]>
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 | ||
143bede5 | 282 | void 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 | ||
9eaed21e | 899 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 900 | { |
c6e30871 | 901 | struct btrfs_device *device, *next; |
443f24fe | 902 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 903 | |
dfe25020 CM |
904 | mutex_lock(&uuid_mutex); |
905 | again: | |
46224705 | 906 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 907 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
908 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
909 | &device->dev_state)) { | |
401e29c1 AJ |
910 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
911 | &device->dev_state) && | |
912 | (!latest_dev || | |
913 | device->generation > latest_dev->generation)) { | |
443f24fe | 914 | latest_dev = device; |
a6b0d5c8 | 915 | } |
2b82032c | 916 | continue; |
a6b0d5c8 | 917 | } |
2b82032c | 918 | |
8dabb742 SB |
919 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
920 | /* | |
921 | * In the first step, keep the device which has | |
922 | * the correct fsid and the devid that is used | |
923 | * for the dev_replace procedure. | |
924 | * In the second step, the dev_replace state is | |
925 | * read from the device tree and it is known | |
926 | * whether the procedure is really active or | |
927 | * not, which means whether this device is | |
928 | * used or whether it should be removed. | |
929 | */ | |
401e29c1 AJ |
930 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
931 | &device->dev_state)) { | |
8dabb742 SB |
932 | continue; |
933 | } | |
934 | } | |
2b82032c | 935 | if (device->bdev) { |
d4d77629 | 936 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
937 | device->bdev = NULL; |
938 | fs_devices->open_devices--; | |
939 | } | |
ebbede42 | 940 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 941 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 942 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
943 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
944 | &device->dev_state)) | |
8dabb742 | 945 | fs_devices->rw_devices--; |
2b82032c | 946 | } |
e4404d6e YZ |
947 | list_del_init(&device->dev_list); |
948 | fs_devices->num_devices--; | |
55de4803 | 949 | free_device(device); |
dfe25020 | 950 | } |
2b82032c YZ |
951 | |
952 | if (fs_devices->seed) { | |
953 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
954 | goto again; |
955 | } | |
956 | ||
443f24fe | 957 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 958 | |
dfe25020 | 959 | mutex_unlock(&uuid_mutex); |
dfe25020 | 960 | } |
a0af469b | 961 | |
f06c5965 | 962 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
963 | { |
964 | struct btrfs_device *device; | |
965 | ||
9f5316c1 | 966 | device = container_of(head, struct btrfs_device, rcu); |
55de4803 | 967 | free_device(device); |
1f78160c XG |
968 | } |
969 | ||
14238819 AJ |
970 | static void btrfs_close_bdev(struct btrfs_device *device) |
971 | { | |
08ffcae8 DS |
972 | if (!device->bdev) |
973 | return; | |
974 | ||
ebbede42 | 975 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
976 | sync_blockdev(device->bdev); |
977 | invalidate_bdev(device->bdev); | |
978 | } | |
979 | ||
08ffcae8 | 980 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
981 | } |
982 | ||
0ccd0528 | 983 | static void btrfs_prepare_close_one_device(struct btrfs_device *device) |
f448341a AJ |
984 | { |
985 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
986 | struct btrfs_device *new_device; | |
987 | struct rcu_string *name; | |
988 | ||
989 | if (device->bdev) | |
990 | fs_devices->open_devices--; | |
991 | ||
ebbede42 | 992 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
993 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
994 | list_del_init(&device->dev_alloc_list); | |
995 | fs_devices->rw_devices--; | |
996 | } | |
997 | ||
e6e674bd | 998 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
999 | fs_devices->missing_devices--; |
1000 | ||
1001 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
1002 | device->uuid); | |
1003 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1004 | ||
1005 | /* Safe because we are under uuid_mutex */ | |
1006 | if (device->name) { | |
1007 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1008 | BUG_ON(!name); /* -ENOMEM */ | |
1009 | rcu_assign_pointer(new_device->name, name); | |
1010 | } | |
1011 | ||
1012 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1013 | new_device->fs_devices = device->fs_devices; | |
f448341a AJ |
1014 | } |
1015 | ||
2b82032c | 1016 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1017 | { |
2037a093 | 1018 | struct btrfs_device *device, *tmp; |
0ccd0528 AJ |
1019 | struct list_head pending_put; |
1020 | ||
1021 | INIT_LIST_HEAD(&pending_put); | |
e4404d6e | 1022 | |
2b82032c YZ |
1023 | if (--fs_devices->opened > 0) |
1024 | return 0; | |
8a4b83cc | 1025 | |
c9513edb | 1026 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1027 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
0ccd0528 AJ |
1028 | btrfs_prepare_close_one_device(device); |
1029 | list_add(&device->dev_list, &pending_put); | |
8a4b83cc | 1030 | } |
c9513edb XG |
1031 | mutex_unlock(&fs_devices->device_list_mutex); |
1032 | ||
0ccd0528 AJ |
1033 | /* |
1034 | * btrfs_show_devname() is using the device_list_mutex, | |
1035 | * sometimes call to blkdev_put() leads vfs calling | |
1036 | * into this func. So do put outside of device_list_mutex, | |
1037 | * as of now. | |
1038 | */ | |
1039 | while (!list_empty(&pending_put)) { | |
1040 | device = list_first_entry(&pending_put, | |
1041 | struct btrfs_device, dev_list); | |
1042 | list_del(&device->dev_list); | |
1043 | btrfs_close_bdev(device); | |
f06c5965 | 1044 | call_rcu(&device->rcu, free_device_rcu); |
0ccd0528 AJ |
1045 | } |
1046 | ||
e4404d6e YZ |
1047 | WARN_ON(fs_devices->open_devices); |
1048 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1049 | fs_devices->opened = 0; |
1050 | fs_devices->seeding = 0; | |
2b82032c | 1051 | |
8a4b83cc CM |
1052 | return 0; |
1053 | } | |
1054 | ||
2b82032c YZ |
1055 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1056 | { | |
e4404d6e | 1057 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1058 | int ret; |
1059 | ||
1060 | mutex_lock(&uuid_mutex); | |
1061 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
1062 | if (!fs_devices->opened) { |
1063 | seed_devices = fs_devices->seed; | |
1064 | fs_devices->seed = NULL; | |
1065 | } | |
2b82032c | 1066 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1067 | |
1068 | while (seed_devices) { | |
1069 | fs_devices = seed_devices; | |
1070 | seed_devices = fs_devices->seed; | |
1071 | __btrfs_close_devices(fs_devices); | |
1072 | free_fs_devices(fs_devices); | |
1073 | } | |
2b82032c YZ |
1074 | return ret; |
1075 | } | |
1076 | ||
e4404d6e YZ |
1077 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
1078 | fmode_t flags, void *holder) | |
8a4b83cc | 1079 | { |
8a4b83cc | 1080 | struct list_head *head = &fs_devices->devices; |
8a4b83cc | 1081 | struct btrfs_device *device; |
443f24fe | 1082 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1083 | int ret = 0; |
8a4b83cc | 1084 | |
d4d77629 TH |
1085 | flags |= FMODE_EXCL; |
1086 | ||
c6e30871 | 1087 | list_for_each_entry(device, head, dev_list) { |
f63e0cca | 1088 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1089 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1090 | continue; |
a0af469b | 1091 | |
9f050db4 AJ |
1092 | if (!latest_dev || |
1093 | device->generation > latest_dev->generation) | |
1094 | latest_dev = device; | |
8a4b83cc | 1095 | } |
a0af469b | 1096 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1097 | ret = -EINVAL; |
a0af469b CM |
1098 | goto out; |
1099 | } | |
2b82032c | 1100 | fs_devices->opened = 1; |
443f24fe | 1101 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1102 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1103 | out: |
2b82032c YZ |
1104 | return ret; |
1105 | } | |
1106 | ||
f8e10cd3 AJ |
1107 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1108 | { | |
1109 | struct btrfs_device *dev1, *dev2; | |
1110 | ||
1111 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1112 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1113 | ||
1114 | if (dev1->devid < dev2->devid) | |
1115 | return -1; | |
1116 | else if (dev1->devid > dev2->devid) | |
1117 | return 1; | |
1118 | return 0; | |
1119 | } | |
1120 | ||
2b82032c | 1121 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1122 | fmode_t flags, void *holder) |
2b82032c YZ |
1123 | { |
1124 | int ret; | |
1125 | ||
1126 | mutex_lock(&uuid_mutex); | |
1127 | if (fs_devices->opened) { | |
e4404d6e YZ |
1128 | fs_devices->opened++; |
1129 | ret = 0; | |
2b82032c | 1130 | } else { |
f8e10cd3 | 1131 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
15916de8 | 1132 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1133 | } |
8a4b83cc | 1134 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1135 | return ret; |
1136 | } | |
1137 | ||
c9162bdf | 1138 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1139 | { |
1140 | kunmap(page); | |
1141 | put_page(page); | |
1142 | } | |
1143 | ||
c9162bdf OS |
1144 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1145 | struct page **page, | |
1146 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1147 | { |
1148 | void *p; | |
1149 | pgoff_t index; | |
1150 | ||
1151 | /* make sure our super fits in the device */ | |
1152 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1153 | return 1; | |
1154 | ||
1155 | /* make sure our super fits in the page */ | |
1156 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1157 | return 1; | |
1158 | ||
1159 | /* make sure our super doesn't straddle pages on disk */ | |
1160 | index = bytenr >> PAGE_SHIFT; | |
1161 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1162 | return 1; | |
1163 | ||
1164 | /* pull in the page with our super */ | |
1165 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1166 | index, GFP_KERNEL); | |
1167 | ||
1168 | if (IS_ERR_OR_NULL(*page)) | |
1169 | return 1; | |
1170 | ||
1171 | p = kmap(*page); | |
1172 | ||
1173 | /* align our pointer to the offset of the super block */ | |
1174 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1175 | ||
1176 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1177 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1178 | btrfs_release_disk_super(*page); | |
1179 | return 1; | |
1180 | } | |
1181 | ||
1182 | if ((*disk_super)->label[0] && | |
1183 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1184 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1185 | ||
1186 | return 0; | |
1187 | } | |
1188 | ||
6f60cbd3 DS |
1189 | /* |
1190 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1191 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1192 | * is read via pagecache | |
1193 | */ | |
97288f2c | 1194 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1195 | struct btrfs_fs_devices **fs_devices_ret) |
1196 | { | |
1197 | struct btrfs_super_block *disk_super; | |
e124ece5 | 1198 | struct btrfs_device *device; |
8a4b83cc | 1199 | struct block_device *bdev; |
6f60cbd3 | 1200 | struct page *page; |
e124ece5 | 1201 | int ret = 0; |
6f60cbd3 | 1202 | u64 bytenr; |
8a4b83cc | 1203 | |
6f60cbd3 DS |
1204 | /* |
1205 | * we would like to check all the supers, but that would make | |
1206 | * a btrfs mount succeed after a mkfs from a different FS. | |
1207 | * So, we need to add a special mount option to scan for | |
1208 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1209 | */ | |
1210 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1211 | flags |= FMODE_EXCL; |
10f6327b | 1212 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1213 | |
1214 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1215 | if (IS_ERR(bdev)) { |
1216 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1217 | goto error; |
6f60cbd3 DS |
1218 | } |
1219 | ||
05a5c55d AJ |
1220 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
1221 | ret = -EINVAL; | |
6f60cbd3 | 1222 | goto error_bdev_put; |
05a5c55d | 1223 | } |
6f60cbd3 | 1224 | |
3acbcbfc | 1225 | device = device_list_add(path, disk_super); |
e124ece5 AJ |
1226 | if (IS_ERR(device)) |
1227 | ret = PTR_ERR(device); | |
1228 | else | |
1229 | *fs_devices_ret = device->fs_devices; | |
6f60cbd3 | 1230 | |
6cf86a00 | 1231 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1232 | |
1233 | error_bdev_put: | |
d4d77629 | 1234 | blkdev_put(bdev, flags); |
8a4b83cc | 1235 | error: |
beaf8ab3 | 1236 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1237 | return ret; |
1238 | } | |
0b86a832 | 1239 | |
6d07bcec MX |
