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> |
f8e10cd3 | 30 | #include <linux/list_sort.h> |
53b381b3 | 31 | #include <asm/div64.h> |
0b86a832 CM |
32 | #include "ctree.h" |
33 | #include "extent_map.h" | |
34 | #include "disk-io.h" | |
35 | #include "transaction.h" | |
36 | #include "print-tree.h" | |
37 | #include "volumes.h" | |
53b381b3 | 38 | #include "raid56.h" |
8b712842 | 39 | #include "async-thread.h" |
21adbd5c | 40 | #include "check-integrity.h" |
606686ee | 41 | #include "rcu-string.h" |
3fed40cc | 42 | #include "math.h" |
8dabb742 | 43 | #include "dev-replace.h" |
99994cde | 44 | #include "sysfs.h" |
0b86a832 | 45 | |
af902047 ZL |
46 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
47 | [BTRFS_RAID_RAID10] = { | |
48 | .sub_stripes = 2, | |
49 | .dev_stripes = 1, | |
50 | .devs_max = 0, /* 0 == as many as possible */ | |
51 | .devs_min = 4, | |
8789f4fe | 52 | .tolerated_failures = 1, |
af902047 ZL |
53 | .devs_increment = 2, |
54 | .ncopies = 2, | |
55 | }, | |
56 | [BTRFS_RAID_RAID1] = { | |
57 | .sub_stripes = 1, | |
58 | .dev_stripes = 1, | |
59 | .devs_max = 2, | |
60 | .devs_min = 2, | |
8789f4fe | 61 | .tolerated_failures = 1, |
af902047 ZL |
62 | .devs_increment = 2, |
63 | .ncopies = 2, | |
64 | }, | |
65 | [BTRFS_RAID_DUP] = { | |
66 | .sub_stripes = 1, | |
67 | .dev_stripes = 2, | |
68 | .devs_max = 1, | |
69 | .devs_min = 1, | |
8789f4fe | 70 | .tolerated_failures = 0, |
af902047 ZL |
71 | .devs_increment = 1, |
72 | .ncopies = 2, | |
73 | }, | |
74 | [BTRFS_RAID_RAID0] = { | |
75 | .sub_stripes = 1, | |
76 | .dev_stripes = 1, | |
77 | .devs_max = 0, | |
78 | .devs_min = 2, | |
8789f4fe | 79 | .tolerated_failures = 0, |
af902047 ZL |
80 | .devs_increment = 1, |
81 | .ncopies = 1, | |
82 | }, | |
83 | [BTRFS_RAID_SINGLE] = { | |
84 | .sub_stripes = 1, | |
85 | .dev_stripes = 1, | |
86 | .devs_max = 1, | |
87 | .devs_min = 1, | |
8789f4fe | 88 | .tolerated_failures = 0, |
af902047 ZL |
89 | .devs_increment = 1, |
90 | .ncopies = 1, | |
91 | }, | |
92 | [BTRFS_RAID_RAID5] = { | |
93 | .sub_stripes = 1, | |
94 | .dev_stripes = 1, | |
95 | .devs_max = 0, | |
96 | .devs_min = 2, | |
8789f4fe | 97 | .tolerated_failures = 1, |
af902047 ZL |
98 | .devs_increment = 1, |
99 | .ncopies = 2, | |
100 | }, | |
101 | [BTRFS_RAID_RAID6] = { | |
102 | .sub_stripes = 1, | |
103 | .dev_stripes = 1, | |
104 | .devs_max = 0, | |
105 | .devs_min = 3, | |
8789f4fe | 106 | .tolerated_failures = 2, |
af902047 ZL |
107 | .devs_increment = 1, |
108 | .ncopies = 3, | |
109 | }, | |
110 | }; | |
111 | ||
fb75d857 | 112 | const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { |
af902047 ZL |
113 | [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10, |
114 | [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1, | |
115 | [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP, | |
116 | [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0, | |
117 | [BTRFS_RAID_SINGLE] = 0, | |
118 | [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5, | |
119 | [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6, | |
120 | }; | |
121 | ||
621292ba DS |
122 | /* |
123 | * Table to convert BTRFS_RAID_* to the error code if minimum number of devices | |
124 | * condition is not met. Zero means there's no corresponding | |
125 | * BTRFS_ERROR_DEV_*_NOT_MET value. | |
126 | */ | |
127 | const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = { | |
128 | [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, | |
129 | [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, | |
130 | [BTRFS_RAID_DUP] = 0, | |
131 | [BTRFS_RAID_RAID0] = 0, | |
132 | [BTRFS_RAID_SINGLE] = 0, | |
133 | [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, | |
134 | [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, | |
135 | }; | |
136 | ||
2b82032c | 137 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 138 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 139 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 140 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 141 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 142 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
143 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
144 | enum btrfs_map_op op, | |
145 | u64 logical, u64 *length, | |
146 | struct btrfs_bio **bbio_ret, | |
147 | int mirror_num, int need_raid_map); | |
2b82032c | 148 | |
9c6b1c4d DS |
149 | /* |
150 | * Device locking | |
151 | * ============== | |
152 | * | |
153 | * There are several mutexes that protect manipulation of devices and low-level | |
154 | * structures like chunks but not block groups, extents or files | |
155 | * | |
156 | * uuid_mutex (global lock) | |
157 | * ------------------------ | |
158 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
159 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
160 | * device) or requested by the device= mount option | |
161 | * | |
162 | * the mutex can be very coarse and can cover long-running operations | |
163 | * | |
164 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
165 | * seeding, structure cloning, openning/closing devices at mount/umount time | |
166 | * | |
167 | * global::fs_devs - add, remove, updates to the global list | |
168 | * | |
169 | * does not protect: manipulation of the fs_devices::devices list! | |
170 | * | |
171 | * btrfs_device::name - renames (write side), read is RCU | |
172 | * | |
173 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
174 | * ------------------------------------------------ | |
175 | * protects updates to fs_devices::devices, ie. adding and deleting | |
176 | * | |
177 | * simple list traversal with read-only actions can be done with RCU protection | |
178 | * | |
179 | * may be used to exclude some operations from running concurrently without any | |
180 | * modifications to the list (see write_all_supers) | |
181 | * | |
182 | * volume_mutex | |
183 | * ------------ | |
184 | * coarse lock owned by a mounted filesystem; used to exclude some operations | |
185 | * that cannot run in parallel and affect the higher-level properties of the | |
186 | * filesystem like: device add/deleting/resize/replace, or balance | |
187 | * | |
188 | * balance_mutex | |
189 | * ------------- | |
190 | * protects balance structures (status, state) and context accessed from | |
191 | * several places (internally, ioctl) | |
192 | * | |
193 | * chunk_mutex | |
194 | * ----------- | |
195 | * protects chunks, adding or removing during allocation, trim or when a new | |
196 | * device is added/removed | |
197 | * | |
198 | * cleaner_mutex | |
199 | * ------------- | |
200 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
201 | * cleaning together with relocation or delayed iputs | |
202 | * | |
203 | * | |
204 | * Lock nesting | |
205 | * ============ | |
206 | * | |
207 | * uuid_mutex | |
208 | * volume_mutex | |
209 | * device_list_mutex | |
210 | * chunk_mutex | |
211 | * balance_mutex | |
212 | */ | |
213 | ||
67a2c45e | 214 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 215 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
216 | struct list_head *btrfs_get_fs_uuids(void) |
217 | { | |
218 | return &fs_uuids; | |
219 | } | |
8a4b83cc | 220 | |
2dfeca9b DS |
221 | /* |
222 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
223 | * @fsid: if not NULL, copy the uuid to fs_devices::fsid | |
224 | * | |
225 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
226 | * The returned struct is not linked onto any lists and can be destroyed with | |
227 | * kfree() right away. | |
228 | */ | |
229 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
2208a378 ID |
230 | { |
231 | struct btrfs_fs_devices *fs_devs; | |
232 | ||
78f2c9e6 | 233 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
234 | if (!fs_devs) |
235 | return ERR_PTR(-ENOMEM); | |
236 | ||
237 | mutex_init(&fs_devs->device_list_mutex); | |
238 | ||
239 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 240 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
241 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
242 | INIT_LIST_HEAD(&fs_devs->list); | |
2208a378 ID |
243 | if (fsid) |
244 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 ID |
245 | |
246 | return fs_devs; | |
247 | } | |
248 | ||
48dae9cf DS |
249 | static void free_device(struct btrfs_device *device) |
250 | { | |
251 | rcu_string_free(device->name); | |
252 | bio_put(device->flush_bio); | |
253 | kfree(device); | |
254 | } | |
255 | ||
e4404d6e YZ |
256 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
257 | { | |
258 | struct btrfs_device *device; | |
259 | WARN_ON(fs_devices->opened); | |
260 | while (!list_empty(&fs_devices->devices)) { | |
261 | device = list_entry(fs_devices->devices.next, | |
262 | struct btrfs_device, dev_list); | |
263 | list_del(&device->dev_list); | |
55de4803 | 264 | free_device(device); |
e4404d6e YZ |
265 | } |
266 | kfree(fs_devices); | |
267 | } | |
268 | ||
b8b8ff59 LC |
269 | static void btrfs_kobject_uevent(struct block_device *bdev, |
270 | enum kobject_action action) | |
271 | { | |
272 | int ret; | |
273 | ||
274 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
275 | if (ret) | |
efe120a0 | 276 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
277 | action, |
278 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
279 | &disk_to_dev(bdev->bd_disk)->kobj); | |
280 | } | |
281 | ||
ffc5a379 | 282 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
283 | { |
284 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 285 | |
2b82032c YZ |
286 | while (!list_empty(&fs_uuids)) { |
287 | fs_devices = list_entry(fs_uuids.next, | |
288 | struct btrfs_fs_devices, list); | |
289 | list_del(&fs_devices->list); | |
e4404d6e | 290 | free_fs_devices(fs_devices); |
8a4b83cc | 291 | } |
8a4b83cc CM |
292 | } |
293 | ||
48dae9cf DS |
294 | /* |
295 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
296 | * Returned struct is not linked onto any lists and must be destroyed using | |
297 | * free_device. | |
298 | */ | |
12bd2fc0 ID |
299 | static struct btrfs_device *__alloc_device(void) |
300 | { | |
301 | struct btrfs_device *dev; | |
302 | ||
78f2c9e6 | 303 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
304 | if (!dev) |
305 | return ERR_PTR(-ENOMEM); | |
306 | ||
e0ae9994 DS |
307 | /* |
308 | * Preallocate a bio that's always going to be used for flushing device | |
309 | * barriers and matches the device lifespan | |
310 | */ | |
311 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
312 | if (!dev->flush_bio) { | |
313 | kfree(dev); | |
314 | return ERR_PTR(-ENOMEM); | |
315 | } | |
e0ae9994 | 316 | |
12bd2fc0 ID |
317 | INIT_LIST_HEAD(&dev->dev_list); |
318 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 319 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
320 | |
321 | spin_lock_init(&dev->io_lock); | |
322 | ||
12bd2fc0 | 323 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 324 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 325 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 326 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 327 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
328 | |
329 | return dev; | |
330 | } | |
331 | ||
35c70103 DS |
332 | /* |
333 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
334 | * return NULL. | |
335 | * | |
336 | * If devid and uuid are both specified, the match must be exact, otherwise | |
337 | * only devid is used. | |
338 | */ | |
339 | static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices, | |
340 | u64 devid, const u8 *uuid) | |
8a4b83cc | 341 | { |
35c70103 | 342 | struct list_head *head = &fs_devices->devices; |
8a4b83cc | 343 | struct btrfs_device *dev; |
8a4b83cc | 344 | |
c6e30871 | 345 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 346 | if (dev->devid == devid && |
8f18cf13 | 347 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 348 | return dev; |
a443755f | 349 | } |
8a4b83cc CM |
350 | } |
351 | return NULL; | |
352 | } | |
353 | ||
a1b32a59 | 354 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 355 | { |
8a4b83cc CM |
356 | struct btrfs_fs_devices *fs_devices; |
357 | ||
c6e30871 | 358 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
359 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
360 | return fs_devices; | |
361 | } | |
362 | return NULL; | |
363 | } | |
364 | ||
beaf8ab3 SB |
365 | static int |
366 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
367 | int flush, struct block_device **bdev, | |
368 | struct buffer_head **bh) | |
369 | { | |
370 | int ret; | |
371 | ||
372 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
373 | ||
374 | if (IS_ERR(*bdev)) { | |
375 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
376 | goto error; |
377 | } | |
378 | ||
379 | if (flush) | |
380 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 381 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
382 | if (ret) { |
383 | blkdev_put(*bdev, flags); | |
384 | goto error; | |
385 | } | |
386 | invalidate_bdev(*bdev); | |
387 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
388 | if (IS_ERR(*bh)) { |
389 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
390 | blkdev_put(*bdev, flags); |
391 | goto error; | |
392 | } | |
393 | ||
394 | return 0; | |
395 | ||
396 | error: | |
397 | *bdev = NULL; | |
398 | *bh = NULL; | |
399 | return ret; | |
400 | } | |
401 | ||
ffbd517d CM |
402 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
403 | struct bio *head, struct bio *tail) | |
404 | { | |
405 | ||
406 | struct bio *old_head; | |
407 | ||
408 | old_head = pending_bios->head; | |
409 | pending_bios->head = head; | |
410 | if (pending_bios->tail) | |
411 | tail->bi_next = old_head; | |
412 | else | |
413 | pending_bios->tail = tail; | |
414 | } | |
415 | ||
8b712842 CM |
416 | /* |
417 | * we try to collect pending bios for a device so we don't get a large | |
418 | * number of procs sending bios down to the same device. This greatly | |
419 | * improves the schedulers ability to collect and merge the bios. | |
420 | * | |
421 | * But, it also turns into a long list of bios to process and that is sure | |
422 | * to eventually make the worker thread block. The solution here is to | |
423 | * make some progress and then put this work struct back at the end of | |
424 | * the list if the block device is congested. This way, multiple devices | |
425 | * can make progress from a single worker thread. | |
426 | */ | |
143bede5 | 427 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 428 | { |
0b246afa | 429 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
430 | struct bio *pending; |
431 | struct backing_dev_info *bdi; | |
ffbd517d | 432 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
433 | struct bio *tail; |
434 | struct bio *cur; | |
435 | int again = 0; | |
ffbd517d | 436 | unsigned long num_run; |
d644d8a1 | 437 | unsigned long batch_run = 0; |
b765ead5 | 438 | unsigned long last_waited = 0; |
d84275c9 | 439 | int force_reg = 0; |
0e588859 | 440 | int sync_pending = 0; |
211588ad CM |
441 | struct blk_plug plug; |
442 | ||
443 | /* | |
444 | * this function runs all the bios we've collected for | |
445 | * a particular device. We don't want to wander off to | |
446 | * another device without first sending all of these down. | |
447 | * So, setup a plug here and finish it off before we return | |
448 | */ | |
449 | blk_start_plug(&plug); | |
8b712842 | 450 | |
efa7c9f9 | 451 | bdi = device->bdev->bd_bdi; |
b64a2851 | 452 | |
8b712842 CM |
453 | loop: |
454 | spin_lock(&device->io_lock); | |
455 | ||
a6837051 | 456 | loop_lock: |
d84275c9 | 457 | num_run = 0; |
ffbd517d | 458 | |
8b712842 CM |
459 | /* take all the bios off the list at once and process them |
460 | * later on (without the lock held). But, remember the | |
461 | * tail and other pointers so the bios can be properly reinserted | |
462 | * into the list if we hit congestion | |
463 | */ | |
d84275c9 | 464 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 465 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
466 | force_reg = 1; |
467 | } else { | |
ffbd517d | 468 | pending_bios = &device->pending_bios; |
d84275c9 CM |
469 | force_reg = 0; |
470 | } | |
ffbd517d CM |
471 | |
472 | pending = pending_bios->head; | |
473 | tail = pending_bios->tail; | |
8b712842 | 474 | WARN_ON(pending && !tail); |
8b712842 CM |
475 | |
476 | /* | |
477 | * if pending was null this time around, no bios need processing | |
478 | * at all and we can stop. Otherwise it'll loop back up again | |
479 | * and do an additional check so no bios are missed. | |
480 | * | |
481 | * device->running_pending is used to synchronize with the | |
482 | * schedule_bio code. | |
483 | */ | |
ffbd517d CM |
484 | if (device->pending_sync_bios.head == NULL && |
485 | device->pending_bios.head == NULL) { | |
8b712842 CM |
486 | again = 0; |
487 | device->running_pending = 0; | |
ffbd517d CM |
488 | } else { |
489 | again = 1; | |
490 | device->running_pending = 1; | |
8b712842 | 491 | } |
ffbd517d CM |
492 | |
493 | pending_bios->head = NULL; | |
494 | pending_bios->tail = NULL; | |
495 | ||
8b712842 CM |
496 | spin_unlock(&device->io_lock); |
497 | ||
d397712b | 498 | while (pending) { |
ffbd517d CM |
499 | |
500 | rmb(); | |
d84275c9 CM |
501 | /* we want to work on both lists, but do more bios on the |
502 | * sync list than the regular list | |
503 | */ | |
504 | if ((num_run > 32 && | |
505 | pending_bios != &device->pending_sync_bios && | |
506 | device->pending_sync_bios.head) || | |
507 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
508 | device->pending_bios.head)) { | |
ffbd517d CM |
509 | spin_lock(&device->io_lock); |
510 | requeue_list(pending_bios, pending, tail); | |
511 | goto loop_lock; | |
512 | } | |
513 | ||
8b712842 CM |
514 | cur = pending; |
515 | pending = pending->bi_next; | |
516 | cur->bi_next = NULL; | |
b64a2851 | 517 | |
dac56212 | 518 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 519 | |
2ab1ba68 CM |
520 | /* |
521 | * if we're doing the sync list, record that our | |
522 | * plug has some sync requests on it | |
523 | * | |
524 | * If we're doing the regular list and there are | |
525 | * sync requests sitting around, unplug before | |
526 | * we add more | |
527 | */ | |
528 | if (pending_bios == &device->pending_sync_bios) { | |
529 | sync_pending = 1; | |
530 | } else if (sync_pending) { | |
531 | blk_finish_plug(&plug); | |
532 | blk_start_plug(&plug); | |
533 | sync_pending = 0; | |
534 | } | |
535 | ||
4e49ea4a | 536 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
537 | num_run++; |
538 | batch_run++; | |
853d8ec4 DS |
539 | |
540 | cond_resched(); | |
8b712842 CM |
541 | |
542 | /* | |
543 | * we made progress, there is more work to do and the bdi | |
544 | * is now congested. Back off and let other work structs | |
545 | * run instead | |
546 | */ | |
57fd5a5f | 547 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 548 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 549 | struct io_context *ioc; |
8b712842 | 550 | |
b765ead5 CM |
551 | ioc = current->io_context; |
552 | ||
553 | /* | |
554 | * the main goal here is that we don't want to | |
555 | * block if we're going to be able to submit | |
556 | * more requests without blocking. | |
557 | * | |
558 | * This code does two great things, it pokes into | |
559 | * the elevator code from a filesystem _and_ | |
560 | * it makes assumptions about how batching works. | |
561 | */ | |
562 | if (ioc && ioc->nr_batch_requests > 0 && | |
563 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
564 | (last_waited == 0 || | |
565 | ioc->last_waited == last_waited)) { | |
566 | /* | |
567 | * we want to go through our batch of | |
568 | * requests and stop. So, we copy out | |
569 | * the ioc->last_waited time and test | |
570 | * against it before looping | |
571 | */ | |
572 | last_waited = ioc->last_waited; | |
853d8ec4 | 573 | cond_resched(); |
b765ead5 CM |
574 | continue; |
575 | } | |
8b712842 | 576 | spin_lock(&device->io_lock); |
ffbd517d | 577 | requeue_list(pending_bios, pending, tail); |
a6837051 | 578 | device->running_pending = 1; |
8b712842 CM |
579 | |
580 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
581 | btrfs_queue_work(fs_info->submit_workers, |
582 | &device->work); | |
8b712842 CM |
583 | goto done; |
584 | } | |
585 | } | |
ffbd517d | 586 | |
51684082 CM |
587 | cond_resched(); |
588 | if (again) | |
589 | goto loop; | |
590 | ||
591 | spin_lock(&device->io_lock); | |
592 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
593 | goto loop_lock; | |
594 | spin_unlock(&device->io_lock); | |
595 | ||
8b712842 | 596 | done: |
211588ad | 597 | blk_finish_plug(&plug); |
8b712842 CM |
598 | } |
599 | ||
b2950863 | 600 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
601 | { |
602 | struct btrfs_device *device; | |
603 | ||
604 | device = container_of(work, struct btrfs_device, work); | |
605 | run_scheduled_bios(device); | |
606 | } | |
607 | ||
d8367db3 AJ |
608 | /* |
609 | * Search and remove all stale (devices which are not mounted) devices. | |
610 | * When both inputs are NULL, it will search and release all stale devices. | |
611 | * path: Optional. When provided will it release all unmounted devices | |
612 | * matching this path only. | |
613 | * skip_dev: Optional. Will skip this device when searching for the stale | |
614 | * devices. | |
615 | */ | |
616 | static void btrfs_free_stale_devices(const char *path, | |
617 | struct btrfs_device *skip_dev) | |
4fde46f0 | 618 | { |
38cf665d AJ |
619 | struct btrfs_fs_devices *fs_devs, *tmp_fs_devs; |
620 | struct btrfs_device *dev, *tmp_dev; | |
4fde46f0 | 621 | |
38cf665d | 622 | list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, list) { |
4fde46f0 AJ |
623 | |
624 | if (fs_devs->opened) | |
625 | continue; | |
4fde46f0 | 626 | |
38cf665d AJ |
627 | list_for_each_entry_safe(dev, tmp_dev, |
628 | &fs_devs->devices, dev_list) { | |
522f1b45 | 629 | int not_found = 0; |
4fde46f0 | 630 | |
d8367db3 AJ |
631 | if (skip_dev && skip_dev == dev) |
632 | continue; | |
633 | if (path && !dev->name) | |
4fde46f0 AJ |
634 | continue; |
635 | ||
4fde46f0 | 636 | rcu_read_lock(); |
d8367db3 | 637 | if (path) |
522f1b45 | 638 | not_found = strcmp(rcu_str_deref(dev->name), |
d8367db3 | 639 | path); |
4fde46f0 | 640 | rcu_read_unlock(); |
38cf665d AJ |
641 | if (not_found) |
642 | continue; | |
4fde46f0 | 643 | |
4fde46f0 AJ |
644 | /* delete the stale device */ |
645 | if (fs_devs->num_devices == 1) { | |
646 | btrfs_sysfs_remove_fsid(fs_devs); | |
647 | list_del(&fs_devs->list); | |
648 | free_fs_devices(fs_devs); | |
fd649f10 | 649 | break; |
4fde46f0 AJ |
650 | } else { |
651 | fs_devs->num_devices--; | |
652 | list_del(&dev->dev_list); | |
55de4803 | 653 | free_device(dev); |
4fde46f0 | 654 | } |
4fde46f0 AJ |
655 | } |
656 | } | |
657 | } | |
658 | ||
0fb08bcc AJ |
659 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
660 | struct btrfs_device *device, fmode_t flags, | |
661 | void *holder) | |
662 | { | |
663 | struct request_queue *q; | |
664 | struct block_device *bdev; | |
665 | struct buffer_head *bh; | |
666 | struct btrfs_super_block *disk_super; | |
667 | u64 devid; | |
668 | int ret; | |
669 | ||
670 | if (device->bdev) | |
671 | return -EINVAL; | |
672 | if (!device->name) | |
673 | return -EINVAL; | |
674 | ||
675 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
676 | &bdev, &bh); | |
677 | if (ret) | |
678 | return ret; | |
679 | ||
680 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
681 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
682 | if (devid != device->devid) | |
683 | goto error_brelse; | |
684 | ||
685 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
686 | goto error_brelse; | |
687 | ||
688 | device->generation = btrfs_super_generation(disk_super); | |
689 | ||
690 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
ebbede42 | 691 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
692 | fs_devices->seeding = 1; |
693 | } else { | |
ebbede42 AJ |
694 | if (bdev_read_only(bdev)) |
695 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
696 | else | |
697 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
698 | } |
699 | ||
700 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
701 | if (!blk_queue_nonrot(q)) |
702 | fs_devices->rotating = 1; | |
703 | ||
704 | device->bdev = bdev; | |
e12c9621 | 705 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
706 | device->mode = flags; |
707 | ||
708 | fs_devices->open_devices++; | |
ebbede42 AJ |
709 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
710 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 711 | fs_devices->rw_devices++; |
b1b8e386 | 712 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
713 | } |
714 | brelse(bh); | |
715 | ||
716 | return 0; | |
717 | ||
718 | error_brelse: | |
719 | brelse(bh); | |
720 | blkdev_put(bdev, flags); | |
721 | ||
722 | return -EINVAL; | |
723 | } | |
724 | ||
60999ca4 DS |
725 | /* |
726 | * Add new device to list of registered devices | |
727 | * | |
728 | * Returns: | |
e124ece5 AJ |
729 | * device pointer which was just added or updated when successful |
730 | * error pointer when failed | |
60999ca4 | 731 | */ |
e124ece5 | 732 | static noinline struct btrfs_device *device_list_add(const char *path, |
3acbcbfc | 733 | struct btrfs_super_block *disk_super) |
8a4b83cc CM |
734 | { |
735 | struct btrfs_device *device; | |
736 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 737 | struct rcu_string *name; |
8a4b83cc | 738 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 739 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
8a4b83cc CM |
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)) | |
e124ece5 | 745 | return ERR_CAST(fs_devices); |
2208a378 | 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 | 756 | if (fs_devices->opened) |
e124ece5 | 757 | return ERR_PTR(-EBUSY); |
2b82032c | 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 */ |
e124ece5 | 763 | return device; |
8a4b83cc | 764 | } |
606686ee JB |
765 | |
766 | name = rcu_string_strdup(path, GFP_NOFS); | |
767 | if (!name) { | |
55de4803 | 768 | free_device(device); |
e124ece5 | 769 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 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 | ||
2b82032c | 778 | device->fs_devices = fs_devices; |
d8367db3 | 779 | btrfs_free_stale_devices(path, device); |
327f18cc AJ |
780 | |
781 | if (disk_super->label[0]) | |
782 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
783 | disk_super->label, devid, found_transid, path); | |
784 | else | |
785 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
786 | disk_super->fsid, devid, found_transid, path); | |
787 | ||
606686ee | 788 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
789 | /* |
790 | * When FS is already mounted. | |
791 | * 1. If you are here and if the device->name is NULL that | |
792 | * means this device was missing at time of FS mount. | |
793 | * 2. If you are here and if the device->name is different | |
794 | * from 'path' that means either | |
795 | * a. The same device disappeared and reappeared with | |
796 | * different name. or | |
797 | * b. The missing-disk-which-was-replaced, has | |
798 | * reappeared now. | |
799 | * | |
800 | * We must allow 1 and 2a above. But 2b would be a spurious | |
801 | * and unintentional. | |
802 | * | |
803 | * Further in case of 1 and 2a above, the disk at 'path' | |
804 | * would have missed some transaction when it was away and | |
805 | * in case of 2a the stale bdev has to be updated as well. | |
806 | * 2b must not be allowed at all time. | |
807 | */ | |
808 | ||
809 | /* | |
0f23ae74 CM |
810 | * For now, we do allow update to btrfs_fs_device through the |
811 | * btrfs dev scan cli after FS has been mounted. We're still | |
812 | * tracking a problem where systems fail mount by subvolume id | |
813 | * when we reject replacement on a mounted FS. | |
b96de000 | 814 | */ |
0f23ae74 | 815 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
816 | /* |
817 | * That is if the FS is _not_ mounted and if you | |
818 | * are here, that means there is more than one | |
819 | * disk with same uuid and devid.We keep the one | |
820 | * with larger generation number or the last-in if | |
821 | * generation are equal. | |
822 | */ | |
e124ece5 | 823 | return ERR_PTR(-EEXIST); |
77bdae4d | 824 | } |
b96de000 | 825 | |
606686ee | 826 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc | 827 | if (!name) |
e124ece5 | 828 | return ERR_PTR(-ENOMEM); |
606686ee JB |
829 | rcu_string_free(device->name); |
830 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 831 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 832 | fs_devices->missing_devices--; |
e6e674bd | 833 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 834 | } |
8a4b83cc CM |
835 | } |
836 | ||
77bdae4d AJ |
837 | /* |
838 | * Unmount does not free the btrfs_device struct but would zero | |
839 | * generation along with most of the other members. So just update | |
840 | * it back. We need it to pick the disk with largest generation | |
841 | * (as above). | |
842 | */ | |
843 | if (!fs_devices->opened) | |
844 | device->generation = found_transid; | |
845 | ||
f2788d2f AJ |
846 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
847 | ||
e124ece5 | 848 | return device; |
8a4b83cc CM |
849 | } |
850 | ||
e4404d6e YZ |
851 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
852 | { | |
853 | struct btrfs_fs_devices *fs_devices; | |
854 | struct btrfs_device *device; | |
855 | struct btrfs_device *orig_dev; | |
856 | ||
2208a378 ID |
857 | fs_devices = alloc_fs_devices(orig->fsid); |
858 | if (IS_ERR(fs_devices)) | |
859 | return fs_devices; | |
e4404d6e | 860 | |
adbbb863 | 861 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 862 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 863 | |
46224705 | 864 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 865 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
866 | struct rcu_string *name; |
867 | ||
12bd2fc0 ID |
868 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
869 | orig_dev->uuid); | |
870 | if (IS_ERR(device)) | |
e4404d6e YZ |
871 | goto error; |
872 | ||
606686ee JB |
873 | /* |
874 | * This is ok to do without rcu read locked because we hold the | |
875 | * uuid mutex so nothing we touch in here is going to disappear. | |
876 | */ | |
e755f780 | 877 | if (orig_dev->name) { |
78f2c9e6 DS |
878 | name = rcu_string_strdup(orig_dev->name->str, |
879 | GFP_KERNEL); | |
e755f780 | 880 | if (!name) { |
55de4803 | 881 | free_device(device); |
e755f780 AJ |
882 | goto error; |
883 | } | |
884 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 885 | } |
e4404d6e | 886 | |
e4404d6e YZ |
887 | list_add(&device->dev_list, &fs_devices->devices); |
888 | device->fs_devices = fs_devices; | |
889 | fs_devices->num_devices++; | |
890 | } | |
adbbb863 | 891 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
892 | return fs_devices; |
893 | error: | |
adbbb863 | 894 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
895 | free_fs_devices(fs_devices); |
896 | return ERR_PTR(-ENOMEM); | |
897 | } | |
898 | ||
9b99b115 AJ |
899 | /* |
900 | * After we have read the system tree and know devids belonging to | |
901 | * this filesystem, remove the device which does not belong there. | |
902 | */ | |
903 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 904 | { |
c6e30871 | 905 | struct btrfs_device *device, *next; |
443f24fe | 906 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 907 | |
dfe25020 CM |
908 | mutex_lock(&uuid_mutex); |
909 | again: | |
46224705 | 910 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 911 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
912 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
913 | &device->dev_state)) { | |
401e29c1 AJ |
914 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
915 | &device->dev_state) && | |
916 | (!latest_dev || | |
917 | device->generation > latest_dev->generation)) { | |
443f24fe | 918 | latest_dev = device; |
a6b0d5c8 | 919 | } |
2b82032c | 920 | continue; |
a6b0d5c8 | 921 | } |
2b82032c | 922 | |
8dabb742 SB |
923 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
924 | /* | |
925 | * In the first step, keep the device which has | |
926 | * the correct fsid and the devid that is used | |
927 | * for the dev_replace procedure. | |
928 | * In the second step, the dev_replace state is | |
929 | * read from the device tree and it is known | |
930 | * whether the procedure is really active or | |
931 | * not, which means whether this device is | |
932 | * used or whether it should be removed. | |
933 | */ | |
401e29c1 AJ |
934 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
935 | &device->dev_state)) { | |
8dabb742 SB |
936 | continue; |
937 | } | |
938 | } | |
2b82032c | 939 | if (device->bdev) { |
d4d77629 | 940 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
941 | device->bdev = NULL; |
942 | fs_devices->open_devices--; | |
943 | } | |
ebbede42 | 944 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 945 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 946 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
947 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
948 | &device->dev_state)) | |
8dabb742 | 949 | fs_devices->rw_devices--; |
2b82032c | 950 | } |
e4404d6e YZ |
951 | list_del_init(&device->dev_list); |
952 | fs_devices->num_devices--; | |
55de4803 | 953 | free_device(device); |
dfe25020 | 954 | } |
2b82032c YZ |
955 | |
956 | if (fs_devices->seed) { | |
957 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
958 | goto again; |
959 | } | |
960 | ||
443f24fe | 961 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 962 | |
dfe25020 | 963 | mutex_unlock(&uuid_mutex); |
dfe25020 | 964 | } |
a0af469b | 965 | |
f06c5965 | 966 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
967 | { |
968 | struct btrfs_device *device; | |
969 | ||
9f5316c1 | 970 | device = container_of(head, struct btrfs_device, rcu); |
55de4803 | 971 | free_device(device); |
1f78160c XG |
972 | } |
973 | ||
14238819 AJ |
974 | static void btrfs_close_bdev(struct btrfs_device *device) |
975 | { | |
08ffcae8 DS |
976 | if (!device->bdev) |
977 | return; | |
978 | ||
ebbede42 | 979 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
980 | sync_blockdev(device->bdev); |
981 | invalidate_bdev(device->bdev); | |
982 | } | |
983 | ||
08ffcae8 | 984 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
985 | } |
986 | ||
0ccd0528 | 987 | static void btrfs_prepare_close_one_device(struct btrfs_device *device) |
f448341a AJ |
988 | { |
989 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
990 | struct btrfs_device *new_device; | |
