices) { |
7055 | mutex_lock(&fs_devices->device_list_mutex); | |
7056 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7057 | device->fs_info = fs_info; |
29cc83f6 LB |
7058 | mutex_unlock(&fs_devices->device_list_mutex); |
7059 | ||
7060 | fs_devices = fs_devices->seed; | |
7061 | } | |
cb517eab MX |
7062 | } |
7063 | ||
733f4fbb SB |
7064 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7065 | { | |
7066 | int i; | |
7067 | ||
7068 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7069 | btrfs_dev_stat_reset(dev, i); | |
7070 | } | |
7071 | ||
7072 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7073 | { | |
7074 | struct btrfs_key key; | |
7075 | struct btrfs_key found_key; | |
7076 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7077 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7078 | struct extent_buffer *eb; | |
7079 | int slot; | |
7080 | int ret = 0; | |
7081 | struct btrfs_device *device; | |
7082 | struct btrfs_path *path = NULL; | |
7083 | int i; | |
7084 | ||
7085 | path = btrfs_alloc_path(); | |
7086 | if (!path) { | |
7087 | ret = -ENOMEM; | |
7088 | goto out; | |
7089 | } | |
7090 | ||
7091 | mutex_lock(&fs_devices->device_list_mutex); | |
7092 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7093 | int item_size; | |
7094 | struct btrfs_dev_stats_item *ptr; | |
7095 | ||
242e2956 DS |
7096 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7097 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7098 | key.offset = device->devid; |
7099 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7100 | if (ret) { | |
733f4fbb SB |
7101 | __btrfs_reset_dev_stats(device); |
7102 | device->dev_stats_valid = 1; | |
7103 | btrfs_release_path(path); | |
7104 | continue; | |
7105 | } | |
7106 | slot = path->slots[0]; | |
7107 | eb = path->nodes[0]; | |
7108 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7109 | item_size = btrfs_item_size_nr(eb, slot); | |
7110 | ||
7111 | ptr = btrfs_item_ptr(eb, slot, | |
7112 | struct btrfs_dev_stats_item); | |
7113 | ||
7114 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7115 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7116 | btrfs_dev_stat_set(device, i, | |
7117 | btrfs_dev_stats_value(eb, ptr, i)); | |
7118 | else | |
7119 | btrfs_dev_stat_reset(device, i); | |
7120 | } | |
7121 | ||
7122 | device->dev_stats_valid = 1; | |
7123 | btrfs_dev_stat_print_on_load(device); | |
7124 | btrfs_release_path(path); | |
7125 | } | |
7126 | mutex_unlock(&fs_devices->device_list_mutex); | |
7127 | ||
7128 | out: | |
7129 | btrfs_free_path(path); | |
7130 | return ret < 0 ? ret : 0; | |
7131 | } | |
7132 | ||
7133 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6bccf3ab | 7134 | struct btrfs_fs_info *fs_info, |
733f4fbb SB |
7135 | struct btrfs_device *device) |
7136 | { | |
6bccf3ab | 7137 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7138 | struct btrfs_path *path; |
7139 | struct btrfs_key key; | |
7140 | struct extent_buffer *eb; | |
7141 | struct btrfs_dev_stats_item *ptr; | |
7142 | int ret; | |
7143 | int i; | |
7144 | ||
242e2956 DS |
7145 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7146 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7147 | key.offset = device->devid; |
7148 | ||
7149 | path = btrfs_alloc_path(); | |
fa252992 DS |
7150 | if (!path) |
7151 | return -ENOMEM; | |
733f4fbb SB |
7152 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7153 | if (ret < 0) { | |
0b246afa | 7154 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7155 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7156 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7157 | goto out; |
7158 | } | |
7159 | ||
7160 | if (ret == 0 && | |
7161 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7162 | /* need to delete old one and insert a new one */ | |
7163 | ret = btrfs_del_item(trans, dev_root, path); | |
7164 | if (ret != 0) { | |
0b246afa | 7165 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7166 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7167 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7168 | goto out; |
7169 | } | |
7170 | ret = 1; | |
7171 | } | |
7172 | ||
7173 | if (ret == 1) { | |
7174 | /* need to insert a new item */ | |
7175 | btrfs_release_path(path); | |
7176 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7177 | &key, sizeof(*ptr)); | |