1240 | /* helper to account the used device space in the range */ |
1241 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1242 | u64 end, u64 *length) | |
1243 | { | |
1244 | struct btrfs_key key; | |
fb456252 | 1245 | struct btrfs_root *root = device->fs_info->dev_root; |
6d07bcec MX |
1246 | struct btrfs_dev_extent *dev_extent; |
1247 | struct btrfs_path *path; | |
1248 | u64 extent_end; | |
1249 | int ret; | |
1250 | int slot; | |
1251 | struct extent_buffer *l; | |
1252 | ||
1253 | *length = 0; | |
1254 | ||
401e29c1 AJ |
1255 | if (start >= device->total_bytes || |
1256 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) | |
6d07bcec MX |
1257 | return 0; |
1258 | ||
1259 | path = btrfs_alloc_path(); | |
1260 | if (!path) | |
1261 | return -ENOMEM; | |
e4058b54 | 1262 | path->reada = READA_FORWARD; |
6d07bcec MX |
1263 | |
1264 | key.objectid = device->devid; | |
1265 | key.offset = start; | |
1266 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1267 | ||
1268 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1269 | if (ret < 0) | |
1270 | goto out; | |
1271 | if (ret > 0) { | |
1272 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1273 | if (ret < 0) | |
1274 | goto out; | |
1275 | } | |
1276 | ||
1277 | while (1) { | |
1278 | l = path->nodes[0]; | |
1279 | slot = path->slots[0]; | |
1280 | if (slot >= btrfs_header_nritems(l)) { | |
1281 | ret = btrfs_next_leaf(root, path); | |
1282 | if (ret == 0) | |
1283 | continue; | |
1284 | if (ret < 0) | |
1285 | goto out; | |
1286 | ||
1287 | break; | |
1288 | } | |
1289 | btrfs_item_key_to_cpu(l, &key, slot); | |
1290 | ||
1291 | if (key.objectid < device->devid) | |
1292 | goto next; | |
1293 | ||
1294 | if (key.objectid > device->devid) | |
1295 | break; | |
1296 | ||
962a298f | 1297 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1298 | goto next; |
1299 | ||
1300 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1301 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1302 | dev_extent); | |
1303 | if (key.offset <= start && extent_end > end) { | |
1304 | *length = end - start + 1; | |
1305 | break; | |
1306 | } else if (key.offset <= start && extent_end > start) | |
1307 | *length += extent_end - start; | |
1308 | else if (key.offset > start && extent_end <= end) | |
1309 | *length += extent_end - key.offset; | |
1310 | else if (key.offset > start && key.offset <= end) { | |
1311 | *length += end - key.offset + 1; | |
1312 | break; | |
1313 | } else if (key.offset > end) | |
1314 | break; | |
1315 | ||
1316 | next: | |
1317 | path->slots[0]++; | |
1318 | } | |
1319 | ret = 0; | |
1320 | out: | |
1321 | btrfs_free_path(path); | |
1322 | return ret; | |
1323 | } | |
1324 | ||
499f377f | 1325 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1326 | struct btrfs_device *device, |
1327 | u64 *start, u64 len) | |
1328 | { | |
fb456252 | 1329 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1330 | struct extent_map *em; |
499f377f | 1331 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1332 | int ret = 0; |
1b984508 | 1333 | u64 physical_start = *start; |
6df9a95e | 1334 | |
499f377f JM |
1335 | if (transaction) |
1336 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1337 | again: |
1338 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1339 | struct map_lookup *map; |
1340 | int i; | |
1341 | ||
95617d69 | 1342 | map = em->map_lookup; |
6df9a95e | 1343 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1344 | u64 end; |
1345 | ||
6df9a95e JB |
1346 | if (map->stripes[i].dev != device) |
1347 | continue; | |
1b984508 | 1348 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1349 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1350 | physical_start) |
6df9a95e | 1351 | continue; |
c152b63e FM |
1352 | /* |
1353 | * Make sure that while processing the pinned list we do | |
1354 | * not override our *start with a lower value, because | |
1355 | * we can have pinned chunks that fall within this | |
1356 | * device hole and that have lower physical addresses | |
1357 | * than the pending chunks we processed before. If we | |
1358 | * do not take this special care we can end up getting | |
1359 | * 2 pending chunks that start at the same physical | |
1360 | * device offsets because the end offset of a pinned | |
1361 | * chunk can be equal to the start offset of some | |
1362 | * pending chunk. | |
1363 | */ | |
1364 | end = map->stripes[i].physical + em->orig_block_len; | |
1365 | if (end > *start) { | |
1366 | *start = end; | |
1367 | ret = 1; | |
1368 | } | |
6df9a95e JB |
1369 | } |
1370 | } | |
499f377f JM |
1371 | if (search_list != &fs_info->pinned_chunks) { |
1372 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1373 | goto again; |
1374 | } | |
6df9a95e JB |
1375 | |
1376 | return ret; | |
1377 | } | |
1378 | ||
1379 | ||
0b86a832 | 1380 | /* |
499f377f JM |
1381 | * find_free_dev_extent_start - find free space in the specified device |
1382 | * @device: the device which we search the free space in | |
1383 | * @num_bytes: the size of the free space that we need | |
1384 | * @search_start: the position from which to begin the search | |
1385 | * @start: store the start of the free space. | |
1386 | * @len: the size of the free space. that we find, or the size | |
1387 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1388 | * |
0b86a832 CM |
1389 | * this uses a pretty simple search, the expectation is that it is |
1390 | * called very infrequently and that a given device has a small number | |
1391 | * of extents | |
7bfc837d MX |
1392 | * |
1393 | * @start is used to store the start of the free space if we find. But if we | |
1394 | * don't find suitable free space, it will be used to store the start position | |
1395 | * of the max free space. | |
1396 | * | |
1397 | * @len is used to store the size of the free space that we find. | |
1398 | * But if we don't find suitable free space, it is used to store the size of | |
1399 | * the max free space. | |
0b86a832 | 1400 | */ |
499f377f JM |
1401 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1402 | struct btrfs_device *device, u64 num_bytes, | |
1403 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1404 | { |
0b246afa JM |
1405 | struct btrfs_fs_info *fs_info = device->fs_info; |
1406 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1407 | struct btrfs_key key; |
7bfc837d | 1408 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1409 | struct btrfs_path *path; |
7bfc837d MX |
1410 | u64 hole_size; |
1411 | u64 max_hole_start; | |
1412 | u64 max_hole_size; | |
1413 | u64 extent_end; | |
0b86a832 CM |
1414 | u64 search_end = device->total_bytes; |
1415 | int ret; | |
7bfc837d | 1416 | int slot; |
0b86a832 | 1417 | struct extent_buffer *l; |
8cdc7c5b FM |
1418 | |
1419 | /* | |
1420 | * We don't want to overwrite the superblock on the drive nor any area | |
1421 | * used by the boot loader (grub for example), so we make sure to start | |
1422 | * at an offset of at least 1MB. | |
1423 | */ | |
0d0c71b3 | 1424 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1425 | |
6df9a95e JB |
1426 | path = btrfs_alloc_path(); |
1427 | if (!path) | |
1428 | return -ENOMEM; | |
f2ab7618 | 1429 | |
7bfc837d MX |
1430 | max_hole_start = search_start; |
1431 | max_hole_size = 0; | |
1432 | ||
f2ab7618 | 1433 | again: |
401e29c1 AJ |
1434 | if (search_start >= search_end || |
1435 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1436 | ret = -ENOSPC; |
6df9a95e | 1437 | goto out; |
7bfc837d MX |
1438 | } |
1439 | ||
e4058b54 | 1440 | path->reada = READA_FORWARD; |
6df9a95e JB |
1441 | path->search_commit_root = 1; |
1442 | path->skip_locking = 1; | |
7bfc837d | 1443 | |
0b86a832 CM |
1444 | key.objectid = device->devid; |
1445 | key.offset = search_start; | |
1446 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1447 | |
125ccb0a | 1448 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1449 | if (ret < 0) |
7bfc837d | 1450 | goto out; |
1fcbac58 YZ |
1451 | if (ret > 0) { |
1452 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1453 | if (ret < 0) | |
7bfc837d | 1454 | goto out; |
1fcbac58 | 1455 | } |
7bfc837d | 1456 | |
0b86a832 CM |
1457 | while (1) { |
1458 | l = path->nodes[0]; | |
1459 | slot = path->slots[0]; | |
1460 | if (slot >= btrfs_header_nritems(l)) { | |
1461 | ret = btrfs_next_leaf(root, path); | |
1462 | if (ret == 0) | |
1463 | continue; | |
1464 | if (ret < 0) | |
7bfc837d MX |
1465 | goto out; |
1466 | ||
1467 | break; | |
0b86a832 CM |
1468 | } |
1469 | btrfs_item_key_to_cpu(l, &key, slot); | |
1470 | ||
1471 | if (key.objectid < device->devid) | |
1472 | goto next; | |
1473 | ||
1474 | if (key.objectid > device->devid) | |
7bfc837d | 1475 | break; |
0b86a832 | 1476 | |
962a298f | 1477 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1478 | goto next; |
9779b72f | 1479 | |
7bfc837d MX |
1480 | if (key.offset > search_start) { |
1481 | hole_size = key.offset - search_start; | |
9779b72f | 1482 | |
6df9a95e JB |
1483 | /* |
1484 | * Have to check before we set max_hole_start, otherwise | |
1485 | * we could end up sending back this offset anyway. | |
1486 | */ | |
499f377f | 1487 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1488 | &search_start, |
1b984508 FL |
1489 | hole_size)) { |
1490 | if (key.offset >= search_start) { | |
1491 | hole_size = key.offset - search_start; | |
1492 | } else { | |
1493 | WARN_ON_ONCE(1); | |
1494 | hole_size = 0; | |
1495 | } | |
1496 | } | |
6df9a95e | 1497 | |
7bfc837d MX |
1498 | if (hole_size > max_hole_size) { |
1499 | max_hole_start = search_start; | |
1500 | max_hole_size = hole_size; | |
1501 | } | |
9779b72f | 1502 | |
7bfc837d MX |
1503 | /* |
1504 | * If this free space is greater than which we need, | |
1505 | * it must be the max free space that we have found | |
1506 | * until now, so max_hole_start must point to the start | |
1507 | * of this free space and the length of this free space | |
1508 | * is stored in max_hole_size. Thus, we return | |
1509 | * max_hole_start and max_hole_size and go back to the | |
1510 | * caller. | |
1511 | */ | |
1512 | if (hole_size >= num_bytes) { | |
1513 | ret = 0; | |
1514 | goto out; | |
0b86a832 CM |
1515 | } |
1516 | } | |
0b86a832 | 1517 | |
0b86a832 | 1518 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1519 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1520 | dev_extent); | |
1521 | if (extent_end > search_start) | |
1522 | search_start = extent_end; | |
0b86a832 CM |
1523 | next: |
1524 | path->slots[0]++; | |
1525 | cond_resched(); | |
1526 | } | |
0b86a832 | 1527 | |
38c01b96 | 1528 | /* |
1529 | * At this point, search_start should be the end of | |
1530 | * allocated dev extents, and when shrinking the device, | |
1531 | * search_end may be smaller than search_start. | |
1532 | */ | |
f2ab7618 | 1533 | if (search_end > search_start) { |
38c01b96 | 1534 | hole_size = search_end - search_start; |
1535 | ||
499f377f | 1536 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1537 | hole_size)) { |
1538 | btrfs_release_path(path); | |
1539 | goto again; | |
1540 | } | |
0b86a832 | 1541 | |
f2ab7618 ZL |
1542 | if (hole_size > max_hole_size) { |
1543 | max_hole_start = search_start; | |
1544 | max_hole_size = hole_size; | |
1545 | } | |
6df9a95e JB |
1546 | } |
1547 | ||
7bfc837d | 1548 | /* See above. */ |
f2ab7618 | 1549 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1550 | ret = -ENOSPC; |
1551 | else | |
1552 | ret = 0; | |
1553 | ||
1554 | out: | |
2b82032c | 1555 | btrfs_free_path(path); |
7bfc837d | 1556 | *start = max_hole_start; |
b2117a39 | 1557 | if (len) |
7bfc837d | 1558 | *len = max_hole_size; |
0b86a832 CM |
1559 | return ret; |
1560 | } | |
1561 | ||
499f377f JM |
1562 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1563 | struct btrfs_device *device, u64 num_bytes, | |
1564 | u64 *start, u64 *len) | |
1565 | { | |
499f377f | 1566 | /* FIXME use last free of some kind */ |
499f377f | 1567 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1568 | num_bytes, 0, start, len); |
499f377f JM |
1569 | } |
1570 | ||
b2950863 | 1571 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1572 | struct btrfs_device *device, |
2196d6e8 | 1573 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1574 | { |
0b246afa JM |
1575 | struct btrfs_fs_info *fs_info = device->fs_info; |
1576 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1577 | int ret; |
1578 | struct btrfs_path *path; | |
8f18cf13 | 1579 | struct btrfs_key key; |
a061fc8d CM |
1580 | struct btrfs_key found_key; |
1581 | struct extent_buffer *leaf = NULL; | |
1582 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1583 | |
1584 | path = btrfs_alloc_path(); | |
1585 | if (!path) | |
1586 | return -ENOMEM; | |
1587 | ||