991 | struct rcu_string *name; | |
992 | ||
993 | if (device->bdev) | |
994 | fs_devices->open_devices--; | |
995 | ||
ebbede42 | 996 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
997 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
998 | list_del_init(&device->dev_alloc_list); | |
999 | fs_devices->rw_devices--; | |
1000 | } | |
1001 | ||
e6e674bd | 1002 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1003 | fs_devices->missing_devices--; |
1004 | ||
1005 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
1006 | device->uuid); | |
1007 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1008 | ||
1009 | /* Safe because we are under uuid_mutex */ | |
1010 | if (device->name) { | |
1011 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1012 | BUG_ON(!name); /* -ENOMEM */ | |
1013 | rcu_assign_pointer(new_device->name, name); | |
1014 | } | |
1015 | ||
1016 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1017 | new_device->fs_devices = device->fs_devices; | |
f448341a AJ |
1018 | } |
1019 | ||
2b82032c | 1020 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1021 | { |
2037a093 | 1022 | struct btrfs_device *device, *tmp; |
0ccd0528 AJ |
1023 | struct list_head pending_put; |
1024 | ||
1025 | INIT_LIST_HEAD(&pending_put); | |
e4404d6e | 1026 | |
2b82032c YZ |
1027 | if (--fs_devices->opened > 0) |
1028 | return 0; | |
8a4b83cc | 1029 | |
c9513edb | 1030 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1031 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
0ccd0528 AJ |
1032 | btrfs_prepare_close_one_device(device); |
1033 | list_add(&device->dev_list, &pending_put); | |
8a4b83cc | 1034 | } |
c9513edb XG |
1035 | mutex_unlock(&fs_devices->device_list_mutex); |
1036 | ||
0ccd0528 AJ |
1037 | /* |
1038 | * btrfs_show_devname() is using the device_list_mutex, | |
1039 | * sometimes call to blkdev_put() leads vfs calling | |
1040 | * into this func. So do put outside of device_list_mutex, | |
1041 | * as of now. | |
1042 | */ | |
1043 | while (!list_empty(&pending_put)) { | |
1044 | device = list_first_entry(&pending_put, | |
1045 | struct btrfs_device, dev_list); | |
1046 | list_del(&device->dev_list); | |
1047 | btrfs_close_bdev(device); | |
f06c5965 | 1048 | call_rcu(&device->rcu, free_device_rcu); |
0ccd0528 AJ |
1049 | } |
1050 | ||
e4404d6e YZ |
1051 | WARN_ON(fs_devices->open_devices); |
1052 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1053 | fs_devices->opened = 0; |
1054 | fs_devices->seeding = 0; | |
2b82032c | 1055 | |
8a4b83cc CM |
1056 | return 0; |
1057 | } | |
1058 | ||
2b82032c YZ |
1059 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1060 | { | |
e4404d6e | 1061 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1062 | int ret; |
1063 | ||
1064 | mutex_lock(&uuid_mutex); | |
1065 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
1066 | if (!fs_devices->opened) { |
1067 | seed_devices = fs_devices->seed; | |
1068 | fs_devices->seed = NULL; | |
1069 | } | |
2b82032c | 1070 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1071 | |
1072 | while (seed_devices) { | |
1073 | fs_devices = seed_devices; | |
1074 | seed_devices = fs_devices->seed; | |
1075 | __btrfs_close_devices(fs_devices); | |
1076 | free_fs_devices(fs_devices); | |
1077 | } | |
2b82032c YZ |
1078 | return ret; |
1079 | } | |
1080 | ||
e4404d6e YZ |
1081 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
1082 | fmode_t flags, void *holder) | |
8a4b83cc | 1083 | { |
8a4b83cc | 1084 | struct list_head *head = &fs_devices->devices; |
8a4b83cc | 1085 | struct btrfs_device *device; |
443f24fe | 1086 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1087 | int ret = 0; |
8a4b83cc | 1088 | |
d4d77629 TH |
1089 | flags |= FMODE_EXCL; |
1090 | ||
c6e30871 | 1091 | list_for_each_entry(device, head, dev_list) { |
f63e0cca | 1092 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1093 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1094 | continue; |
a0af469b | 1095 | |
9f050db4 AJ |
1096 | if (!latest_dev || |
1097 | device->generation > latest_dev->generation) | |
1098 | latest_dev = device; | |
8a4b83cc | 1099 | } |
a0af469b | 1100 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1101 | ret = -EINVAL; |
a0af469b CM |
1102 | goto out; |
1103 | } | |
2b82032c | 1104 | fs_devices->opened = 1; |
443f24fe | 1105 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1106 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1107 | out: |
2b82032c YZ |
1108 | return ret; |
1109 | } | |
1110 | ||
f8e10cd3 AJ |
1111 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1112 | { | |
1113 | struct btrfs_device *dev1, *dev2; | |
1114 | ||
1115 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1116 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1117 | ||
1118 | if (dev1->devid < dev2->devid) | |
1119 | return -1; | |
1120 | else if (dev1->devid > dev2->devid) | |
1121 | return 1; | |
1122 | return 0; | |
1123 | } | |
1124 | ||
2b82032c | 1125 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1126 | fmode_t flags, void *holder) |
2b82032c YZ |
1127 | { |
1128 | int ret; | |
1129 | ||
1130 | mutex_lock(&uuid_mutex); | |
1131 | if (fs_devices->opened) { | |
e4404d6e YZ |
1132 | fs_devices->opened++; |
1133 | ret = 0; | |
2b82032c | 1134 | } else { |
f8e10cd3 | 1135 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
15916de8 | 1136 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1137 | } |
8a4b83cc | 1138 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1139 | return ret; |
1140 | } | |
1141 | ||
c9162bdf | 1142 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1143 | { |
1144 | kunmap(page); | |
1145 | put_page(page); | |
1146 | } | |
1147 | ||
c9162bdf OS |
1148 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1149 | struct page **page, | |
1150 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1151 | { |
1152 | void *p; | |
1153 | pgoff_t index; | |
1154 | ||
1155 | /* make sure our super fits in the device */ | |
1156 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1157 | return 1; | |
1158 | ||
1159 | /* make sure our super fits in the page */ | |
1160 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1161 | return 1; | |
1162 | ||
1163 | /* make sure our super doesn't straddle pages on disk */ | |
1164 | index = bytenr >> PAGE_SHIFT; | |
1165 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1166 | return 1; | |
1167 | ||
1168 | /* pull in the page with our super */ | |
1169 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1170 | index, GFP_KERNEL); | |
1171 | ||
1172 | if (IS_ERR_OR_NULL(*page)) | |
1173 | return 1; | |
1174 | ||
1175 | p = kmap(*page); | |
1176 | ||
1177 | /* align our pointer to the offset of the super block */ | |
1178 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1179 | ||
1180 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1181 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1182 | btrfs_release_disk_super(*page); | |
1183 | return 1; | |
1184 | } | |
1185 | ||
1186 | if ((*disk_super)->label[0] && | |
1187 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1188 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1189 | ||
1190 | return 0; | |
1191 | } | |
1192 | ||
6f60cbd3 DS |
1193 | /* |
1194 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1195 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1196 | * is read via pagecache | |
1197 | */ | |
97288f2c | 1198 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1199 | struct btrfs_fs_devices **fs_devices_ret) |
1200 | { | |
1201 | struct btrfs_super_block *disk_super; | |
e124ece5 | 1202 | struct btrfs_device *device; |
8a4b83cc | 1203 | struct block_device *bdev; |
6f60cbd3 | 1204 | struct page *page; |
e124ece5 | 1205 | int ret = 0; |
6f60cbd3 | 1206 | u64 bytenr; |
8a4b83cc | 1207 | |
6f60cbd3 DS |
1208 | /* |
1209 | * we would like to check all the supers, but that would make | |
1210 | * a btrfs mount succeed after a mkfs from a different FS. | |
1211 | * So, we need to add a special mount option to scan for | |
1212 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1213 | */ | |
1214 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1215 | flags |= FMODE_EXCL; |
10f6327b | 1216 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1217 | |
1218 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1219 | if (IS_ERR(bdev)) { |
1220 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1221 | goto error; |
6f60cbd3 DS |
1222 | } |
1223 | ||
05a5c55d AJ |
1224 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
1225 | ret = -EINVAL; | |
6f60cbd3 | 1226 | goto error_bdev_put; |
05a5c55d | 1227 | } |
6f60cbd3 | 1228 | |
3acbcbfc | 1229 | device = device_list_add(path, disk_super); |
e124ece5 AJ |
1230 | if (IS_ERR(device)) |
1231 | ret = PTR_ERR(device); | |
1232 | else | |
1233 | *fs_devices_ret = device->fs_devices; | |
6f60cbd3 | 1234 | |
6cf86a00 | 1235 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1236 | |
1237 | error_bdev_put: | |
d4d77629 | 1238 | blkdev_put(bdev, flags); |
8a4b83cc | 1239 | error: |
beaf8ab3 | 1240 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1241 | return ret; |
1242 | } | |
0b86a832 | 1243 | |
6d07bcec MX |
1244 | /* helper to account the used device space in the range */ |
1245 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1246 | u64 end, u64 *length) | |
1247 | { | |
1248 | struct btrfs_key key; | |
fb456252 | 1249 | struct btrfs_root *root = device->fs_info->dev_root; |
6d07bcec MX |
1250 | struct btrfs_dev_extent *dev_extent; |
1251 | struct btrfs_path *path; | |
1252 | u64 extent_end; | |
1253 | int ret; | |
1254 | int slot; | |
1255 | struct extent_buffer *l; | |
1256 | ||
1257 | *length = 0; | |
1258 | ||
401e29c1 AJ |
1259 | if (start >= device->total_bytes || |
1260 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) | |
6d07bcec MX |
1261 | return 0; |
1262 | ||
1263 | path = btrfs_alloc_path(); | |
1264 | if (!path) | |
1265 | return -ENOMEM; | |
e4058b54 | 1266 | path->reada = READA_FORWARD; |
6d07bcec MX |
1267 | |
1268 | key.objectid = device->devid; | |
1269 | key.offset = start; | |
1270 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1271 | ||
1272 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1273 | if (ret < 0) | |
1274 | goto out; | |
1275 | if (ret > 0) { | |
1276 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1277 | if (ret < 0) | |
1278 | goto out; | |
1279 | } | |
1280 | ||
1281 | while (1) { | |
1282 | l = path->nodes[0]; | |
1283 | slot = path->slots[0]; | |
1284 | if (slot >= btrfs_header_nritems(l)) { | |
1285 | ret = btrfs_next_leaf(root, path); | |
1286 | if (ret == 0) | |
1287 | continue; | |
1288 | if (ret < 0) | |
1289 | goto out; | |
1290 | ||
1291 | break; | |
1292 | } | |
1293 | btrfs_item_key_to_cpu(l, &key, slot); | |
1294 | ||
1295 | if (key.objectid < device->devid) | |
1296 | goto next; | |
1297 | ||
1298 | if (key.objectid > device->devid) | |
1299 | break; | |
1300 | ||
962a298f | 1301 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1302 | goto next; |
1303 | ||
1304 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1305 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1306 | dev_extent); | |
1307 | if (key.offset <= start && extent_end > end) { | |
1308 | *length = end - start + 1; | |
1309 | break; | |
1310 | } else if (key.offset <= start && extent_end > start) | |
1311 | *length += extent_end - start; | |
1312 | else if (key.offset > start && extent_end <= end) | |
1313 | *length += extent_end - key.offset; | |
1314 | else if (key.offset > start && key.offset <= end) { | |
1315 | *length += end - key.offset + 1; | |
1316 | break; | |
1317 | } else if (key.offset > end) | |
1318 | break; | |
1319 | ||
1320 | next: | |
1321 | path->slots[0]++; | |
1322 | } | |
1323 | ret = 0; | |
1324 | out: | |
1325 | btrfs_free_path(path); | |
1326 | return ret; | |
1327 | } | |
1328 | ||
499f377f | 1329 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1330 | struct btrfs_device *device, |
1331 | u64 *start, u64 len) | |
1332 | { | |
fb456252 | 1333 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1334 | struct extent_map *em; |
499f377f | 1335 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1336 | int ret = 0; |
1b984508 | 1337 | u64 physical_start = *start; |
6df9a95e | 1338 | |
499f377f JM |
1339 | if (transaction) |
1340 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1341 | again: |
1342 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1343 | struct map_lookup *map; |
1344 | int i; | |
1345 | ||
95617d69 | 1346 | map = em->map_lookup; |
6df9a95e | 1347 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1348 | u64 end; |
1349 | ||
6df9a95e JB |
1350 | if (map->stripes[i].dev != device) |
1351 | continue; | |
1b984508 | 1352 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1353 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1354 | physical_start) |
6df9a95e | 1355 | continue; |
c152b63e FM |
1356 | /* |
1357 | * Make sure that while processing the pinned list we do | |
1358 | * not override our *start with a lower value, because | |
1359 | * we can have pinned chunks that fall within this | |
1360 | * device hole and that have lower physical addresses | |
1361 | * than the pending chunks we processed before. If we | |
1362 | * do not take this special care we can end up getting | |
1363 | * 2 pending chunks that start at the same physical | |
1364 | * device offsets because the end offset of a pinned | |
1365 | * chunk can be equal to the start offset of some | |
1366 | * pending chunk. | |
1367 | */ | |
1368 | end = map->stripes[i].physical + em->orig_block_len; | |
1369 | if (end > *start) { | |
1370 | *start = end; | |
1371 | ret = 1; | |
1372 | } | |
6df9a95e JB |
1373 | } |
1374 | } | |
499f377f JM |
1375 | if (search_list != &fs_info->pinned_chunks) { |
1376 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1377 | goto again; |
1378 | } | |
6df9a95e JB |
1379 | |
1380 | return ret; | |
1381 | } | |
1382 | ||
1383 | ||
0b86a832 | 1384 | /* |
499f377f JM |
1385 | * find_free_dev_extent_start - find free space in the specified device |
1386 | * @device: the device which we search the free space in | |
1387 | * @num_bytes: the size of the free space that we need | |
1388 | * @search_start: the position from which to begin the search | |
1389 | * @start: store the start of the free space. | |
1390 | * @len: the size of the free space. that we find, or the size | |
1391 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1392 | * |
0b86a832 CM |
1393 | * this uses a pretty simple search, the expectation is that it is |
1394 | * called very infrequently and that a given device has a small number | |
1395 | * of extents | |
7bfc837d MX |
1396 | * |
1397 | * @start is used to store the start of the free space if we find. But if we | |
1398 | * don't find suitable free space, it will be used to store the start position | |
1399 | * of the max free space. | |
1400 | * | |
1401 | * @len is used to store the size of the free space that we find. | |
1402 | * But if we don't find suitable free space, it is used to store the size of | |
1403 | * the max free space. | |
0b86a832 | 1404 | */ |
499f377f JM |
1405 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1406 | struct btrfs_device *device, u64 num_bytes, | |
1407 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1408 | { |
0b246afa JM |
1409 | struct btrfs_fs_info *fs_info = device->fs_info; |
1410 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1411 | struct btrfs_key key; |
7bfc837d | 1412 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1413 | struct btrfs_path *path; |
7bfc837d MX |
1414 | u64 hole_size; |
1415 | u64 max_hole_start; | |
1416 | u64 max_hole_size; | |
1417 | u64 extent_end; | |
0b86a832 CM |
1418 | u64 search_end = device->total_bytes; |
1419 | int ret; | |
7bfc837d | 1420 | int slot; |
0b86a832 | 1421 | struct extent_buffer *l; |
8cdc7c5b FM |
1422 | |
1423 | /* | |
1424 | * We don't want to overwrite the superblock on the drive nor any area | |
1425 | * used by the boot loader (grub for example), so we make sure to start | |
1426 | * at an offset of at least 1MB. | |
1427 | */ | |
0d0c71b3 | 1428 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1429 | |
6df9a95e JB |
1430 | path = btrfs_alloc_path(); |
1431 | if (!path) | |
1432 | return -ENOMEM; | |
f2ab7618 | 1433 | |
7bfc837d MX |
1434 | max_hole_start = search_start; |
1435 | max_hole_size = 0; | |
1436 | ||
f2ab7618 | 1437 | again: |
401e29c1 AJ |
1438 | if (search_start >= search_end || |
1439 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1440 | ret = -ENOSPC; |
6df9a95e | 1441 | goto out; |
7bfc837d MX |
1442 | } |
1443 | ||
e4058b54 | 1444 | path->reada = READA_FORWARD; |
6df9a95e JB |
1445 | path->search_commit_root = 1; |
1446 | path->skip_locking = 1; | |
7bfc837d | 1447 | |
0b86a832 CM |
1448 | key.objectid = device->devid; |
1449 | key.offset = search_start; | |
1450 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1451 | |
125ccb0a | 1452 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1453 | if (ret < 0) |
7bfc837d | 1454 | goto out; |
1fcbac58 YZ |
1455 | if (ret > 0) { |
1456 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1457 | if (ret < 0) | |
7bfc837d | 1458 | goto out; |
1fcbac58 | 1459 | } |
7bfc837d | 1460 | |
0b86a832 CM |
1461 | while (1) { |
1462 | l = path->nodes[0]; | |
1463 | slot = path->slots[0]; | |
1464 | if (slot >= btrfs_header_nritems(l)) { | |
1465 | ret = btrfs_next_leaf(root, path); | |
1466 | if (ret == 0) | |
1467 | continue; | |
1468 | if (ret < 0) | |
7bfc837d MX |
1469 | goto out; |
1470 | ||
1471 | break; | |
0b86a832 CM |
1472 | } |
1473 | btrfs_item_key_to_cpu(l, &key, slot); | |
1474 | ||
1475 | if (key.objectid < device->devid) | |
1476 | goto next; | |
1477 | ||
1478 | if (key.objectid > device->devid) | |
7bfc837d | 1479 | break; |
0b86a832 | 1480 | |
962a298f | 1481 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1482 | goto next; |
9779b72f | 1483 | |
7bfc837d MX |
1484 | if (key.offset > search_start) { |
1485 | hole_size = key.offset - search_start; | |
9779b72f | 1486 | |
6df9a95e JB |
1487 | /* |
1488 | * Have to check before we set max_hole_start, otherwise | |
1489 | * we could end up sending back this offset anyway. | |
1490 | */ | |
499f377f | 1491 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1492 | &search_start, |
1b984508 FL |
1493 | hole_size)) { |
1494 | if (key.offset >= search_start) { | |
1495 | hole_size = key.offset - search_start; | |
1496 | } else { | |
1497 | WARN_ON_ONCE(1); | |
1498 | hole_size = 0; | |
1499 | } | |
1500 | } | |
6df9a95e | 1501 | |
7bfc837d MX |
1502 | if (hole_size > max_hole_size) { |
1503 | max_hole_start = search_start; | |
1504 | max_hole_size = hole_size; | |
1505 | } | |
9779b72f | 1506 | |
7bfc837d MX |
1507 | /* |
1508 | * If this free space is greater than which we need, | |
1509 | * it must be the max free space that we have found | |
1510 | * until now, so max_hole_start must point to the start | |
1511 | * of this free space and the length of this free space | |
1512 | * is stored in max_hole_size. Thus, we return | |
1513 | * max_hole_start and max_hole_size and go back to the | |
1514 | * caller. | |
1515 | */ | |
1516 | if (hole_size >= num_bytes) { | |
1517 | ret = 0; | |
1518 | goto out; | |
0b86a832 CM |
1519 | } |
1520 | } | |
0b86a832 | 1521 | |
0b86a832 | 1522 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1523 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1524 | dev_extent); | |
1525 | if (extent_end > search_start) | |
1526 | search_start = extent_end; | |
0b86a832 CM |
1527 | next: |
1528 | path->slots[0]++; | |
1529 | cond_resched(); | |
1530 | } | |
0b86a832 | 1531 | |
38c01b96 | 1532 | /* |
1533 | * At this point, search_start should be the end of | |
1534 | * allocated dev extents, and when shrinking the device, | |
1535 | * search_end may be smaller than search_start. | |
1536 | */ | |
f2ab7618 | 1537 | if (search_end > search_start) { |
38c01b96 | 1538 | hole_size = search_end - search_start; |
1539 | ||
499f377f | 1540 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1541 | hole_size)) { |
1542 | btrfs_release_path(path); | |
1543 | goto again; | |
1544 | } | |
0b86a832 | 1545 | |
f2ab7618 ZL |
1546 | if (hole_size > max_hole_size) { |
1547 | max_hole_start = search_start; | |
1548 | max_hole_size = hole_size; | |
1549 | } | |
6df9a95e JB |
1550 | } |
1551 | ||
7bfc837d | 1552 | /* See above. */ |
f2ab7618 | 1553 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1554 | ret = -ENOSPC; |
1555 | else | |
1556 | ret = 0; | |
1557 | ||
1558 | out: | |
2b82032c | 1559 | btrfs_free_path(path); |
7bfc837d | 1560 | *start = max_hole_start; |
b2117a39 | 1561 | if (len) |
7bfc837d | 1562 | *len = max_hole_size; |
0b86a832 CM |
1563 | return ret; |
1564 | } | |
1565 | ||
499f377f JM |
1566 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1567 | struct btrfs_device *device, u64 num_bytes, | |
1568 | u64 *start, u64 *len) | |
1569 | { | |
499f377f | 1570 | /* FIXME use last free of some kind */ |
499f377f | 1571 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1572 | num_bytes, 0, start, len); |
499f377f JM |
1573 | } |
1574 | ||
b2950863 | 1575 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1576 | struct btrfs_device *device, |
2196d6e8 | 1577 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1578 | { |
0b246afa JM |
1579 | struct btrfs_fs_info *fs_info = device->fs_info; |
1580 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1581 | int ret; |
1582 | struct btrfs_path *path; | |
8f18cf13 | 1583 | struct btrfs_key key; |
a061fc8d CM |
1584 | struct btrfs_key found_key; |
1585 | struct extent_buffer *leaf = NULL; | |
1586 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1587 | |
1588 | path = btrfs_alloc_path(); | |
1589 | if (!path) | |
1590 | return -ENOMEM; | |
1591 | ||
1592 | key.objectid = device->devid; | |
1593 | key.offset = start; | |
1594 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1595 | again: |
8f18cf13 | 1596 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1597 | if (ret > 0) { |
1598 | ret = btrfs_previous_item(root, path, key.objectid, | |
1599 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1600 | if (ret) |
1601 | goto out; | |
a061fc8d CM |
1602 | leaf = path->nodes[0]; |
1603 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1604 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1605 | struct btrfs_dev_extent); | |
1606 | BUG_ON(found_key.offset > start || found_key.offset + | |
1607 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1608 | key = found_key; |
1609 | btrfs_release_path(path); | |
1610 | goto again; | |
a061fc8d CM |
1611 | } else if (ret == 0) { |
1612 | leaf = path->nodes[0]; | |
1613 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1614 | struct btrfs_dev_extent); | |
79787eaa | 1615 | } else { |
0b246afa | 1616 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1617 | goto out; |
a061fc8d | 1618 | } |
8f18cf13 | 1619 | |
2196d6e8 MX |
1620 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1621 | ||
8f18cf13 | 1622 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1623 | if (ret) { |
0b246afa JM |
1624 | btrfs_handle_fs_error(fs_info, ret, |
1625 | "Failed to remove dev extent item"); | |
13212b54 | 1626 | } else { |
3204d33c | 1627 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1628 | } |
b0b802d7 | 1629 | out: |
8f18cf13 CM |
1630 | btrfs_free_path(path); |
1631 | return ret; | |
1632 | } | |
1633 | ||
48a3b636 ES |
1634 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1635 | struct btrfs_device *device, | |
48a3b636 | 1636 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1637 | { |
1638 | int ret; | |
1639 | struct btrfs_path *path; | |
0b246afa JM |
1640 | struct btrfs_fs_info *fs_info = device->fs_info; |
1641 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1642 | struct btrfs_dev_extent *extent; |
1643 | struct extent_buffer *leaf; | |
1644 | struct btrfs_key key; | |
1645 | ||
e12c9621 | 1646 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1647 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1648 | path = btrfs_alloc_path(); |
1649 | if (!path) | |
1650 | return -ENOMEM; | |
1651 | ||
0b86a832 | 1652 | key.objectid = device->devid; |
2b82032c | 1653 | key.offset = start; |
0b86a832 CM |
1654 | key.type = BTRFS_DEV_EXTENT_KEY; |
1655 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1656 | sizeof(*extent)); | |
2cdcecbc MF |
1657 | if (ret) |
1658 | goto out; | |
0b86a832 CM |
1659 | |
1660 | leaf = path->nodes[0]; | |
1661 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1662 | struct btrfs_dev_extent); | |
b5d9071c NB |
1663 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1664 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1665 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1666 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1667 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1668 | ||
0b86a832 CM |
1669 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1670 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1671 | out: |
0b86a832 CM |
1672 | btrfs_free_path(path); |
1673 | return ret; | |
1674 | } | |
1675 | ||
6df9a95e | 1676 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1677 | { |
6df9a95e JB |
1678 | struct extent_map_tree *em_tree; |
1679 | struct extent_map *em; | |
1680 | struct rb_node *n; | |
1681 | u64 ret = 0; | |
0b86a832 | 1682 | |
6df9a95e JB |
1683 | em_tree = &fs_info->mapping_tree.map_tree; |
1684 | read_lock(&em_tree->lock); | |
1685 | n = rb_last(&em_tree->map); | |
1686 | if (n) { | |
1687 | em = rb_entry(n, struct extent_map, rb_node); | |
1688 | ret = em->start + em->len; | |
0b86a832 | 1689 | } |
6df9a95e JB |
1690 | read_unlock(&em_tree->lock); |
1691 | ||
0b86a832 CM |
1692 | return ret; |
1693 | } | |
1694 | ||
53f10659 ID |
1695 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1696 | u64 *devid_ret) | |
0b86a832 CM |
1697 | { |
1698 | int ret; | |
1699 | struct btrfs_key key; | |
1700 | struct btrfs_key found_key; | |
2b82032c YZ |
1701 | struct btrfs_path *path; |
1702 | ||
2b82032c YZ |
1703 | path = btrfs_alloc_path(); |
1704 | if (!path) | |
1705 | return -ENOMEM; | |
0b86a832 CM |
1706 | |
1707 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1708 | key.type = BTRFS_DEV_ITEM_KEY; | |
1709 | key.offset = (u64)-1; | |
1710 | ||
53f10659 | 1711 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1712 | if (ret < 0) |
1713 | goto error; | |
1714 | ||
79787eaa | 1715 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1716 | |
53f10659 ID |
1717 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1718 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1719 | BTRFS_DEV_ITEM_KEY); |
1720 | if (ret) { | |
53f10659 | 1721 | *devid_ret = 1; |
0b86a832 CM |
1722 | } else { |
1723 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1724 | path->slots[0]); | |
53f10659 | 1725 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1726 | } |
1727 | ret = 0; | |
1728 | error: | |
2b82032c | 1729 | btrfs_free_path(path); |
0b86a832 CM |
1730 | return ret; |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * the device information is stored in the chunk root | |
1735 | * the btrfs_device struct should be fully filled in | |
1736 | */ | |
c74a0b02 | 1737 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
5b4aacef | 1738 | struct btrfs_fs_info *fs_info, |
48a3b636 | 1739 | struct btrfs_device *device) |
0b86a832 | 1740 | { |
5b4aacef | 1741 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
1742 | int ret; |
1743 | struct btrfs_path *path; | |
1744 | struct btrfs_dev_item *dev_item; | |
1745 | struct extent_buffer *leaf; | |
1746 | struct btrfs_key key; | |
1747 | unsigned long ptr; | |
0b86a832 | 1748 | |
0b86a832 CM |
1749 | path = btrfs_alloc_path(); |
1750 | if (!path) | |
1751 | return -ENOMEM; | |
1752 | ||
0b86a832 CM |
1753 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1754 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1755 | key.offset = device->devid; |
0b86a832 CM |
1756 | |
1757 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1758 | sizeof(*dev_item)); |
0b86a832 CM |
1759 | if (ret) |
1760 | goto out; | |
1761 | ||
1762 | leaf = path->nodes[0]; | |
1763 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1764 | ||
1765 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1766 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1767 | btrfs_set_device_type(leaf, dev_item, device->type); |
1768 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1769 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1770 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1771 | btrfs_set_device_total_bytes(leaf, dev_item, |
1772 | btrfs_device_get_disk_total_bytes(device)); | |
1773 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1774 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1775 | btrfs_set_device_group(leaf, dev_item, 0); |
1776 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1777 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1778 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1779 | |
410ba3a2 | 1780 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1781 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1782 | ptr = btrfs_device_fsid(dev_item); |
44880fdc | 1783 | write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE); |
0b86a832 | 1784 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1785 | |
2b82032c | 1786 | ret = 0; |
0b86a832 CM |
1787 | out: |
1788 | btrfs_free_path(path); | |
1789 | return ret; | |
1790 | } | |
8f18cf13 | 1791 | |
5a1972bd QW |
1792 | /* |
1793 | * Function to update ctime/mtime for a given device path. | |
1794 | * Mainly used for ctime/mtime based probe like libblkid. | |
1795 | */ | |
da353f6b | 1796 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1797 | { |
1798 | struct file *filp; | |
1799 | ||
1800 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1801 | if (IS_ERR(filp)) |
5a1972bd QW |
1802 | return; |
1803 | file_update_time(filp); | |
1804 | filp_close(filp, NULL); | |
5a1972bd QW |
1805 | } |
1806 | ||
5b4aacef | 1807 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1808 | struct btrfs_device *device) |
1809 | { | |
5b4aacef | 1810 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1811 | int ret; |
1812 | struct btrfs_path *path; | |
a061fc8d | 1813 | struct btrfs_key key; |
a061fc8d CM |
1814 | struct btrfs_trans_handle *trans; |
1815 | ||
a061fc8d CM |
1816 | path = btrfs_alloc_path(); |
1817 | if (!path) | |
1818 | return -ENOMEM; | |
1819 | ||
a22285a6 | 1820 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1821 | if (IS_ERR(trans)) { |
1822 | btrfs_free_path(path); | |
1823 | return PTR_ERR(trans); | |
1824 | } | |
a061fc8d CM |
1825 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1826 | key.type = BTRFS_DEV_ITEM_KEY; | |
1827 | key.offset = device->devid; | |
1828 | ||
1829 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1830 | if (ret) { |
1831 | if (ret > 0) | |
1832 | ret = -ENOENT; | |
1833 | btrfs_abort_transaction(trans, ret); | |
1834 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1835 | goto out; |
1836 | } | |
1837 | ||
1838 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1839 | if (ret) { |
1840 | btrfs_abort_transaction(trans, ret); | |
1841 | btrfs_end_transaction(trans); | |
1842 | } | |
1843 | ||
a061fc8d CM |
1844 | out: |
1845 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1846 | if (!ret) |
1847 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1848 | return ret; |
1849 | } | |
1850 | ||
3cc31a0d DS |
1851 | /* |
1852 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1853 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1854 | * replace. | |
1855 | */ | |
1856 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1857 | u64 num_devices) | |
a061fc8d | 1858 | { |
a061fc8d | 1859 | u64 all_avail; |
de98ced9 | 1860 | unsigned seq; |
418775a2 | 1861 | int i; |
a061fc8d | 1862 | |
de98ced9 | 1863 | do { |
bd45ffbc | 1864 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1865 | |
bd45ffbc AJ |
1866 | all_avail = fs_info->avail_data_alloc_bits | |
1867 | fs_info->avail_system_alloc_bits | | |
1868 | fs_info->avail_metadata_alloc_bits; | |
1869 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1870 | |
418775a2 DS |
1871 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1872 | if (!(all_avail & btrfs_raid_group[i])) | |
1873 | continue; | |
a061fc8d | 1874 | |
418775a2 DS |
1875 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1876 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1877 | |
418775a2 DS |
1878 | if (ret) |
1879 | return ret; | |
1880 | } | |
53b381b3 DW |
1881 | } |
1882 | ||
bd45ffbc | 1883 | return 0; |
f1fa7f26 AJ |
1884 | } |
1885 | ||
c9162bdf OS |
1886 | static struct btrfs_device * btrfs_find_next_active_device( |
1887 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1888 | { |
2b82032c | 1889 | struct btrfs_device *next_device; |
88acff64 AJ |
1890 | |
1891 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1892 | if (next_device != device && | |
e6e674bd AJ |
1893 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1894 | && next_device->bdev) | |
88acff64 AJ |
1895 | return next_device; |
1896 | } | |
1897 | ||
1898 | return NULL; | |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1903 | * and replace it with the provided or the next active device, in the context | |
1904 | * where this function called, there should be always be another device (or | |
1905 | * this_dev) which is active. | |
1906 | */ | |
1907 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1908 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1909 | { | |
1910 | struct btrfs_device *next_device; | |
1911 | ||
1912 | if (this_dev) | |
1913 | next_device = this_dev; | |
1914 | else | |
1915 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1916 | device); | |
1917 | ASSERT(next_device); | |
1918 | ||
1919 | if (fs_info->sb->s_bdev && | |
1920 | (fs_info->sb->s_bdev == device->bdev)) | |
1921 | fs_info->sb->s_bdev = next_device->bdev; | |
1922 | ||
1923 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1924 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1925 | } | |
1926 | ||
da353f6b DS |
1927 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1928 | u64 devid) | |
f1fa7f26 AJ |
1929 | { |
1930 | struct btrfs_device *device; | |
1f78160c | 1931 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1932 | u64 num_devices; |
a061fc8d CM |
1933 | int ret = 0; |
1934 | ||
2c997384 | 1935 | mutex_lock(&fs_info->volume_mutex); |
a061fc8d CM |
1936 | mutex_lock(&uuid_mutex); |
1937 | ||
0b246afa | 1938 | num_devices = fs_info->fs_devices->num_devices; |