7178 | if (ret < 0) { | |
0b246afa | 7179 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7180 | "insert dev_stats item for device %s failed %d", |
7181 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7182 | goto out; |
7183 | } | |
7184 | } | |
7185 | ||
7186 | eb = path->nodes[0]; | |
7187 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7188 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7189 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7190 | btrfs_dev_stat_read(device, i)); | |
7191 | btrfs_mark_buffer_dirty(eb); | |
7192 | ||
7193 | out: | |
7194 | btrfs_free_path(path); | |
7195 | return ret; | |
7196 | } | |
7197 | ||
7198 | /* | |
7199 | * called from commit_transaction. Writes all changed device stats to disk. | |
7200 | */ | |
7201 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7202 | struct btrfs_fs_info *fs_info) | |
7203 | { | |
733f4fbb SB |
7204 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7205 | struct btrfs_device *device; | |
addc3fa7 | 7206 | int stats_cnt; |
733f4fbb SB |
7207 | int ret = 0; |
7208 | ||
7209 | mutex_lock(&fs_devices->device_list_mutex); | |
7210 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7211 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7212 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7213 | continue; |
7214 | ||
9deae968 NB |
7215 | |
7216 | /* | |
7217 | * There is a LOAD-LOAD control dependency between the value of | |
7218 | * dev_stats_ccnt and updating the on-disk values which requires | |
7219 | * reading the in-memory counters. Such control dependencies | |
7220 | * require explicit read memory barriers. | |
7221 | * | |
7222 | * This memory barriers pairs with smp_mb__before_atomic in | |
7223 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7224 | * barrier implied by atomic_xchg in | |
7225 | * btrfs_dev_stats_read_and_reset | |
7226 | */ | |
7227 | smp_rmb(); | |
7228 | ||
6bccf3ab | 7229 | ret = update_dev_stat_item(trans, fs_info, device); |
733f4fbb | 7230 | if (!ret) |
addc3fa7 | 7231 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7232 | } |
7233 | mutex_unlock(&fs_devices->device_list_mutex); | |
7234 | ||
7235 | return ret; | |
7236 | } | |
7237 | ||
442a4f63 SB |
7238 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7239 | { | |
7240 | btrfs_dev_stat_inc(dev, index); | |
7241 | btrfs_dev_stat_print_on_error(dev); | |
7242 | } | |
7243 | ||
48a3b636 | 7244 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7245 | { |
733f4fbb SB |
7246 | if (!dev->dev_stats_valid) |
7247 | return; | |
fb456252 | 7248 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7249 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7250 | rcu_str_deref(dev->name), |
442a4f63 SB |
7251 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7252 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7253 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7254 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7255 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7256 | } |
c11d2c23 | 7257 | |
733f4fbb SB |
7258 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7259 | { | |
a98cdb85 SB |
7260 | int i; |
7261 | ||
7262 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7263 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7264 | break; | |
7265 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7266 | return; /* all values == 0, suppress message */ | |
7267 | ||
fb456252 | 7268 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7269 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7270 | rcu_str_deref(dev->name), |
733f4fbb SB |
7271 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7272 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7273 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7274 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7275 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7276 | } | |
7277 | ||
2ff7e61e | 7278 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7279 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7280 | { |
7281 | struct btrfs_device *dev; | |
0b246afa | 7282 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7283 | int i; |
7284 | ||
7285 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7286 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7287 | mutex_unlock(&fs_devices->device_list_mutex); |