1588 | key.objectid = device->devid; | |
1589 | key.offset = start; | |
1590 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1591 | again: |
8f18cf13 | 1592 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1593 | if (ret > 0) { |
1594 | ret = btrfs_previous_item(root, path, key.objectid, | |
1595 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1596 | if (ret) |
1597 | goto out; | |
a061fc8d CM |
1598 | leaf = path->nodes[0]; |
1599 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1600 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1601 | struct btrfs_dev_extent); | |
1602 | BUG_ON(found_key.offset > start || found_key.offset + | |
1603 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1604 | key = found_key; |
1605 | btrfs_release_path(path); | |
1606 | goto again; | |
a061fc8d CM |
1607 | } else if (ret == 0) { |
1608 | leaf = path->nodes[0]; | |
1609 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1610 | struct btrfs_dev_extent); | |
79787eaa | 1611 | } else { |
0b246afa | 1612 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1613 | goto out; |
a061fc8d | 1614 | } |
8f18cf13 | 1615 | |
2196d6e8 MX |
1616 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1617 | ||
8f18cf13 | 1618 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1619 | if (ret) { |
0b246afa JM |
1620 | btrfs_handle_fs_error(fs_info, ret, |
1621 | "Failed to remove dev extent item"); | |
13212b54 | 1622 | } else { |
3204d33c | 1623 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1624 | } |
b0b802d7 | 1625 | out: |
8f18cf13 CM |
1626 | btrfs_free_path(path); |
1627 | return ret; | |
1628 | } | |
1629 | ||
48a3b636 ES |
1630 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1631 | struct btrfs_device *device, | |
48a3b636 | 1632 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1633 | { |
1634 | int ret; | |
1635 | struct btrfs_path *path; | |
0b246afa JM |
1636 | struct btrfs_fs_info *fs_info = device->fs_info; |
1637 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1638 | struct btrfs_dev_extent *extent; |
1639 | struct extent_buffer *leaf; | |
1640 | struct btrfs_key key; | |
1641 | ||
e12c9621 | 1642 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1643 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1644 | path = btrfs_alloc_path(); |
1645 | if (!path) | |
1646 | return -ENOMEM; | |
1647 | ||
0b86a832 | 1648 | key.objectid = device->devid; |
2b82032c | 1649 | key.offset = start; |
0b86a832 CM |
1650 | key.type = BTRFS_DEV_EXTENT_KEY; |
1651 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1652 | sizeof(*extent)); | |
2cdcecbc MF |
1653 | if (ret) |
1654 | goto out; | |
0b86a832 CM |
1655 | |
1656 | leaf = path->nodes[0]; | |
1657 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1658 | struct btrfs_dev_extent); | |
b5d9071c NB |
1659 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1660 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1661 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1662 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1663 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1664 | ||
0b86a832 CM |
1665 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1666 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1667 | out: |
0b86a832 CM |
1668 | btrfs_free_path(path); |
1669 | return ret; | |
1670 | } | |
1671 | ||
6df9a95e | 1672 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1673 | { |
6df9a95e JB |
1674 | struct extent_map_tree *em_tree; |
1675 | struct extent_map *em; | |
1676 | struct rb_node *n; | |
1677 | u64 ret = 0; | |
0b86a832 | 1678 | |
6df9a95e JB |
1679 | em_tree = &fs_info->mapping_tree.map_tree; |
1680 | read_lock(&em_tree->lock); | |
1681 | n = rb_last(&em_tree->map); | |
1682 | if (n) { | |
1683 | em = rb_entry(n, struct extent_map, rb_node); | |
1684 | ret = em->start + em->len; | |
0b86a832 | 1685 | } |
6df9a95e JB |
1686 | read_unlock(&em_tree->lock); |
1687 | ||
0b86a832 CM |
1688 | return ret; |
1689 | } | |
1690 | ||
53f10659 ID |
1691 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1692 | u64 *devid_ret) | |
0b86a832 CM |
1693 | { |
1694 | int ret; | |
1695 | struct btrfs_key key; | |
1696 | struct btrfs_key found_key; | |
2b82032c YZ |
1697 | struct btrfs_path *path; |
1698 | ||
2b82032c YZ |
1699 | path = btrfs_alloc_path(); |
1700 | if (!path) | |
1701 | return -ENOMEM; | |
0b86a832 CM |
1702 | |
1703 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1704 | key.type = BTRFS_DEV_ITEM_KEY; | |
1705 | key.offset = (u64)-1; | |
1706 | ||
53f10659 | 1707 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1708 | if (ret < 0) |
1709 | goto error; | |
1710 | ||
79787eaa | 1711 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1712 | |
53f10659 ID |
1713 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1714 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1715 | BTRFS_DEV_ITEM_KEY); |
1716 | if (ret) { | |
53f10659 | 1717 | *devid_ret = 1; |
0b86a832 CM |
1718 | } else { |
1719 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1720 | path->slots[0]); | |
53f10659 | 1721 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1722 | } |
1723 | ret = 0; | |
1724 | error: | |
2b82032c | 1725 | btrfs_free_path(path); |
0b86a832 CM |
1726 | return ret; |
1727 | } | |
1728 | ||
1729 | /* | |
1730 | * the device information is stored in the chunk root | |
1731 | * the btrfs_device struct should be fully filled in | |
1732 | */ | |
c74a0b02 | 1733 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
5b4aacef | 1734 | struct btrfs_fs_info *fs_info, |
48a3b636 | 1735 | struct btrfs_device *device) |
0b86a832 | 1736 | { |
5b4aacef | 1737 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
1738 | int ret; |
1739 | struct btrfs_path *path; | |
1740 | struct btrfs_dev_item *dev_item; | |
1741 | struct extent_buffer *leaf; | |
1742 | struct btrfs_key key; | |
1743 | unsigned long ptr; | |
0b86a832 | 1744 | |
0b86a832 CM |
1745 | path = btrfs_alloc_path(); |
1746 | if (!path) | |
1747 | return -ENOMEM; | |
1748 | ||
0b86a832 CM |
1749 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1750 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1751 | key.offset = device->devid; |
0b86a832 CM |
1752 | |
1753 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1754 | sizeof(*dev_item)); |
0b86a832 CM |
1755 | if (ret) |
1756 | goto out; | |
1757 | ||
1758 | leaf = path->nodes[0]; | |
1759 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1760 | ||
1761 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1762 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1763 | btrfs_set_device_type(leaf, dev_item, device->type); |
1764 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1765 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1766 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1767 | btrfs_set_device_total_bytes(leaf, dev_item, |
1768 | btrfs_device_get_disk_total_bytes(device)); | |
1769 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1770 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1771 | btrfs_set_device_group(leaf, dev_item, 0); |
1772 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1773 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1774 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1775 | |
410ba3a2 | 1776 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1777 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1778 | ptr = btrfs_device_fsid(dev_item); |
44880fdc | 1779 | write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE); |
0b86a832 | 1780 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1781 | |
2b82032c | 1782 | ret = 0; |
0b86a832 CM |
1783 | out: |
1784 | btrfs_free_path(path); | |
1785 | return ret; | |
1786 | } | |
8f18cf13 | 1787 | |
5a1972bd QW |
1788 | /* |
1789 | * Function to update ctime/mtime for a given device path. | |
1790 | * Mainly used for ctime/mtime based probe like libblkid. | |
1791 | */ | |
da353f6b | 1792 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1793 | { |
1794 | struct file *filp; | |
1795 | ||
1796 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1797 | if (IS_ERR(filp)) |
5a1972bd QW |
1798 | return; |
1799 | file_update_time(filp); | |
1800 | filp_close(filp, NULL); | |
5a1972bd QW |
1801 | } |
1802 | ||
5b4aacef | 1803 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1804 | struct btrfs_device *device) |
1805 | { | |
5b4aacef | 1806 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1807 | int ret; |
1808 | struct btrfs_path *path; | |
a061fc8d | 1809 | struct btrfs_key key; |
a061fc8d CM |
1810 | struct btrfs_trans_handle *trans; |
1811 | ||
a061fc8d CM |
1812 | path = btrfs_alloc_path(); |
1813 | if (!path) | |
1814 | return -ENOMEM; | |
1815 | ||
a22285a6 | 1816 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1817 | if (IS_ERR(trans)) { |
1818 | btrfs_free_path(path); | |
1819 | return PTR_ERR(trans); | |
1820 | } | |
a061fc8d CM |
1821 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1822 | key.type = BTRFS_DEV_ITEM_KEY; | |
1823 | key.offset = device->devid; | |
1824 | ||
1825 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1826 | if (ret) { |
1827 | if (ret > 0) | |
1828 | ret = -ENOENT; | |
1829 | btrfs_abort_transaction(trans, ret); | |
1830 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1831 | goto out; |
1832 | } | |
1833 | ||
1834 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1835 | if (ret) { |
1836 | btrfs_abort_transaction(trans, ret); | |
1837 | btrfs_end_transaction(trans); | |
1838 | } | |
1839 | ||
a061fc8d CM |
1840 | out: |
1841 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1842 | if (!ret) |
1843 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1844 | return ret; |
1845 | } | |
1846 | ||
3cc31a0d DS |
1847 | /* |
1848 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1849 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1850 | * replace. | |
1851 | */ | |
1852 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1853 | u64 num_devices) | |
a061fc8d | 1854 | { |
a061fc8d | 1855 | u64 all_avail; |
de98ced9 | 1856 | unsigned seq; |
418775a2 | 1857 | int i; |
a061fc8d | 1858 | |
de98ced9 | 1859 | do { |
bd45ffbc | 1860 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1861 | |
bd45ffbc AJ |
1862 | all_avail = fs_info->avail_data_alloc_bits | |
1863 | fs_info->avail_system_alloc_bits | | |
1864 | fs_info->avail_metadata_alloc_bits; | |
1865 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1866 | |
418775a2 DS |
1867 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1868 | if (!(all_avail & btrfs_raid_group[i])) | |
1869 | continue; | |
a061fc8d | 1870 | |
418775a2 DS |
1871 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1872 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1873 | |
418775a2 DS |
1874 | if (ret) |
1875 | return ret; | |
1876 | } | |
53b381b3 DW |
1877 | } |
1878 | ||
bd45ffbc | 1879 | return 0; |
f1fa7f26 AJ |
1880 | } |
1881 | ||
c9162bdf OS |
1882 | static struct btrfs_device * btrfs_find_next_active_device( |
1883 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1884 | { |
2b82032c | 1885 | struct btrfs_device *next_device; |
88acff64 AJ |
1886 | |
1887 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1888 | if (next_device != device && | |
e6e674bd AJ |
1889 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1890 | && next_device->bdev) | |
88acff64 AJ |
1891 | return next_device; |
1892 | } | |
1893 | ||
1894 | return NULL; | |
1895 | } | |
1896 | ||
1897 | /* | |
1898 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1899 | * and replace it with the provided or the next active device, in the context | |
1900 | * where this function called, there should be always be another device (or | |
1901 | * this_dev) which is active. | |
1902 | */ | |
1903 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1904 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1905 | { | |
1906 | struct btrfs_device *next_device; | |
1907 | ||
1908 | if (this_dev) | |
1909 | next_device = this_dev; | |
1910 | else | |
1911 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1912 | device); | |
1913 | ASSERT(next_device); | |
1914 | ||
1915 | if (fs_info->sb->s_bdev && | |
1916 | (fs_info->sb->s_bdev == device->bdev)) | |
1917 | fs_info->sb->s_bdev = next_device->bdev; | |
1918 | ||
1919 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1920 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1921 | } | |
1922 | ||
da353f6b DS |
1923 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1924 | u64 devid) | |
f1fa7f26 AJ |
1925 | { |
1926 | struct btrfs_device *device; | |
1f78160c | 1927 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1928 | u64 num_devices; |
a061fc8d CM |