7e79cb86 | 1939 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
0b246afa | 1940 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
8dabb742 SB |
1941 | WARN_ON(num_devices < 1); |
1942 | num_devices--; | |
1943 | } | |
7e79cb86 | 1944 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
8dabb742 | 1945 | |
0b246afa | 1946 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1947 | if (ret) |
a061fc8d | 1948 | goto out; |
a061fc8d | 1949 | |
2ff7e61e JM |
1950 | ret = btrfs_find_device_by_devspec(fs_info, devid, device_path, |
1951 | &device); | |
24fc572f | 1952 | if (ret) |
53b381b3 | 1953 | goto out; |
dfe25020 | 1954 | |
401e29c1 | 1955 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1956 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1957 | goto out; |
63a212ab SB |
1958 | } |
1959 | ||
ebbede42 AJ |
1960 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1961 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1962 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1963 | goto out; |
2b82032c YZ |
1964 | } |
1965 | ||
ebbede42 | 1966 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1967 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1968 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1969 | device->fs_devices->rw_devices--; |
34441361 | 1970 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 1971 | } |
a061fc8d | 1972 | |
d7901554 | 1973 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1974 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1975 | mutex_lock(&uuid_mutex); |
a061fc8d | 1976 | if (ret) |
9b3517e9 | 1977 | goto error_undo; |
a061fc8d | 1978 | |
63a212ab SB |
1979 | /* |
1980 | * TODO: the superblock still includes this device in its num_devices | |
1981 | * counter although write_all_supers() is not locked out. This | |
1982 | * could give a filesystem state which requires a degraded mount. | |
1983 | */ | |
0b246afa | 1984 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1985 | if (ret) |
9b3517e9 | 1986 | goto error_undo; |
a061fc8d | 1987 | |
e12c9621 | 1988 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 1989 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1990 | |
1991 | /* | |
1992 | * the device list mutex makes sure that we don't change | |
1993 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1994 | * the device supers. Whoever is writing all supers, should |
1995 | * lock the device list mutex before getting the number of | |
1996 | * devices in the super block (super_copy). Conversely, | |
1997 | * whoever updates the number of devices in the super block | |
1998 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1999 | */ |
1f78160c XG |
2000 | |
2001 | cur_devices = device->fs_devices; | |
0b246afa | 2002 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c | 2003 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2004 | |
e4404d6e | 2005 | device->fs_devices->num_devices--; |
02db0844 | 2006 | device->fs_devices->total_devices--; |
2b82032c | 2007 | |
e6e674bd | 2008 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
3a7d55c8 | 2009 | device->fs_devices->missing_devices--; |
cd02dca5 | 2010 | |
0b246afa | 2011 | btrfs_assign_next_active_device(fs_info, device, NULL); |
2b82032c | 2012 | |
0bfaa9c5 | 2013 | if (device->bdev) { |
e4404d6e | 2014 | device->fs_devices->open_devices--; |
0bfaa9c5 | 2015 | /* remove sysfs entry */ |
0b246afa | 2016 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
0bfaa9c5 | 2017 | } |
99994cde | 2018 | |
0b246afa JM |
2019 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2020 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
2021 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2022 | |
cea67ab9 JM |
2023 | /* |
2024 | * at this point, the device is zero sized and detached from | |
2025 | * the devices list. All that's left is to zero out the old | |
2026 | * supers and free the device. | |
2027 | */ | |
ebbede42 | 2028 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2029 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2030 | ||
2031 | btrfs_close_bdev(device); | |
f06c5965 | 2032 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2033 | |
1f78160c | 2034 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2035 | struct btrfs_fs_devices *fs_devices; |
0b246afa | 2036 | fs_devices = fs_info->fs_devices; |
e4404d6e | 2037 | while (fs_devices) { |
8321cf25 RS |
2038 | if (fs_devices->seed == cur_devices) { |
2039 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2040 | break; |
8321cf25 | 2041 | } |
e4404d6e | 2042 | fs_devices = fs_devices->seed; |
2b82032c | 2043 | } |
1f78160c | 2044 | cur_devices->seed = NULL; |
1f78160c | 2045 | __btrfs_close_devices(cur_devices); |
1f78160c | 2046 | free_fs_devices(cur_devices); |
2b82032c YZ |
2047 | } |
2048 | ||
a061fc8d CM |
2049 | out: |
2050 | mutex_unlock(&uuid_mutex); | |
2c997384 | 2051 | mutex_unlock(&fs_info->volume_mutex); |
a061fc8d | 2052 | return ret; |
24fc572f | 2053 | |
9b3517e9 | 2054 | error_undo: |
ebbede42 | 2055 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2056 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2057 | list_add(&device->dev_alloc_list, |
0b246afa | 2058 | &fs_info->fs_devices->alloc_list); |
c3929c36 | 2059 | device->fs_devices->rw_devices++; |
34441361 | 2060 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2061 | } |
24fc572f | 2062 | goto out; |
a061fc8d CM |
2063 | } |
2064 | ||
084b6e7c QW |
2065 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
2066 | struct btrfs_device *srcdev) | |
e93c89c1 | 2067 | { |
d51908ce AJ |
2068 | struct btrfs_fs_devices *fs_devices; |
2069 | ||
a32bf9a3 | 2070 | lockdep_assert_held(&fs_info->fs_devices->device_list_mutex); |
1357272f | 2071 | |
25e8e911 AJ |
2072 | /* |
2073 | * in case of fs with no seed, srcdev->fs_devices will point | |
2074 | * to fs_devices of fs_info. However when the dev being replaced is | |
2075 | * a seed dev it will point to the seed's local fs_devices. In short | |
2076 | * srcdev will have its correct fs_devices in both the cases. | |
2077 | */ | |
2078 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2079 | |
e93c89c1 | 2080 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2081 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2082 | fs_devices->num_devices--; |
e6e674bd | 2083 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2084 | fs_devices->missing_devices--; |
e93c89c1 | 2085 | |
ebbede42 | 2086 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2087 | fs_devices->rw_devices--; |
1357272f | 2088 | |
82372bc8 | 2089 | if (srcdev->bdev) |
d51908ce | 2090 | fs_devices->open_devices--; |
084b6e7c QW |
2091 | } |
2092 | ||
2093 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2094 | struct btrfs_device *srcdev) | |
2095 | { | |
2096 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2097 | |
ebbede42 | 2098 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2099 | /* zero out the old super if it is writable */ |
2100 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2101 | } | |
14238819 AJ |
2102 | |
2103 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2104 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2105 | |
94d5f0c2 AJ |
2106 | /* if this is no devs we rather delete the fs_devices */ |
2107 | if (!fs_devices->num_devices) { | |
2108 | struct btrfs_fs_devices *tmp_fs_devices; | |
2109 | ||
6dd38f81 AJ |
2110 | /* |
2111 | * On a mounted FS, num_devices can't be zero unless it's a | |
2112 | * seed. In case of a seed device being replaced, the replace | |
2113 | * target added to the sprout FS, so there will be no more | |
2114 | * device left under the seed FS. | |
2115 | */ | |
2116 | ASSERT(fs_devices->seeding); | |
2117 | ||
94d5f0c2 AJ |
2118 | tmp_fs_devices = fs_info->fs_devices; |
2119 | while (tmp_fs_devices) { | |
2120 | if (tmp_fs_devices->seed == fs_devices) { | |
2121 | tmp_fs_devices->seed = fs_devices->seed; | |
2122 | break; | |
2123 | } | |
2124 | tmp_fs_devices = tmp_fs_devices->seed; | |
2125 | } | |
2126 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2127 | __btrfs_close_devices(fs_devices); |
2128 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2129 | } |
e93c89c1 SB |
2130 | } |
2131 | ||
2132 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2133 | struct btrfs_device *tgtdev) | |
2134 | { | |
67a2c45e | 2135 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2136 | WARN_ON(!tgtdev); |
2137 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2138 | |
32576040 | 2139 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2140 | |
779bf3fe | 2141 | if (tgtdev->bdev) |
e93c89c1 | 2142 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2143 | |
e93c89c1 | 2144 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2145 | |
88acff64 | 2146 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2147 | |
e93c89c1 | 2148 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2149 | |
2150 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2151 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2152 | |
2153 | /* | |
2154 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2155 | * may lead to a call to btrfs_show_devname() which will try | |
2156 | * to hold device_list_mutex. And here this device | |
2157 | * is already out of device list, so we don't have to hold | |
2158 | * the device_list_mutex lock. | |
2159 | */ | |
2160 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2161 | |
2162 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2163 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2164 | } |
2165 | ||
2ff7e61e | 2166 | static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2167 | const char *device_path, |
48a3b636 | 2168 | struct btrfs_device **device) |
7ba15b7d SB |
2169 | { |
2170 | int ret = 0; | |
2171 | struct btrfs_super_block *disk_super; | |
2172 | u64 devid; | |
2173 | u8 *dev_uuid; | |
2174 | struct block_device *bdev; | |
2175 | struct buffer_head *bh; | |
2176 | ||
2177 | *device = NULL; | |
2178 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2179 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2180 | if (ret) |
2181 | return ret; | |
2182 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2183 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2184 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2185 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2186 | brelse(bh); |
2187 | if (!*device) | |
2188 | ret = -ENOENT; | |
2189 | blkdev_put(bdev, FMODE_READ); | |
2190 | return ret; | |
2191 | } | |
2192 | ||
2ff7e61e | 2193 | int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2194 | const char *device_path, |
7ba15b7d SB |
2195 | struct btrfs_device **device) |
2196 | { | |
2197 | *device = NULL; | |
2198 | if (strcmp(device_path, "missing") == 0) { | |
2199 | struct list_head *devices; | |
2200 | struct btrfs_device *tmp; | |
2201 | ||
0b246afa | 2202 | devices = &fs_info->fs_devices->devices; |
7ba15b7d SB |
2203 | /* |
2204 | * It is safe to read the devices since the volume_mutex | |
2205 | * is held by the caller. | |
2206 | */ | |
2207 | list_for_each_entry(tmp, devices, dev_list) { | |
e12c9621 AJ |
2208 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2209 | &tmp->dev_state) && !tmp->bdev) { | |
7ba15b7d SB |
2210 | *device = tmp; |
2211 | break; | |
2212 | } | |
2213 | } | |
2214 | ||
d74a6259 AJ |
2215 | if (!*device) |
2216 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2217 | |
2218 | return 0; | |
2219 | } else { | |
2ff7e61e | 2220 | return btrfs_find_device_by_path(fs_info, device_path, device); |
7ba15b7d SB |
2221 | } |
2222 | } | |
2223 | ||
5c5c0df0 DS |
2224 | /* |
2225 | * Lookup a device given by device id, or the path if the id is 0. | |
2226 | */ | |
2ff7e61e | 2227 | int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid, |
da353f6b DS |
2228 | const char *devpath, |
2229 | struct btrfs_device **device) | |
24e0474b AJ |
2230 | { |
2231 | int ret; | |
2232 | ||
5c5c0df0 | 2233 | if (devid) { |
24e0474b | 2234 | ret = 0; |
0b246afa | 2235 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2236 | if (!*device) |
2237 | ret = -ENOENT; | |
2238 | } else { | |
5c5c0df0 | 2239 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2240 | return -EINVAL; |
2241 | ||
2ff7e61e | 2242 | ret = btrfs_find_device_missing_or_by_path(fs_info, devpath, |
24e0474b AJ |
2243 | device); |
2244 | } | |
2245 | return ret; | |
2246 | } | |
2247 | ||
2b82032c YZ |
2248 | /* |
2249 | * does all the dirty work required for changing file system's UUID. | |
2250 | */ | |
2ff7e61e | 2251 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2252 | { |
0b246afa | 2253 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2254 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2255 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2256 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2257 | struct btrfs_device *device; |
2258 | u64 super_flags; | |
2259 | ||
a32bf9a3 | 2260 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2261 | if (!fs_devices->seeding) |
2b82032c YZ |
2262 | return -EINVAL; |
2263 | ||
2dfeca9b | 2264 | seed_devices = alloc_fs_devices(NULL); |
2208a378 ID |
2265 | if (IS_ERR(seed_devices)) |
2266 | return PTR_ERR(seed_devices); | |
2b82032c | 2267 | |
e4404d6e YZ |
2268 | old_devices = clone_fs_devices(fs_devices); |
2269 | if (IS_ERR(old_devices)) { | |
2270 | kfree(seed_devices); | |
2271 | return PTR_ERR(old_devices); | |
2b82032c | 2272 | } |
e4404d6e | 2273 | |
2b82032c YZ |
2274 | list_add(&old_devices->list, &fs_uuids); |
2275 | ||
e4404d6e YZ |
2276 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2277 | seed_devices->opened = 1; | |
2278 | INIT_LIST_HEAD(&seed_devices->devices); | |
2279 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2280 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2281 | |
0b246afa | 2282 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c XG |
2283 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2284 | synchronize_rcu); | |
2196d6e8 MX |
2285 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2286 | device->fs_devices = seed_devices; | |
c9513edb | 2287 | |
34441361 | 2288 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2289 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2290 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2291 | |
2b82032c YZ |
2292 | fs_devices->seeding = 0; |
2293 | fs_devices->num_devices = 0; | |
2294 | fs_devices->open_devices = 0; | |
69611ac8 | 2295 | fs_devices->missing_devices = 0; |
69611ac8 | 2296 | fs_devices->rotating = 0; |
e4404d6e | 2297 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2298 | |
2299 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2300 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2301 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
0b246afa | 2302 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
f7171750 | 2303 | |
2b82032c YZ |
2304 | super_flags = btrfs_super_flags(disk_super) & |
2305 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2306 | btrfs_set_super_flags(disk_super, super_flags); | |
2307 | ||
2308 | return 0; | |
2309 | } | |
2310 | ||
2311 | /* | |
01327610 | 2312 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2313 | */ |
2314 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2315 | struct btrfs_fs_info *fs_info) |
2b82032c | 2316 | { |
5b4aacef | 2317 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2318 | struct btrfs_path *path; |
2319 | struct extent_buffer *leaf; | |
2320 | struct btrfs_dev_item *dev_item; | |
2321 | struct btrfs_device *device; | |
2322 | struct btrfs_key key; | |
44880fdc | 2323 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2324 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2325 | u64 devid; | |
2326 | int ret; | |
2327 | ||
2328 | path = btrfs_alloc_path(); | |
2329 | if (!path) | |
2330 | return -ENOMEM; | |
2331 | ||
2b82032c YZ |
2332 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2333 | key.offset = 0; | |
2334 | key.type = BTRFS_DEV_ITEM_KEY; | |
2335 | ||
2336 | while (1) { | |
2337 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2338 | if (ret < 0) | |
2339 | goto error; | |
2340 | ||
2341 | leaf = path->nodes[0]; | |
2342 | next_slot: | |
2343 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2344 | ret = btrfs_next_leaf(root, path); | |
2345 | if (ret > 0) | |
2346 | break; | |
2347 | if (ret < 0) | |
2348 | goto error; | |
2349 | leaf = path->nodes[0]; | |
2350 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2351 | btrfs_release_path(path); |
2b82032c YZ |
2352 | continue; |
2353 | } | |
2354 | ||
2355 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2356 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2357 | key.type != BTRFS_DEV_ITEM_KEY) | |
2358 | break; | |
2359 | ||
2360 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2361 | struct btrfs_dev_item); | |
2362 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2363 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2364 | BTRFS_UUID_SIZE); |
1473b24e | 2365 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2366 | BTRFS_FSID_SIZE); |
0b246afa | 2367 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2368 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2369 | |
2370 | if (device->fs_devices->seeding) { | |
2371 | btrfs_set_device_generation(leaf, dev_item, | |
2372 | device->generation); | |
2373 | btrfs_mark_buffer_dirty(leaf); | |
2374 | } | |
2375 | ||
2376 | path->slots[0]++; | |
2377 | goto next_slot; | |
2378 | } | |
2379 | ret = 0; | |
2380 | error: | |
2381 | btrfs_free_path(path); | |
2382 | return ret; | |
2383 | } | |
2384 | ||
da353f6b | 2385 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2386 | { |
5112febb | 2387 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2388 | struct request_queue *q; |
788f20eb CM |
2389 | struct btrfs_trans_handle *trans; |
2390 | struct btrfs_device *device; | |
2391 | struct block_device *bdev; | |
788f20eb | 2392 | struct list_head *devices; |
0b246afa | 2393 | struct super_block *sb = fs_info->sb; |
606686ee | 2394 | struct rcu_string *name; |
3c1dbdf5 | 2395 | u64 tmp; |
2b82032c | 2396 | int seeding_dev = 0; |
788f20eb | 2397 | int ret = 0; |
7132a262 | 2398 | bool unlocked = false; |
788f20eb | 2399 | |
bc98a42c | 2400 | if (sb_rdonly(sb) && !fs_info->fs_devices->seeding) |
f8c5d0b4 | 2401 | return -EROFS; |
788f20eb | 2402 | |
a5d16333 | 2403 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2404 | fs_info->bdev_holder); |
7f59203a JB |
2405 | if (IS_ERR(bdev)) |
2406 | return PTR_ERR(bdev); | |
a2135011 | 2407 | |
0b246afa | 2408 | if (fs_info->fs_devices->seeding) { |
2b82032c YZ |
2409 | seeding_dev = 1; |
2410 | down_write(&sb->s_umount); | |
2411 | mutex_lock(&uuid_mutex); | |
2412 | } | |
2413 | ||
8c8bee1d | 2414 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2415 | |
0b246afa | 2416 | devices = &fs_info->fs_devices->devices; |
d25628bd | 2417 | |
0b246afa | 2418 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
c6e30871 | 2419 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2420 | if (device->bdev == bdev) { |
2421 | ret = -EEXIST; | |
d25628bd | 2422 | mutex_unlock( |
0b246afa | 2423 | &fs_info->fs_devices->device_list_mutex); |
2b82032c | 2424 | goto error; |
788f20eb CM |
2425 | } |
2426 | } | |
0b246afa | 2427 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2428 | |
0b246afa | 2429 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2430 | if (IS_ERR(device)) { |
788f20eb | 2431 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2432 | ret = PTR_ERR(device); |
2b82032c | 2433 | goto error; |
788f20eb CM |
2434 | } |
2435 | ||
78f2c9e6 | 2436 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2437 | if (!name) { |
2b82032c | 2438 | ret = -ENOMEM; |
5c4cf6c9 | 2439 | goto error_free_device; |
788f20eb | 2440 | } |
606686ee | 2441 | rcu_assign_pointer(device->name, name); |
2b82032c | 2442 | |
a22285a6 | 2443 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2444 | if (IS_ERR(trans)) { |
98d5dc13 | 2445 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2446 | goto error_free_device; |
98d5dc13 TI |
2447 | } |
2448 | ||
d5e2003c | 2449 | q = bdev_get_queue(bdev); |
ebbede42 | 2450 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2451 | device->generation = trans->transid; |
0b246afa JM |
2452 | device->io_width = fs_info->sectorsize; |
2453 | device->io_align = fs_info->sectorsize; | |
2454 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2455 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2456 | fs_info->sectorsize); | |
2cc3c559 | 2457 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2458 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2459 | device->fs_info = fs_info; |
788f20eb | 2460 | device->bdev = bdev; |
e12c9621 | 2461 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2462 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2463 | device->mode = FMODE_EXCL; |
27087f37 | 2464 | device->dev_stats_valid = 1; |
9f6d2510 | 2465 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2466 | |
2b82032c | 2467 | if (seeding_dev) { |
1751e8a6 | 2468 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2469 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2470 | if (ret) { |
2471 | btrfs_abort_transaction(trans, ret); | |
2472 | goto error_trans; | |
2473 | } | |
2b82032c | 2474 | } |
788f20eb | 2475 | |
0b246afa | 2476 | device->fs_devices = fs_info->fs_devices; |
e5e9a520 | 2477 | |
0b246afa | 2478 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
34441361 | 2479 | mutex_lock(&fs_info->chunk_mutex); |
0b246afa | 2480 | list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices); |
2b82032c | 2481 | list_add(&device->dev_alloc_list, |
0b246afa JM |
2482 | &fs_info->fs_devices->alloc_list); |
2483 | fs_info->fs_devices->num_devices++; | |
2484 | fs_info->fs_devices->open_devices++; | |
2485 | fs_info->fs_devices->rw_devices++; | |
2486 | fs_info->fs_devices->total_devices++; | |
2487 | fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2488 | |
a5ed45f8 | 2489 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2490 | |
e884f4f0 | 2491 | if (!blk_queue_nonrot(q)) |
0b246afa | 2492 | fs_info->fs_devices->rotating = 1; |
c289811c | 2493 | |
0b246afa JM |
2494 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2495 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
7dfb8be1 | 2496 | round_down(tmp + device->total_bytes, fs_info->sectorsize)); |
788f20eb | 2497 | |
0b246afa JM |
2498 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2499 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2500 | |
2501 | /* add sysfs device entry */ | |
0b246afa | 2502 | btrfs_sysfs_add_device_link(fs_info->fs_devices, device); |
0d39376a | 2503 | |
2196d6e8 MX |
2504 | /* |
2505 | * we've got more storage, clear any full flags on the space | |
2506 | * infos | |
2507 | */ | |
0b246afa | 2508 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2509 | |
34441361 | 2510 | mutex_unlock(&fs_info->chunk_mutex); |
0b246afa | 2511 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2512 | |
2b82032c | 2513 | if (seeding_dev) { |
34441361 | 2514 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2515 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2516 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2517 | if (ret) { |
66642832 | 2518 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2519 | goto error_sysfs; |
005d6427 | 2520 | } |
2196d6e8 MX |
2521 | } |
2522 | ||
c74a0b02 | 2523 | ret = btrfs_add_dev_item(trans, fs_info, device); |
2196d6e8 | 2524 | if (ret) { |
66642832 | 2525 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2526 | goto error_sysfs; |
2196d6e8 MX |
2527 | } |
2528 | ||
2529 | if (seeding_dev) { | |
2530 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2531 | ||
0b246afa | 2532 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2533 | if (ret) { |
66642832 | 2534 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2535 | goto error_sysfs; |
005d6427 | 2536 | } |
b2373f25 AJ |
2537 | |
2538 | /* Sprouting would change fsid of the mounted root, | |
2539 | * so rename the fsid on the sysfs | |
2540 | */ | |
2541 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa JM |
2542 | fs_info->fsid); |
2543 | if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf)) | |
2544 | btrfs_warn(fs_info, | |
2545 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2546 | } |
2547 | ||
3a45bb20 | 2548 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2549 | |
2b82032c YZ |
2550 | if (seeding_dev) { |
2551 | mutex_unlock(&uuid_mutex); | |
2552 | up_write(&sb->s_umount); | |
7132a262 | 2553 | unlocked = true; |
788f20eb | 2554 | |
79787eaa JM |
2555 | if (ret) /* transaction commit */ |
2556 | return ret; | |
2557 | ||
2ff7e61e | 2558 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2559 | if (ret < 0) |
0b246afa | 2560 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2561 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2562 | trans = btrfs_attach_transaction(root); |
2563 | if (IS_ERR(trans)) { | |
2564 | if (PTR_ERR(trans) == -ENOENT) | |
2565 | return 0; | |
7132a262 AJ |
2566 | ret = PTR_ERR(trans); |
2567 | trans = NULL; | |
2568 | goto error_sysfs; | |
671415b7 | 2569 | } |
3a45bb20 | 2570 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2571 | } |
c9e9f97b | 2572 | |
5a1972bd QW |
2573 | /* Update ctime/mtime for libblkid */ |
2574 | update_dev_time(device_path); | |
2b82032c | 2575 | return ret; |
79787eaa | 2576 | |
d31c32f6 AJ |
2577 | error_sysfs: |
2578 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); | |
79787eaa | 2579 | error_trans: |
0af2c4bf | 2580 | if (seeding_dev) |
1751e8a6 | 2581 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2582 | if (trans) |
2583 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2584 | error_free_device: |
55de4803 | 2585 | free_device(device); |
2b82032c | 2586 | error: |
e525fd89 | 2587 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2588 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2589 | mutex_unlock(&uuid_mutex); |
2590 | up_write(&sb->s_umount); | |
2591 | } | |
c9e9f97b | 2592 | return ret; |
788f20eb CM |
2593 | } |
2594 | ||
2ff7e61e | 2595 | int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, |
da353f6b | 2596 | const char *device_path, |
1c43366d | 2597 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2598 | struct btrfs_device **device_out) |
2599 | { | |
e93c89c1 SB |
2600 | struct btrfs_device *device; |
2601 | struct block_device *bdev; | |
e93c89c1 SB |
2602 | struct list_head *devices; |
2603 | struct rcu_string *name; | |
12bd2fc0 | 2604 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2605 | int ret = 0; |
2606 | ||
2607 | *device_out = NULL; | |
1c43366d MX |
2608 | if (fs_info->fs_devices->seeding) { |
2609 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2610 | return -EINVAL; |
1c43366d | 2611 | } |
e93c89c1 SB |
2612 | |
2613 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2614 | fs_info->bdev_holder); | |
1c43366d MX |
2615 | if (IS_ERR(bdev)) { |
2616 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2617 | return PTR_ERR(bdev); |
1c43366d | 2618 | } |
e93c89c1 SB |
2619 | |
2620 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2621 | ||
2622 | devices = &fs_info->fs_devices->devices; | |
2623 | list_for_each_entry(device, devices, dev_list) { | |
2624 | if (device->bdev == bdev) { | |
5d163e0e JM |
2625 | btrfs_err(fs_info, |
2626 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2627 | ret = -EEXIST; |
2628 | goto error; | |
2629 | } | |
2630 | } | |
2631 | ||
1c43366d | 2632 | |
7cc8e58d MX |
2633 | if (i_size_read(bdev->bd_inode) < |
2634 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2635 | btrfs_err(fs_info, |
2636 | "target device is smaller than source device!"); | |
1c43366d MX |
2637 | ret = -EINVAL; |
2638 | goto error; | |
2639 | } | |
2640 | ||
2641 | ||
12bd2fc0 ID |
2642 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2643 | if (IS_ERR(device)) { | |
2644 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2645 | goto error; |
2646 | } | |
2647 | ||
6165572c | 2648 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
e93c89c1 | 2649 | if (!name) { |
55de4803 | 2650 | free_device(device); |
e93c89c1 SB |
2651 | ret = -ENOMEM; |
2652 | goto error; | |
2653 | } | |
2654 | rcu_assign_pointer(device->name, name); | |
2655 | ||
0b246afa | 2656 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
ebbede42 | 2657 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
e93c89c1 | 2658 | device->generation = 0; |
0b246afa JM |
2659 | device->io_width = fs_info->sectorsize; |
2660 | device->io_align = fs_info->sectorsize; | |
2661 | device->sector_size = fs_info->sectorsize; | |
7cc8e58d MX |
2662 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2663 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2664 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 | 2665 | device->commit_total_bytes = srcdev->commit_total_bytes; |
ce7213c7 | 2666 | device->commit_bytes_used = device->bytes_used; |
fb456252 | 2667 | device->fs_info = fs_info; |
e93c89c1 | 2668 | device->bdev = bdev; |
e12c9621 | 2669 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2670 | set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
e93c89c1 | 2671 | device->mode = FMODE_EXCL; |
27087f37 | 2672 | device->dev_stats_valid = 1; |
9f6d2510 | 2673 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
e93c89c1 SB |
2674 | device->fs_devices = fs_info->fs_devices; |
2675 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2676 | fs_info->fs_devices->num_devices++; | |
2677 | fs_info->fs_devices->open_devices++; | |
0b246afa | 2678 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 SB |
2679 | |
2680 | *device_out = device; | |
2681 | return ret; | |
2682 | ||
2683 | error: | |
2684 | blkdev_put(bdev, FMODE_EXCL); | |
2685 | return ret; | |
2686 | } | |
2687 | ||
d397712b CM |
2688 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2689 | struct btrfs_device *device) | |
0b86a832 CM |
2690 | { |
2691 | int ret; | |
2692 | struct btrfs_path *path; | |
0b246afa | 2693 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2694 | struct btrfs_dev_item *dev_item; |
2695 | struct extent_buffer *leaf; | |
2696 | struct btrfs_key key; | |
2697 | ||
0b86a832 CM |
2698 | path = btrfs_alloc_path(); |
2699 | if (!path) | |
2700 | return -ENOMEM; | |
2701 | ||
2702 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2703 | key.type = BTRFS_DEV_ITEM_KEY; | |
2704 | key.offset = device->devid; | |
2705 | ||
2706 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2707 | if (ret < 0) | |
2708 | goto out; | |
2709 | ||
2710 | if (ret > 0) { | |
2711 | ret = -ENOENT; | |
2712 | goto out; | |
2713 | } | |
2714 | ||
2715 | leaf = path->nodes[0]; | |
2716 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2717 | ||
2718 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2719 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2720 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2721 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2722 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2723 | btrfs_set_device_total_bytes(leaf, dev_item, |
2724 | btrfs_device_get_disk_total_bytes(device)); | |
2725 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2726 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2727 | btrfs_mark_buffer_dirty(leaf); |
2728 | ||
2729 | out: | |
2730 | btrfs_free_path(path); | |
2731 | return ret; | |
2732 | } | |
2733 | ||
2196d6e8 | 2734 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2735 | struct btrfs_device *device, u64 new_size) |
2736 | { | |
0b246afa JM |
2737 | struct btrfs_fs_info *fs_info = device->fs_info; |
2738 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2739 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2740 | u64 old_total; |
2741 | u64 diff; | |
8f18cf13 | 2742 | |
ebbede42 | 2743 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2744 | return -EACCES; |
2196d6e8 | 2745 | |
7dfb8be1 NB |
2746 | new_size = round_down(new_size, fs_info->sectorsize); |
2747 | ||
34441361 | 2748 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2749 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2750 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2751 | |
63a212ab | 2752 | if (new_size <= device->total_bytes || |
401e29c1 | 2753 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2754 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2755 | return -EINVAL; |
2196d6e8 | 2756 | } |
2b82032c | 2757 | |
0b246afa | 2758 | fs_devices = fs_info->fs_devices; |
2b82032c | 2759 | |
7dfb8be1 NB |
2760 | btrfs_set_super_total_bytes(super_copy, |
2761 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2762 | device->fs_devices->total_rw_bytes += diff; |
2763 | ||
7cc8e58d MX |
2764 | btrfs_device_set_total_bytes(device, new_size); |
2765 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2766 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2767 | if (list_empty(&device->resized_list)) |
2768 | list_add_tail(&device->resized_list, | |
2769 | &fs_devices->resized_devices); | |
34441361 | 2770 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2771 | |
8f18cf13 CM |
2772 | return btrfs_update_device(trans, device); |
2773 | } | |
2774 | ||
2775 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
408fbf19 | 2776 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2777 | { |
5b4aacef | 2778 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2779 | int ret; |
2780 | struct btrfs_path *path; | |
2781 | struct btrfs_key key; | |
2782 | ||
8f18cf13 CM |
2783 | path = btrfs_alloc_path(); |
2784 | if (!path) | |
2785 | return -ENOMEM; | |
2786 | ||
408fbf19 | 2787 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2788 | key.offset = chunk_offset; |
2789 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2790 | ||
2791 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2792 | if (ret < 0) |
2793 | goto out; | |
2794 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2795 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2796 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2797 | ret = -ENOENT; |
2798 | goto out; | |
2799 | } | |
8f18cf13 CM |
2800 | |
2801 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2802 | if (ret < 0) |
0b246afa JM |
2803 | btrfs_handle_fs_error(fs_info, ret, |
2804 | "Failed to delete chunk item."); | |
79787eaa | 2805 | out: |
8f18cf13 | 2806 | btrfs_free_path(path); |
65a246c5 | 2807 | return ret; |
8f18cf13 CM |
2808 | } |
2809 | ||
408fbf19 | 2810 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2811 | { |
0b246afa | 2812 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2813 | struct btrfs_disk_key *disk_key; |