7288 | ||
7289 | if (!dev) { | |
0b246afa | 7290 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7291 | return -ENODEV; |
733f4fbb | 7292 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7293 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7294 | return -ENODEV; |
b27f7c0c | 7295 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7296 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7297 | if (stats->nr_items > i) | |
7298 | stats->values[i] = | |
7299 | btrfs_dev_stat_read_and_reset(dev, i); | |
7300 | else | |
7301 | btrfs_dev_stat_reset(dev, i); | |
7302 | } | |
7303 | } else { | |
7304 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7305 | if (stats->nr_items > i) | |
7306 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7307 | } | |
7308 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7309 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7310 | return 0; | |
7311 | } | |
a8a6dab7 | 7312 | |
da353f6b | 7313 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7314 | { |
7315 | struct buffer_head *bh; | |
7316 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7317 | int copy_num; |
a8a6dab7 | 7318 | |
12b1c263 AJ |
7319 | if (!bdev) |
7320 | return; | |
a8a6dab7 | 7321 | |
12b1c263 AJ |
7322 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7323 | copy_num++) { | |
a8a6dab7 | 7324 | |
12b1c263 AJ |
7325 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7326 | continue; | |
7327 | ||
7328 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7329 | ||
7330 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7331 | set_buffer_dirty(bh); | |
7332 | sync_dirty_buffer(bh); | |
7333 | brelse(bh); | |
7334 | } | |
7335 | ||
7336 | /* Notify udev that device has changed */ | |
7337 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7338 | ||
7339 | /* Update ctime/mtime for device path for libblkid */ | |
7340 | update_dev_time(device_path); | |
a8a6dab7 | 7341 | } |
935e5cc9 MX |
7342 | |
7343 | /* | |
7344 | * Update the size of all devices, which is used for writing out the | |
7345 | * super blocks. | |
7346 | */ | |
7347 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7348 | { | |
7349 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7350 | struct btrfs_device *curr, *next; | |
7351 | ||
7352 | if (list_empty(&fs_devices->resized_devices)) | |
7353 | return; | |
7354 | ||
7355 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7356 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7357 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7358 | resized_list) { | |
7359 | list_del_init(&curr->resized_list); | |
7360 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7361 | } | |
34441361 | 7362 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7363 | mutex_unlock(&fs_devices->device_list_mutex); |
7364 | } | |
ce7213c7 MX |
7365 | |
7366 | /* Must be invoked during the transaction commit */ | |
2ff7e61e | 7367 | void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info, |
ce7213c7 MX |
7368 | struct btrfs_transaction *transaction) |
7369 | { | |
7370 | struct extent_map *em; | |
7371 | struct map_lookup *map; | |
7372 | struct btrfs_device *dev; | |
7373 | int i; | |
7374 | ||
7375 | if (list_empty(&transaction->pending_chunks)) | |
7376 | return; | |
7377 | ||
7378 | /* In order to kick the device replace finish process */ | |
34441361 | 7379 | mutex_lock(&fs_info->chunk_mutex); |
ce7213c7 | 7380 | list_for_each_entry(em, &transaction->pending_chunks, list) { |
95617d69 | 7381 | map = em->map_lookup; |
ce7213c7 MX |
7382 | |
7383 | for (i = 0; i < map->num_stripes; i++) { | |
7384 | dev = map->stripes[i].dev; | |
7385 | dev->commit_bytes_used = dev->bytes_used; | |
7386 | } | |
7387 | } | |
34441361 | 7388 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7389 | } |
5a13f430 AJ |
7390 | |
7391 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7392 | { | |
7393 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7394 | while (fs_devices) { | |
7395 | fs_devices->fs_info = fs_info; | |
7396 | fs_devices = fs_devices->seed; | |
7397 | } | |
7398 | } | |
7399 | ||
7400 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7401 | { | |
7402 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7403 | while (fs_devices) { | |
7404 | fs_devices->fs_info = NULL; | |
7405 | fs_devices = fs_devices->seed; | |
7406 | } | |
7407 | } |