1929 | int ret = 0; |
1930 | ||
2c997384 | 1931 | mutex_lock(&fs_info->volume_mutex); |
a061fc8d CM |
1932 | mutex_lock(&uuid_mutex); |
1933 | ||
0b246afa JM |
1934 | num_devices = fs_info->fs_devices->num_devices; |
1935 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
1936 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
8dabb742 SB |
1937 | WARN_ON(num_devices < 1); |
1938 | num_devices--; | |
1939 | } | |
0b246afa | 1940 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
8dabb742 | 1941 | |
0b246afa | 1942 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1943 | if (ret) |
a061fc8d | 1944 | goto out; |
a061fc8d | 1945 | |
2ff7e61e JM |
1946 | ret = btrfs_find_device_by_devspec(fs_info, devid, device_path, |
1947 | &device); | |
24fc572f | 1948 | if (ret) |
53b381b3 | 1949 | goto out; |
dfe25020 | 1950 | |
401e29c1 | 1951 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1952 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1953 | goto out; |
63a212ab SB |
1954 | } |
1955 | ||
ebbede42 AJ |
1956 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1957 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1958 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1959 | goto out; |
2b82032c YZ |
1960 | } |
1961 | ||
ebbede42 | 1962 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1963 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1964 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1965 | device->fs_devices->rw_devices--; |
34441361 | 1966 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 1967 | } |
a061fc8d | 1968 | |
d7901554 | 1969 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1970 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1971 | mutex_lock(&uuid_mutex); |
a061fc8d | 1972 | if (ret) |
9b3517e9 | 1973 | goto error_undo; |
a061fc8d | 1974 | |
63a212ab SB |
1975 | /* |
1976 | * TODO: the superblock still includes this device in its num_devices | |
1977 | * counter although write_all_supers() is not locked out. This | |
1978 | * could give a filesystem state which requires a degraded mount. | |
1979 | */ | |
0b246afa | 1980 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1981 | if (ret) |
9b3517e9 | 1982 | goto error_undo; |
a061fc8d | 1983 | |
e12c9621 | 1984 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 1985 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1986 | |
1987 | /* | |
1988 | * the device list mutex makes sure that we don't change | |
1989 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1990 | * the device supers. Whoever is writing all supers, should |
1991 | * lock the device list mutex before getting the number of | |
1992 | * devices in the super block (super_copy). Conversely, | |
1993 | * whoever updates the number of devices in the super block | |
1994 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1995 | */ |
1f78160c XG |
1996 | |
1997 | cur_devices = device->fs_devices; | |
0b246afa | 1998 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c | 1999 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2000 | |
e4404d6e | 2001 | device->fs_devices->num_devices--; |
02db0844 | 2002 | device->fs_devices->total_devices--; |
2b82032c | 2003 | |
e6e674bd | 2004 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
3a7d55c8 | 2005 | device->fs_devices->missing_devices--; |
cd02dca5 | 2006 | |
0b246afa | 2007 | btrfs_assign_next_active_device(fs_info, device, NULL); |
2b82032c | 2008 | |
0bfaa9c5 | 2009 | if (device->bdev) { |
e4404d6e | 2010 | device->fs_devices->open_devices--; |
0bfaa9c5 | 2011 | /* remove sysfs entry */ |
0b246afa | 2012 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
0bfaa9c5 | 2013 | } |
99994cde | 2014 | |
0b246afa JM |
2015 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2016 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
2017 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2018 | |
cea67ab9 JM |
2019 | /* |
2020 | * at this point, the device is zero sized and detached from | |
2021 | * the devices list. All that's left is to zero out the old | |
2022 | * supers and free the device. | |
2023 | */ | |
ebbede42 | 2024 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2025 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2026 | ||
2027 | btrfs_close_bdev(device); | |
f06c5965 | 2028 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2029 | |
1f78160c | 2030 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2031 | struct btrfs_fs_devices *fs_devices; |
0b246afa | 2032 | fs_devices = fs_info->fs_devices; |
e4404d6e | 2033 | while (fs_devices) { |
8321cf25 RS |
2034 | if (fs_devices->seed == cur_devices) { |
2035 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2036 | break; |
8321cf25 | 2037 | } |
e4404d6e | 2038 | fs_devices = fs_devices->seed; |
2b82032c | 2039 | } |
1f78160c | 2040 | cur_devices->seed = NULL; |
1f78160c | 2041 | __btrfs_close_devices(cur_devices); |
1f78160c | 2042 | free_fs_devices(cur_devices); |
2b82032c YZ |
2043 | } |
2044 | ||
a061fc8d CM |
2045 | out: |
2046 | mutex_unlock(&uuid_mutex); | |
2c997384 | 2047 | mutex_unlock(&fs_info->volume_mutex); |
a061fc8d | 2048 | return ret; |
24fc572f | 2049 | |
9b3517e9 | 2050 | error_undo: |
ebbede42 | 2051 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2052 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2053 | list_add(&device->dev_alloc_list, |
0b246afa | 2054 | &fs_info->fs_devices->alloc_list); |
c3929c36 | 2055 | device->fs_devices->rw_devices++; |
34441361 | 2056 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2057 | } |
24fc572f | 2058 | goto out; |
a061fc8d CM |
2059 | } |
2060 | ||
084b6e7c QW |
2061 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
2062 | struct btrfs_device *srcdev) | |
e93c89c1 | 2063 | { |
d51908ce AJ |
2064 | struct btrfs_fs_devices *fs_devices; |
2065 | ||
e93c89c1 | 2066 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 2067 | |
25e8e911 AJ |
2068 | /* |
2069 | * in case of fs with no seed, srcdev->fs_devices will point | |
2070 | * to fs_devices of fs_info. However when the dev being replaced is | |
2071 | * a seed dev it will point to the seed's local fs_devices. In short | |
2072 | * srcdev will have its correct fs_devices in both the cases. | |
2073 | */ | |
2074 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2075 | |
e93c89c1 | 2076 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2077 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2078 | fs_devices->num_devices--; |
e6e674bd | 2079 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2080 | fs_devices->missing_devices--; |
e93c89c1 | 2081 | |
ebbede42 | 2082 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2083 | fs_devices->rw_devices--; |
1357272f | 2084 | |
82372bc8 | 2085 | if (srcdev->bdev) |
d51908ce | 2086 | fs_devices->open_devices--; |
084b6e7c QW |
2087 | } |
2088 | ||
2089 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2090 | struct btrfs_device *srcdev) | |
2091 | { | |
2092 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2093 | |
ebbede42 | 2094 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2095 | /* zero out the old super if it is writable */ |
2096 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2097 | } | |
14238819 AJ |
2098 | |
2099 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2100 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2101 | |
94d5f0c2 AJ |
2102 | /* if this is no devs we rather delete the fs_devices */ |
2103 | if (!fs_devices->num_devices) { | |
2104 | struct btrfs_fs_devices *tmp_fs_devices; | |
2105 | ||
6dd38f81 AJ |
2106 | /* |
2107 | * On a mounted FS, num_devices can't be zero unless it's a | |
2108 | * seed. In case of a seed device being replaced, the replace | |
2109 | * target added to the sprout FS, so there will be no more | |
2110 | * device left under the seed FS. | |
2111 | */ | |
2112 | ASSERT(fs_devices->seeding); | |
2113 | ||
94d5f0c2 AJ |
2114 | tmp_fs_devices = fs_info->fs_devices; |
2115 | while (tmp_fs_devices) { | |
2116 | if (tmp_fs_devices->seed == fs_devices) { | |
2117 | tmp_fs_devices->seed = fs_devices->seed; | |
2118 | break; | |
2119 | } | |
2120 | tmp_fs_devices = tmp_fs_devices->seed; | |
2121 | } | |
2122 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2123 | __btrfs_close_devices(fs_devices); |
2124 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2125 | } |
e93c89c1 SB |
2126 | } |
2127 | ||
2128 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2129 | struct btrfs_device *tgtdev) | |
2130 | { | |
67a2c45e | 2131 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2132 | WARN_ON(!tgtdev); |
2133 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2134 | |
32576040 | 2135 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2136 | |
779bf3fe | 2137 | if (tgtdev->bdev) |
e93c89c1 | 2138 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2139 | |
e93c89c1 | 2140 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2141 | |
88acff64 | 2142 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2143 | |
e93c89c1 | 2144 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2145 | |
2146 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2147 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2148 | |
2149 | /* | |
2150 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2151 | * may lead to a call to btrfs_show_devname() which will try | |
2152 | * to hold device_list_mutex. And here this device | |
2153 | * is already out of device list, so we don't have to hold | |
2154 | * the device_list_mutex lock. | |
2155 | */ | |
2156 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2157 | |
2158 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2159 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2160 | } |
2161 | ||
2ff7e61e | 2162 | static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2163 | const char *device_path, |
48a3b636 | 2164 | struct btrfs_device **device) |
7ba15b7d SB |
2165 | { |
2166 | int ret = 0; | |
2167 | struct btrfs_super_block *disk_super; | |
2168 | u64 devid; | |
2169 | u8 *dev_uuid; | |
2170 | struct block_device *bdev; | |
2171 | struct buffer_head *bh; | |
2172 | ||
2173 | *device = NULL; | |
2174 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2175 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2176 | if (ret) |
2177 | return ret; | |
2178 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2179 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2180 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2181 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2182 | brelse(bh); |
2183 | if (!*device) | |
2184 | ret = -ENOENT; | |
2185 | blkdev_put(bdev, FMODE_READ); | |
2186 | return ret; | |
2187 | } | |
2188 | ||
2ff7e61e | 2189 | int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2190 | const char *device_path, |
7ba15b7d SB |
2191 | struct btrfs_device **device) |
2192 | { | |
2193 | *device = NULL; | |
2194 | if (strcmp(device_path, "missing") == 0) { | |
2195 | struct list_head *devices; | |
2196 | struct btrfs_device *tmp; | |
2197 | ||
0b246afa | 2198 | devices = &fs_info->fs_devices->devices; |
7ba15b7d SB |
2199 | /* |
2200 | * It is safe to read the devices since the volume_mutex | |
2201 | * is held by the caller. | |
2202 | */ | |
2203 | list_for_each_entry(tmp, devices, dev_list) { | |
e12c9621 AJ |
2204 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2205 | &tmp->dev_state) && !tmp->bdev) { | |
7ba15b7d SB |
2206 | *device = tmp; |
2207 | break; | |
2208 | } | |
2209 | } | |
2210 | ||
d74a6259 AJ |
2211 | if (!*device) |
2212 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2213 | |
2214 | return 0; | |
2215 | } else { | |
2ff7e61e | 2216 | return btrfs_find_device_by_path(fs_info, device_path, device); |
7ba15b7d SB |
2217 | } |
2218 | } | |
2219 | ||
5c5c0df0 DS |
2220 | /* |
2221 | * Lookup a device given by device id, or the path if the id is 0. | |
2222 | */ | |
2ff7e61e | 2223 | int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid, |
da353f6b DS |
2224 | const char *devpath, |
2225 | struct btrfs_device **device) | |
24e0474b AJ |
2226 | { |
2227 | int ret; | |
2228 | ||
5c5c0df0 | 2229 | if (devid) { |
24e0474b | 2230 | ret = 0; |
0b246afa | 2231 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2232 | if (!*device) |
2233 | ret = -ENOENT; | |
2234 | } else { | |
5c5c0df0 | 2235 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2236 | return -EINVAL; |
2237 | ||
2ff7e61e | 2238 | ret = btrfs_find_device_missing_or_by_path(fs_info, devpath, |
24e0474b AJ |
2239 | device); |
2240 | } | |
2241 | return ret; | |
2242 | } | |
2243 | ||
2b82032c YZ |
2244 | /* |
2245 | * does all the dirty work required for changing file system's UUID. | |