2814 | struct btrfs_chunk *chunk; | |
2815 | u8 *ptr; | |
2816 | int ret = 0; | |
2817 | u32 num_stripes; | |
2818 | u32 array_size; | |
2819 | u32 len = 0; | |
2820 | u32 cur; | |
2821 | struct btrfs_key key; | |
2822 | ||
34441361 | 2823 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2824 | array_size = btrfs_super_sys_array_size(super_copy); |
2825 | ||
2826 | ptr = super_copy->sys_chunk_array; | |
2827 | cur = 0; | |
2828 | ||
2829 | while (cur < array_size) { | |
2830 | disk_key = (struct btrfs_disk_key *)ptr; | |
2831 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2832 | ||
2833 | len = sizeof(*disk_key); | |
2834 | ||
2835 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2836 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2837 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2838 | len += btrfs_chunk_item_size(num_stripes); | |
2839 | } else { | |
2840 | ret = -EIO; | |
2841 | break; | |
2842 | } | |
408fbf19 | 2843 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2844 | key.offset == chunk_offset) { |
2845 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2846 | array_size -= len; | |
2847 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2848 | } else { | |
2849 | ptr += len; | |
2850 | cur += len; | |
2851 | } | |
2852 | } | |
34441361 | 2853 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2854 | return ret; |
2855 | } | |
2856 | ||
592d92ee LB |
2857 | static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info, |
2858 | u64 logical, u64 length) | |
2859 | { | |
2860 | struct extent_map_tree *em_tree; | |
2861 | struct extent_map *em; | |
2862 | ||
2863 | em_tree = &fs_info->mapping_tree.map_tree; | |
2864 | read_lock(&em_tree->lock); | |
2865 | em = lookup_extent_mapping(em_tree, logical, length); | |
2866 | read_unlock(&em_tree->lock); | |
2867 | ||
2868 | if (!em) { | |
2869 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2870 | logical, length); | |
2871 | return ERR_PTR(-EINVAL); | |
2872 | } | |
2873 | ||
2874 | if (em->start > logical || em->start + em->len < logical) { | |
2875 | btrfs_crit(fs_info, | |
2876 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2877 | logical, length, em->start, em->start + em->len); | |
2878 | free_extent_map(em); | |
2879 | return ERR_PTR(-EINVAL); | |
2880 | } | |
2881 | ||
2882 | /* callers are responsible for dropping em's ref. */ | |
2883 | return em; | |
2884 | } | |
2885 | ||
47ab2a6c | 2886 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
5b4aacef | 2887 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2888 | { |
8f18cf13 CM |
2889 | struct extent_map *em; |
2890 | struct map_lookup *map; | |
2196d6e8 | 2891 | u64 dev_extent_len = 0; |
47ab2a6c | 2892 | int i, ret = 0; |
0b246afa | 2893 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2894 | |
592d92ee LB |
2895 | em = get_chunk_map(fs_info, chunk_offset, 1); |
2896 | if (IS_ERR(em)) { | |
47ab2a6c JB |
2897 | /* |
2898 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2899 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2900 | * do anything we still error out. |
2901 | */ | |
2902 | ASSERT(0); | |
592d92ee | 2903 | return PTR_ERR(em); |
47ab2a6c | 2904 | } |
95617d69 | 2905 | map = em->map_lookup; |
34441361 | 2906 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 2907 | check_system_chunk(trans, fs_info, map->type); |
34441361 | 2908 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2909 | |
57ba4cb8 FM |
2910 | /* |
2911 | * Take the device list mutex to prevent races with the final phase of | |
2912 | * a device replace operation that replaces the device object associated | |
2913 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2914 | */ | |
2915 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2916 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2917 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2918 | ret = btrfs_free_dev_extent(trans, device, |
2919 | map->stripes[i].physical, | |
2920 | &dev_extent_len); | |
47ab2a6c | 2921 | if (ret) { |
57ba4cb8 | 2922 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2923 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2924 | goto out; |
2925 | } | |
a061fc8d | 2926 | |
2196d6e8 | 2927 | if (device->bytes_used > 0) { |
34441361 | 2928 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2929 | btrfs_device_set_bytes_used(device, |
2930 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2931 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2932 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2933 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2934 | } |
a061fc8d | 2935 | |
dfe25020 CM |
2936 | if (map->stripes[i].dev) { |
2937 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2938 | if (ret) { |
57ba4cb8 | 2939 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2940 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2941 | goto out; |
2942 | } | |
dfe25020 | 2943 | } |
8f18cf13 | 2944 | } |
57ba4cb8 FM |
2945 | mutex_unlock(&fs_devices->device_list_mutex); |
2946 | ||
408fbf19 | 2947 | ret = btrfs_free_chunk(trans, fs_info, chunk_offset); |
47ab2a6c | 2948 | if (ret) { |
66642832 | 2949 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2950 | goto out; |
2951 | } | |
8f18cf13 | 2952 | |
6bccf3ab | 2953 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2954 | |
8f18cf13 | 2955 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2956 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2957 | if (ret) { |
66642832 | 2958 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2959 | goto out; |
2960 | } | |
8f18cf13 CM |
2961 | } |
2962 | ||
6bccf3ab | 2963 | ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em); |
47ab2a6c | 2964 | if (ret) { |
66642832 | 2965 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2966 | goto out; |
2967 | } | |
2b82032c | 2968 | |
47ab2a6c | 2969 | out: |
2b82032c YZ |
2970 | /* once for us */ |
2971 | free_extent_map(em); | |
47ab2a6c JB |
2972 | return ret; |
2973 | } | |
2b82032c | 2974 | |
5b4aacef | 2975 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2976 | { |
5b4aacef | 2977 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2978 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2979 | int ret; |
2b82032c | 2980 | |
67c5e7d4 FM |
2981 | /* |
2982 | * Prevent races with automatic removal of unused block groups. | |
2983 | * After we relocate and before we remove the chunk with offset | |
2984 | * chunk_offset, automatic removal of the block group can kick in, | |
2985 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2986 | * | |
2987 | * Make sure to acquire this mutex before doing a tree search (dev | |
2988 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2989 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2990 | * we release the path used to search the chunk/dev tree and before | |
2991 | * the current task acquires this mutex and calls us. | |
2992 | */ | |
a32bf9a3 | 2993 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 2994 | |
0b246afa | 2995 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2996 | if (ret) |
2997 | return -ENOSPC; | |
2998 | ||
2999 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 3000 | btrfs_scrub_pause(fs_info); |
0b246afa | 3001 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3002 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3003 | if (ret) |
3004 | return ret; | |
3005 | ||
75cb379d JM |
3006 | /* |
3007 | * We add the kobjects here (and after forcing data chunk creation) | |
3008 | * since relocation is the only place we'll create chunks of a new | |
3009 | * type at runtime. The only place where we'll remove the last | |
3010 | * chunk of a type is the call immediately below this one. Even | |
3011 | * so, we're protected against races with the cleaner thread since | |
3012 | * we're covered by the delete_unused_bgs_mutex. | |
3013 | */ | |
3014 | btrfs_add_raid_kobjects(fs_info); | |
3015 | ||
19c4d2f9 CM |
3016 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3017 | chunk_offset); | |
3018 | if (IS_ERR(trans)) { | |
3019 | ret = PTR_ERR(trans); | |
3020 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3021 | return ret; | |
3022 | } | |
3023 | ||
47ab2a6c | 3024 | /* |
19c4d2f9 CM |
3025 | * step two, delete the device extents and the |
3026 | * chunk tree entries | |
47ab2a6c | 3027 | */ |
5b4aacef | 3028 | ret = btrfs_remove_chunk(trans, fs_info, chunk_offset); |
3a45bb20 | 3029 | btrfs_end_transaction(trans); |
19c4d2f9 | 3030 | return ret; |
2b82032c YZ |
3031 | } |
3032 | ||
2ff7e61e | 3033 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3034 | { |
0b246afa | 3035 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3036 | struct btrfs_path *path; |
3037 | struct extent_buffer *leaf; | |
3038 | struct btrfs_chunk *chunk; | |
3039 | struct btrfs_key key; | |
3040 | struct btrfs_key found_key; | |
2b82032c | 3041 | u64 chunk_type; |
ba1bf481 JB |
3042 | bool retried = false; |
3043 | int failed = 0; | |
2b82032c YZ |
3044 | int ret; |
3045 | ||
3046 | path = btrfs_alloc_path(); | |
3047 | if (!path) | |
3048 | return -ENOMEM; | |
3049 | ||
ba1bf481 | 3050 | again: |
2b82032c YZ |
3051 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3052 | key.offset = (u64)-1; | |
3053 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3054 | ||
3055 | while (1) { | |
0b246afa | 3056 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3057 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3058 | if (ret < 0) { |
0b246afa | 3059 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3060 | goto error; |
67c5e7d4 | 3061 | } |
79787eaa | 3062 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3063 | |
3064 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3065 | key.type); | |
67c5e7d4 | 3066 | if (ret) |
0b246afa | 3067 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3068 | if (ret < 0) |
3069 | goto error; | |
3070 | if (ret > 0) | |
3071 | break; | |
1a40e23b | 3072 | |
2b82032c YZ |
3073 | leaf = path->nodes[0]; |
3074 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3075 | |
2b82032c YZ |
3076 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3077 | struct btrfs_chunk); | |
3078 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3079 | btrfs_release_path(path); |
8f18cf13 | 3080 | |
2b82032c | 3081 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3082 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3083 | if (ret == -ENOSPC) |
3084 | failed++; | |
14586651 HS |
3085 | else |
3086 | BUG_ON(ret); | |
2b82032c | 3087 | } |
0b246afa | 3088 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3089 | |
2b82032c YZ |
3090 | if (found_key.offset == 0) |
3091 | break; | |
3092 | key.offset = found_key.offset - 1; | |
3093 | } | |
3094 | ret = 0; | |
ba1bf481 JB |
3095 | if (failed && !retried) { |
3096 | failed = 0; | |
3097 | retried = true; | |
3098 | goto again; | |
fae7f21c | 3099 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3100 | ret = -ENOSPC; |
3101 | } | |
2b82032c YZ |
3102 | error: |
3103 | btrfs_free_path(path); | |
3104 | return ret; | |
8f18cf13 CM |
3105 | } |
3106 | ||
a6f93c71 LB |
3107 | /* |
3108 | * return 1 : allocate a data chunk successfully, | |
3109 | * return <0: errors during allocating a data chunk, | |
3110 | * return 0 : no need to allocate a data chunk. | |
3111 | */ | |
3112 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3113 | u64 chunk_offset) | |
3114 | { | |
3115 | struct btrfs_block_group_cache *cache; | |
3116 | u64 bytes_used; | |
3117 | u64 chunk_type; | |
3118 | ||
3119 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3120 | ASSERT(cache); | |
3121 | chunk_type = cache->flags; | |
3122 | btrfs_put_block_group(cache); | |
3123 | ||
3124 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3125 | spin_lock(&fs_info->data_sinfo->lock); | |
3126 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3127 | spin_unlock(&fs_info->data_sinfo->lock); | |
3128 | ||
3129 | if (!bytes_used) { | |
3130 | struct btrfs_trans_handle *trans; | |
3131 | int ret; | |
3132 | ||
3133 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3134 | if (IS_ERR(trans)) | |
3135 | return PTR_ERR(trans); | |
3136 | ||
3137 | ret = btrfs_force_chunk_alloc(trans, fs_info, | |
3138 | BTRFS_BLOCK_GROUP_DATA); | |
3139 | btrfs_end_transaction(trans); | |
3140 | if (ret < 0) | |
3141 | return ret; | |
3142 | ||
75cb379d JM |
3143 | btrfs_add_raid_kobjects(fs_info); |
3144 | ||
a6f93c71 LB |
3145 | return 1; |
3146 | } | |
3147 | } | |
3148 | return 0; | |
3149 | } | |
3150 | ||
6bccf3ab | 3151 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3152 | struct btrfs_balance_control *bctl) |
3153 | { | |
6bccf3ab | 3154 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3155 | struct btrfs_trans_handle *trans; |
3156 | struct btrfs_balance_item *item; | |
3157 | struct btrfs_disk_balance_args disk_bargs; | |
3158 | struct btrfs_path *path; | |
3159 | struct extent_buffer *leaf; | |
3160 | struct btrfs_key key; | |
3161 | int ret, err; | |
3162 | ||
3163 | path = btrfs_alloc_path(); | |
3164 | if (!path) | |
3165 | return -ENOMEM; | |
3166 | ||
3167 | trans = btrfs_start_transaction(root, 0); | |
3168 | if (IS_ERR(trans)) { | |
3169 | btrfs_free_path(path); | |
3170 | return PTR_ERR(trans); | |
3171 | } | |
3172 | ||
3173 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3174 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3175 | key.offset = 0; |
3176 | ||
3177 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3178 | sizeof(*item)); | |
3179 | if (ret) | |
3180 | goto out; | |
3181 | ||
3182 | leaf = path->nodes[0]; | |
3183 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3184 | ||
b159fa28 | 3185 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3186 | |
3187 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3188 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3189 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3190 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3191 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3192 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3193 | ||
3194 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3195 | ||
3196 | btrfs_mark_buffer_dirty(leaf); | |
3197 | out: | |
3198 | btrfs_free_path(path); | |
3a45bb20 | 3199 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3200 | if (err && !ret) |
3201 | ret = err; | |
3202 | return ret; | |
3203 | } | |
3204 | ||
6bccf3ab | 3205 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3206 | { |
6bccf3ab | 3207 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3208 | struct btrfs_trans_handle *trans; |
3209 | struct btrfs_path *path; | |
3210 | struct btrfs_key key; | |
3211 | int ret, err; | |
3212 | ||
3213 | path = btrfs_alloc_path(); | |
3214 | if (!path) | |
3215 | return -ENOMEM; | |
3216 | ||
3217 | trans = btrfs_start_transaction(root, 0); | |
3218 | if (IS_ERR(trans)) { | |
3219 | btrfs_free_path(path); | |
3220 | return PTR_ERR(trans); | |
3221 | } | |
3222 | ||
3223 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3224 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3225 | key.offset = 0; |
3226 | ||
3227 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3228 | if (ret < 0) | |
3229 | goto out; | |
3230 | if (ret > 0) { | |
3231 | ret = -ENOENT; | |
3232 | goto out; | |
3233 | } | |
3234 | ||
3235 | ret = btrfs_del_item(trans, root, path); | |
3236 | out: | |
3237 | btrfs_free_path(path); | |
3a45bb20 | 3238 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3239 | if (err && !ret) |
3240 | ret = err; | |
3241 | return ret; | |
3242 | } | |
3243 | ||
59641015 ID |
3244 | /* |
3245 | * This is a heuristic used to reduce the number of chunks balanced on | |
3246 | * resume after balance was interrupted. | |
3247 | */ | |
3248 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3249 | { | |
3250 | /* | |
3251 | * Turn on soft mode for chunk types that were being converted. | |
3252 | */ | |
3253 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3254 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3255 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3256 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3257 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3258 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3259 | ||
3260 | /* | |
3261 | * Turn on usage filter if is not already used. The idea is | |
3262 | * that chunks that we have already balanced should be | |
3263 | * reasonably full. Don't do it for chunks that are being | |
3264 | * converted - that will keep us from relocating unconverted | |
3265 | * (albeit full) chunks. | |
3266 | */ | |
3267 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3268 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3269 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3270 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3271 | bctl->data.usage = 90; | |
3272 | } | |
3273 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3274 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3275 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3276 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3277 | bctl->sys.usage = 90; | |
3278 | } | |
3279 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3280 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3281 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3282 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3283 | bctl->meta.usage = 90; | |
3284 | } | |
3285 | } | |
3286 | ||
c9e9f97b ID |
3287 | /* |
3288 | * Should be called with both balance and volume mutexes held to | |
3289 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3290 | * restriper. Same goes for unset_balance_control. | |
3291 | */ | |
3292 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3293 | { | |
3294 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3295 | ||
3296 | BUG_ON(fs_info->balance_ctl); | |
3297 | ||
3298 | spin_lock(&fs_info->balance_lock); | |
3299 | fs_info->balance_ctl = bctl; | |
3300 | spin_unlock(&fs_info->balance_lock); | |
3301 | } | |
3302 | ||
3303 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3304 | { | |
3305 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3306 | ||
3307 | BUG_ON(!fs_info->balance_ctl); | |
3308 | ||
3309 | spin_lock(&fs_info->balance_lock); | |
3310 | fs_info->balance_ctl = NULL; | |
3311 | spin_unlock(&fs_info->balance_lock); | |
3312 | ||
3313 | kfree(bctl); | |
3314 | } | |
3315 | ||
ed25e9b2 ID |
3316 | /* |
3317 | * Balance filters. Return 1 if chunk should be filtered out | |
3318 | * (should not be balanced). | |
3319 | */ | |
899c81ea | 3320 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3321 | struct btrfs_balance_args *bargs) |
3322 | { | |
899c81ea ID |
3323 | chunk_type = chunk_to_extended(chunk_type) & |
3324 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3325 | |
899c81ea | 3326 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3327 | return 0; |
3328 | ||
3329 | return 1; | |
3330 | } | |
3331 | ||
dba72cb3 | 3332 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3333 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3334 | { |
3335 | struct btrfs_block_group_cache *cache; | |
3336 | u64 chunk_used; | |
3337 | u64 user_thresh_min; | |
3338 | u64 user_thresh_max; | |
3339 | int ret = 1; | |
3340 | ||
3341 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3342 | chunk_used = btrfs_block_group_used(&cache->item); | |
3343 | ||
3344 | if (bargs->usage_min == 0) | |
3345 | user_thresh_min = 0; | |
3346 | else | |
3347 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3348 | bargs->usage_min); | |
3349 | ||
3350 | if (bargs->usage_max == 0) | |
3351 | user_thresh_max = 1; | |
3352 | else if (bargs->usage_max > 100) | |
3353 | user_thresh_max = cache->key.offset; | |
3354 | else | |
3355 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3356 | bargs->usage_max); | |
3357 | ||
3358 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3359 | ret = 0; | |
3360 | ||
3361 | btrfs_put_block_group(cache); | |
3362 | return ret; | |
3363 | } | |
3364 | ||
dba72cb3 | 3365 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3366 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3367 | { |
3368 | struct btrfs_block_group_cache *cache; | |
3369 | u64 chunk_used, user_thresh; | |
3370 | int ret = 1; | |
3371 | ||
3372 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3373 | chunk_used = btrfs_block_group_used(&cache->item); | |
3374 | ||
bc309467 | 3375 | if (bargs->usage_min == 0) |
3e39cea6 | 3376 | user_thresh = 1; |
a105bb88 ID |
3377 | else if (bargs->usage > 100) |
3378 | user_thresh = cache->key.offset; | |
3379 | else | |
3380 | user_thresh = div_factor_fine(cache->key.offset, | |
3381 | bargs->usage); | |
3382 | ||
5ce5b3c0 ID |
3383 | if (chunk_used < user_thresh) |
3384 | ret = 0; | |
3385 | ||
3386 | btrfs_put_block_group(cache); | |
3387 | return ret; | |
3388 | } | |
3389 | ||
409d404b ID |
3390 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3391 | struct btrfs_chunk *chunk, | |
3392 | struct btrfs_balance_args *bargs) | |
3393 | { | |
3394 | struct btrfs_stripe *stripe; | |
3395 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3396 | int i; | |
3397 | ||
3398 | for (i = 0; i < num_stripes; i++) { | |
3399 | stripe = btrfs_stripe_nr(chunk, i); | |
3400 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3401 | return 0; | |
3402 | } | |
3403 | ||
3404 | return 1; | |
3405 | } | |
3406 | ||
94e60d5a ID |
3407 | /* [pstart, pend) */ |
3408 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3409 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3410 | struct btrfs_balance_args *bargs) |
3411 | { | |
3412 | struct btrfs_stripe *stripe; | |
3413 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3414 | u64 stripe_offset; | |
3415 | u64 stripe_length; | |
3416 | int factor; | |
3417 | int i; | |
3418 | ||
3419 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3420 | return 0; | |
3421 | ||
3422 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3423 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3424 | factor = num_stripes / 2; | |
3425 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3426 | factor = num_stripes - 1; | |
3427 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3428 | factor = num_stripes - 2; | |
3429 | } else { | |
3430 | factor = num_stripes; | |
3431 | } | |
94e60d5a ID |
3432 | |
3433 | for (i = 0; i < num_stripes; i++) { | |
3434 | stripe = btrfs_stripe_nr(chunk, i); | |
3435 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3436 | continue; | |
3437 | ||
3438 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3439 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3440 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3441 | |
3442 | if (stripe_offset < bargs->pend && | |
3443 | stripe_offset + stripe_length > bargs->pstart) | |
3444 | return 0; | |
3445 | } | |
3446 | ||
3447 | return 1; | |
3448 | } | |
3449 | ||
ea67176a ID |
3450 | /* [vstart, vend) */ |
3451 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3452 | struct btrfs_chunk *chunk, | |
3453 | u64 chunk_offset, | |
3454 | struct btrfs_balance_args *bargs) | |
3455 | { | |
3456 | if (chunk_offset < bargs->vend && | |
3457 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3458 | /* at least part of the chunk is inside this vrange */ | |
3459 | return 0; | |
3460 | ||
3461 | return 1; | |
3462 | } | |
3463 | ||
dee32d0a GAP |
3464 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3465 | struct btrfs_chunk *chunk, | |
3466 | struct btrfs_balance_args *bargs) | |
3467 | { | |
3468 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3469 | ||
3470 | if (bargs->stripes_min <= num_stripes | |
3471 | && num_stripes <= bargs->stripes_max) | |
3472 | return 0; | |
3473 | ||
3474 | return 1; | |
3475 | } | |
3476 | ||
899c81ea | 3477 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3478 | struct btrfs_balance_args *bargs) |
3479 | { | |
3480 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3481 | return 0; | |
3482 | ||
899c81ea ID |
3483 | chunk_type = chunk_to_extended(chunk_type) & |
3484 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3485 | |
899c81ea | 3486 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3487 | return 1; |
3488 | ||
3489 | return 0; | |
3490 | } | |
3491 | ||
2ff7e61e | 3492 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3493 | struct extent_buffer *leaf, |
3494 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3495 | { | |
0b246afa | 3496 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3497 | struct btrfs_balance_args *bargs = NULL; |
3498 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3499 | ||
3500 | /* type filter */ | |
3501 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3502 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3503 | return 0; | |
3504 | } | |
3505 | ||
3506 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3507 | bargs = &bctl->data; | |
3508 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3509 | bargs = &bctl->sys; | |
3510 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3511 | bargs = &bctl->meta; | |
3512 | ||
ed25e9b2 ID |
3513 | /* profiles filter */ |
3514 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3515 | chunk_profiles_filter(chunk_type, bargs)) { | |
3516 | return 0; | |
5ce5b3c0 ID |
3517 | } |
3518 | ||
3519 | /* usage filter */ | |
3520 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3521 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3522 | return 0; |
bc309467 | 3523 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3524 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3525 | return 0; |
409d404b ID |
3526 | } |
3527 | ||
3528 | /* devid filter */ | |
3529 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3530 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3531 | return 0; | |
94e60d5a ID |
3532 | } |
3533 | ||
3534 | /* drange filter, makes sense only with devid filter */ | |
3535 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3536 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3537 | return 0; |
ea67176a ID |
3538 | } |
3539 | ||
3540 | /* vrange filter */ | |
3541 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3542 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3543 | return 0; | |
ed25e9b2 ID |
3544 | } |
3545 | ||
dee32d0a GAP |
3546 | /* stripes filter */ |
3547 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3548 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3549 | return 0; | |
3550 | } | |
3551 | ||
cfa4c961 ID |
3552 | /* soft profile changing mode */ |
3553 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3554 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3555 | return 0; | |
3556 | } | |
3557 | ||
7d824b6f DS |
3558 | /* |
3559 | * limited by count, must be the last filter | |
3560 | */ | |
3561 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3562 | if (bargs->limit == 0) | |
3563 | return 0; | |
3564 | else | |
3565 | bargs->limit--; | |
12907fc7 DS |
3566 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3567 | /* | |
3568 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3569 | * determined here because we do not have the global information |
12907fc7 DS |
3570 | * about the count of all chunks that satisfy the filters. |
3571 | */ | |
3572 | if (bargs->limit_max == 0) | |
3573 | return 0; | |
3574 | else | |
3575 | bargs->limit_max--; | |
7d824b6f DS |
3576 | } |
3577 | ||
f43ffb60 ID |
3578 | return 1; |
3579 | } | |
3580 | ||
c9e9f97b | 3581 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3582 | { |
19a39dce | 3583 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3584 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3585 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3586 | struct list_head *devices; | |
ec44a35c CM |
3587 | struct btrfs_device *device; |
3588 | u64 old_size; | |
3589 | u64 size_to_free; | |
12907fc7 | 3590 | u64 chunk_type; |
f43ffb60 | 3591 | struct btrfs_chunk *chunk; |
5a488b9d | 3592 | struct btrfs_path *path = NULL; |
ec44a35c | 3593 | struct btrfs_key key; |
ec44a35c | 3594 | struct btrfs_key found_key; |
c9e9f97b | 3595 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3596 | struct extent_buffer *leaf; |
3597 | int slot; | |
c9e9f97b ID |
3598 | int ret; |
3599 | int enospc_errors = 0; | |
19a39dce | 3600 | bool counting = true; |
12907fc7 | 3601 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3602 | u64 limit_data = bctl->data.limit; |
3603 | u64 limit_meta = bctl->meta.limit; | |
3604 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3605 | u32 count_data = 0; |
3606 | u32 count_meta = 0; | |
3607 | u32 count_sys = 0; | |
2c9fe835 | 3608 | int chunk_reserved = 0; |
ec44a35c | 3609 | |
ec44a35c | 3610 | /* step one make some room on all the devices */ |
c9e9f97b | 3611 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3612 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3613 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3614 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3615 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3616 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3617 | btrfs_device_get_total_bytes(device) - |
3618 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3619 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3620 | continue; |
3621 | ||
3622 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3623 | if (ret == -ENOSPC) |
3624 | break; | |
5a488b9d LB |
3625 | if (ret) { |
3626 | /* btrfs_shrink_device never returns ret > 0 */ | |
3627 | WARN_ON(ret > 0); | |
3628 | goto error; | |
3629 | } | |
ec44a35c | 3630 | |
a22285a6 | 3631 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3632 | if (IS_ERR(trans)) { |
3633 | ret = PTR_ERR(trans); | |
3634 | btrfs_info_in_rcu(fs_info, | |
3635 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3636 | rcu_str_deref(device->name), ret, | |
3637 | old_size, old_size - size_to_free); | |
3638 | goto error; | |
3639 | } | |
ec44a35c CM |
3640 | |
3641 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3642 | if (ret) { |
3a45bb20 | 3643 | btrfs_end_transaction(trans); |
5a488b9d LB |
3644 | /* btrfs_grow_device never returns ret > 0 */ |
3645 | WARN_ON(ret > 0); | |
3646 | btrfs_info_in_rcu(fs_info, | |
3647 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3648 | rcu_str_deref(device->name), ret, | |
3649 | old_size, old_size - size_to_free); | |
3650 | goto error; | |
3651 | } | |
ec44a35c | 3652 | |
3a45bb20 | 3653 | btrfs_end_transaction(trans); |
ec44a35c CM |
3654 | } |
3655 | ||
3656 | /* step two, relocate all the chunks */ | |
3657 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3658 | if (!path) { |
3659 | ret = -ENOMEM; | |
3660 | goto error; | |
3661 | } | |
19a39dce ID |
3662 | |
3663 | /* zero out stat counters */ | |
3664 | spin_lock(&fs_info->balance_lock); | |
3665 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3666 | spin_unlock(&fs_info->balance_lock); | |
3667 | again: | |
7d824b6f | 3668 | if (!counting) { |
12907fc7 DS |
3669 | /* |
3670 | * The single value limit and min/max limits use the same bytes | |
3671 | * in the | |
3672 | */ | |
7d824b6f DS |
3673 | bctl->data.limit = limit_data; |
3674 | bctl->meta.limit = limit_meta; | |
3675 | bctl->sys.limit = limit_sys; | |
3676 | } | |
ec44a35c CM |
3677 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3678 | key.offset = (u64)-1; | |
3679 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3680 | ||
d397712b | 3681 | while (1) { |
19a39dce | 3682 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3683 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3684 | ret = -ECANCELED; |
3685 | goto error; | |
3686 | } | |
3687 | ||
67c5e7d4 | 3688 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3689 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3690 | if (ret < 0) { |
3691 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3692 | goto error; |
67c5e7d4 | 3693 | } |
ec44a35c CM |
3694 | |
3695 | /* | |
3696 | * this shouldn't happen, it means the last relocate | |
3697 | * failed | |
3698 | */ | |
3699 | if (ret == 0) | |
c9e9f97b | 3700 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3701 | |
3702 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3703 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3704 | if (ret) { |
67c5e7d4 | 3705 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3706 | ret = 0; |
ec44a35c | 3707 | break; |
c9e9f97b | 3708 | } |
7d9eb12c | 3709 | |
f43ffb60 ID |
3710 | leaf = path->nodes[0]; |
3711 | slot = path->slots[0]; | |
3712 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3713 | |
67c5e7d4 FM |
3714 | if (found_key.objectid != key.objectid) { |
3715 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3716 | break; |
67c5e7d4 | 3717 | } |
7d9eb12c | 3718 | |
f43ffb60 | 3719 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3720 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3721 | |
19a39dce ID |
3722 | if (!counting) { |
3723 | spin_lock(&fs_info->balance_lock); | |
3724 | bctl->stat.considered++; | |
3725 | spin_unlock(&fs_info->balance_lock); | |
3726 | } | |
3727 | ||
2ff7e61e | 3728 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3729 | found_key.offset); |
2c9fe835 | 3730 | |
b3b4aa74 | 3731 | btrfs_release_path(path); |
67c5e7d4 FM |
3732 | if (!ret) { |
3733 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3734 | goto loop; |
67c5e7d4 | 3735 | } |
f43ffb60 | 3736 | |
19a39dce | 3737 | if (counting) { |
67c5e7d4 | 3738 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3739 | spin_lock(&fs_info->balance_lock); |
3740 | bctl->stat.expected++; | |
3741 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3742 | |
3743 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3744 | count_data++; | |
3745 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3746 | count_sys++; | |
3747 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3748 | count_meta++; | |
3749 | ||
3750 | goto loop; | |
3751 | } | |
3752 | ||
3753 | /* | |
3754 | * Apply limit_min filter, no need to check if the LIMITS | |
3755 | * filter is used, limit_min is 0 by default | |
3756 | */ | |
3757 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3758 | count_data < bctl->data.limit_min) | |
3759 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3760 | count_meta < bctl->meta.limit_min) | |
3761 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3762 | count_sys < bctl->sys.limit_min)) { | |
3763 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3764 | goto loop; |
3765 | } | |
3766 | ||
a6f93c71 LB |
3767 | if (!chunk_reserved) { |
3768 | /* | |
3769 | * We may be relocating the only data chunk we have, | |