2246 | */ | |
2ff7e61e | 2247 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2248 | { |
0b246afa | 2249 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2250 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2251 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2252 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2253 | struct btrfs_device *device; |
2254 | u64 super_flags; | |
2255 | ||
2256 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2257 | if (!fs_devices->seeding) |
2b82032c YZ |
2258 | return -EINVAL; |
2259 | ||
2dfeca9b | 2260 | seed_devices = alloc_fs_devices(NULL); |
2208a378 ID |
2261 | if (IS_ERR(seed_devices)) |
2262 | return PTR_ERR(seed_devices); | |
2b82032c | 2263 | |
e4404d6e YZ |
2264 | old_devices = clone_fs_devices(fs_devices); |
2265 | if (IS_ERR(old_devices)) { | |
2266 | kfree(seed_devices); | |
2267 | return PTR_ERR(old_devices); | |
2b82032c | 2268 | } |
e4404d6e | 2269 | |
2b82032c YZ |
2270 | list_add(&old_devices->list, &fs_uuids); |
2271 | ||
e4404d6e YZ |
2272 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2273 | seed_devices->opened = 1; | |
2274 | INIT_LIST_HEAD(&seed_devices->devices); | |
2275 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2276 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2277 | |
0b246afa | 2278 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c XG |
2279 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2280 | synchronize_rcu); | |
2196d6e8 MX |
2281 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2282 | device->fs_devices = seed_devices; | |
c9513edb | 2283 | |
34441361 | 2284 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2285 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2286 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2287 | |
2b82032c YZ |
2288 | fs_devices->seeding = 0; |
2289 | fs_devices->num_devices = 0; | |
2290 | fs_devices->open_devices = 0; | |
69611ac8 | 2291 | fs_devices->missing_devices = 0; |
69611ac8 | 2292 | fs_devices->rotating = 0; |
e4404d6e | 2293 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2294 | |
2295 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2296 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2297 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
0b246afa | 2298 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
f7171750 | 2299 | |
2b82032c YZ |
2300 | super_flags = btrfs_super_flags(disk_super) & |
2301 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2302 | btrfs_set_super_flags(disk_super, super_flags); | |
2303 | ||
2304 | return 0; | |
2305 | } | |
2306 | ||
2307 | /* | |
01327610 | 2308 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2309 | */ |
2310 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2311 | struct btrfs_fs_info *fs_info) |
2b82032c | 2312 | { |
5b4aacef | 2313 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2314 | struct btrfs_path *path; |
2315 | struct extent_buffer *leaf; | |
2316 | struct btrfs_dev_item *dev_item; | |
2317 | struct btrfs_device *device; | |
2318 | struct btrfs_key key; | |
44880fdc | 2319 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2320 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2321 | u64 devid; | |
2322 | int ret; | |
2323 | ||
2324 | path = btrfs_alloc_path(); | |
2325 | if (!path) | |
2326 | return -ENOMEM; | |
2327 | ||
2b82032c YZ |
2328 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2329 | key.offset = 0; | |
2330 | key.type = BTRFS_DEV_ITEM_KEY; | |
2331 | ||
2332 | while (1) { | |
2333 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2334 | if (ret < 0) | |
2335 | goto error; | |
2336 | ||
2337 | leaf = path->nodes[0]; | |
2338 | next_slot: | |
2339 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2340 | ret = btrfs_next_leaf(root, path); | |
2341 | if (ret > 0) | |
2342 | break; | |
2343 | if (ret < 0) | |
2344 | goto error; | |
2345 | leaf = path->nodes[0]; | |
2346 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2347 | btrfs_release_path(path); |
2b82032c YZ |
2348 | continue; |
2349 | } | |
2350 | ||
2351 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2352 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2353 | key.type != BTRFS_DEV_ITEM_KEY) | |
2354 | break; | |
2355 | ||
2356 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2357 | struct btrfs_dev_item); | |
2358 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2359 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2360 | BTRFS_UUID_SIZE); |
1473b24e | 2361 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2362 | BTRFS_FSID_SIZE); |
0b246afa | 2363 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2364 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2365 | |
2366 | if (device->fs_devices->seeding) { | |
2367 | btrfs_set_device_generation(leaf, dev_item, | |
2368 | device->generation); | |
2369 | btrfs_mark_buffer_dirty(leaf); | |
2370 | } | |
2371 | ||
2372 | path->slots[0]++; | |
2373 | goto next_slot; | |
2374 | } | |
2375 | ret = 0; | |
2376 | error: | |
2377 | btrfs_free_path(path); | |
2378 | return ret; | |
2379 | } | |
2380 | ||
da353f6b | 2381 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2382 | { |
5112febb | 2383 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2384 | struct request_queue *q; |
788f20eb CM |
2385 | struct btrfs_trans_handle *trans; |
2386 | struct btrfs_device *device; | |
2387 | struct block_device *bdev; | |
788f20eb | 2388 | struct list_head *devices; |
0b246afa | 2389 | struct super_block *sb = fs_info->sb; |
606686ee | 2390 | struct rcu_string *name; |
3c1dbdf5 | 2391 | u64 tmp; |
2b82032c | 2392 | int seeding_dev = 0; |
788f20eb | 2393 | int ret = 0; |
7132a262 | 2394 | bool unlocked = false; |
788f20eb | 2395 | |
bc98a42c | 2396 | if (sb_rdonly(sb) && !fs_info->fs_devices->seeding) |
f8c5d0b4 | 2397 | return -EROFS; |
788f20eb | 2398 | |
a5d16333 | 2399 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2400 | fs_info->bdev_holder); |
7f59203a JB |
2401 | if (IS_ERR(bdev)) |
2402 | return PTR_ERR(bdev); | |
a2135011 | 2403 | |
0b246afa | 2404 | if (fs_info->fs_devices->seeding) { |
2b82032c YZ |
2405 | seeding_dev = 1; |
2406 | down_write(&sb->s_umount); | |
2407 | mutex_lock(&uuid_mutex); | |
2408 | } | |
2409 | ||
8c8bee1d | 2410 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2411 | |
0b246afa | 2412 | devices = &fs_info->fs_devices->devices; |
d25628bd | 2413 | |
0b246afa | 2414 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
c6e30871 | 2415 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2416 | if (device->bdev == bdev) { |
2417 | ret = -EEXIST; | |
d25628bd | 2418 | mutex_unlock( |
0b246afa | 2419 | &fs_info->fs_devices->device_list_mutex); |
2b82032c | 2420 | goto error; |
788f20eb CM |
2421 | } |
2422 | } | |
0b246afa | 2423 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2424 | |
0b246afa | 2425 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2426 | if (IS_ERR(device)) { |
788f20eb | 2427 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2428 | ret = PTR_ERR(device); |
2b82032c | 2429 | goto error; |
788f20eb CM |
2430 | } |
2431 | ||
78f2c9e6 | 2432 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2433 | if (!name) { |
2b82032c | 2434 | ret = -ENOMEM; |
5c4cf6c9 | 2435 | goto error_free_device; |
788f20eb | 2436 | } |
606686ee | 2437 | rcu_assign_pointer(device->name, name); |
2b82032c | 2438 | |
a22285a6 | 2439 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2440 | if (IS_ERR(trans)) { |
98d5dc13 | 2441 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2442 | goto error_free_device; |
98d5dc13 TI |
2443 | } |
2444 | ||
d5e2003c | 2445 | q = bdev_get_queue(bdev); |
ebbede42 | 2446 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2447 | device->generation = trans->transid; |
0b246afa JM |
2448 | device->io_width = fs_info->sectorsize; |
2449 | device->io_align = fs_info->sectorsize; | |
2450 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2451 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2452 | fs_info->sectorsize); | |
2cc3c559 | 2453 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2454 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2455 | device->fs_info = fs_info; |
788f20eb | 2456 | device->bdev = bdev; |
e12c9621 | 2457 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2458 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2459 | device->mode = FMODE_EXCL; |
27087f37 | 2460 | device->dev_stats_valid = 1; |
9f6d2510 | 2461 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2462 | |
2b82032c | 2463 | if (seeding_dev) { |
1751e8a6 | 2464 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2465 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2466 | if (ret) { |
2467 | btrfs_abort_transaction(trans, ret); | |
2468 | goto error_trans; | |
2469 | } | |
2b82032c | 2470 | } |
788f20eb | 2471 | |
0b246afa | 2472 | device->fs_devices = fs_info->fs_devices; |
e5e9a520 | 2473 | |
0b246afa | 2474 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
34441361 | 2475 | mutex_lock(&fs_info->chunk_mutex); |
0b246afa | 2476 | list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices); |
2b82032c | 2477 | list_add(&device->dev_alloc_list, |
0b246afa JM |
2478 | &fs_info->fs_devices->alloc_list); |
2479 | fs_info->fs_devices->num_devices++; | |
2480 | fs_info->fs_devices->open_devices++; | |
2481 | fs_info->fs_devices->rw_devices++; | |
2482 | fs_info->fs_devices->total_devices++; | |
2483 | fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2484 | |
a5ed45f8 | 2485 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2486 | |
e884f4f0 | 2487 | if (!blk_queue_nonrot(q)) |
0b246afa | 2488 | fs_info->fs_devices->rotating = 1; |
c289811c | 2489 | |
0b246afa JM |
2490 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2491 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
7dfb8be1 | 2492 | round_down(tmp + device->total_bytes, fs_info->sectorsize)); |
788f20eb | 2493 | |
0b246afa JM |
2494 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2495 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2496 | |
2497 | /* add sysfs device entry */ | |
0b246afa | 2498 | btrfs_sysfs_add_device_link(fs_info->fs_devices, device); |
0d39376a | 2499 | |
2196d6e8 MX |
2500 | /* |
2501 | * we've got more storage, clear any full flags on the space | |
2502 | * infos | |
2503 | */ | |
0b246afa | 2504 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2505 | |
34441361 | 2506 | mutex_unlock(&fs_info->chunk_mutex); |
0b246afa | 2507 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2508 | |
2b82032c | 2509 | if (seeding_dev) { |
34441361 | 2510 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2511 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2512 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2513 | if (ret) { |
66642832 | 2514 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2515 | goto error_sysfs; |
005d6427 | 2516 | } |
2196d6e8 MX |
2517 | } |
2518 | ||
c74a0b02 | 2519 | ret = btrfs_add_dev_item(trans, fs_info, device); |
2196d6e8 | 2520 | if (ret) { |
66642832 | 2521 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2522 | goto error_sysfs; |
2196d6e8 MX |
2523 | } |
2524 | ||
2525 | if (seeding_dev) { | |
2526 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2527 | ||
0b246afa | 2528 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2529 | if (ret) { |
66642832 | 2530 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2531 | goto error_sysfs; |
005d6427 | 2532 | } |
b2373f25 AJ |
2533 | |
2534 | /* Sprouting would change fsid of the mounted root, | |
2535 | * so rename the fsid on the sysfs | |
2536 | */ | |
2537 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa JM |
2538 | fs_info->fsid); |
2539 | if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf)) | |
2540 | btrfs_warn(fs_info, | |
2541 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2542 | } |
2543 | ||
3a45bb20 | 2544 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2545 | |
2b82032c YZ |
2546 | if (seeding_dev) { |
2547 | mutex_unlock(&uuid_mutex); | |
2548 | up_write(&sb->s_umount); | |
7132a262 | 2549 | unlocked = true; |
788f20eb | 2550 | |
79787eaa JM |
2551 | if (ret) /* transaction commit */ |
2552 | return ret; | |
2553 | ||
2ff7e61e | 2554 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2555 | if (ret < 0) |
0b246afa | 2556 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2557 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2558 | trans = btrfs_attach_transaction(root); |
2559 | if (IS_ERR(trans)) { | |
2560 | if (PTR_ERR(trans) == -ENOENT) | |
2561 | return 0; | |
7132a262 AJ |
2562 | ret = PTR_ERR(trans); |
2563 | trans = NULL; | |
2564 | goto error_sysfs; | |
671415b7 | 2565 | } |
3a45bb20 | 2566 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2567 | } |
c9e9f97b | 2568 | |
5a1972bd QW |
2569 | /* Update ctime/mtime for libblkid */ |
2570 | update_dev_time(device_path); | |
2b82032c | 2571 | return ret; |
79787eaa | 2572 | |
d31c32f6 AJ |
2573 | error_sysfs: |
2574 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); | |
79787eaa | 2575 | error_trans: |
0af2c4bf | 2576 | if (seeding_dev) |
1751e8a6 | 2577 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2578 | if (trans) |
2579 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2580 | error_free_device: |
55de4803 | 2581 | free_device(device); |
2b82032c | 2582 | error: |
e525fd89 | 2583 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2584 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2585 | mutex_unlock(&uuid_mutex); |
2586 | up_write(&sb->s_umount); | |
2587 | } | |
c9e9f97b | 2588 | return ret; |
788f20eb CM |
2589 | } |
2590 | ||
2ff7e61e | 2591 | int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, |
da353f6b | 2592 | const char *device_path, |
1c43366d | 2593 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2594 | struct btrfs_device **device_out) |
2595 | { | |
e93c89c1 SB |
2596 | struct btrfs_device *device; |
2597 | struct block_device *bdev; | |
e93c89c1 SB |
2598 | struct list_head *devices; |
2599 | struct rcu_string *name; | |
12bd2fc0 | 2600 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2601 | int ret = 0; |
2602 | ||
2603 | *device_out = NULL; | |
1c43366d MX |
2604 | if (fs_info->fs_devices->seeding) { |
2605 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2606 | return -EINVAL; |
1c43366d | 2607 | } |
e93c89c1 SB |
2608 | |
2609 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2610 | fs_info->bdev_holder); | |
1c43366d MX |
2611 | if (IS_ERR(bdev)) { |
2612 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2613 | return PTR_ERR(bdev); |
1c43366d | 2614 | } |
e93c89c1 SB |
2615 | |
2616 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2617 | ||
2618 | devices = &fs_info->fs_devices->devices; | |
2619 | list_for_each_entry(device, devices, dev_list) { | |
2620 | if (device->bdev == bdev) { | |
5d163e0e JM |
2621 | btrfs_err(fs_info, |
2622 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2623 | ret = -EEXIST; |
2624 | goto error; | |
2625 | } | |
2626 | } | |
2627 | ||
1c43366d | 2628 | |
7cc8e58d MX |
2629 | if (i_size_read(bdev->bd_inode) < |
2630 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2631 | btrfs_err(fs_info, |
2632 | "target device is smaller than source device!"); | |
1c43366d MX |
2633 | ret = -EINVAL; |
2634 | goto error; | |
2635 | } | |
2636 | ||
2637 | ||
12bd2fc0 ID |
2638 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2639 | if (IS_ERR(device)) { | |
2640 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2641 | goto error; |
2642 | } | |
2643 | ||
6165572c | 2644 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
e93c89c1 | 2645 | if (!name) { |
55de4803 | 2646 | free_device(device); |
e93c89c1 SB |
2647 | ret = -ENOMEM; |
2648 | goto error; | |
2649 | } | |
2650 | rcu_assign_pointer(device->name, name); | |
2651 | ||
0b246afa | 2652 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
ebbede42 | 2653 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
e93c89c1 | 2654 | device->generation = 0; |
0b246afa JM |
2655 | device->io_width = fs_info->sectorsize; |
2656 | device->io_align = fs_info->sectorsize; | |
2657 | device->sector_size = fs_info->sectorsize; | |
7cc8e58d MX |
2658 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2659 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2660 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2661 | ASSERT(list_empty(&srcdev->resized_list)); |
2662 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2663 | device->commit_bytes_used = device->bytes_used; |
fb456252 | 2664 | device->fs_info = fs_info; |
e93c89c1 | 2665 | device->bdev = bdev; |
e12c9621 | 2666 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2667 | set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
e93c89c1 | 2668 | device->mode = FMODE_EXCL; |
27087f37 | 2669 | device->dev_stats_valid = 1; |
9f6d2510 | 2670 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
e93c89c1 SB |
2671 | device->fs_devices = fs_info->fs_devices; |
2672 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2673 | fs_info->fs_devices->num_devices++; | |
2674 | fs_info->fs_devices->open_devices++; | |
0b246afa | 2675 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 SB |
2676 | |
2677 | *device_out = device; | |
2678 | return ret; | |
2679 | ||
2680 | error: | |
2681 | blkdev_put(bdev, FMODE_EXCL); | |
2682 | return ret; | |
2683 | } | |
2684 | ||
d397712b CM |
2685 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2686 | struct btrfs_device *device) | |
0b86a832 CM |
2687 | { |
2688 | int ret; | |
2689 | struct btrfs_path *path; | |
0b246afa | 2690 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2691 | struct btrfs_dev_item *dev_item; |
2692 | struct extent_buffer *leaf; | |
2693 | struct btrfs_key key; | |
2694 | ||
0b86a832 CM |
2695 | path = btrfs_alloc_path(); |
2696 | if (!path) | |
2697 | return -ENOMEM; | |
2698 | ||
2699 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2700 | key.type = BTRFS_DEV_ITEM_KEY; | |
2701 | key.offset = device->devid; | |
2702 | ||
2703 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2704 | if (ret < 0) | |
2705 | goto out; | |
2706 | ||
2707 | if (ret > 0) { | |
2708 | ret = -ENOENT; | |
2709 | goto out; | |
2710 | } | |
2711 | ||
2712 | leaf = path->nodes[0]; | |
2713 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2714 | ||
2715 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2716 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2717 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2718 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2719 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2720 | btrfs_set_device_total_bytes(leaf, dev_item, |
2721 | btrfs_device_get_disk_total_bytes(device)); | |
2722 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2723 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2724 | btrfs_mark_buffer_dirty(leaf); |
2725 | ||
2726 | out: | |
2727 | btrfs_free_path(path); | |
2728 | return ret; | |
2729 | } | |
2730 | ||
2196d6e8 | 2731 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2732 | struct btrfs_device *device, u64 new_size) |
2733 | { | |
0b246afa JM |
2734 | struct btrfs_fs_info *fs_info = device->fs_info; |
2735 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2736 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2737 | u64 old_total; |
2738 | u64 diff; | |
8f18cf13 | 2739 | |
ebbede42 | 2740 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2741 | return -EACCES; |
2196d6e8 | 2742 | |
7dfb8be1 NB |
2743 | new_size = round_down(new_size, fs_info->sectorsize); |
2744 | ||
34441361 | 2745 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2746 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2747 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2748 | |
63a212ab | 2749 | if (new_size <= device->total_bytes || |
401e29c1 | 2750 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2751 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2752 | return -EINVAL; |
2196d6e8 | 2753 | } |
2b82032c | 2754 | |
0b246afa | 2755 | fs_devices = fs_info->fs_devices; |
2b82032c | 2756 | |
7dfb8be1 NB |
2757 | btrfs_set_super_total_bytes(super_copy, |
2758 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2759 | device->fs_devices->total_rw_bytes += diff; |
2760 | ||
7cc8e58d MX |
2761 | btrfs_device_set_total_bytes(device, new_size); |
2762 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2763 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2764 | if (list_empty(&device->resized_list)) |
2765 | list_add_tail(&device->resized_list, | |
2766 | &fs_devices->resized_devices); | |
34441361 | 2767 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2768 | |
8f18cf13 CM |
2769 | return btrfs_update_device(trans, device); |
2770 | } | |
2771 | ||
2772 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
408fbf19 | 2773 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2774 | { |
5b4aacef | 2775 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2776 | int ret; |
2777 | struct btrfs_path *path; | |
2778 | struct btrfs_key key; | |
2779 | ||
8f18cf13 CM |
2780 | path = btrfs_alloc_path(); |
2781 | if (!path) | |
2782 | return -ENOMEM; | |
2783 | ||
408fbf19 | 2784 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2785 | key.offset = chunk_offset; |
2786 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2787 | ||
2788 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2789 | if (ret < 0) |
2790 | goto out; | |
2791 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2792 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2793 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2794 | ret = -ENOENT; |
2795 | goto out; | |
2796 | } | |
8f18cf13 CM |
2797 | |
2798 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2799 | if (ret < 0) |
0b246afa JM |
2800 | btrfs_handle_fs_error(fs_info, ret, |
2801 | "Failed to delete chunk item."); | |
79787eaa | 2802 | out: |
8f18cf13 | 2803 | btrfs_free_path(path); |
65a246c5 | 2804 | return ret; |
8f18cf13 CM |
2805 | } |
2806 | ||
408fbf19 | 2807 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2808 | { |
0b246afa | 2809 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2810 | struct btrfs_disk_key *disk_key; |
2811 | struct btrfs_chunk *chunk; | |
2812 | u8 *ptr; | |
2813 | int ret = 0; | |
2814 | u32 num_stripes; | |
2815 | u32 array_size; | |
2816 | u32 len = 0; | |
2817 | u32 cur; | |
2818 | struct btrfs_key key; | |
2819 | ||
34441361 | 2820 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2821 | array_size = btrfs_super_sys_array_size(super_copy); |
2822 | ||
2823 | ptr = super_copy->sys_chunk_array; | |
2824 | cur = 0; | |
2825 | ||
2826 | while (cur < array_size) { | |
2827 | disk_key = (struct btrfs_disk_key *)ptr; | |
2828 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2829 | ||
2830 | len = sizeof(*disk_key); | |
2831 | ||
2832 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2833 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2834 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2835 | len += btrfs_chunk_item_size(num_stripes); | |
2836 | } else { | |
2837 | ret = -EIO; | |
2838 | break; | |
2839 | } | |
408fbf19 | 2840 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2841 | key.offset == chunk_offset) { |
2842 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2843 | array_size -= len; | |
2844 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2845 | } else { | |
2846 | ptr += len; | |
2847 | cur += len; | |
2848 | } | |
2849 | } | |
34441361 | 2850 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2851 | return ret; |
2852 | } | |
2853 | ||
592d92ee LB |
2854 | static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info, |
2855 | u64 logical, u64 length) | |
2856 | { | |
2857 | struct extent_map_tree *em_tree; | |
2858 | struct extent_map *em; | |
2859 | ||
2860 | em_tree = &fs_info->mapping_tree.map_tree; | |
2861 | read_lock(&em_tree->lock); | |
2862 | em = lookup_extent_mapping(em_tree, logical, length); | |
2863 | read_unlock(&em_tree->lock); | |
2864 | ||
2865 | if (!em) { | |
2866 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2867 | logical, length); | |
2868 | return ERR_PTR(-EINVAL); | |
2869 | } | |
2870 | ||
2871 | if (em->start > logical || em->start + em->len < logical) { | |
2872 | btrfs_crit(fs_info, | |
2873 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2874 | logical, length, em->start, em->start + em->len); | |
2875 | free_extent_map(em); | |
2876 | return ERR_PTR(-EINVAL); | |
2877 | } | |
2878 | ||
2879 | /* callers are responsible for dropping em's ref. */ | |
2880 | return em; | |
2881 | } | |
2882 | ||
47ab2a6c | 2883 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
5b4aacef | 2884 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2885 | { |
8f18cf13 CM |
2886 | struct extent_map *em; |
2887 | struct map_lookup *map; | |
2196d6e8 | 2888 | u64 dev_extent_len = 0; |
47ab2a6c | 2889 | int i, ret = 0; |
0b246afa | 2890 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2891 | |
592d92ee LB |
2892 | em = get_chunk_map(fs_info, chunk_offset, 1); |
2893 | if (IS_ERR(em)) { | |
47ab2a6c JB |
2894 | /* |
2895 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2896 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2897 | * do anything we still error out. |
2898 | */ | |
2899 | ASSERT(0); | |