3770 | * which could potentially end up with losing data's | |
3771 | * raid profile, so lets allocate an empty one in | |
3772 | * advance. | |
3773 | */ | |
3774 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3775 | found_key.offset); | |
2c9fe835 ZL |
3776 | if (ret < 0) { |
3777 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3778 | goto error; | |
a6f93c71 LB |
3779 | } else if (ret == 1) { |
3780 | chunk_reserved = 1; | |
2c9fe835 | 3781 | } |
2c9fe835 ZL |
3782 | } |
3783 | ||
5b4aacef | 3784 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3785 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3786 | if (ret && ret != -ENOSPC) |
3787 | goto error; | |
19a39dce | 3788 | if (ret == -ENOSPC) { |
c9e9f97b | 3789 | enospc_errors++; |
19a39dce ID |
3790 | } else { |
3791 | spin_lock(&fs_info->balance_lock); | |
3792 | bctl->stat.completed++; | |
3793 | spin_unlock(&fs_info->balance_lock); | |
3794 | } | |
f43ffb60 | 3795 | loop: |
795a3321 ID |
3796 | if (found_key.offset == 0) |
3797 | break; | |
ba1bf481 | 3798 | key.offset = found_key.offset - 1; |
ec44a35c | 3799 | } |
c9e9f97b | 3800 | |
19a39dce ID |
3801 | if (counting) { |
3802 | btrfs_release_path(path); | |
3803 | counting = false; | |
3804 | goto again; | |
3805 | } | |
ec44a35c CM |
3806 | error: |
3807 | btrfs_free_path(path); | |
c9e9f97b | 3808 | if (enospc_errors) { |
efe120a0 | 3809 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3810 | enospc_errors); |
c9e9f97b ID |
3811 | if (!ret) |
3812 | ret = -ENOSPC; | |
3813 | } | |
3814 | ||
ec44a35c CM |
3815 | return ret; |
3816 | } | |
3817 | ||
0c460c0d ID |
3818 | /** |
3819 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3820 | * @flags: profile to validate | |
3821 | * @extended: if true @flags is treated as an extended profile | |
3822 | */ | |
3823 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3824 | { | |
3825 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3826 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3827 | ||
3828 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3829 | ||
3830 | /* 1) check that all other bits are zeroed */ | |
3831 | if (flags & ~mask) | |
3832 | return 0; | |
3833 | ||
3834 | /* 2) see if profile is reduced */ | |
3835 | if (flags == 0) | |
3836 | return !extended; /* "0" is valid for usual profiles */ | |
3837 | ||
3838 | /* true if exactly one bit set */ | |
3839 | return (flags & (flags - 1)) == 0; | |
3840 | } | |
3841 | ||
837d5b6e ID |
3842 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3843 | { | |
a7e99c69 ID |
3844 | /* cancel requested || normal exit path */ |
3845 | return atomic_read(&fs_info->balance_cancel_req) || | |
3846 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3847 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3848 | } |
3849 | ||
c9e9f97b ID |
3850 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3851 | { | |
0940ebf6 ID |
3852 | int ret; |
3853 | ||
c9e9f97b | 3854 | unset_balance_control(fs_info); |
6bccf3ab | 3855 | ret = del_balance_item(fs_info); |
0f788c58 | 3856 | if (ret) |
34d97007 | 3857 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f | 3858 | |
171938e5 | 3859 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
c9e9f97b ID |
3860 | } |
3861 | ||
bdcd3c97 AM |
3862 | /* Non-zero return value signifies invalidity */ |
3863 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3864 | u64 allowed) | |
3865 | { | |
3866 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3867 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3868 | (bctl_arg->target & ~allowed))); | |
3869 | } | |
3870 | ||
c9e9f97b ID |
3871 | /* |
3872 | * Should be called with both balance and volume mutexes held | |
3873 | */ | |
3874 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3875 | struct btrfs_ioctl_balance_args *bargs) | |
3876 | { | |
3877 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
14506127 | 3878 | u64 meta_target, data_target; |
f43ffb60 | 3879 | u64 allowed; |
e4837f8f | 3880 | int mixed = 0; |
c9e9f97b | 3881 | int ret; |
8dabb742 | 3882 | u64 num_devices; |
de98ced9 | 3883 | unsigned seq; |
c9e9f97b | 3884 | |
837d5b6e | 3885 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3886 | atomic_read(&fs_info->balance_pause_req) || |
3887 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3888 | ret = -EINVAL; |
3889 | goto out; | |
3890 | } | |
3891 | ||
e4837f8f ID |
3892 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3893 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3894 | mixed = 1; | |
3895 | ||
f43ffb60 ID |
3896 | /* |
3897 | * In case of mixed groups both data and meta should be picked, | |
3898 | * and identical options should be given for both of them. | |
3899 | */ | |
e4837f8f ID |
3900 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3901 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3902 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3903 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3904 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3905 | btrfs_err(fs_info, |
3906 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3907 | ret = -EINVAL; |
3908 | goto out; | |
3909 | } | |
3910 | } | |
3911 | ||
8dabb742 | 3912 | num_devices = fs_info->fs_devices->num_devices; |
7e79cb86 | 3913 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
8dabb742 SB |
3914 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3915 | BUG_ON(num_devices < 1); | |
3916 | num_devices--; | |
3917 | } | |
7e79cb86 | 3918 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
88be159c AH |
3919 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3920 | if (num_devices > 1) | |
e4d8ec0f | 3921 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3922 | if (num_devices > 2) |
3923 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3924 | if (num_devices > 3) | |
3925 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3926 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3927 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3928 | btrfs_err(fs_info, |
3929 | "unable to start balance with target data profile %llu", | |
3930 | bctl->data.target); | |
e4d8ec0f ID |
3931 | ret = -EINVAL; |
3932 | goto out; | |
3933 | } | |
bdcd3c97 | 3934 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3935 | btrfs_err(fs_info, |
5d163e0e JM |
3936 | "unable to start balance with target metadata profile %llu", |
3937 | bctl->meta.target); | |
e4d8ec0f ID |
3938 | ret = -EINVAL; |
3939 | goto out; | |
3940 | } | |
bdcd3c97 | 3941 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3942 | btrfs_err(fs_info, |
5d163e0e JM |
3943 | "unable to start balance with target system profile %llu", |
3944 | bctl->sys.target); | |
e4d8ec0f ID |
3945 | ret = -EINVAL; |
3946 | goto out; | |
3947 | } | |
3948 | ||
e4d8ec0f ID |
3949 | /* allow to reduce meta or sys integrity only if force set */ |
3950 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3951 | BTRFS_BLOCK_GROUP_RAID10 | |
3952 | BTRFS_BLOCK_GROUP_RAID5 | | |
3953 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3954 | do { |
3955 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3956 | ||
3957 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3958 | (fs_info->avail_system_alloc_bits & allowed) && | |
3959 | !(bctl->sys.target & allowed)) || | |
3960 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3961 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3962 | !(bctl->meta.target & allowed))) { | |
3963 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3964 | btrfs_info(fs_info, |
3965 | "force reducing metadata integrity"); | |
de98ced9 | 3966 | } else { |
5d163e0e JM |
3967 | btrfs_err(fs_info, |
3968 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3969 | ret = -EINVAL; |
3970 | goto out; | |
3971 | } | |
e4d8ec0f | 3972 | } |
de98ced9 | 3973 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3974 | |
14506127 AB |
3975 | /* if we're not converting, the target field is uninitialized */ |
3976 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3977 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3978 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3979 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3980 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3981 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 3982 | btrfs_warn(fs_info, |
5d163e0e | 3983 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
14506127 | 3984 | meta_target, data_target); |
ee592d07 ST |
3985 | } |
3986 | ||
6bccf3ab | 3987 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3988 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3989 | goto out; |
3990 | ||
59641015 ID |
3991 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3992 | BUG_ON(ret == -EEXIST); | |
3993 | set_balance_control(bctl); | |
3994 | } else { | |
3995 | BUG_ON(ret != -EEXIST); | |
3996 | spin_lock(&fs_info->balance_lock); | |
3997 | update_balance_args(bctl); | |
3998 | spin_unlock(&fs_info->balance_lock); | |
3999 | } | |
c9e9f97b | 4000 | |
837d5b6e | 4001 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
4002 | mutex_unlock(&fs_info->balance_mutex); |
4003 | ||
4004 | ret = __btrfs_balance(fs_info); | |
4005 | ||
4006 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 4007 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
4008 | |
4009 | if (bargs) { | |
4010 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 4011 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
4012 | } |
4013 | ||
3a01aa7a ID |
4014 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4015 | balance_need_close(fs_info)) { | |
4016 | __cancel_balance(fs_info); | |
4017 | } | |
4018 | ||
837d5b6e | 4019 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4020 | |
4021 | return ret; | |
4022 | out: | |
59641015 ID |
4023 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
4024 | __cancel_balance(fs_info); | |
ed0fb78f | 4025 | else { |
59641015 | 4026 | kfree(bctl); |
171938e5 | 4027 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
ed0fb78f | 4028 | } |
59641015 ID |
4029 | return ret; |
4030 | } | |
4031 | ||
4032 | static int balance_kthread(void *data) | |
4033 | { | |
2b6ba629 | 4034 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4035 | int ret = 0; |
59641015 ID |
4036 | |
4037 | mutex_lock(&fs_info->volume_mutex); | |
4038 | mutex_lock(&fs_info->balance_mutex); | |
4039 | ||
2b6ba629 | 4040 | if (fs_info->balance_ctl) { |
efe120a0 | 4041 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 4042 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 4043 | } |
59641015 ID |
4044 | |
4045 | mutex_unlock(&fs_info->balance_mutex); | |
4046 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 4047 | |
59641015 ID |
4048 | return ret; |
4049 | } | |
4050 | ||
2b6ba629 ID |
4051 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4052 | { | |
4053 | struct task_struct *tsk; | |
4054 | ||
4055 | spin_lock(&fs_info->balance_lock); | |
4056 | if (!fs_info->balance_ctl) { | |
4057 | spin_unlock(&fs_info->balance_lock); | |
4058 | return 0; | |
4059 | } | |
4060 | spin_unlock(&fs_info->balance_lock); | |
4061 | ||
3cdde224 | 4062 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 4063 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
4064 | return 0; |
4065 | } | |
4066 | ||
4067 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 4068 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4069 | } |
4070 | ||
68310a5e | 4071 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4072 | { |
59641015 ID |
4073 | struct btrfs_balance_control *bctl; |
4074 | struct btrfs_balance_item *item; | |
4075 | struct btrfs_disk_balance_args disk_bargs; | |
4076 | struct btrfs_path *path; | |
4077 | struct extent_buffer *leaf; | |
4078 | struct btrfs_key key; | |
4079 | int ret; | |
4080 | ||
4081 | path = btrfs_alloc_path(); | |
4082 | if (!path) | |
4083 | return -ENOMEM; | |
4084 | ||
59641015 | 4085 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4086 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4087 | key.offset = 0; |
4088 | ||
68310a5e | 4089 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4090 | if (ret < 0) |
68310a5e | 4091 | goto out; |
59641015 ID |
4092 | if (ret > 0) { /* ret = -ENOENT; */ |
4093 | ret = 0; | |
68310a5e ID |
4094 | goto out; |
4095 | } | |
4096 | ||
4097 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4098 | if (!bctl) { | |
4099 | ret = -ENOMEM; | |
4100 | goto out; | |
59641015 ID |
4101 | } |
4102 | ||
4103 | leaf = path->nodes[0]; | |
4104 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4105 | ||
68310a5e ID |
4106 | bctl->fs_info = fs_info; |
4107 | bctl->flags = btrfs_balance_flags(leaf, item); | |
4108 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4109 | |
4110 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4111 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4112 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4113 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4114 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4115 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4116 | ||
171938e5 | 4117 | WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)); |
ed0fb78f | 4118 | |
68310a5e ID |
4119 | mutex_lock(&fs_info->volume_mutex); |
4120 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 4121 | |
68310a5e ID |
4122 | set_balance_control(bctl); |
4123 | ||
4124 | mutex_unlock(&fs_info->balance_mutex); | |
4125 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4126 | out: |
4127 | btrfs_free_path(path); | |
ec44a35c CM |
4128 | return ret; |
4129 | } | |
4130 | ||
837d5b6e ID |
4131 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4132 | { | |
4133 | int ret = 0; | |
4134 | ||
4135 | mutex_lock(&fs_info->balance_mutex); | |
4136 | if (!fs_info->balance_ctl) { | |
4137 | mutex_unlock(&fs_info->balance_mutex); | |
4138 | return -ENOTCONN; | |
4139 | } | |
4140 | ||
4141 | if (atomic_read(&fs_info->balance_running)) { | |
4142 | atomic_inc(&fs_info->balance_pause_req); | |
4143 | mutex_unlock(&fs_info->balance_mutex); | |
4144 | ||
4145 | wait_event(fs_info->balance_wait_q, | |
4146 | atomic_read(&fs_info->balance_running) == 0); | |
4147 | ||
4148 | mutex_lock(&fs_info->balance_mutex); | |
4149 | /* we are good with balance_ctl ripped off from under us */ | |
4150 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4151 | atomic_dec(&fs_info->balance_pause_req); | |
4152 | } else { | |
4153 | ret = -ENOTCONN; | |
4154 | } | |
4155 | ||
4156 | mutex_unlock(&fs_info->balance_mutex); | |
4157 | return ret; | |
4158 | } | |
4159 | ||
a7e99c69 ID |
4160 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4161 | { | |
bc98a42c | 4162 | if (sb_rdonly(fs_info->sb)) |
e649e587 ID |
4163 | return -EROFS; |
4164 | ||
a7e99c69 ID |
4165 | mutex_lock(&fs_info->balance_mutex); |
4166 | if (!fs_info->balance_ctl) { | |
4167 | mutex_unlock(&fs_info->balance_mutex); | |
4168 | return -ENOTCONN; | |
4169 | } | |
4170 | ||
4171 | atomic_inc(&fs_info->balance_cancel_req); | |
4172 | /* | |
4173 | * if we are running just wait and return, balance item is | |
4174 | * deleted in btrfs_balance in this case | |
4175 | */ | |
4176 | if (atomic_read(&fs_info->balance_running)) { | |
4177 | mutex_unlock(&fs_info->balance_mutex); | |
4178 | wait_event(fs_info->balance_wait_q, | |
4179 | atomic_read(&fs_info->balance_running) == 0); | |
4180 | mutex_lock(&fs_info->balance_mutex); | |
4181 | } else { | |
4182 | /* __cancel_balance needs volume_mutex */ | |
4183 | mutex_unlock(&fs_info->balance_mutex); | |
4184 | mutex_lock(&fs_info->volume_mutex); | |
4185 | mutex_lock(&fs_info->balance_mutex); | |
4186 | ||
4187 | if (fs_info->balance_ctl) | |
4188 | __cancel_balance(fs_info); | |
4189 | ||
4190 | mutex_unlock(&fs_info->volume_mutex); | |
4191 | } | |
4192 | ||
4193 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4194 | atomic_dec(&fs_info->balance_cancel_req); | |
4195 | mutex_unlock(&fs_info->balance_mutex); | |
4196 | return 0; | |
4197 | } | |
4198 | ||
803b2f54 SB |
4199 | static int btrfs_uuid_scan_kthread(void *data) |
4200 | { | |
4201 | struct btrfs_fs_info *fs_info = data; | |
4202 | struct btrfs_root *root = fs_info->tree_root; | |
4203 | struct btrfs_key key; | |
803b2f54 SB |
4204 | struct btrfs_path *path = NULL; |
4205 | int ret = 0; | |
4206 | struct extent_buffer *eb; | |
4207 | int slot; | |
4208 | struct btrfs_root_item root_item; | |
4209 | u32 item_size; | |
f45388f3 | 4210 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4211 | |
4212 | path = btrfs_alloc_path(); | |
4213 | if (!path) { | |
4214 | ret = -ENOMEM; | |
4215 | goto out; | |
4216 | } | |
4217 | ||
4218 | key.objectid = 0; | |
4219 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4220 | key.offset = 0; | |
4221 | ||
803b2f54 | 4222 | while (1) { |
7c829b72 AJ |
4223 | ret = btrfs_search_forward(root, &key, path, |
4224 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4225 | if (ret) { |
4226 | if (ret > 0) | |
4227 | ret = 0; | |
4228 | break; | |
4229 | } | |
4230 | ||
4231 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4232 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4233 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4234 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4235 | goto skip; | |
4236 | ||
4237 | eb = path->nodes[0]; | |
4238 | slot = path->slots[0]; | |
4239 | item_size = btrfs_item_size_nr(eb, slot); | |
4240 | if (item_size < sizeof(root_item)) | |
4241 | goto skip; | |
4242 | ||
803b2f54 SB |
4243 | read_extent_buffer(eb, &root_item, |
4244 | btrfs_item_ptr_offset(eb, slot), | |
4245 | (int)sizeof(root_item)); | |
4246 | if (btrfs_root_refs(&root_item) == 0) | |
4247 | goto skip; | |
f45388f3 FDBM |
4248 | |
4249 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4250 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4251 | if (trans) | |
4252 | goto update_tree; | |
4253 | ||
4254 | btrfs_release_path(path); | |
803b2f54 SB |
4255 | /* |
4256 | * 1 - subvol uuid item | |
4257 | * 1 - received_subvol uuid item | |
4258 | */ | |
4259 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4260 | if (IS_ERR(trans)) { | |
4261 | ret = PTR_ERR(trans); | |
4262 | break; | |
4263 | } | |
f45388f3 FDBM |
4264 | continue; |
4265 | } else { | |
4266 | goto skip; | |
4267 | } | |
4268 | update_tree: | |
4269 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
6bccf3ab | 4270 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4271 | root_item.uuid, |
4272 | BTRFS_UUID_KEY_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 | ||
4281 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
6bccf3ab | 4282 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4283 | root_item.received_uuid, |
4284 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4285 | key.objectid); | |
4286 | if (ret < 0) { | |
efe120a0 | 4287 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4288 | ret); |
803b2f54 SB |
4289 | break; |
4290 | } | |
4291 | } | |
4292 | ||
f45388f3 | 4293 | skip: |
803b2f54 | 4294 | if (trans) { |
3a45bb20 | 4295 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4296 | trans = NULL; |
803b2f54 SB |
4297 | if (ret) |
4298 | break; | |
4299 | } | |
4300 | ||
803b2f54 SB |
4301 | btrfs_release_path(path); |
4302 | if (key.offset < (u64)-1) { | |
4303 | key.offset++; | |
4304 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4305 | key.offset = 0; | |
4306 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4307 | } else if (key.objectid < (u64)-1) { | |
4308 | key.offset = 0; | |
4309 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4310 | key.objectid++; | |
4311 | } else { | |
4312 | break; | |
4313 | } | |
4314 | cond_resched(); | |
4315 | } | |
4316 | ||
4317 | out: | |
4318 | btrfs_free_path(path); | |
f45388f3 | 4319 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4320 | btrfs_end_transaction(trans); |
803b2f54 | 4321 | if (ret) |
efe120a0 | 4322 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4323 | else |
afcdd129 | 4324 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4325 | up(&fs_info->uuid_tree_rescan_sem); |
4326 | return 0; | |
4327 | } | |
4328 | ||
70f80175 SB |
4329 | /* |
4330 | * Callback for btrfs_uuid_tree_iterate(). | |
4331 | * returns: | |
4332 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4333 | * < 0 if an error occurred. |
70f80175 SB |
4334 | * > 0 if the check failed, which means the caller shall remove the entry. |
4335 | */ | |
4336 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4337 | u8 *uuid, u8 type, u64 subid) | |
4338 | { | |
4339 | struct btrfs_key key; | |
4340 | int ret = 0; | |
4341 | struct btrfs_root *subvol_root; | |
4342 | ||
4343 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4344 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4345 | goto out; | |
4346 | ||
4347 | key.objectid = subid; | |
4348 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4349 | key.offset = (u64)-1; | |
4350 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4351 | if (IS_ERR(subvol_root)) { | |
4352 | ret = PTR_ERR(subvol_root); | |
4353 | if (ret == -ENOENT) | |
4354 | ret = 1; | |
4355 | goto out; | |
4356 | } | |
4357 | ||
4358 | switch (type) { | |
4359 | case BTRFS_UUID_KEY_SUBVOL: | |
4360 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4361 | ret = 1; | |
4362 | break; | |
4363 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4364 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4365 | BTRFS_UUID_SIZE)) | |
4366 | ret = 1; | |
4367 | break; | |
4368 | } | |
4369 | ||
4370 | out: | |
4371 | return ret; | |
4372 | } | |
4373 | ||
4374 | static int btrfs_uuid_rescan_kthread(void *data) | |
4375 | { | |
4376 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4377 | int ret; | |
4378 | ||
4379 | /* | |
4380 | * 1st step is to iterate through the existing UUID tree and | |
4381 | * to delete all entries that contain outdated data. | |
4382 | * 2nd step is to add all missing entries to the UUID tree. | |
4383 | */ | |
4384 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4385 | if (ret < 0) { | |
efe120a0 | 4386 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4387 | up(&fs_info->uuid_tree_rescan_sem); |
4388 | return ret; | |
4389 | } | |
4390 | return btrfs_uuid_scan_kthread(data); | |
4391 | } | |
4392 | ||
f7a81ea4 SB |
4393 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4394 | { | |
4395 | struct btrfs_trans_handle *trans; | |
4396 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4397 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4398 | struct task_struct *task; |
4399 | int ret; | |
f7a81ea4 SB |
4400 | |
4401 | /* | |
4402 | * 1 - root node | |
4403 | * 1 - root item | |
4404 | */ | |
4405 | trans = btrfs_start_transaction(tree_root, 2); | |
4406 | if (IS_ERR(trans)) | |
4407 | return PTR_ERR(trans); | |
4408 | ||
4409 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4410 | BTRFS_UUID_TREE_OBJECTID); | |
4411 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4412 | ret = PTR_ERR(uuid_root); |
66642832 | 4413 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4414 | btrfs_end_transaction(trans); |
6d13f549 | 4415 | return ret; |
f7a81ea4 SB |
4416 | } |
4417 | ||
4418 | fs_info->uuid_root = uuid_root; | |
4419 | ||
3a45bb20 | 4420 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4421 | if (ret) |
4422 | return ret; | |
4423 | ||
4424 | down(&fs_info->uuid_tree_rescan_sem); | |
4425 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4426 | if (IS_ERR(task)) { | |
70f80175 | 4427 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4428 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4429 | up(&fs_info->uuid_tree_rescan_sem); |
4430 | return PTR_ERR(task); | |
4431 | } | |
4432 | ||
4433 | return 0; | |
f7a81ea4 | 4434 | } |
803b2f54 | 4435 | |
70f80175 SB |
4436 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4437 | { | |
4438 | struct task_struct *task; | |
4439 | ||
4440 | down(&fs_info->uuid_tree_rescan_sem); | |
4441 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4442 | if (IS_ERR(task)) { | |
4443 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4444 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4445 | up(&fs_info->uuid_tree_rescan_sem); |
4446 | return PTR_ERR(task); | |
4447 | } | |
4448 | ||
4449 | return 0; | |
4450 | } | |
4451 | ||
8f18cf13 CM |
4452 | /* |
4453 | * shrinking a device means finding all of the device extents past | |
4454 | * the new size, and then following the back refs to the chunks. | |
4455 | * The chunk relocation code actually frees the device extent | |
4456 | */ | |
4457 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4458 | { | |
0b246afa JM |
4459 | struct btrfs_fs_info *fs_info = device->fs_info; |
4460 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4461 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4462 | struct btrfs_dev_extent *dev_extent = NULL; |
4463 | struct btrfs_path *path; | |
4464 | u64 length; | |
8f18cf13 CM |
4465 | u64 chunk_offset; |
4466 | int ret; | |
4467 | int slot; | |
ba1bf481 JB |
4468 | int failed = 0; |
4469 | bool retried = false; | |
53e489bc | 4470 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4471 | struct extent_buffer *l; |
4472 | struct btrfs_key key; | |
0b246afa | 4473 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4474 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4475 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4476 | u64 diff; |
4477 | ||
4478 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4479 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4480 | |
401e29c1 | 4481 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4482 | return -EINVAL; |
4483 | ||
8f18cf13 CM |
4484 | path = btrfs_alloc_path(); |
4485 | if (!path) | |
4486 | return -ENOMEM; | |
4487 | ||
e4058b54 | 4488 | path->reada = READA_FORWARD; |
8f18cf13 | 4489 | |
34441361 | 4490 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4491 | |
7cc8e58d | 4492 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4493 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4494 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4495 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4496 | } |
34441361 | 4497 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4498 | |
ba1bf481 | 4499 | again: |
8f18cf13 CM |
4500 | key.objectid = device->devid; |
4501 | key.offset = (u64)-1; | |
4502 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4503 | ||
213e64da | 4504 | do { |
0b246afa | 4505 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4506 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4507 | if (ret < 0) { |
0b246afa | 4508 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4509 | goto done; |
67c5e7d4 | 4510 | } |
8f18cf13 CM |
4511 | |
4512 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4513 | if (ret) |
0b246afa | 4514 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4515 | if (ret < 0) |
4516 | goto done; | |
4517 | if (ret) { | |
4518 | ret = 0; | |
b3b4aa74 | 4519 | btrfs_release_path(path); |
bf1fb512 | 4520 | break; |
8f18cf13 CM |
4521 | } |
4522 | ||
4523 | l = path->nodes[0]; | |
4524 | slot = path->slots[0]; | |
4525 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4526 | ||
ba1bf481 | 4527 | if (key.objectid != device->devid) { |
0b246afa | 4528 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4529 | btrfs_release_path(path); |
bf1fb512 | 4530 | break; |
ba1bf481 | 4531 | } |
8f18cf13 CM |
4532 | |
4533 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4534 | length = btrfs_dev_extent_length(l, dev_extent); | |
4535 | ||
ba1bf481 | 4536 | if (key.offset + length <= new_size) { |
0b246afa | 4537 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4538 | btrfs_release_path(path); |
d6397bae | 4539 | break; |
ba1bf481 | 4540 | } |
8f18cf13 | 4541 | |
8f18cf13 | 4542 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4543 | btrfs_release_path(path); |
8f18cf13 | 4544 | |
a6f93c71 LB |
4545 | /* |
4546 | * We may be relocating the only data chunk we have, | |
4547 | * which could potentially end up with losing data's | |
4548 | * raid profile, so lets allocate an empty one in | |
4549 | * advance. | |
4550 | */ | |
4551 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4552 | if (ret < 0) { | |
4553 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4554 | goto done; | |
4555 | } | |
4556 | ||
0b246afa JM |
4557 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4558 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4559 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4560 | goto done; |
ba1bf481 JB |
4561 | if (ret == -ENOSPC) |
4562 | failed++; | |
213e64da | 4563 | } while (key.offset-- > 0); |
ba1bf481 JB |
4564 | |
4565 | if (failed && !retried) { | |
4566 | failed = 0; | |
4567 | retried = true; | |
4568 | goto again; | |
4569 | } else if (failed && retried) { | |
4570 | ret = -ENOSPC; | |
ba1bf481 | 4571 | goto done; |
8f18cf13 CM |
4572 | } |
4573 | ||
d6397bae | 4574 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4575 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4576 | if (IS_ERR(trans)) { |
4577 | ret = PTR_ERR(trans); | |
4578 | goto done; | |
4579 | } | |
4580 | ||
34441361 | 4581 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4582 | |
4583 | /* | |
4584 | * We checked in the above loop all device extents that were already in | |
4585 | * the device tree. However before we have updated the device's | |
4586 | * total_bytes to the new size, we might have had chunk allocations that | |
4587 | * have not complete yet (new block groups attached to transaction | |
4588 | * handles), and therefore their device extents were not yet in the | |
4589 | * device tree and we missed them in the loop above. So if we have any | |
4590 | * pending chunk using a device extent that overlaps the device range | |
4591 | * that we can not use anymore, commit the current transaction and | |
4592 | * repeat the search on the device tree - this way we guarantee we will | |
4593 | * not have chunks using device extents that end beyond 'new_size'. | |
4594 | */ | |
4595 | if (!checked_pending_chunks) { | |
4596 | u64 start = new_size; | |
4597 | u64 len = old_size - new_size; | |
4598 | ||
499f377f JM |
4599 | if (contains_pending_extent(trans->transaction, device, |
4600 | &start, len)) { | |
34441361 | 4601 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4602 | checked_pending_chunks = true; |
4603 | failed = 0; | |
4604 | retried = false; | |
3a45bb20 | 4605 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4606 | if (ret) |
4607 | goto done; | |
4608 | goto again; | |
4609 | } | |
4610 | } | |
4611 | ||
7cc8e58d | 4612 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4613 | if (list_empty(&device->resized_list)) |
4614 | list_add_tail(&device->resized_list, | |
0b246afa | 4615 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4616 | |
d6397bae | 4617 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4618 | btrfs_set_super_total_bytes(super_copy, |
4619 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4620 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4621 | |
4622 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4623 | ret = btrfs_update_device(trans, device); | |
3a45bb20 | 4624 | btrfs_end_transaction(trans); |
8f18cf13 CM |
4625 | done: |
4626 | btrfs_free_path(path); | |
53e489bc | 4627 | if (ret) { |
34441361 | 4628 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4629 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4630 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4631 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4632 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4633 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4634 | } |
8f18cf13 CM |
4635 | return ret; |
4636 | } | |
4637 | ||
2ff7e61e | 4638 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4639 | struct btrfs_key *key, |
4640 | struct btrfs_chunk *chunk, int item_size) | |
4641 | { | |
0b246afa | 4642 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4643 | struct btrfs_disk_key disk_key; |
4644 | u32 array_size; | |
4645 | u8 *ptr; | |
4646 | ||
34441361 | 4647 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4648 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4649 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4650 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4651 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4652 | return -EFBIG; |
fe48a5c0 | 4653 | } |
0b86a832 CM |
4654 | |
4655 | ptr = super_copy->sys_chunk_array + array_size; | |
4656 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4657 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4658 | ptr += sizeof(disk_key); | |
4659 | memcpy(ptr, chunk, item_size); | |
4660 | item_size += sizeof(disk_key); | |
4661 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4662 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4663 | |
0b86a832 CM |
4664 | return 0; |
4665 | } | |
4666 | ||
73c5de00 AJ |
4667 | /* |
4668 | * sort the devices in descending order by max_avail, total_avail | |
4669 | */ | |
4670 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4671 | { |
73c5de00 AJ |
4672 | const struct btrfs_device_info *di_a = a; |
4673 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4674 | |
73c5de00 | 4675 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4676 | return -1; |
73c5de00 | 4677 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4678 | return 1; |
73c5de00 AJ |
4679 | if (di_a->total_avail > di_b->total_avail) |
4680 | return -1; | |
4681 | if (di_a->total_avail < di_b->total_avail) | |
4682 | return 1; | |
4683 | return 0; | |
b2117a39 | 4684 | } |
0b86a832 | 4685 | |
53b381b3 DW |
4686 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4687 | { | |
ffe2d203 | 4688 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4689 | return; |
4690 | ||
ceda0864 | 4691 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4692 | } |
4693 | ||
062d4d1f | 4694 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
4695 | - sizeof(struct btrfs_chunk)) \ |
4696 | / sizeof(struct btrfs_stripe) + 1) | |
4697 | ||
4698 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4699 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4700 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4701 | / sizeof(struct btrfs_stripe) + 1) | |
4702 | ||
73c5de00 | 4703 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4704 | u64 start, u64 type) |
b2117a39 | 4705 | { |
2ff7e61e | 4706 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4707 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4708 | struct btrfs_device *device; |
73c5de00 AJ |
4709 | struct map_lookup *map = NULL; |
4710 | struct extent_map_tree *em_tree; | |
4711 | struct extent_map *em; | |
4712 | struct btrfs_device_info *devices_info = NULL; | |
4713 | u64 total_avail; | |
4714 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4715 | int data_stripes; /* number of stripes that count for |
4716 | block group size */ | |
73c5de00 AJ |
4717 | int sub_stripes; /* sub_stripes info for map */ |
4718 | int dev_stripes; /* stripes per dev */ | |
4719 | int devs_max; /* max devs to use */ | |
4720 | int devs_min; /* min devs needed */ | |
4721 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4722 | int ncopies; /* how many copies to data has */ | |
4723 | int ret; | |
4724 | u64 max_stripe_size; | |
4725 | u64 max_chunk_size; | |
4726 | u64 stripe_size; | |
4727 | u64 num_bytes; | |
4728 | int ndevs; | |
4729 | int i; | |
4730 | int j; | |
31e50229 | 4731 | int index; |
593060d7 | 4732 | |
0c460c0d | 4733 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4734 | |
4117f207 QW |
4735 | if (list_empty(&fs_devices->alloc_list)) { |
4736 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4737 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4738 | return -ENOSPC; |
4117f207 | 4739 | } |
b2117a39 | 4740 | |
3e72ee88 | 4741 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4742 | |
31e50229 LB |
4743 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4744 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4745 | devs_max = btrfs_raid_array[index].devs_max; | |
4746 | devs_min = btrfs_raid_array[index].devs_min; | |
4747 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4748 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4749 | |
9b3f68b9 | 4750 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4751 | max_stripe_size = SZ_1G; |
73c5de00 | 4752 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 | 4753 | if (!devs_max) |
062d4d1f | 4754 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 4755 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4756 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4757 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4758 | max_stripe_size = SZ_1G; | |
1100373f | 4759 | else |
ee22184b | 4760 | max_stripe_size = SZ_256M; |
73c5de00 | 4761 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 4762 | if (!devs_max) |
062d4d1f | 4763 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 4764 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4765 | max_stripe_size = SZ_32M; |
73c5de00 | 4766 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4767 | if (!devs_max) |
4768 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4769 | } else { |
351fd353 | 4770 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4771 | type); |
4772 | BUG_ON(1); | |
9b3f68b9 CM |
4773 | } |
4774 | ||
2b82032c YZ |
4775 | /* we don't want a chunk larger than 10% of writeable space */ |
4776 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4777 | max_chunk_size); | |
9b3f68b9 | 4778 | |
31e818fe | 4779 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4780 | GFP_NOFS); |
4781 | if (!devices_info) | |
4782 | return -ENOMEM; | |
0cad8a11 | 4783 | |
9f680ce0 | 4784 | /* |
73c5de00 AJ |
4785 | * in the first pass through the devices list, we gather information |
4786 | * about the available holes on each device. | |
9f680ce0 | 4787 | */ |
73c5de00 | 4788 | ndevs = 0; |
ebcc9301 | 4789 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4790 | u64 max_avail; |
4791 | u64 dev_offset; | |
b2117a39 | 4792 | |
ebbede42 | 4793 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4794 | WARN(1, KERN_ERR |
efe120a0 | 4795 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4796 | continue; |
4797 | } | |
b2117a39 | 4798 | |
e12c9621 AJ |
4799 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4800 | &device->dev_state) || | |
401e29c1 | 4801 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4802 | continue; |
b2117a39 | 4803 | |
73c5de00 AJ |
4804 | if (device->total_bytes > device->bytes_used) |
4805 | total_avail = device->total_bytes - device->bytes_used; | |
4806 | else | |
4807 | total_avail = 0; | |
38c01b96 | 4808 | |
4809 | /* If there is no space on this device, skip it. */ | |
4810 | if (total_avail == 0) | |
4811 | continue; | |
b2117a39 | 4812 | |
6df9a95e | 4813 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4814 | max_stripe_size * dev_stripes, |
4815 | &dev_offset, &max_avail); | |
4816 | if (ret && ret != -ENOSPC) | |
4817 | goto error; | |
b2117a39 | 4818 | |
73c5de00 AJ |
4819 | if (ret == 0) |
4820 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4821 | |
4117f207 QW |
4822 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4823 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4824 | btrfs_debug(info, | |
4825 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4826 | __func__, device->devid, max_avail, | |
4827 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4828 | continue; |
4117f207 | 4829 | } |
b2117a39 | 4830 | |
063d006f ES |
4831 | if (ndevs == fs_devices->rw_devices) { |
4832 | WARN(1, "%s: found more than %llu devices\n", | |
4833 | __func__, fs_devices->rw_devices); | |
4834 | break; | |
4835 | } | |
73c5de00 AJ |
4836 | devices_info[ndevs].dev_offset = dev_offset; |
4837 | devices_info[ndevs].max_avail = max_avail; | |
4838 | devices_info[ndevs].total_avail = total_avail; | |
4839 | devices_info[ndevs].dev = device; | |
4840 | ++ndevs; | |
4841 | } | |
b2117a39 | 4842 | |
73c5de00 AJ |
4843 | /* |
4844 | * now sort the devices by hole size / available space | |
4845 | */ | |
4846 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4847 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4848 | |
73c5de00 | 4849 | /* round down to number of usable stripes */ |
e5600fd6 | 4850 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4851 | |
ba89b802 | 4852 | if (ndevs < devs_min) { |
73c5de00 | 4853 | ret = -ENOSPC; |
4117f207 QW |
4854 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4855 | btrfs_debug(info, | |
4856 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 4857 | __func__, ndevs, devs_min); |
4117f207 | 4858 | } |
73c5de00 | 4859 | goto error; |
b2117a39 | 4860 | } |
9f680ce0 | 4861 | |
f148ef4d NB |
4862 | ndevs = min(ndevs, devs_max); |
4863 | ||
73c5de00 | 4864 | /* |
92e222df HK |
4865 | * The primary goal is to maximize the number of stripes, so use as |
4866 | * many devices as possible, even if the stripes are not maximum sized. | |
4867 | * | |
4868 | * The DUP profile stores more than one stripe per device, the | |
4869 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4870 | */ |
92e222df | 4871 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4872 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4873 | |
53b381b3 DW |
4874 | /* |
4875 | * this will have to be fixed for RAID1 and RAID10 over | |
4876 | * more drives | |
4877 | */ | |
4878 | data_stripes = num_stripes / ncopies; | |
4879 | ||
500ceed8 | 4880 | if (type & BTRFS_BLOCK_GROUP_RAID5) |
53b381b3 | 4881 | data_stripes = num_stripes - 1; |
500ceed8 NB |
4882 | |
4883 | if (type & BTRFS_BLOCK_GROUP_RAID6) | |
53b381b3 | 4884 | data_stripes = num_stripes - 2; |
86db2578 CM |
4885 | |
4886 | /* | |
4887 | * Use the number of data stripes to figure out how big this chunk | |
4888 | * is really going to be in terms of logical address space, | |
4889 | * and compare that answer with the max chunk size | |
4890 | */ | |
4891 | if (stripe_size * data_stripes > max_chunk_size) { | |
b8b93add | 4892 | stripe_size = div_u64(max_chunk_size, data_stripes); |
86db2578 CM |
4893 | |
4894 | /* bump the answer up to a 16MB boundary */ | |
793ff2c8 | 4895 | stripe_size = round_up(stripe_size, SZ_16M); |
86db2578 | 4896 | |
793ff2c8 QW |
4897 | /* |
4898 | * But don't go higher than the limits we found while searching | |
4899 | * for free extents | |
86db2578 | 4900 | */ |
793ff2c8 QW |
4901 | stripe_size = min(devices_info[ndevs - 1].max_avail, |
4902 | stripe_size); | |
86db2578 CM |
4903 | } |
4904 | ||
37db63a4 | 4905 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 4906 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4907 | |
4908 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4909 | if (!map) { | |
4910 | ret = -ENOMEM; | |
4911 | goto error; | |
4912 | } | |
4913 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4914 | |
73c5de00 AJ |
4915 | for (i = 0; i < ndevs; ++i) { |
4916 | for (j = 0; j < dev_stripes; ++j) { | |
4917 | int s = i * dev_stripes + j; | |
4918 | map->stripes[s].dev = devices_info[i].dev; | |
4919 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4920 | j * stripe_size; | |
6324fbf3 | 4921 | } |
6324fbf3 | 4922 | } |
500ceed8 NB |
4923 | map->stripe_len = BTRFS_STRIPE_LEN; |
4924 | map->io_align = BTRFS_STRIPE_LEN; | |
4925 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4926 | map->type = type; |
2b82032c | 4927 | map->sub_stripes = sub_stripes; |
0b86a832 | 4928 | |
53b381b3 | 4929 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4930 | |
6bccf3ab | 4931 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4932 | |
172ddd60 | 4933 | em = alloc_extent_map(); |
2b82032c | 4934 | if (!em) { |
298a8f9c | 4935 | kfree(map); |
b2117a39 MX |
4936 | ret = -ENOMEM; |
4937 | goto error; | |
593060d7 | 4938 | } |
298a8f9c | 4939 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4940 | em->map_lookup = map; |
2b82032c | 4941 | em->start = start; |
73c5de00 | 4942 | em->len = num_bytes; |
2b82032c YZ |
4943 | em->block_start = 0; |
4944 | em->block_len = em->len; | |
6df9a95e | 4945 | em->orig_block_len = stripe_size; |
593060d7 | 4946 | |
0b246afa | 4947 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4948 | write_lock(&em_tree->lock); |
09a2a8f9 | 4949 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 4950 | if (ret) { |
1efb72a3 | 4951 | write_unlock(&em_tree->lock); |
0f5d42b2 | 4952 | free_extent_map(em); |
1dd4602f | 4953 | goto error; |
0f5d42b2 | 4954 | } |
0b86a832 | 4955 | |
1efb72a3 NB |
4956 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
4957 | refcount_inc(&em->refs); | |
4958 | write_unlock(&em_tree->lock); | |
4959 | ||
0174484d | 4960 | ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes); |
6df9a95e JB |
4961 | if (ret) |
4962 | goto error_del_extent; | |
2b82032c | 4963 | |
7cc8e58d MX |
4964 | for (i = 0; i < map->num_stripes; i++) { |
4965 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4966 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4967 | } | |
43530c46 | 4968 | |
a5ed45f8 | 4969 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4970 | |
0f5d42b2 | 4971 | free_extent_map(em); |
0b246afa | 4972 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4973 | |
b2117a39 | 4974 | kfree(devices_info); |
2b82032c | 4975 | return 0; |
b2117a39 | 4976 | |
6df9a95e | 4977 | error_del_extent: |
0f5d42b2 JB |
4978 | write_lock(&em_tree->lock); |
4979 | remove_extent_mapping(em_tree, em); | |
4980 | write_unlock(&em_tree->lock); | |
4981 | ||
4982 | /* One for our allocation */ | |
4983 | free_extent_map(em); | |
4984 | /* One for the tree reference */ | |
4985 | free_extent_map(em); | |
495e64f4 FM |
4986 | /* One for the pending_chunks list reference */ |
4987 | free_extent_map(em); | |
b2117a39 | 4988 | error: |
b2117a39 MX |
4989 | kfree(devices_info); |
4990 | return ret; | |
2b82032c YZ |
4991 | } |
4992 | ||
6df9a95e | 4993 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
6bccf3ab | 4994 | struct btrfs_fs_info *fs_info, |
6df9a95e | 4995 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4996 | { |
6bccf3ab JM |
4997 | struct btrfs_root *extent_root = fs_info->extent_root; |
4998 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4999 | struct btrfs_key key; |
2b82032c YZ |
5000 | struct btrfs_device *device; |
5001 | struct btrfs_chunk *chunk; | |
5002 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5003 | struct extent_map *em; |
5004 | struct map_lookup *map; | |
5005 | size_t item_size; | |
5006 | u64 dev_offset; | |
5007 | u64 stripe_size; | |
5008 | int i = 0; | |
140e639f | 5009 | int ret = 0; |
2b82032c | 5010 | |
592d92ee LB |
5011 | em = get_chunk_map(fs_info, chunk_offset, chunk_size); |
5012 | if (IS_ERR(em)) | |
5013 | return PTR_ERR(em); | |
6df9a95e | 5014 | |
95617d69 | 5015 | map = em->map_lookup; |
6df9a95e JB |
5016 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5017 | stripe_size = em->orig_block_len; | |
5018 | ||
2b82032c | 5019 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5020 | if (!chunk) { |
5021 | ret = -ENOMEM; | |
5022 | goto out; | |
5023 | } | |
5024 | ||
50460e37 FM |
5025 | /* |
5026 | * Take the device list mutex to prevent races with the final phase of | |
5027 | * a device replace operation that replaces the device object associated | |
5028 | * with the map's stripes, because the device object's id can change | |
5029 | * at any time during that final phase of the device replace operation | |
5030 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5031 | */ | |
0b246afa | 5032 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5033 | for (i = 0; i < map->num_stripes; i++) { |
5034 | device = map->stripes[i].dev; | |
5035 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5036 | |
0b86a832 | 5037 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5038 | if (ret) |
50460e37 | 5039 | break; |
b5d9071c NB |
5040 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5041 | dev_offset, stripe_size); | |
6df9a95e | 5042 | if (ret) |
50460e37 FM |
5043 | break; |
5044 | } | |
5045 | if (ret) { | |
0b246afa | 5046 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5047 | goto out; |
2b82032c YZ |
5048 | } |
5049 | ||
2b82032c | 5050 | stripe = &chunk->stripe; |
6df9a95e JB |
5051 | for (i = 0; i < map->num_stripes; i++) { |
5052 | device = map->stripes[i].dev; | |
5053 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5054 | |
e17cade2 CM |
5055 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5056 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5057 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5058 | stripe++; |
0b86a832 | 5059 | } |
0b246afa | 5060 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5061 | |
2b82032c | 5062 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5063 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5064 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5065 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5066 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5067 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5068 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5069 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5070 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5071 | |
2b82032c YZ |
5072 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5073 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5074 | key.offset = chunk_offset; | |
0b86a832 | 5075 | |
2b82032c | 5076 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5077 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5078 | /* | |
5079 | * TODO: Cleanup of inserted chunk root in case of | |
5080 | * failure. | |
5081 | */ | |
2ff7e61e | 5082 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5083 | } |
1abe9b8a | 5084 | |
6df9a95e | 5085 | out: |
0b86a832 | 5086 | kfree(chunk); |
6df9a95e | 5087 | free_extent_map(em); |
4ed1d16e | 5088 | return ret; |
2b82032c | 5089 | } |
0b86a832 | 5090 | |
2b82032c YZ |
5091 | /* |
5092 | * Chunk allocation falls into two parts. The first part does works | |
5093 | * that make the new allocated chunk useable, but not do any operation | |
5094 | * that modifies the chunk tree. The second part does the works that | |
5095 | * require modifying the chunk tree. This division is important for the | |
5096 | * bootstrap process of adding storage to a seed btrfs. | |
5097 | */ | |
5098 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 5099 | struct btrfs_fs_info *fs_info, u64 type) |
2b82032c YZ |
5100 | { |
5101 | u64 chunk_offset; | |
2b82032c | 5102 | |
a32bf9a3 | 5103 | lockdep_assert_held(&fs_info->chunk_mutex); |
0b246afa | 5104 | chunk_offset = find_next_chunk(fs_info); |
72b468c8 | 5105 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5106 | } |
5107 | ||
d397712b | 5108 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5109 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5110 | { |
5111 | u64 chunk_offset; | |
5112 | u64 sys_chunk_offset; | |
2b82032c | 5113 | u64 alloc_profile; |
2b82032c YZ |
5114 | int ret; |
5115 | ||
6df9a95e | 5116 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5117 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5118 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5119 | if (ret) |
5120 | return ret; | |
2b82032c | 5121 | |
0b246afa | 5122 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5123 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5124 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5125 | return ret; |
2b82032c YZ |
5126 | } |
5127 | ||
d20983b4 MX |
5128 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5129 | { | |
5130 | int max_errors; | |
5131 | ||
5132 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5133 | BTRFS_BLOCK_GROUP_RAID10 | | |
5134 | BTRFS_BLOCK_GROUP_RAID5 | | |
5135 | BTRFS_BLOCK_GROUP_DUP)) { | |
5136 | max_errors = 1; | |
5137 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5138 | max_errors = 2; | |
5139 | } else { | |
5140 | max_errors = 0; | |
005d6427 | 5141 | } |
2b82032c | 5142 | |
d20983b4 | 5143 | return max_errors; |
2b82032c YZ |
5144 | } |
5145 | ||
2ff7e61e | 5146 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5147 | { |
5148 | struct extent_map *em; | |
5149 | struct map_lookup *map; | |
2b82032c | 5150 | int readonly = 0; |
d20983b4 | 5151 | int miss_ndevs = 0; |
2b82032c YZ |
5152 | int i; |
5153 | ||
592d92ee LB |
5154 | em = get_chunk_map(fs_info, chunk_offset, 1); |
5155 | if (IS_ERR(em)) | |
2b82032c YZ |
5156 | return 1; |
5157 | ||
95617d69 | 5158 | map = em->map_lookup; |
2b82032c | 5159 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5160 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5161 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5162 | miss_ndevs++; |
5163 | continue; | |
5164 | } | |
ebbede42 AJ |
5165 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5166 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5167 | readonly = 1; |
d20983b4 | 5168 | goto end; |
2b82032c YZ |
5169 | } |
5170 | } | |
d20983b4 MX |
5171 | |
5172 | /* | |
5173 | * If the number of missing devices is larger than max errors, | |
5174 | * we can not write the data into that chunk successfully, so | |
5175 | * set it readonly. | |
5176 | */ | |
5177 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5178 | readonly = 1; | |
5179 | end: | |
0b86a832 | 5180 | free_extent_map(em); |
2b82032c | 5181 | return readonly; |
0b86a832 CM |
5182 | } |
5183 | ||
5184 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5185 | { | |
a8067e02 | 5186 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5187 | } |
5188 | ||
5189 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5190 | { | |
5191 | struct extent_map *em; | |
5192 | ||
d397712b | 5193 | while (1) { |
890871be | 5194 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5195 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5196 | if (em) | |
5197 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5198 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5199 | if (!em) |
5200 | break; | |
0b86a832 CM |
5201 | /* once for us */ |
5202 | free_extent_map(em); | |
5203 | /* once for the tree */ | |
5204 | free_extent_map(em); | |
5205 | } | |
5206 | } | |
5207 | ||
5d964051 | 5208 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5209 | { |
5210 | struct extent_map *em; | |
5211 | struct map_lookup *map; | |
f188591e CM |
5212 | int ret; |
5213 | ||
592d92ee LB |
5214 | em = get_chunk_map(fs_info, logical, len); |
5215 | if (IS_ERR(em)) | |
5216 | /* | |
5217 | * We could return errors for these cases, but that could get | |
5218 | * ugly and we'd probably do the same thing which is just not do | |
5219 | * anything else and exit, so return 1 so the callers don't try | |
5220 | * to use other copies. | |
5221 | */ | |
fb7669b5 | 5222 | return 1; |
fb7669b5 | 5223 | |
95617d69 | 5224 | map = em->map_lookup; |
f188591e CM |
5225 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5226 | ret = map->num_stripes; | |
321aecc6 CM |
5227 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5228 | ret = map->sub_stripes; | |
53b381b3 DW |
5229 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5230 | ret = 2; | |
5231 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5232 | /* |
5233 | * There could be two corrupted data stripes, we need | |
5234 | * to loop retry in order to rebuild the correct data. | |
5235 | * | |
5236 | * Fail a stripe at a time on every retry except the | |
5237 | * stripe under reconstruction. | |
5238 | */ | |
5239 | ret = map->num_stripes; | |
f188591e CM |
5240 | else |
5241 | ret = 1; | |
5242 | free_extent_map(em); | |
ad6d620e | 5243 | |
7e79cb86 | 5244 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
6fad823f LB |
5245 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5246 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5247 | ret++; |
7e79cb86 | 5248 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
ad6d620e | 5249 | |
f188591e CM |
5250 | return ret; |
5251 | } | |
5252 | ||
2ff7e61e | 5253 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5254 | u64 logical) |
5255 | { | |
5256 | struct extent_map *em; | |
5257 | struct map_lookup *map; | |
0b246afa | 5258 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5259 | |
592d92ee | 5260 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5261 | |
69f03f13 NB |
5262 | if (!WARN_ON(IS_ERR(em))) { |
5263 | map = em->map_lookup; | |
5264 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5265 | len = map->stripe_len * nr_data_stripes(map); | |
5266 | free_extent_map(em); | |
5267 | } | |
53b381b3 DW |
5268 | return len; |
5269 | } | |
5270 | ||
e4ff5fb5 | 5271 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5272 | { |
5273 | struct extent_map *em; | |
5274 | struct map_lookup *map; | |
53b381b3 DW |
5275 | int ret = 0; |
5276 | ||
592d92ee | 5277 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5278 | |
69f03f13 NB |
5279 | if(!WARN_ON(IS_ERR(em))) { |
5280 | map = em->map_lookup; | |
5281 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5282 | ret = 1; | |
5283 | free_extent_map(em); | |
5284 | } | |
53b381b3 DW |
5285 | return ret; |
5286 | } | |
5287 | ||
30d9861f | 5288 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5289 | struct map_lookup *map, int first, |
8ba0ae78 | 5290 | int dev_replace_is_ongoing) |
dfe25020 CM |
5291 | { |
5292 | int i; | |
99f92a7c | 5293 | int num_stripes; |
8ba0ae78 | 5294 | int preferred_mirror; |
30d9861f SB |
5295 | int tolerance; |
5296 | struct btrfs_device *srcdev; | |
5297 | ||
99f92a7c AJ |
5298 | ASSERT((map->type & |
5299 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5300 | ||
5301 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5302 | num_stripes = map->sub_stripes; | |
5303 | else | |
5304 | num_stripes = map->num_stripes; | |
5305 | ||
8ba0ae78 AJ |
5306 | preferred_mirror = first + current->pid % num_stripes; |
5307 | ||
30d9861f SB |
5308 | if (dev_replace_is_ongoing && |
5309 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5310 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5311 | srcdev = fs_info->dev_replace.srcdev; | |
5312 | else | |
5313 | srcdev = NULL; | |
5314 | ||
5315 | /* | |
5316 | * try to avoid the drive that is the source drive for a | |
5317 | * dev-replace procedure, only choose it if no other non-missing | |
5318 | * mirror is available | |
5319 | */ | |
5320 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5321 | if (map->stripes[preferred_mirror].dev->bdev && |
5322 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5323 | return preferred_mirror; | |
99f92a7c | 5324 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5325 | if (map->stripes[i].dev->bdev && |
5326 | (tolerance || map->stripes[i].dev != srcdev)) | |
5327 | return i; | |
5328 | } | |
dfe25020 | 5329 | } |
30d9861f | 5330 | |
dfe25020 CM |
5331 | /* we couldn't find one that doesn't fail. Just return something |
5332 | * and the io error handling code will clean up eventually | |
5333 | */ | |
8ba0ae78 | 5334 | return preferred_mirror; |
dfe25020 CM |
5335 | } |
5336 | ||
53b381b3 DW |
5337 | static inline int parity_smaller(u64 a, u64 b) |
5338 | { | |
5339 | return a > b; | |
5340 | } | |
5341 | ||
5342 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5343 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5344 | { |
5345 | struct btrfs_bio_stripe s; | |
5346 | int i; | |
5347 | u64 l; | |
5348 | int again = 1; | |
5349 | ||
5350 | while (again) { | |
5351 | again = 0; | |
cc7539ed | 5352 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5353 | if (parity_smaller(bbio->raid_map[i], |
5354 | bbio->raid_map[i+1])) { | |
53b381b3 | 5355 | s = bbio->stripes[i]; |
8e5cfb55 | 5356 | l = bbio->raid_map[i]; |
53b381b3 | 5357 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5358 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5359 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5360 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5361 | |
53b381b3 DW |
5362 | again = 1; |
5363 | } | |
5364 | } | |
5365 | } | |
5366 | } | |
5367 | ||
6e9606d2 ZL |
5368 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5369 | { | |
5370 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5371 | /* the size of the btrfs_bio */ |
6e9606d2 | 5372 | sizeof(struct btrfs_bio) + |
e57cf21e | 5373 | /* plus the variable array for the stripes */ |
6e9606d2 | 5374 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5375 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5376 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5377 | /* |
5378 | * plus the raid_map, which includes both the tgt dev | |
5379 | * and the stripes | |
5380 | */ | |
5381 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5382 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5383 | |
5384 | atomic_set(&bbio->error, 0); | |
140475ae | 5385 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5386 | |
5387 | return bbio; | |
5388 | } | |
5389 | ||
5390 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5391 | { | |
140475ae ER |
5392 | WARN_ON(!refcount_read(&bbio->refs)); |
5393 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5394 | } |
5395 | ||
5396 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5397 | { | |
5398 | if (!bbio) | |
5399 | return; | |
140475ae | 5400 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5401 | kfree(bbio); |
5402 | } | |
5403 | ||
0b3d4cd3 LB |
5404 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5405 | /* | |
5406 | * Please note that, discard won't be sent to target device of device | |
5407 | * replace. | |
5408 | */ | |
5409 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5410 | u64 logical, u64 length, | |
5411 | struct btrfs_bio **bbio_ret) | |
5412 | { | |
5413 | struct extent_map *em; | |
5414 | struct map_lookup *map; | |
5415 | struct btrfs_bio *bbio; | |
5416 | u64 offset; | |
5417 | u64 stripe_nr; | |
5418 | u64 stripe_nr_end; | |
5419 | u64 stripe_end_offset; | |
5420 | u64 stripe_cnt; | |
5421 | u64 stripe_len; | |
5422 | u64 stripe_offset; | |
5423 | u64 num_stripes; | |
5424 | u32 stripe_index; | |
5425 | u32 factor = 0; | |
5426 | u32 sub_stripes = 0; | |
5427 | u64 stripes_per_dev = 0; | |
5428 | u32 remaining_stripes = 0; | |
5429 | u32 last_stripe = 0; | |
5430 | int ret = 0; | |
5431 | int i; | |
5432 | ||
5433 | /* discard always return a bbio */ | |
5434 | ASSERT(bbio_ret); | |
5435 | ||
5436 | em = get_chunk_map(fs_info, logical, length); | |
5437 | if (IS_ERR(em)) | |
5438 | return PTR_ERR(em); | |
5439 | ||
5440 | map = em->map_lookup; | |
5441 | /* we don't discard raid56 yet */ | |
5442 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5443 | ret = -EOPNOTSUPP; | |
5444 | goto out; | |
5445 | } | |
5446 | ||
5447 | offset = logical - em->start; | |
5448 | length = min_t(u64, em->len - offset, length); | |
5449 | ||
5450 | stripe_len = map->stripe_len; | |
5451 | /* | |
5452 | * stripe_nr counts the total number of stripes we have to stride | |
5453 | * to get to this block | |
5454 | */ | |
5455 | stripe_nr = div64_u64(offset, stripe_len); | |
5456 | ||
5457 | /* stripe_offset is the offset of this block in its stripe */ | |
5458 | stripe_offset = offset - stripe_nr * stripe_len; | |
5459 | ||
5460 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5461 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5462 | stripe_cnt = stripe_nr_end - stripe_nr; |
5463 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5464 | (offset + length); | |
5465 | /* | |
5466 | * after this, stripe_nr is the number of stripes on this | |
5467 | * device we have to walk to find the data, and stripe_index is | |
5468 | * the number of our device in the stripe array | |
5469 | */ | |
5470 | num_stripes = 1; | |
5471 | stripe_index = 0; | |
5472 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5473 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5474 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5475 | sub_stripes = 1; | |
5476 | else | |
5477 | sub_stripes = map->sub_stripes; | |
5478 | ||
5479 | factor = map->num_stripes / sub_stripes; | |
5480 | num_stripes = min_t(u64, map->num_stripes, | |
5481 | sub_stripes * stripe_cnt); | |
5482 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5483 | stripe_index *= sub_stripes; | |
5484 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5485 | &remaining_stripes); | |
5486 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5487 | last_stripe *= sub_stripes; | |
5488 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5489 | BTRFS_BLOCK_GROUP_DUP)) { | |
5490 | num_stripes = map->num_stripes; | |
5491 | } else { | |
5492 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5493 | &stripe_index); | |
5494 | } | |
5495 | ||
5496 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5497 | if (!bbio) { | |
5498 | ret = -ENOMEM; | |
5499 | goto out; | |
5500 | } | |
5501 | ||
5502 | for (i = 0; i < num_stripes; i++) { | |
5503 | bbio->stripes[i].physical = | |
5504 | map->stripes[stripe_index].physical + | |
5505 | stripe_offset + stripe_nr * map->stripe_len; | |
5506 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5507 | ||
5508 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5509 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5510 | bbio->stripes[i].length = stripes_per_dev * | |
5511 | map->stripe_len; | |
5512 | ||
5513 | if (i / sub_stripes < remaining_stripes) | |
5514 | bbio->stripes[i].length += | |
5515 | map->stripe_len; | |
5516 | ||
5517 | /* | |
5518 | * Special for the first stripe and | |
5519 | * the last stripe: | |
5520 | * | |
5521 | * |-------|...|-------| | |
5522 | * |----------| | |
5523 | * off end_off | |
5524 | */ | |
5525 | if (i < sub_stripes) | |
5526 | bbio->stripes[i].length -= | |
5527 | stripe_offset; | |
5528 | ||
5529 | if (stripe_index >= last_stripe && | |
5530 | stripe_index <= (last_stripe + | |
5531 | sub_stripes - 1)) | |
5532 | bbio->stripes[i].length -= | |
5533 | stripe_end_offset; | |
5534 | ||
5535 | if (i == sub_stripes - 1) | |
5536 | stripe_offset = 0; | |
5537 | } else { | |
5538 | bbio->stripes[i].length = length; | |
5539 | } | |
5540 | ||
5541 | stripe_index++; | |
5542 | if (stripe_index == map->num_stripes) { | |
5543 | stripe_index = 0; | |
5544 | stripe_nr++; | |
5545 | } | |
5546 | } | |
5547 | ||
5548 | *bbio_ret = bbio; | |
5549 | bbio->map_type = map->type; | |
5550 | bbio->num_stripes = num_stripes; | |
5551 | out: | |
5552 | free_extent_map(em); | |
5553 | return ret; | |
5554 | } | |
5555 | ||
5ab56090 LB |
5556 | /* |
5557 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5558 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5559 | * left cursor can also be read from the target drive. | |
5560 | * | |
5561 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5562 | * array of stripes. | |
5563 | * For READ, it also needs to be supported using the same mirror number. | |
5564 | * | |
5565 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5566 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5567 | * case. | |
5568 | */ | |
5569 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5570 | u64 logical, u64 length, | |
5571 | u64 srcdev_devid, int *mirror_num, | |
5572 | u64 *physical) | |
5573 | { | |
5574 | struct btrfs_bio *bbio = NULL; | |
5575 | int num_stripes; | |
5576 | int index_srcdev = 0; | |
5577 | int found = 0; | |
5578 | u64 physical_of_found = 0; | |
5579 | int i; | |
5580 | int ret = 0; | |
5581 | ||
5582 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5583 | logical, &length, &bbio, 0, 0); | |
5584 | if (ret) { | |
5585 | ASSERT(bbio == NULL); | |
5586 | return ret; | |
5587 | } | |
5588 | ||
5589 | num_stripes = bbio->num_stripes; | |
5590 | if (*mirror_num > num_stripes) { | |
5591 | /* | |
5592 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5593 | * that means that the requested area is not left of the left | |
5594 | * cursor | |
5595 | */ | |
5596 | btrfs_put_bbio(bbio); | |
5597 | return -EIO; | |
5598 | } | |
5599 | ||
5600 | /* | |
5601 | * process the rest of the function using the mirror_num of the source | |
5602 | * drive. Therefore look it up first. At the end, patch the device | |
5603 | * pointer to the one of the target drive. | |
5604 | */ | |
5605 | for (i = 0; i < num_stripes; i++) { | |
5606 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5607 | continue; | |
5608 | ||
5609 | /* | |
5610 | * In case of DUP, in order to keep it simple, only add the | |
5611 | * mirror with the lowest physical address | |
5612 | */ | |
5613 | if (found && | |
5614 | physical_of_found <= bbio->stripes[i].physical) | |
5615 | continue; | |
5616 | ||
5617 | index_srcdev = i; | |
5618 | found = 1; | |
5619 | physical_of_found = bbio->stripes[i].physical; | |
5620 | } | |
5621 | ||
5622 | btrfs_put_bbio(bbio); | |
5623 | ||
5624 | ASSERT(found); | |
5625 | if (!found) | |
5626 | return -EIO; | |
5627 | ||
5628 | *mirror_num = index_srcdev + 1; | |
5629 | *physical = physical_of_found; | |
5630 | return ret; | |
5631 | } | |
5632 | ||
73c0f228 LB |
5633 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5634 | struct btrfs_bio **bbio_ret, | |
5635 | struct btrfs_dev_replace *dev_replace, | |
5636 | int *num_stripes_ret, int *max_errors_ret) | |
5637 | { | |
5638 | struct btrfs_bio *bbio = *bbio_ret; | |