592d92ee | 2900 | return PTR_ERR(em); |
47ab2a6c | 2901 | } |
95617d69 | 2902 | map = em->map_lookup; |
34441361 | 2903 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 2904 | check_system_chunk(trans, fs_info, map->type); |
34441361 | 2905 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2906 | |
57ba4cb8 FM |
2907 | /* |
2908 | * Take the device list mutex to prevent races with the final phase of | |
2909 | * a device replace operation that replaces the device object associated | |
2910 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2911 | */ | |
2912 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2913 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2914 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2915 | ret = btrfs_free_dev_extent(trans, device, |
2916 | map->stripes[i].physical, | |
2917 | &dev_extent_len); | |
47ab2a6c | 2918 | if (ret) { |
57ba4cb8 | 2919 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2920 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2921 | goto out; |
2922 | } | |
a061fc8d | 2923 | |
2196d6e8 | 2924 | if (device->bytes_used > 0) { |
34441361 | 2925 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2926 | btrfs_device_set_bytes_used(device, |
2927 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2928 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2929 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2930 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2931 | } |
a061fc8d | 2932 | |
dfe25020 CM |
2933 | if (map->stripes[i].dev) { |
2934 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2935 | if (ret) { |
57ba4cb8 | 2936 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2937 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2938 | goto out; |
2939 | } | |
dfe25020 | 2940 | } |
8f18cf13 | 2941 | } |
57ba4cb8 FM |
2942 | mutex_unlock(&fs_devices->device_list_mutex); |
2943 | ||
408fbf19 | 2944 | ret = btrfs_free_chunk(trans, fs_info, chunk_offset); |
47ab2a6c | 2945 | if (ret) { |
66642832 | 2946 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2947 | goto out; |
2948 | } | |
8f18cf13 | 2949 | |
6bccf3ab | 2950 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2951 | |
8f18cf13 | 2952 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2953 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2954 | if (ret) { |
66642832 | 2955 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2956 | goto out; |
2957 | } | |
8f18cf13 CM |
2958 | } |
2959 | ||
6bccf3ab | 2960 | ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em); |
47ab2a6c | 2961 | if (ret) { |
66642832 | 2962 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2963 | goto out; |
2964 | } | |
2b82032c | 2965 | |
47ab2a6c | 2966 | out: |
2b82032c YZ |
2967 | /* once for us */ |
2968 | free_extent_map(em); | |
47ab2a6c JB |
2969 | return ret; |
2970 | } | |
2b82032c | 2971 | |
5b4aacef | 2972 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2973 | { |
5b4aacef | 2974 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2975 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2976 | int ret; |
2b82032c | 2977 | |
67c5e7d4 FM |
2978 | /* |
2979 | * Prevent races with automatic removal of unused block groups. | |
2980 | * After we relocate and before we remove the chunk with offset | |
2981 | * chunk_offset, automatic removal of the block group can kick in, | |
2982 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2983 | * | |
2984 | * Make sure to acquire this mutex before doing a tree search (dev | |
2985 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2986 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2987 | * we release the path used to search the chunk/dev tree and before | |
2988 | * the current task acquires this mutex and calls us. | |
2989 | */ | |
0b246afa | 2990 | ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex)); |
67c5e7d4 | 2991 | |
0b246afa | 2992 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2993 | if (ret) |
2994 | return -ENOSPC; | |
2995 | ||
2996 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 2997 | btrfs_scrub_pause(fs_info); |
0b246afa | 2998 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2999 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3000 | if (ret) |
3001 | return ret; | |
3002 | ||
19c4d2f9 CM |
3003 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3004 | chunk_offset); | |
3005 | if (IS_ERR(trans)) { | |
3006 | ret = PTR_ERR(trans); | |
3007 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3008 | return ret; | |
3009 | } | |
3010 | ||
47ab2a6c | 3011 | /* |
19c4d2f9 CM |
3012 | * step two, delete the device extents and the |
3013 | * chunk tree entries | |
47ab2a6c | 3014 | */ |
5b4aacef | 3015 | ret = btrfs_remove_chunk(trans, fs_info, chunk_offset); |
3a45bb20 | 3016 | btrfs_end_transaction(trans); |
19c4d2f9 | 3017 | return ret; |
2b82032c YZ |
3018 | } |
3019 | ||
2ff7e61e | 3020 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3021 | { |
0b246afa | 3022 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3023 | struct btrfs_path *path; |
3024 | struct extent_buffer *leaf; | |
3025 | struct btrfs_chunk *chunk; | |
3026 | struct btrfs_key key; | |
3027 | struct btrfs_key found_key; | |
2b82032c | 3028 | u64 chunk_type; |
ba1bf481 JB |
3029 | bool retried = false; |
3030 | int failed = 0; | |
2b82032c YZ |
3031 | int ret; |
3032 | ||
3033 | path = btrfs_alloc_path(); | |
3034 | if (!path) | |
3035 | return -ENOMEM; | |
3036 | ||
ba1bf481 | 3037 | again: |
2b82032c YZ |
3038 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3039 | key.offset = (u64)-1; | |
3040 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3041 | ||
3042 | while (1) { | |
0b246afa | 3043 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3044 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3045 | if (ret < 0) { |
0b246afa | 3046 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3047 | goto error; |
67c5e7d4 | 3048 | } |
79787eaa | 3049 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3050 | |
3051 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3052 | key.type); | |
67c5e7d4 | 3053 | if (ret) |
0b246afa | 3054 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3055 | if (ret < 0) |
3056 | goto error; | |
3057 | if (ret > 0) | |
3058 | break; | |
1a40e23b | 3059 | |
2b82032c YZ |
3060 | leaf = path->nodes[0]; |
3061 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3062 | |
2b82032c YZ |
3063 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3064 | struct btrfs_chunk); | |
3065 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3066 | btrfs_release_path(path); |
8f18cf13 | 3067 | |
2b82032c | 3068 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3069 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3070 | if (ret == -ENOSPC) |
3071 | failed++; | |
14586651 HS |
3072 | else |
3073 | BUG_ON(ret); | |
2b82032c | 3074 | } |
0b246afa | 3075 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3076 | |
2b82032c YZ |
3077 | if (found_key.offset == 0) |
3078 | break; | |
3079 | key.offset = found_key.offset - 1; | |
3080 | } | |
3081 | ret = 0; | |
ba1bf481 JB |
3082 | if (failed && !retried) { |
3083 | failed = 0; | |
3084 | retried = true; | |
3085 | goto again; | |
fae7f21c | 3086 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3087 | ret = -ENOSPC; |
3088 | } | |
2b82032c YZ |
3089 | error: |
3090 | btrfs_free_path(path); | |
3091 | return ret; | |
8f18cf13 CM |
3092 | } |
3093 | ||
a6f93c71 LB |
3094 | /* |
3095 | * return 1 : allocate a data chunk successfully, | |
3096 | * return <0: errors during allocating a data chunk, | |
3097 | * return 0 : no need to allocate a data chunk. | |
3098 | */ | |
3099 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3100 | u64 chunk_offset) | |
3101 | { | |
3102 | struct btrfs_block_group_cache *cache; | |
3103 | u64 bytes_used; | |
3104 | u64 chunk_type; | |
3105 | ||
3106 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3107 | ASSERT(cache); | |
3108 | chunk_type = cache->flags; | |
3109 | btrfs_put_block_group(cache); | |
3110 | ||
3111 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3112 | spin_lock(&fs_info->data_sinfo->lock); | |
3113 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3114 | spin_unlock(&fs_info->data_sinfo->lock); | |
3115 | ||
3116 | if (!bytes_used) { | |
3117 | struct btrfs_trans_handle *trans; | |
3118 | int ret; | |
3119 | ||
3120 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3121 | if (IS_ERR(trans)) | |
3122 | return PTR_ERR(trans); | |
3123 | ||
3124 | ret = btrfs_force_chunk_alloc(trans, fs_info, | |
3125 | BTRFS_BLOCK_GROUP_DATA); | |
3126 | btrfs_end_transaction(trans); | |
3127 | if (ret < 0) | |
3128 | return ret; | |
3129 | ||
3130 | return 1; | |
3131 | } | |
3132 | } | |
3133 | return 0; | |
3134 | } | |
3135 | ||
6bccf3ab | 3136 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3137 | struct btrfs_balance_control *bctl) |
3138 | { | |
6bccf3ab | 3139 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3140 | struct btrfs_trans_handle *trans; |
3141 | struct btrfs_balance_item *item; | |
3142 | struct btrfs_disk_balance_args disk_bargs; | |
3143 | struct btrfs_path *path; | |
3144 | struct extent_buffer *leaf; | |
3145 | struct btrfs_key key; | |
3146 | int ret, err; | |
3147 | ||
3148 | path = btrfs_alloc_path(); | |
3149 | if (!path) | |
3150 | return -ENOMEM; | |
3151 | ||
3152 | trans = btrfs_start_transaction(root, 0); | |
3153 | if (IS_ERR(trans)) { | |
3154 | btrfs_free_path(path); | |
3155 | return PTR_ERR(trans); | |
3156 | } | |
3157 | ||
3158 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3159 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3160 | key.offset = 0; |
3161 | ||
3162 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3163 | sizeof(*item)); | |
3164 | if (ret) | |
3165 | goto out; | |
3166 | ||
3167 | leaf = path->nodes[0]; | |
3168 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3169 | ||
b159fa28 | 3170 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3171 | |
3172 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3173 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3174 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3175 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3176 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3177 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3178 | ||
3179 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3180 | ||
3181 | btrfs_mark_buffer_dirty(leaf); | |
3182 | out: | |
3183 | btrfs_free_path(path); | |
3a45bb20 | 3184 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3185 | if (err && !ret) |
3186 | ret = err; | |
3187 | return ret; | |
3188 | } | |
3189 | ||
6bccf3ab | 3190 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3191 | { |
6bccf3ab | 3192 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3193 | struct btrfs_trans_handle *trans; |
3194 | struct btrfs_path *path; | |
3195 | struct btrfs_key key; | |
3196 | int ret, err; | |
3197 | ||
3198 | path = btrfs_alloc_path(); | |
3199 | if (!path) | |
3200 | return -ENOMEM; | |
3201 | ||
3202 | trans = btrfs_start_transaction(root, 0); | |
3203 | if (IS_ERR(trans)) { | |
3204 | btrfs_free_path(path); | |
3205 | return PTR_ERR(trans); | |
3206 | } | |
3207 | ||
3208 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3209 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3210 | key.offset = 0; |
3211 | ||
3212 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3213 | if (ret < 0) | |
3214 | goto out; | |
3215 | if (ret > 0) { | |
3216 | ret = -ENOENT; | |
3217 | goto out; | |
3218 | } | |
3219 | ||
3220 | ret = btrfs_del_item(trans, root, path); | |
3221 | out: | |
3222 | btrfs_free_path(path); | |
3a45bb20 | 3223 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3224 | if (err && !ret) |
3225 | ret = err; | |
3226 | return ret; | |
3227 | } | |
3228 | ||
59641015 ID |
3229 | /* |
3230 | * This is a heuristic used to reduce the number of chunks balanced on | |
3231 | * resume after balance was interrupted. | |
3232 | */ | |
3233 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3234 | { | |
3235 | /* | |
3236 | * Turn on soft mode for chunk types that were being converted. | |
3237 | */ | |
3238 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3239 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3240 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3241 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3242 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3243 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3244 | ||
3245 | /* | |
3246 | * Turn on usage filter if is not already used. The idea is | |
3247 | * that chunks that we have already balanced should be | |