5639 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5640 | int tgtdev_indexes = 0; | |
5641 | int num_stripes = *num_stripes_ret; | |
5642 | int max_errors = *max_errors_ret; | |
5643 | int i; | |
5644 | ||
5645 | if (op == BTRFS_MAP_WRITE) { | |
5646 | int index_where_to_add; | |
5647 | ||
5648 | /* | |
5649 | * duplicate the write operations while the dev replace | |
5650 | * procedure is running. Since the copying of the old disk to | |
5651 | * the new disk takes place at run time while the filesystem is | |
5652 | * mounted writable, the regular write operations to the old | |
5653 | * disk have to be duplicated to go to the new disk as well. | |
5654 | * | |
5655 | * Note that device->missing is handled by the caller, and that | |
5656 | * the write to the old disk is already set up in the stripes | |
5657 | * array. | |
5658 | */ | |
5659 | index_where_to_add = num_stripes; | |
5660 | for (i = 0; i < num_stripes; i++) { | |
5661 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5662 | /* write to new disk, too */ | |
5663 | struct btrfs_bio_stripe *new = | |
5664 | bbio->stripes + index_where_to_add; | |
5665 | struct btrfs_bio_stripe *old = | |
5666 | bbio->stripes + i; | |
5667 | ||
5668 | new->physical = old->physical; | |
5669 | new->length = old->length; | |
5670 | new->dev = dev_replace->tgtdev; | |
5671 | bbio->tgtdev_map[i] = index_where_to_add; | |
5672 | index_where_to_add++; | |
5673 | max_errors++; | |
5674 | tgtdev_indexes++; | |
5675 | } | |
5676 | } | |
5677 | num_stripes = index_where_to_add; | |
5678 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5679 | int index_srcdev = 0; | |
5680 | int found = 0; | |
5681 | u64 physical_of_found = 0; | |
5682 | ||
5683 | /* | |
5684 | * During the dev-replace procedure, the target drive can also | |
5685 | * be used to read data in case it is needed to repair a corrupt | |
5686 | * block elsewhere. This is possible if the requested area is | |
5687 | * left of the left cursor. In this area, the target drive is a | |
5688 | * full copy of the source drive. | |
5689 | */ | |
5690 | for (i = 0; i < num_stripes; i++) { | |
5691 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5692 | /* | |
5693 | * In case of DUP, in order to keep it simple, | |
5694 | * only add the mirror with the lowest physical | |
5695 | * address | |
5696 | */ | |
5697 | if (found && | |
5698 | physical_of_found <= | |
5699 | bbio->stripes[i].physical) | |
5700 | continue; | |
5701 | index_srcdev = i; | |
5702 | found = 1; | |
5703 | physical_of_found = bbio->stripes[i].physical; | |
5704 | } | |
5705 | } | |
5706 | if (found) { | |
5707 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5708 | bbio->stripes + num_stripes; | |
5709 | ||
5710 | tgtdev_stripe->physical = physical_of_found; | |
5711 | tgtdev_stripe->length = | |
5712 | bbio->stripes[index_srcdev].length; | |
5713 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5714 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5715 | ||
5716 | tgtdev_indexes++; | |
5717 | num_stripes++; | |
5718 | } | |
5719 | } | |
5720 | ||
5721 | *num_stripes_ret = num_stripes; | |
5722 | *max_errors_ret = max_errors; | |
5723 | bbio->num_tgtdevs = tgtdev_indexes; | |
5724 | *bbio_ret = bbio; | |
5725 | } | |
5726 | ||
2b19a1fe LB |
5727 | static bool need_full_stripe(enum btrfs_map_op op) |
5728 | { | |
5729 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5730 | } | |
5731 | ||
cf8cddd3 CH |
5732 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5733 | enum btrfs_map_op op, | |
f2d8d74d | 5734 | u64 logical, u64 *length, |
a1d3c478 | 5735 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5736 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5737 | { |
5738 | struct extent_map *em; | |
5739 | struct map_lookup *map; | |
0b86a832 | 5740 | u64 offset; |
593060d7 CM |
5741 | u64 stripe_offset; |
5742 | u64 stripe_nr; | |
53b381b3 | 5743 | u64 stripe_len; |
9d644a62 | 5744 | u32 stripe_index; |
cea9e445 | 5745 | int i; |
de11cc12 | 5746 | int ret = 0; |
f2d8d74d | 5747 | int num_stripes; |
a236aed1 | 5748 | int max_errors = 0; |
2c8cdd6e | 5749 | int tgtdev_indexes = 0; |
a1d3c478 | 5750 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5751 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5752 | int dev_replace_is_ongoing = 0; | |
5753 | int num_alloc_stripes; | |
ad6d620e SB |
5754 | int patch_the_first_stripe_for_dev_replace = 0; |
5755 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5756 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5757 | |
0b3d4cd3 LB |
5758 | if (op == BTRFS_MAP_DISCARD) |
5759 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5760 | *length, bbio_ret); | |
5761 | ||
592d92ee LB |
5762 | em = get_chunk_map(fs_info, logical, *length); |
5763 | if (IS_ERR(em)) | |
5764 | return PTR_ERR(em); | |
0b86a832 | 5765 | |
95617d69 | 5766 | map = em->map_lookup; |
0b86a832 | 5767 | offset = logical - em->start; |
593060d7 | 5768 | |
53b381b3 | 5769 | stripe_len = map->stripe_len; |
593060d7 CM |
5770 | stripe_nr = offset; |
5771 | /* | |
5772 | * stripe_nr counts the total number of stripes we have to stride | |
5773 | * to get to this block | |
5774 | */ | |
47c5713f | 5775 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5776 | |
53b381b3 | 5777 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5778 | if (offset < stripe_offset) { |
5d163e0e JM |
5779 | btrfs_crit(fs_info, |
5780 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5781 | stripe_offset, offset, em->start, logical, |
5782 | stripe_len); | |
5783 | free_extent_map(em); | |
5784 | return -EINVAL; | |
5785 | } | |
593060d7 CM |
5786 | |
5787 | /* stripe_offset is the offset of this block in its stripe*/ | |
5788 | stripe_offset = offset - stripe_offset; | |
5789 | ||
53b381b3 | 5790 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5791 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5792 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5793 | raid56_full_stripe_start = offset; | |
5794 | ||
5795 | /* allow a write of a full stripe, but make sure we don't | |
5796 | * allow straddling of stripes | |
5797 | */ | |
47c5713f DS |
5798 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5799 | full_stripe_len); | |
53b381b3 DW |
5800 | raid56_full_stripe_start *= full_stripe_len; |
5801 | } | |
5802 | ||
0b3d4cd3 | 5803 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5804 | u64 max_len; |
5805 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5806 | For other RAID types and for RAID[56] reads, just allow a single | |
5807 | stripe (on a single disk). */ | |
ffe2d203 | 5808 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5809 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5810 | max_len = stripe_len * nr_data_stripes(map) - |
5811 | (offset - raid56_full_stripe_start); | |
5812 | } else { | |
5813 | /* we limit the length of each bio to what fits in a stripe */ | |
5814 | max_len = stripe_len - stripe_offset; | |
5815 | } | |
5816 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5817 | } else { |
5818 | *length = em->len - offset; | |
5819 | } | |
f2d8d74d | 5820 | |
53b381b3 DW |
5821 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5822 | it cares about is the length */ | |
a1d3c478 | 5823 | if (!bbio_ret) |
cea9e445 CM |
5824 | goto out; |
5825 | ||
7e79cb86 | 5826 | btrfs_dev_replace_read_lock(dev_replace); |
472262f3 SB |
5827 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5828 | if (!dev_replace_is_ongoing) | |
7e79cb86 | 5829 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 LB |
5830 | else |
5831 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5832 | |
ad6d620e | 5833 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5834 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5835 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5836 | dev_replace->srcdev->devid, | |
5837 | &mirror_num, | |
5838 | &physical_to_patch_in_first_stripe); | |
5839 | if (ret) | |
ad6d620e | 5840 | goto out; |
5ab56090 LB |
5841 | else |
5842 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5843 | } else if (mirror_num > map->num_stripes) { |
5844 | mirror_num = 0; | |
5845 | } | |
5846 | ||
f2d8d74d | 5847 | num_stripes = 1; |
cea9e445 | 5848 | stripe_index = 0; |
fce3bb9a | 5849 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5850 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5851 | &stripe_index); | |
de483734 | 5852 | if (!need_full_stripe(op)) |
28e1cc7d | 5853 | mirror_num = 1; |
fce3bb9a | 5854 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 5855 | if (need_full_stripe(op)) |
f2d8d74d | 5856 | num_stripes = map->num_stripes; |
2fff734f | 5857 | else if (mirror_num) |
f188591e | 5858 | stripe_index = mirror_num - 1; |
dfe25020 | 5859 | else { |
30d9861f | 5860 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 5861 | dev_replace_is_ongoing); |
a1d3c478 | 5862 | mirror_num = stripe_index + 1; |
dfe25020 | 5863 | } |
2fff734f | 5864 | |
611f0e00 | 5865 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5866 | if (need_full_stripe(op)) { |
f2d8d74d | 5867 | num_stripes = map->num_stripes; |
a1d3c478 | 5868 | } else if (mirror_num) { |
f188591e | 5869 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5870 | } else { |
5871 | mirror_num = 1; | |
5872 | } | |
2fff734f | 5873 | |
321aecc6 | 5874 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5875 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5876 | |
47c5713f | 5877 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5878 | stripe_index *= map->sub_stripes; |
5879 | ||
de483734 | 5880 | if (need_full_stripe(op)) |
f2d8d74d | 5881 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5882 | else if (mirror_num) |
5883 | stripe_index += mirror_num - 1; | |
dfe25020 | 5884 | else { |
3e74317a | 5885 | int old_stripe_index = stripe_index; |
30d9861f SB |
5886 | stripe_index = find_live_mirror(fs_info, map, |
5887 | stripe_index, | |
30d9861f | 5888 | dev_replace_is_ongoing); |
3e74317a | 5889 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5890 | } |
53b381b3 | 5891 | |
ffe2d203 | 5892 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5893 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5894 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5895 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5896 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5897 | |
5898 | /* RAID[56] write or recovery. Return all stripes */ | |
5899 | num_stripes = map->num_stripes; | |
5900 | max_errors = nr_parity_stripes(map); | |
5901 | ||
53b381b3 DW |
5902 | *length = map->stripe_len; |
5903 | stripe_index = 0; | |
5904 | stripe_offset = 0; | |
5905 | } else { | |
5906 | /* | |
5907 | * Mirror #0 or #1 means the original data block. | |
5908 | * Mirror #2 is RAID5 parity block. | |
5909 | * Mirror #3 is RAID6 Q block. | |
5910 | */ | |
47c5713f DS |
5911 | stripe_nr = div_u64_rem(stripe_nr, |
5912 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5913 | if (mirror_num > 1) |
5914 | stripe_index = nr_data_stripes(map) + | |
5915 | mirror_num - 2; | |
5916 | ||
5917 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5918 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5919 | &stripe_index); | |
de483734 | 5920 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 5921 | mirror_num = 1; |
53b381b3 | 5922 | } |
8790d502 CM |
5923 | } else { |
5924 | /* | |
47c5713f DS |
5925 | * after this, stripe_nr is the number of stripes on this |
5926 | * device we have to walk to find the data, and stripe_index is | |
5927 | * the number of our device in the stripe array | |
8790d502 | 5928 | */ |
47c5713f DS |
5929 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5930 | &stripe_index); | |
a1d3c478 | 5931 | mirror_num = stripe_index + 1; |
8790d502 | 5932 | } |
e042d1ec | 5933 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5934 | btrfs_crit(fs_info, |
5935 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5936 | stripe_index, map->num_stripes); |
5937 | ret = -EINVAL; | |
5938 | goto out; | |
5939 | } | |
cea9e445 | 5940 | |
472262f3 | 5941 | num_alloc_stripes = num_stripes; |
6fad823f | 5942 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5943 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5944 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5945 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5946 | num_alloc_stripes++; |
2c8cdd6e | 5947 | tgtdev_indexes = num_stripes; |
ad6d620e | 5948 | } |
2c8cdd6e | 5949 | |
6e9606d2 | 5950 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5951 | if (!bbio) { |
5952 | ret = -ENOMEM; | |
5953 | goto out; | |
5954 | } | |
6fad823f | 5955 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5956 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5957 | |
8e5cfb55 | 5958 | /* build raid_map */ |
2b19a1fe LB |
5959 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5960 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5961 | u64 tmp; |
9d644a62 | 5962 | unsigned rot; |
8e5cfb55 ZL |
5963 | |
5964 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5965 | sizeof(struct btrfs_bio_stripe) * | |
5966 | num_alloc_stripes + | |
5967 | sizeof(int) * tgtdev_indexes); | |
5968 | ||
5969 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5970 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5971 | |
5972 | /* Fill in the logical address of each stripe */ | |
5973 | tmp = stripe_nr * nr_data_stripes(map); | |
5974 | for (i = 0; i < nr_data_stripes(map); i++) | |
5975 | bbio->raid_map[(i+rot) % num_stripes] = | |
5976 | em->start + (tmp + i) * map->stripe_len; | |
5977 | ||
5978 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5979 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5980 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5981 | RAID6_Q_STRIPE; | |
5982 | } | |
5983 | ||
b89203f7 | 5984 | |
0b3d4cd3 LB |
5985 | for (i = 0; i < num_stripes; i++) { |
5986 | bbio->stripes[i].physical = | |
5987 | map->stripes[stripe_index].physical + | |
5988 | stripe_offset + | |
5989 | stripe_nr * map->stripe_len; | |
5990 | bbio->stripes[i].dev = | |
5991 | map->stripes[stripe_index].dev; | |
5992 | stripe_index++; | |
593060d7 | 5993 | } |
de11cc12 | 5994 | |
2b19a1fe | 5995 | if (need_full_stripe(op)) |
d20983b4 | 5996 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5997 | |
8e5cfb55 ZL |
5998 | if (bbio->raid_map) |
5999 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6000 | |
73c0f228 | 6001 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6002 | need_full_stripe(op)) { |
73c0f228 LB |
6003 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6004 | &max_errors); | |
472262f3 SB |
6005 | } |
6006 | ||
de11cc12 | 6007 | *bbio_ret = bbio; |
10f11900 | 6008 | bbio->map_type = map->type; |
de11cc12 LZ |
6009 | bbio->num_stripes = num_stripes; |
6010 | bbio->max_errors = max_errors; | |
6011 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6012 | |
6013 | /* | |
6014 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6015 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6016 | * available as a mirror | |
6017 | */ | |
6018 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6019 | WARN_ON(num_stripes > 1); | |
6020 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6021 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6022 | bbio->mirror_num = map->num_stripes + 1; | |
6023 | } | |
cea9e445 | 6024 | out: |
73beece9 LB |
6025 | if (dev_replace_is_ongoing) { |
6026 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
7e79cb86 | 6027 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 | 6028 | } |
0b86a832 | 6029 | free_extent_map(em); |
de11cc12 | 6030 | return ret; |
0b86a832 CM |
6031 | } |
6032 | ||
cf8cddd3 | 6033 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6034 | u64 logical, u64 *length, |
a1d3c478 | 6035 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6036 | { |
b3d3fa51 | 6037 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6038 | mirror_num, 0); |
f2d8d74d CM |
6039 | } |
6040 | ||
af8e2d1d | 6041 | /* For Scrub/replace */ |
cf8cddd3 | 6042 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6043 | u64 logical, u64 *length, |
825ad4c9 | 6044 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6045 | { |
825ad4c9 | 6046 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6047 | } |
6048 | ||
ab8d0fc4 | 6049 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
6050 | u64 chunk_start, u64 physical, u64 devid, |
6051 | u64 **logical, int *naddrs, int *stripe_len) | |
6052 | { | |
a512bbf8 YZ |
6053 | struct extent_map *em; |
6054 | struct map_lookup *map; | |
6055 | u64 *buf; | |
6056 | u64 bytenr; | |
6057 | u64 length; | |
6058 | u64 stripe_nr; | |
53b381b3 | 6059 | u64 rmap_len; |
a512bbf8 YZ |
6060 | int i, j, nr = 0; |
6061 | ||
592d92ee LB |
6062 | em = get_chunk_map(fs_info, chunk_start, 1); |
6063 | if (IS_ERR(em)) | |
835d974f | 6064 | return -EIO; |
835d974f | 6065 | |
95617d69 | 6066 | map = em->map_lookup; |
a512bbf8 | 6067 | length = em->len; |
53b381b3 DW |
6068 | rmap_len = map->stripe_len; |
6069 | ||
a512bbf8 | 6070 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6071 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6072 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6073 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6074 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6075 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6076 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6077 | } | |
a512bbf8 | 6078 | |
31e818fe | 6079 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6080 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6081 | |
6082 | for (i = 0; i < map->num_stripes; i++) { | |
6083 | if (devid && map->stripes[i].dev->devid != devid) | |
6084 | continue; | |
6085 | if (map->stripes[i].physical > physical || | |
6086 | map->stripes[i].physical + length <= physical) | |
6087 | continue; | |
6088 | ||
6089 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6090 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6091 | |
6092 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6093 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6094 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6095 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6096 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6097 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6098 | * Alternatively, just use rmap_len below instead of | |
6099 | * map->stripe_len */ | |
6100 | ||
6101 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6102 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6103 | for (j = 0; j < nr; j++) { |
6104 | if (buf[j] == bytenr) | |
6105 | break; | |
6106 | } | |
934d375b CM |
6107 | if (j == nr) { |
6108 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6109 | buf[nr++] = bytenr; |
934d375b | 6110 | } |
a512bbf8 YZ |
6111 | } |
6112 | ||
a512bbf8 YZ |
6113 | *logical = buf; |
6114 | *naddrs = nr; | |
53b381b3 | 6115 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6116 | |
6117 | free_extent_map(em); | |
6118 | return 0; | |
f2d8d74d CM |
6119 | } |
6120 | ||
4246a0b6 | 6121 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6122 | { |
326e1dbb MS |
6123 | bio->bi_private = bbio->private; |
6124 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6125 | bio_endio(bio); |
326e1dbb | 6126 | |
6e9606d2 | 6127 | btrfs_put_bbio(bbio); |
8408c716 MX |
6128 | } |
6129 | ||
4246a0b6 | 6130 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6131 | { |
9be3395b | 6132 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6133 | int is_orig_bio = 0; |
8790d502 | 6134 | |
4e4cbee9 | 6135 | if (bio->bi_status) { |
a1d3c478 | 6136 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6137 | if (bio->bi_status == BLK_STS_IOERR || |
6138 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6139 | unsigned int stripe_index = |
9be3395b | 6140 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6141 | struct btrfs_device *dev; |
442a4f63 SB |
6142 | |
6143 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6144 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6145 | if (dev->bdev) { |
37226b21 | 6146 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6147 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6148 | BTRFS_DEV_STAT_WRITE_ERRS); |
6149 | else | |
1cb34c8e | 6150 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6151 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6152 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6153 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6154 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6155 | } |
442a4f63 SB |
6156 | } |
6157 | } | |
8790d502 | 6158 | |
a1d3c478 | 6159 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6160 | is_orig_bio = 1; |
6161 | ||
c404e0dc MX |
6162 | btrfs_bio_counter_dec(bbio->fs_info); |
6163 | ||
a1d3c478 | 6164 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6165 | if (!is_orig_bio) { |
6166 | bio_put(bio); | |
a1d3c478 | 6167 | bio = bbio->orig_bio; |
7d2b4daa | 6168 | } |
c7b22bb1 | 6169 | |
9be3395b | 6170 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6171 | /* only send an error to the higher layers if it is |
53b381b3 | 6172 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6173 | */ |
a1d3c478 | 6174 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6175 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6176 | } else { |
1259ab75 CM |
6177 | /* |
6178 | * this bio is actually up to date, we didn't | |
6179 | * go over the max number of errors | |
6180 | */ | |
2dbe0c77 | 6181 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6182 | } |
c55f1396 | 6183 | |
4246a0b6 | 6184 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6185 | } else if (!is_orig_bio) { |
8790d502 CM |
6186 | bio_put(bio); |
6187 | } | |
8790d502 CM |
6188 | } |
6189 | ||
8b712842 CM |
6190 | /* |
6191 | * see run_scheduled_bios for a description of why bios are collected for | |
6192 | * async submit. | |
6193 | * | |
6194 | * This will add one bio to the pending list for a device and make sure | |
6195 | * the work struct is scheduled. | |
6196 | */ | |
2ff7e61e | 6197 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6198 | struct bio *bio) |
8b712842 | 6199 | { |
0b246afa | 6200 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6201 | int should_queue = 1; |
ffbd517d | 6202 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6203 | |
e6e674bd AJ |
6204 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) || |
6205 | !device->bdev) { | |
4246a0b6 | 6206 | bio_io_error(bio); |
53b381b3 DW |
6207 | return; |
6208 | } | |
6209 | ||
8b712842 | 6210 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6211 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6212 | btrfsic_submit_bio(bio); |
143bede5 | 6213 | return; |
8b712842 CM |
6214 | } |
6215 | ||
492bb6de | 6216 | WARN_ON(bio->bi_next); |
8b712842 | 6217 | bio->bi_next = NULL; |
8b712842 CM |
6218 | |
6219 | spin_lock(&device->io_lock); | |
67f055c7 | 6220 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6221 | pending_bios = &device->pending_sync_bios; |
6222 | else | |
6223 | pending_bios = &device->pending_bios; | |
8b712842 | 6224 | |
ffbd517d CM |
6225 | if (pending_bios->tail) |
6226 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6227 | |
ffbd517d CM |
6228 | pending_bios->tail = bio; |
6229 | if (!pending_bios->head) | |
6230 | pending_bios->head = bio; | |
8b712842 CM |
6231 | if (device->running_pending) |
6232 | should_queue = 0; | |
6233 | ||
6234 | spin_unlock(&device->io_lock); | |
6235 | ||
6236 | if (should_queue) | |
0b246afa | 6237 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6238 | } |
6239 | ||
2ff7e61e JM |
6240 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6241 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6242 | { |
6243 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6244 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6245 | |
6246 | bio->bi_private = bbio; | |
9be3395b | 6247 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6248 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6249 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6250 | #ifdef DEBUG |
6251 | { | |
6252 | struct rcu_string *name; | |
6253 | ||
6254 | rcu_read_lock(); | |
6255 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6256 | btrfs_debug(fs_info, |
6257 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6258 | bio_op(bio), bio->bi_opf, | |
6259 | (u64)bio->bi_iter.bi_sector, | |
6260 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6261 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6262 | rcu_read_unlock(); |
6263 | } | |
6264 | #endif | |
74d46992 | 6265 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6266 | |
2ff7e61e | 6267 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6268 | |
de1ee92a | 6269 | if (async) |
2ff7e61e | 6270 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6271 | else |
4e49ea4a | 6272 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6273 | } |
6274 | ||
de1ee92a JB |
6275 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6276 | { | |
6277 | atomic_inc(&bbio->error); | |
6278 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6279 | /* Should be the original bio. */ |
8408c716 MX |
6280 | WARN_ON(bio != bbio->orig_bio); |
6281 | ||
9be3395b | 6282 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6283 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6284 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6285 | bio->bi_status = BLK_STS_IOERR; | |
6286 | else | |
6287 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6288 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6289 | } |
6290 | } | |
6291 | ||
58efbc9f OS |
6292 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6293 | int mirror_num, int async_submit) | |
0b86a832 | 6294 | { |
0b86a832 | 6295 | struct btrfs_device *dev; |
8790d502 | 6296 | struct bio *first_bio = bio; |
4f024f37 | 6297 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6298 | u64 length = 0; |
6299 | u64 map_length; | |
0b86a832 | 6300 | int ret; |
08da757d ZL |
6301 | int dev_nr; |
6302 | int total_devs; | |
a1d3c478 | 6303 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6304 | |
4f024f37 | 6305 | length = bio->bi_iter.bi_size; |
0b86a832 | 6306 | map_length = length; |
cea9e445 | 6307 | |
0b246afa | 6308 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6309 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6310 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6311 | if (ret) { |
0b246afa | 6312 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6313 | return errno_to_blk_status(ret); |
c404e0dc | 6314 | } |
cea9e445 | 6315 | |
a1d3c478 | 6316 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6317 | bbio->orig_bio = first_bio; |
6318 | bbio->private = first_bio->bi_private; | |
6319 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6320 | bbio->fs_info = fs_info; |
53b381b3 DW |
6321 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6322 | ||
ad1ba2a0 | 6323 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6324 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6325 | /* In this case, map_length has been set to the length of |
6326 | a single stripe; not the whole write */ | |
37226b21 | 6327 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6328 | ret = raid56_parity_write(fs_info, bio, bbio, |
6329 | map_length); | |
53b381b3 | 6330 | } else { |
2ff7e61e JM |
6331 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6332 | map_length, mirror_num, 1); | |
53b381b3 | 6333 | } |
4245215d | 6334 | |
0b246afa | 6335 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6336 | return errno_to_blk_status(ret); |
53b381b3 DW |
6337 | } |
6338 | ||
cea9e445 | 6339 | if (map_length < length) { |
0b246afa | 6340 | btrfs_crit(fs_info, |
5d163e0e JM |
6341 | "mapping failed logical %llu bio len %llu len %llu", |
6342 | logical, length, map_length); | |
cea9e445 CM |
6343 | BUG(); |
6344 | } | |
a1d3c478 | 6345 | |
08da757d | 6346 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6347 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6348 | if (!dev || !dev->bdev || |
ebbede42 AJ |
6349 | (bio_op(first_bio) == REQ_OP_WRITE && |
6350 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6351 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6352 | continue; |
6353 | } | |
6354 | ||
3aa8e074 | 6355 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6356 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6357 | else |
a1d3c478 | 6358 | bio = first_bio; |
de1ee92a | 6359 | |
2ff7e61e JM |
6360 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6361 | dev_nr, async_submit); | |
8790d502 | 6362 | } |
0b246afa | 6363 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6364 | return BLK_STS_OK; |
0b86a832 CM |
6365 | } |
6366 | ||
aa1b8cd4 | 6367 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6368 | u8 *uuid, u8 *fsid) |
0b86a832 | 6369 | { |
2b82032c YZ |
6370 | struct btrfs_device *device; |
6371 | struct btrfs_fs_devices *cur_devices; | |
6372 | ||
aa1b8cd4 | 6373 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6374 | while (cur_devices) { |
6375 | if (!fsid || | |
44880fdc | 6376 | !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6377 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6378 | if (device) |
6379 | return device; | |
6380 | } | |
6381 | cur_devices = cur_devices->seed; | |
6382 | } | |
6383 | return NULL; | |
0b86a832 CM |
6384 | } |
6385 | ||
2ff7e61e | 6386 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6387 | u64 devid, u8 *dev_uuid) |
6388 | { | |
6389 | struct btrfs_device *device; | |
dfe25020 | 6390 | |
12bd2fc0 ID |
6391 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6392 | if (IS_ERR(device)) | |
adfb69af | 6393 | return device; |
12bd2fc0 ID |
6394 | |
6395 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6396 | device->fs_devices = fs_devices; |
dfe25020 | 6397 | fs_devices->num_devices++; |
12bd2fc0 | 6398 | |
e6e674bd | 6399 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6400 | fs_devices->missing_devices++; |
12bd2fc0 | 6401 | |
dfe25020 CM |
6402 | return device; |
6403 | } | |
6404 | ||
12bd2fc0 ID |
6405 | /** |
6406 | * btrfs_alloc_device - allocate struct btrfs_device | |
6407 | * @fs_info: used only for generating a new devid, can be NULL if | |
6408 | * devid is provided (i.e. @devid != NULL). | |
6409 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6410 | * is generated. | |
6411 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6412 | * is generated. | |
6413 | * | |
6414 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf DS |
6415 | * on error. Returned struct is not linked onto any lists and must be |
6416 | * destroyed with free_device. | |
12bd2fc0 ID |
6417 | */ |
6418 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6419 | const u64 *devid, | |
6420 | const u8 *uuid) | |
6421 | { | |
6422 | struct btrfs_device *dev; | |
6423 | u64 tmp; | |
6424 | ||
fae7f21c | 6425 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6426 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6427 | |
6428 | dev = __alloc_device(); | |
6429 | if (IS_ERR(dev)) | |
6430 | return dev; | |
6431 | ||
6432 | if (devid) | |
6433 | tmp = *devid; | |
6434 | else { | |
6435 | int ret; | |
6436 | ||
6437 | ret = find_next_devid(fs_info, &tmp); | |
6438 | if (ret) { | |
55de4803 | 6439 | free_device(dev); |
12bd2fc0 ID |
6440 | return ERR_PTR(ret); |
6441 | } | |
6442 | } | |
6443 | dev->devid = tmp; | |
6444 | ||
6445 | if (uuid) | |
6446 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6447 | else | |
6448 | generate_random_uuid(dev->uuid); | |
6449 | ||
9e0af237 LB |
6450 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6451 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6452 | |
6453 | return dev; | |
6454 | } | |
6455 | ||
e06cd3dd | 6456 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6457 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6458 | struct extent_buffer *leaf, |
6459 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6460 | { |
0b86a832 | 6461 | u64 length; |
f04b772b | 6462 | u64 stripe_len; |
e06cd3dd LB |
6463 | u16 num_stripes; |
6464 | u16 sub_stripes; | |
6465 | u64 type; | |
0b86a832 | 6466 | |
e17cade2 | 6467 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6468 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6469 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6470 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6471 | type = btrfs_chunk_type(leaf, chunk); | |
6472 | ||
f04b772b | 6473 | if (!num_stripes) { |
0b246afa | 6474 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6475 | num_stripes); |
6476 | return -EIO; | |
6477 | } | |
0b246afa JM |
6478 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6479 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6480 | return -EIO; |
6481 | } | |
0b246afa JM |
6482 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6483 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6484 | btrfs_chunk_sector_size(leaf, chunk)); |
6485 | return -EIO; | |
6486 | } | |
0b246afa JM |
6487 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6488 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6489 | return -EIO; |
6490 | } | |
3d8da678 | 6491 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6492 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6493 | stripe_len); |
6494 | return -EIO; | |
6495 | } | |
6496 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6497 | type) { |
0b246afa | 6498 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6499 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6500 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6501 | btrfs_chunk_type(leaf, chunk)); | |
6502 | return -EIO; | |
6503 | } | |
e06cd3dd LB |
6504 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6505 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6506 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6507 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6508 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6509 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6510 | num_stripes != 1)) { | |
0b246afa | 6511 | btrfs_err(fs_info, |
e06cd3dd LB |
6512 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6513 | num_stripes, sub_stripes, | |
6514 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6515 | return -EIO; | |
6516 | } | |
6517 | ||
6518 | return 0; | |
6519 | } | |
6520 | ||
5a2b8e60 | 6521 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6522 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6523 | { |
2b902dfc AJ |
6524 | if (error) |
6525 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6526 | devid, uuid); | |
6527 | else | |
6528 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6529 | devid, uuid); | |
5a2b8e60 AJ |
6530 | } |
6531 | ||
2ff7e61e | 6532 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6533 | struct extent_buffer *leaf, |
6534 | struct btrfs_chunk *chunk) | |
6535 | { | |
0b246afa | 6536 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6537 | struct map_lookup *map; |
6538 | struct extent_map *em; | |
6539 | u64 logical; | |
6540 | u64 length; | |
e06cd3dd LB |
6541 | u64 devid; |
6542 | u8 uuid[BTRFS_UUID_SIZE]; | |
6543 | int num_stripes; | |
6544 | int ret; | |
6545 | int i; | |
6546 | ||
6547 | logical = key->offset; | |
6548 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6549 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6550 | ||
2ff7e61e | 6551 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6552 | if (ret) |
6553 | return ret; | |
a061fc8d | 6554 | |
890871be | 6555 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6556 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6557 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6558 | |
6559 | /* already mapped? */ | |
6560 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6561 | free_extent_map(em); | |
0b86a832 CM |
6562 | return 0; |
6563 | } else if (em) { | |
6564 | free_extent_map(em); | |
6565 | } | |
0b86a832 | 6566 | |
172ddd60 | 6567 | em = alloc_extent_map(); |
0b86a832 CM |
6568 | if (!em) |
6569 | return -ENOMEM; | |
593060d7 | 6570 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6571 | if (!map) { |
6572 | free_extent_map(em); | |
6573 | return -ENOMEM; | |
6574 | } | |
6575 | ||
298a8f9c | 6576 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6577 | em->map_lookup = map; |
0b86a832 CM |
6578 | em->start = logical; |
6579 | em->len = length; | |
70c8a91c | 6580 | em->orig_start = 0; |
0b86a832 | 6581 | em->block_start = 0; |
c8b97818 | 6582 | em->block_len = em->len; |
0b86a832 | 6583 | |
593060d7 CM |
6584 | map->num_stripes = num_stripes; |
6585 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6586 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6587 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6588 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6589 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6590 | for (i = 0; i < num_stripes; i++) { |
6591 | map->stripes[i].physical = | |
6592 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6593 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6594 | read_extent_buffer(leaf, uuid, (unsigned long) |
6595 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6596 | BTRFS_UUID_SIZE); | |
0b246afa | 6597 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6598 | uuid, NULL); |
3cdde224 | 6599 | if (!map->stripes[i].dev && |
0b246afa | 6600 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6601 | free_extent_map(em); |
2b902dfc | 6602 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6603 | return -ENOENT; |
593060d7 | 6604 | } |
dfe25020 CM |
6605 | if (!map->stripes[i].dev) { |
6606 | map->stripes[i].dev = | |
2ff7e61e JM |
6607 | add_missing_dev(fs_info->fs_devices, devid, |
6608 | uuid); | |
adfb69af | 6609 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6610 | free_extent_map(em); |
adfb69af AJ |
6611 | btrfs_err(fs_info, |
6612 | "failed to init missing dev %llu: %ld", | |
6613 | devid, PTR_ERR(map->stripes[i].dev)); | |
6614 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6615 | } |
2b902dfc | 6616 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6617 | } |
e12c9621 AJ |
6618 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6619 | &(map->stripes[i].dev->dev_state)); | |
6620 | ||
0b86a832 CM |
6621 | } |
6622 | ||
890871be | 6623 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6624 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6625 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6626 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6627 | free_extent_map(em); |
6628 | ||
6629 | return 0; | |
6630 | } | |
6631 | ||
143bede5 | 6632 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6633 | struct btrfs_dev_item *dev_item, |
6634 | struct btrfs_device *device) | |
6635 | { | |
6636 | unsigned long ptr; | |
0b86a832 CM |
6637 | |
6638 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6639 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6640 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6641 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6642 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6643 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6644 | device->type = btrfs_device_type(leaf, dev_item); |
6645 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6646 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6647 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6648 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6649 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6650 | |
410ba3a2 | 6651 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6652 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6653 | } |
6654 | ||
2ff7e61e | 6655 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6656 | u8 *fsid) |
2b82032c YZ |
6657 | { |
6658 | struct btrfs_fs_devices *fs_devices; | |
6659 | int ret; | |
6660 | ||
a32bf9a3 | 6661 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6662 | ASSERT(fsid); |
2b82032c | 6663 | |
0b246afa | 6664 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6665 | while (fs_devices) { |
44880fdc | 6666 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6667 | return fs_devices; |
6668 | ||
2b82032c YZ |
6669 | fs_devices = fs_devices->seed; |
6670 | } | |
6671 | ||
6672 | fs_devices = find_fsid(fsid); | |
6673 | if (!fs_devices) { | |
0b246afa | 6674 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6675 | return ERR_PTR(-ENOENT); |
6676 | ||
6677 | fs_devices = alloc_fs_devices(fsid); | |
6678 | if (IS_ERR(fs_devices)) | |
6679 | return fs_devices; | |
6680 | ||
6681 | fs_devices->seeding = 1; | |
6682 | fs_devices->opened = 1; | |
6683 | return fs_devices; | |
2b82032c | 6684 | } |
e4404d6e YZ |
6685 | |
6686 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6687 | if (IS_ERR(fs_devices)) |
6688 | return fs_devices; | |
2b82032c | 6689 | |
97288f2c | 6690 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
0b246afa | 6691 | fs_info->bdev_holder); |
48d28232 JL |
6692 | if (ret) { |
6693 | free_fs_devices(fs_devices); | |
5f375835 | 6694 | fs_devices = ERR_PTR(ret); |
2b82032c | 6695 | goto out; |
48d28232 | 6696 | } |
2b82032c YZ |
6697 | |
6698 | if (!fs_devices->seeding) { | |
6699 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6700 | free_fs_devices(fs_devices); |
5f375835 | 6701 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6702 | goto out; |
6703 | } | |
6704 | ||
0b246afa JM |
6705 | fs_devices->seed = fs_info->fs_devices->seed; |
6706 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6707 | out: |
5f375835 | 6708 | return fs_devices; |
2b82032c YZ |
6709 | } |
6710 | ||
2ff7e61e | 6711 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6712 | struct extent_buffer *leaf, |
6713 | struct btrfs_dev_item *dev_item) | |
6714 | { | |
0b246afa | 6715 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6716 | struct btrfs_device *device; |
6717 | u64 devid; | |
6718 | int ret; | |
44880fdc | 6719 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6720 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6721 | ||
0b86a832 | 6722 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6723 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6724 | BTRFS_UUID_SIZE); |
1473b24e | 6725 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6726 | BTRFS_FSID_SIZE); |
2b82032c | 6727 | |
44880fdc | 6728 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6729 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6730 | if (IS_ERR(fs_devices)) |
6731 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6732 | } |
6733 | ||
0b246afa | 6734 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6735 | if (!device) { |
c5502451 | 6736 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6737 | btrfs_report_missing_device(fs_info, devid, |
6738 | dev_uuid, true); | |
45dbdbc9 | 6739 | return -ENOENT; |
c5502451 | 6740 | } |
2b82032c | 6741 | |
2ff7e61e | 6742 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6743 | if (IS_ERR(device)) { |
6744 | btrfs_err(fs_info, | |
6745 | "failed to add missing dev %llu: %ld", | |
6746 | devid, PTR_ERR(device)); | |
6747 | return PTR_ERR(device); | |
6748 | } | |
2b902dfc | 6749 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6750 | } else { |
c5502451 | 6751 | if (!device->bdev) { |
2b902dfc AJ |
6752 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6753 | btrfs_report_missing_device(fs_info, | |
6754 | devid, dev_uuid, true); | |
45dbdbc9 | 6755 | return -ENOENT; |
2b902dfc AJ |
6756 | } |
6757 | btrfs_report_missing_device(fs_info, devid, | |
6758 | dev_uuid, false); | |
c5502451 | 6759 | } |
5f375835 | 6760 | |
e6e674bd AJ |
6761 | if (!device->bdev && |
6762 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6763 | /* |
6764 | * this happens when a device that was properly setup | |
6765 | * in the device info lists suddenly goes bad. | |
6766 | * device->bdev is NULL, and so we have to set | |
6767 | * device->missing to one here | |
6768 | */ | |
5f375835 | 6769 | device->fs_devices->missing_devices++; |
e6e674bd | 6770 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6771 | } |
5f375835 MX |
6772 | |
6773 | /* Move the device to its own fs_devices */ | |
6774 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6775 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6776 | &device->dev_state)); | |
5f375835 MX |
6777 | |
6778 | list_move(&device->dev_list, &fs_devices->devices); | |
6779 | device->fs_devices->num_devices--; | |
6780 | fs_devices->num_devices++; | |
6781 | ||
6782 | device->fs_devices->missing_devices--; | |
6783 | fs_devices->missing_devices++; | |
6784 | ||
6785 | device->fs_devices = fs_devices; | |
6786 | } | |
2b82032c YZ |
6787 | } |
6788 | ||
0b246afa | 6789 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6790 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6791 | if (device->generation != |
6792 | btrfs_device_generation(leaf, dev_item)) | |
6793 | return -EINVAL; | |
6324fbf3 | 6794 | } |
0b86a832 CM |
6795 | |
6796 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6797 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6798 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6799 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6800 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6801 | atomic64_add(device->total_bytes - device->bytes_used, |
6802 | &fs_info->free_chunk_space); | |
2bf64758 | 6803 | } |
0b86a832 | 6804 | ret = 0; |
0b86a832 CM |
6805 | return ret; |
6806 | } | |
6807 | ||
6bccf3ab | 6808 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6809 | { |
6bccf3ab | 6810 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6811 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6812 | struct extent_buffer *sb; |
0b86a832 | 6813 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6814 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6815 | u8 *array_ptr; |
6816 | unsigned long sb_array_offset; | |
84eed90f | 6817 | int ret = 0; |
0b86a832 CM |
6818 | u32 num_stripes; |
6819 | u32 array_size; | |
6820 | u32 len = 0; | |
1ffb22cf | 6821 | u32 cur_offset; |
e06cd3dd | 6822 | u64 type; |
84eed90f | 6823 | struct btrfs_key key; |
0b86a832 | 6824 | |
0b246afa | 6825 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6826 | /* |
6827 | * This will create extent buffer of nodesize, superblock size is | |
6828 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6829 | * overallocate but we can keep it as-is, only the first page is used. | |
6830 | */ | |
2ff7e61e | 6831 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6832 | if (IS_ERR(sb)) |
6833 | return PTR_ERR(sb); | |
4db8c528 | 6834 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6835 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6836 | /* |
01327610 | 6837 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6838 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6839 | * pages up-to-date when the page is larger: extent does not cover the |
6840 | * whole page and consequently check_page_uptodate does not find all | |
6841 | * the page's extents up-to-date (the hole beyond sb), | |
6842 | * write_extent_buffer then triggers a WARN_ON. | |
6843 | * | |
6844 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6845 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6846 | * to silence the warning eg. on PowerPC 64. | |
6847 | */ | |
09cbfeaf | 6848 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6849 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6850 | |
a061fc8d | 6851 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6852 | array_size = btrfs_super_sys_array_size(super_copy); |
6853 | ||
1ffb22cf DS |
6854 | array_ptr = super_copy->sys_chunk_array; |
6855 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6856 | cur_offset = 0; | |
0b86a832 | 6857 | |
1ffb22cf DS |
6858 | while (cur_offset < array_size) { |
6859 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6860 | len = sizeof(*disk_key); |
6861 | if (cur_offset + len > array_size) | |
6862 | goto out_short_read; | |
6863 | ||
0b86a832 CM |
6864 | btrfs_disk_key_to_cpu(&key, disk_key); |
6865 | ||
1ffb22cf DS |
6866 | array_ptr += len; |
6867 | sb_array_offset += len; | |
6868 | cur_offset += len; | |
0b86a832 | 6869 | |
0d81ba5d | 6870 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6871 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6872 | /* |
6873 | * At least one btrfs_chunk with one stripe must be | |
6874 | * present, exact stripe count check comes afterwards | |
6875 | */ | |
6876 | len = btrfs_chunk_item_size(1); | |
6877 | if (cur_offset + len > array_size) | |
6878 | goto out_short_read; | |
6879 | ||
6880 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6881 | if (!num_stripes) { |
ab8d0fc4 JM |
6882 | btrfs_err(fs_info, |
6883 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6884 | num_stripes, cur_offset); |
6885 | ret = -EIO; | |
6886 | break; | |
6887 | } | |
6888 | ||
e06cd3dd LB |
6889 | type = btrfs_chunk_type(sb, chunk); |
6890 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6891 | btrfs_err(fs_info, |
e06cd3dd LB |
6892 | "invalid chunk type %llu in sys_array at offset %u", |
6893 | type, cur_offset); | |
6894 | ret = -EIO; | |
6895 | break; | |
6896 | } | |
6897 | ||
e3540eab DS |
6898 | len = btrfs_chunk_item_size(num_stripes); |
6899 | if (cur_offset + len > array_size) | |
6900 | goto out_short_read; | |
6901 | ||
2ff7e61e | 6902 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6903 | if (ret) |
6904 | break; | |
0b86a832 | 6905 | } else { |
ab8d0fc4 JM |
6906 | btrfs_err(fs_info, |
6907 | "unexpected item type %u in sys_array at offset %u", | |
6908 | (u32)key.type, cur_offset); | |
84eed90f CM |
6909 | ret = -EIO; |
6910 | break; | |
0b86a832 | 6911 | } |
1ffb22cf DS |
6912 | array_ptr += len; |
6913 | sb_array_offset += len; | |
6914 | cur_offset += len; | |
0b86a832 | 6915 | } |
d865177a | 6916 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6917 | free_extent_buffer_stale(sb); |
84eed90f | 6918 | return ret; |
e3540eab DS |
6919 | |
6920 | out_short_read: | |
ab8d0fc4 | 6921 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6922 | len, cur_offset); |
d865177a | 6923 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6924 | free_extent_buffer_stale(sb); |
e3540eab | 6925 | return -EIO; |
0b86a832 CM |
6926 | } |
6927 | ||
21634a19 QW |
6928 | /* |
6929 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6930 | * | |
6528b99d AJ |
6931 | * If the @failing_dev is specified, it's accounted as missing. |
6932 | * | |
21634a19 QW |
6933 | * Return true if all chunks meet the minimal RW mount requirements. |
6934 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6935 | */ | |
6528b99d AJ |
6936 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6937 | struct btrfs_device *failing_dev) | |
21634a19 QW |
6938 | { |
6939 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
6940 | struct extent_map *em; | |
6941 | u64 next_start = 0; | |
6942 | bool ret = true; | |
6943 | ||
6944 | read_lock(&map_tree->map_tree.lock); | |
6945 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
6946 | read_unlock(&map_tree->map_tree.lock); | |
6947 | /* No chunk at all? Return false anyway */ | |
6948 | if (!em) { | |
6949 | ret = false; | |
6950 | goto out; | |
6951 | } | |
6952 | while (em) { | |
6953 | struct map_lookup *map; | |
6954 | int missing = 0; | |
6955 | int max_tolerated; | |
6956 | int i; | |
6957 | ||
6958 | map = em->map_lookup; | |
6959 | max_tolerated = | |
6960 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6961 | map->type); | |
6962 | for (i = 0; i < map->num_stripes; i++) { | |
6963 | struct btrfs_device *dev = map->stripes[i].dev; | |
6964 | ||
e6e674bd AJ |
6965 | if (!dev || !dev->bdev || |
6966 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6967 | dev->last_flush_error) |
6968 | missing++; | |
6528b99d AJ |
6969 | else if (failing_dev && failing_dev == dev) |
6970 | missing++; | |
21634a19 QW |
6971 | } |
6972 | if (missing > max_tolerated) { | |
6528b99d AJ |
6973 | if (!failing_dev) |
6974 | btrfs_warn(fs_info, | |
21634a19 QW |
6975 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
6976 | em->start, missing, max_tolerated); | |
6977 | free_extent_map(em); | |
6978 | ret = false; | |
6979 | goto out; | |
6980 | } | |
6981 | next_start = extent_map_end(em); | |
6982 | free_extent_map(em); | |
6983 | ||
6984 | read_lock(&map_tree->map_tree.lock); | |
6985 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
6986 | (u64)(-1) - next_start); | |
6987 | read_unlock(&map_tree->map_tree.lock); | |
6988 | } | |
6989 | out: | |
6990 | return ret; | |
6991 | } | |
6992 | ||
5b4aacef | 6993 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6994 | { |
5b4aacef | 6995 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6996 | struct btrfs_path *path; |
6997 | struct extent_buffer *leaf; | |
6998 | struct btrfs_key key; | |
6999 | struct btrfs_key found_key; | |
7000 | int ret; | |
7001 | int slot; | |
99e3ecfc | 7002 | u64 total_dev = 0; |
0b86a832 | 7003 | |
0b86a832 CM |
7004 | path = btrfs_alloc_path(); |
7005 | if (!path) | |
7006 | return -ENOMEM; | |
7007 | ||
b367e47f | 7008 | mutex_lock(&uuid_mutex); |
34441361 | 7009 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7010 | |
395927a9 FDBM |
7011 | /* |
7012 | * Read all device items, and then all the chunk items. All | |
7013 | * device items are found before any chunk item (their object id | |
7014 | * is smaller than the lowest possible object id for a chunk | |
7015 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7016 | */ |
7017 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7018 | key.offset = 0; | |
7019 | key.type = 0; | |
0b86a832 | 7020 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7021 | if (ret < 0) |
7022 | goto error; | |
d397712b | 7023 | while (1) { |
0b86a832 CM |
7024 | leaf = path->nodes[0]; |
7025 | slot = path->slots[0]; | |
7026 | if (slot >= btrfs_header_nritems(leaf)) { | |
7027 | ret = btrfs_next_leaf(root, path); | |
7028 | if (ret == 0) | |
7029 | continue; | |
7030 | if (ret < 0) | |
7031 | goto error; | |
7032 | break; | |
7033 | } | |
7034 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7035 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7036 | struct btrfs_dev_item *dev_item; | |
7037 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7038 | struct btrfs_dev_item); |
2ff7e61e | 7039 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
7040 | if (ret) |
7041 | goto error; | |
99e3ecfc | 7042 | total_dev++; |
0b86a832 CM |
7043 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7044 | struct btrfs_chunk *chunk; | |
7045 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7046 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7047 | if (ret) |
7048 | goto error; | |
0b86a832 CM |
7049 | } |
7050 | path->slots[0]++; | |
7051 | } | |
99e3ecfc LB |
7052 | |
7053 | /* | |
7054 | * After loading chunk tree, we've got all device information, | |
7055 | * do another round of validation checks. | |
7056 | */ | |
0b246afa JM |
7057 | if (total_dev != fs_info->fs_devices->total_devices) { |
7058 | btrfs_err(fs_info, | |
99e3ecfc | 7059 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7060 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7061 | total_dev); |
7062 | ret = -EINVAL; | |
7063 | goto error; | |
7064 | } | |
0b246afa JM |
7065 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7066 | fs_info->fs_devices->total_rw_bytes) { | |
7067 | btrfs_err(fs_info, | |
99e3ecfc | 7068 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7069 | btrfs_super_total_bytes(fs_info->super_copy), |
7070 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7071 | ret = -EINVAL; |
7072 | goto error; | |
7073 | } | |
0b86a832 CM |
7074 | ret = 0; |
7075 | error: | |
34441361 | 7076 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7077 | mutex_unlock(&uuid_mutex); |
7078 | ||
2b82032c | 7079 | btrfs_free_path(path); |
0b86a832 CM |
7080 | return ret; |
7081 | } | |
442a4f63 | 7082 | |
cb517eab MX |
7083 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7084 | { | |
7085 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7086 | struct btrfs_device *device; | |
7087 | ||
29cc83f6 LB |
7088 | while (fs_devices) { |
7089 | mutex_lock(&fs_devices->device_list_mutex); | |
7090 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7091 | device->fs_info = fs_info; |
29cc83f6 LB |
7092 | mutex_unlock(&fs_devices->device_list_mutex); |
7093 | ||
7094 | fs_devices = fs_devices->seed; | |
7095 | } | |
cb517eab MX |
7096 | } |
7097 | ||
733f4fbb SB |
7098 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7099 | { | |
7100 | int i; | |
7101 | ||
7102 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7103 | btrfs_dev_stat_reset(dev, i); | |
7104 | } | |
7105 | ||
7106 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7107 | { | |
7108 | struct btrfs_key key; | |
7109 | struct btrfs_key found_key; | |
7110 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7111 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7112 | struct extent_buffer *eb; | |
7113 | int slot; | |
7114 | int ret = 0; | |
7115 | struct btrfs_device *device; | |
7116 | struct btrfs_path *path = NULL; | |
7117 | int i; | |
7118 | ||
7119 | path = btrfs_alloc_path(); | |
7120 | if (!path) { | |
7121 | ret = -ENOMEM; | |
7122 | goto out; | |
7123 | } | |
7124 | ||
7125 | mutex_lock(&fs_devices->device_list_mutex); | |
7126 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7127 | int item_size; | |
7128 | struct btrfs_dev_stats_item *ptr; | |
7129 | ||
242e2956 DS |
7130 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7131 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7132 | key.offset = device->devid; |
7133 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7134 | if (ret) { | |
733f4fbb SB |
7135 | __btrfs_reset_dev_stats(device); |
7136 | device->dev_stats_valid = 1; | |
7137 | btrfs_release_path(path); | |
7138 | continue; | |
7139 | } | |
7140 | slot = path->slots[0]; | |
7141 | eb = path->nodes[0]; | |
7142 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7143 | item_size = btrfs_item_size_nr(eb, slot); | |
7144 | ||
7145 | ptr = btrfs_item_ptr(eb, slot, | |
7146 | struct btrfs_dev_stats_item); | |
7147 | ||
7148 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7149 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7150 | btrfs_dev_stat_set(device, i, | |
7151 | btrfs_dev_stats_value(eb, ptr, i)); | |
7152 | else | |
7153 | btrfs_dev_stat_reset(device, i); | |
7154 | } | |
7155 | ||
7156 | device->dev_stats_valid = 1; | |
7157 | btrfs_dev_stat_print_on_load(device); | |
7158 | btrfs_release_path(path); | |
7159 | } | |
7160 | mutex_unlock(&fs_devices->device_list_mutex); | |
7161 | ||
7162 | out: | |
7163 | btrfs_free_path(path); | |
7164 | return ret < 0 ? ret : 0; | |
7165 | } | |
7166 | ||
7167 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6bccf3ab | 7168 | struct btrfs_fs_info *fs_info, |
733f4fbb SB |
7169 | struct btrfs_device *device) |
7170 | { | |
6bccf3ab | 7171 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7172 | struct btrfs_path *path; |
7173 | struct btrfs_key key; | |
7174 | struct extent_buffer *eb; | |
7175 | struct btrfs_dev_stats_item *ptr; | |
7176 | int ret; | |
7177 | int i; | |
7178 | ||
242e2956 DS |
7179 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7180 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7181 | key.offset = device->devid; |
7182 | ||
7183 | path = btrfs_alloc_path(); | |
fa252992 DS |
7184 | if (!path) |
7185 | return -ENOMEM; | |
733f4fbb SB |
7186 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7187 | if (ret < 0) { | |
0b246afa | 7188 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7189 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7190 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7191 | goto out; |
7192 | } | |
7193 | ||
7194 | if (ret == 0 && | |
7195 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7196 | /* need to delete old one and insert a new one */ | |
7197 | ret = btrfs_del_item(trans, dev_root, path); | |
7198 | if (ret != 0) { | |
0b246afa | 7199 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7200 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7201 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7202 | goto out; |
7203 | } | |
7204 | ret = 1; | |
7205 | } | |
7206 | ||
7207 | if (ret == 1) { | |
7208 | /* need to insert a new item */ | |
7209 | btrfs_release_path(path); | |
7210 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7211 | &key, sizeof(*ptr)); | |
7212 | if (ret < 0) { | |
0b246afa | 7213 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7214 | "insert dev_stats item for device %s failed %d", |
7215 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7216 | goto out; |
7217 | } | |
7218 | } | |
7219 | ||
7220 | eb = path->nodes[0]; | |
7221 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7222 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7223 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7224 | btrfs_dev_stat_read(device, i)); | |
7225 | btrfs_mark_buffer_dirty(eb); | |
7226 | ||
7227 | out: | |
7228 | btrfs_free_path(path); | |
7229 | return ret; | |
7230 | } | |
7231 | ||
7232 | /* | |
7233 | * called from commit_transaction. Writes all changed device stats to disk. | |
7234 | */ | |
7235 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7236 | struct btrfs_fs_info *fs_info) | |
7237 | { | |
733f4fbb SB |
7238 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7239 | struct btrfs_device *device; | |
addc3fa7 | 7240 | int stats_cnt; |
733f4fbb SB |
7241 | int ret = 0; |
7242 | ||
7243 | mutex_lock(&fs_devices->device_list_mutex); | |
7244 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7245 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7246 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7247 | continue; |
7248 | ||
9deae968 NB |
7249 | |
7250 | /* | |
7251 | * There is a LOAD-LOAD control dependency between the value of | |
7252 | * dev_stats_ccnt and updating the on-disk values which requires | |
7253 | * reading the in-memory counters. Such control dependencies | |
7254 | * require explicit read memory barriers. | |
7255 | * | |
7256 | * This memory barriers pairs with smp_mb__before_atomic in | |
7257 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7258 | * barrier implied by atomic_xchg in | |
7259 | * btrfs_dev_stats_read_and_reset | |
7260 | */ | |
7261 | smp_rmb(); | |
7262 | ||
6bccf3ab | 7263 | ret = update_dev_stat_item(trans, fs_info, device); |
733f4fbb | 7264 | if (!ret) |
addc3fa7 | 7265 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7266 | } |
7267 | mutex_unlock(&fs_devices->device_list_mutex); | |
7268 | ||
7269 | return ret; | |
7270 | } | |
7271 | ||
442a4f63 SB |
7272 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7273 | { | |
7274 | btrfs_dev_stat_inc(dev, index); | |
7275 | btrfs_dev_stat_print_on_error(dev); | |
7276 | } | |
7277 | ||
48a3b636 | 7278 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7279 | { |
733f4fbb SB |
7280 | if (!dev->dev_stats_valid) |
7281 | return; | |
fb456252 | 7282 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7283 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7284 | rcu_str_deref(dev->name), |
442a4f63 SB |
7285 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7286 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7287 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7288 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7289 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7290 | } |
c11d2c23 | 7291 | |
733f4fbb SB |
7292 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7293 | { | |
a98cdb85 SB |
7294 | int i; |
7295 | ||
7296 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7297 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7298 | break; | |
7299 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7300 | return; /* all values == 0, suppress message */ | |
7301 | ||
fb456252 | 7302 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7303 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7304 | rcu_str_deref(dev->name), |
733f4fbb SB |
7305 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7306 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7307 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7308 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7309 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7310 | } | |
7311 | ||
2ff7e61e | 7312 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7313 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7314 | { |
7315 | struct btrfs_device *dev; | |
0b246afa | 7316 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7317 | int i; |
7318 | ||
7319 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7320 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7321 | mutex_unlock(&fs_devices->device_list_mutex); |
7322 | ||
7323 | if (!dev) { | |
0b246afa | 7324 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7325 | return -ENODEV; |
733f4fbb | 7326 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7327 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7328 | return -ENODEV; |
b27f7c0c | 7329 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7330 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7331 | if (stats->nr_items > i) | |
7332 | stats->values[i] = | |
7333 | btrfs_dev_stat_read_and_reset(dev, i); | |
7334 | else | |
7335 | btrfs_dev_stat_reset(dev, i); | |
7336 | } | |
7337 | } else { | |
7338 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7339 | if (stats->nr_items > i) | |
7340 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7341 | } | |
7342 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7343 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7344 | return 0; | |
7345 | } | |
a8a6dab7 | 7346 | |
da353f6b | 7347 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7348 | { |
7349 | struct buffer_head *bh; | |
7350 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7351 | int copy_num; |
a8a6dab7 | 7352 | |
12b1c263 AJ |
7353 | if (!bdev) |
7354 | return; | |
a8a6dab7 | 7355 | |
12b1c263 AJ |
7356 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7357 | copy_num++) { | |
a8a6dab7 | 7358 | |
12b1c263 AJ |
7359 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7360 | continue; | |
7361 | ||
7362 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7363 | ||
7364 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7365 | set_buffer_dirty(bh); | |
7366 | sync_dirty_buffer(bh); | |
7367 | brelse(bh); | |
7368 | } | |
7369 | ||
7370 | /* Notify udev that device has changed */ | |
7371 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7372 | ||
7373 | /* Update ctime/mtime for device path for libblkid */ | |
7374 | update_dev_time(device_path); | |
a8a6dab7 | 7375 | } |
935e5cc9 MX |
7376 | |
7377 | /* | |
7378 | * Update the size of all devices, which is used for writing out the | |
7379 | * super blocks. | |
7380 | */ | |
7381 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7382 | { | |
7383 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7384 | struct btrfs_device *curr, *next; | |
7385 | ||
7386 | if (list_empty(&fs_devices->resized_devices)) | |
7387 | return; | |
7388 | ||
7389 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7390 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7391 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7392 | resized_list) { | |
7393 | list_del_init(&curr->resized_list); | |
7394 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7395 | } | |
34441361 | 7396 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7397 | mutex_unlock(&fs_devices->device_list_mutex); |
7398 | } | |
ce7213c7 MX |
7399 | |
7400 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7401 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7402 | { |
e9b919b1 | 7403 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7404 | struct extent_map *em; |
7405 | struct map_lookup *map; | |
7406 | struct btrfs_device *dev; | |
7407 | int i; | |
7408 | ||
e9b919b1 | 7409 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7410 | return; |
7411 | ||
7412 | /* In order to kick the device replace finish process */ | |
34441361 | 7413 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7414 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7415 | map = em->map_lookup; |
ce7213c7 MX |
7416 | |
7417 | for (i = 0; i < map->num_stripes; i++) { | |
7418 | dev = map->stripes[i].dev; | |
7419 | dev->commit_bytes_used = dev->bytes_used; | |
7420 | } | |
7421 | } | |
34441361 | 7422 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7423 | } |
5a13f430 AJ |
7424 | |
7425 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7426 | { | |
7427 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7428 | while (fs_devices) { | |
7429 | fs_devices->fs_info = fs_info; | |
7430 | fs_devices = fs_devices->seed; | |
7431 | } | |
7432 | } | |
7433 | ||
7434 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7435 | { | |
7436 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7437 | while (fs_devices) { | |
7438 | fs_devices->fs_info = NULL; | |
7439 | fs_devices = fs_devices->seed; | |
7440 | } | |
7441 | } |