3248 | * reasonably full. Don't do it for chunks that are being | |
3249 | * converted - that will keep us from relocating unconverted | |
3250 | * (albeit full) chunks. | |
3251 | */ | |
3252 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3253 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3254 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3255 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3256 | bctl->data.usage = 90; | |
3257 | } | |
3258 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3259 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3260 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3261 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3262 | bctl->sys.usage = 90; | |
3263 | } | |
3264 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3265 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3266 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3267 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3268 | bctl->meta.usage = 90; | |
3269 | } | |
3270 | } | |
3271 | ||
c9e9f97b ID |
3272 | /* |
3273 | * Should be called with both balance and volume mutexes held to | |
3274 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3275 | * restriper. Same goes for unset_balance_control. | |
3276 | */ | |
3277 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3278 | { | |
3279 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3280 | ||
3281 | BUG_ON(fs_info->balance_ctl); | |
3282 | ||
3283 | spin_lock(&fs_info->balance_lock); | |
3284 | fs_info->balance_ctl = bctl; | |
3285 | spin_unlock(&fs_info->balance_lock); | |
3286 | } | |
3287 | ||
3288 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3289 | { | |
3290 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3291 | ||
3292 | BUG_ON(!fs_info->balance_ctl); | |
3293 | ||
3294 | spin_lock(&fs_info->balance_lock); | |
3295 | fs_info->balance_ctl = NULL; | |
3296 | spin_unlock(&fs_info->balance_lock); | |
3297 | ||
3298 | kfree(bctl); | |
3299 | } | |
3300 | ||
ed25e9b2 ID |
3301 | /* |
3302 | * Balance filters. Return 1 if chunk should be filtered out | |
3303 | * (should not be balanced). | |
3304 | */ | |
899c81ea | 3305 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3306 | struct btrfs_balance_args *bargs) |
3307 | { | |
899c81ea ID |
3308 | chunk_type = chunk_to_extended(chunk_type) & |
3309 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3310 | |
899c81ea | 3311 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3312 | return 0; |
3313 | ||
3314 | return 1; | |
3315 | } | |
3316 | ||
dba72cb3 | 3317 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3318 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3319 | { |
3320 | struct btrfs_block_group_cache *cache; | |
3321 | u64 chunk_used; | |
3322 | u64 user_thresh_min; | |
3323 | u64 user_thresh_max; | |
3324 | int ret = 1; | |
3325 | ||
3326 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3327 | chunk_used = btrfs_block_group_used(&cache->item); | |
3328 | ||
3329 | if (bargs->usage_min == 0) | |
3330 | user_thresh_min = 0; | |
3331 | else | |
3332 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3333 | bargs->usage_min); | |
3334 | ||
3335 | if (bargs->usage_max == 0) | |
3336 | user_thresh_max = 1; | |
3337 | else if (bargs->usage_max > 100) | |
3338 | user_thresh_max = cache->key.offset; | |
3339 | else | |
3340 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3341 | bargs->usage_max); | |
3342 | ||
3343 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3344 | ret = 0; | |
3345 | ||
3346 | btrfs_put_block_group(cache); | |
3347 | return ret; | |
3348 | } | |
3349 | ||
dba72cb3 | 3350 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3351 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3352 | { |
3353 | struct btrfs_block_group_cache *cache; | |
3354 | u64 chunk_used, user_thresh; | |
3355 | int ret = 1; | |
3356 | ||
3357 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3358 | chunk_used = btrfs_block_group_used(&cache->item); | |
3359 | ||
bc309467 | 3360 | if (bargs->usage_min == 0) |
3e39cea6 | 3361 | user_thresh = 1; |
a105bb88 ID |
3362 | else if (bargs->usage > 100) |
3363 | user_thresh = cache->key.offset; | |
3364 | else | |
3365 | user_thresh = div_factor_fine(cache->key.offset, | |
3366 | bargs->usage); | |
3367 | ||
5ce5b3c0 ID |
3368 | if (chunk_used < user_thresh) |
3369 | ret = 0; | |
3370 | ||
3371 | btrfs_put_block_group(cache); | |
3372 | return ret; | |
3373 | } | |
3374 | ||
409d404b ID |
3375 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3376 | struct btrfs_chunk *chunk, | |
3377 | struct btrfs_balance_args *bargs) | |
3378 | { | |
3379 | struct btrfs_stripe *stripe; | |
3380 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3381 | int i; | |
3382 | ||
3383 | for (i = 0; i < num_stripes; i++) { | |
3384 | stripe = btrfs_stripe_nr(chunk, i); | |
3385 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3386 | return 0; | |
3387 | } | |
3388 | ||
3389 | return 1; | |
3390 | } | |
3391 | ||
94e60d5a ID |
3392 | /* [pstart, pend) */ |
3393 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3394 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3395 | struct btrfs_balance_args *bargs) |
3396 | { | |
3397 | struct btrfs_stripe *stripe; | |
3398 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3399 | u64 stripe_offset; | |
3400 | u64 stripe_length; | |
3401 | int factor; | |
3402 | int i; | |
3403 | ||
3404 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3405 | return 0; | |
3406 | ||
3407 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3408 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3409 | factor = num_stripes / 2; | |
3410 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3411 | factor = num_stripes - 1; | |
3412 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3413 | factor = num_stripes - 2; | |
3414 | } else { | |
3415 | factor = num_stripes; | |
3416 | } | |
94e60d5a ID |
3417 | |
3418 | for (i = 0; i < num_stripes; i++) { | |
3419 | stripe = btrfs_stripe_nr(chunk, i); | |
3420 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3421 | continue; | |
3422 | ||
3423 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3424 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3425 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3426 | |
3427 | if (stripe_offset < bargs->pend && | |
3428 | stripe_offset + stripe_length > bargs->pstart) | |
3429 | return 0; | |
3430 | } | |
3431 | ||
3432 | return 1; | |
3433 | } | |
3434 | ||
ea67176a ID |
3435 | /* [vstart, vend) */ |
3436 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3437 | struct btrfs_chunk *chunk, | |
3438 | u64 chunk_offset, | |
3439 | struct btrfs_balance_args *bargs) | |
3440 | { | |
3441 | if (chunk_offset < bargs->vend && | |
3442 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3443 | /* at least part of the chunk is inside this vrange */ | |
3444 | return 0; | |
3445 | ||
3446 | return 1; | |
3447 | } | |
3448 | ||
dee32d0a GAP |
3449 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3450 | struct btrfs_chunk *chunk, | |
3451 | struct btrfs_balance_args *bargs) | |
3452 | { | |
3453 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3454 | ||
3455 | if (bargs->stripes_min <= num_stripes | |
3456 | && num_stripes <= bargs->stripes_max) | |
3457 | return 0; | |
3458 | ||
3459 | return 1; | |
3460 | } | |
3461 | ||
899c81ea | 3462 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3463 | struct btrfs_balance_args *bargs) |
3464 | { | |
3465 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3466 | return 0; | |
3467 | ||
899c81ea ID |
3468 | chunk_type = chunk_to_extended(chunk_type) & |
3469 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3470 | |
899c81ea | 3471 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3472 | return 1; |
3473 | ||
3474 | return 0; | |
3475 | } | |
3476 | ||
2ff7e61e | 3477 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3478 | struct extent_buffer *leaf, |
3479 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3480 | { | |
0b246afa | 3481 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3482 | struct btrfs_balance_args *bargs = NULL; |
3483 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3484 | ||
3485 | /* type filter */ | |
3486 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3487 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3488 | return 0; | |
3489 | } | |
3490 | ||
3491 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3492 | bargs = &bctl->data; | |
3493 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3494 | bargs = &bctl->sys; | |
3495 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3496 | bargs = &bctl->meta; | |
3497 | ||
ed25e9b2 ID |
3498 | /* profiles filter */ |
3499 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3500 | chunk_profiles_filter(chunk_type, bargs)) { | |
3501 | return 0; | |
5ce5b3c0 ID |
3502 | } |
3503 | ||
3504 | /* usage filter */ | |
3505 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3506 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3507 | return 0; |
bc309467 | 3508 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3509 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3510 | return 0; |
409d404b ID |
3511 | } |
3512 | ||
3513 | /* devid filter */ | |
3514 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3515 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3516 | return 0; | |
94e60d5a ID |
3517 | } |
3518 | ||
3519 | /* drange filter, makes sense only with devid filter */ | |
3520 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3521 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3522 | return 0; |
ea67176a ID |
3523 | } |
3524 | ||
3525 | /* vrange filter */ | |
3526 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3527 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3528 | return 0; | |
ed25e9b2 ID |
3529 | } |
3530 | ||
dee32d0a GAP |
3531 | /* stripes filter */ |
3532 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3533 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3534 | return 0; | |
3535 | } | |
3536 | ||
cfa4c961 ID |
3537 | /* soft profile changing mode */ |
3538 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3539 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3540 | return 0; | |
3541 | } | |
3542 | ||
7d824b6f DS |
3543 | /* |
3544 | * limited by count, must be the last filter | |
3545 | */ | |
3546 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3547 | if (bargs->limit == 0) | |
3548 | return 0; | |
3549 | else | |
3550 | bargs->limit--; | |
12907fc7 DS |
3551 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3552 | /* | |
3553 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3554 | * determined here because we do not have the global information |
12907fc7 DS |
3555 | * about the count of all chunks that satisfy the filters. |
3556 | */ | |
3557 | if (bargs->limit_max == 0) | |
3558 | return 0; | |
3559 | else | |
3560 | bargs->limit_max--; | |
7d824b6f DS |
3561 | } |
3562 | ||
f43ffb60 ID |
3563 | return 1; |
3564 | } | |
3565 | ||
c9e9f97b | 3566 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3567 | { |
19a39dce | 3568 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3569 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3570 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3571 | struct list_head *devices; | |
ec44a35c CM |
3572 | struct btrfs_device *device; |
3573 | u64 old_size; | |
3574 | u64 size_to_free; | |
12907fc7 | 3575 | u64 chunk_type; |
f43ffb60 | 3576 | struct btrfs_chunk *chunk; |
5a488b9d | 3577 | struct btrfs_path *path = NULL; |
ec44a35c | 3578 | struct btrfs_key key; |
ec44a35c | 3579 | struct btrfs_key found_key; |
c9e9f97b | 3580 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3581 | struct extent_buffer *leaf; |
3582 | int slot; | |
c9e9f97b ID |
3583 | int ret; |
3584 | int enospc_errors = 0; | |
19a39dce | 3585 | bool counting = true; |
12907fc7 | 3586 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3587 | u64 limit_data = bctl->data.limit; |
3588 | u64 limit_meta = bctl->meta.limit; | |
3589 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3590 | u32 count_data = 0; |
3591 | u32 count_meta = 0; | |
3592 | u32 count_sys = 0; | |
2c9fe835 | 3593 | int chunk_reserved = 0; |
ec44a35c | 3594 | |
ec44a35c | 3595 | /* step one make some room on all the devices */ |
c9e9f97b | 3596 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3597 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3598 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3599 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3600 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3601 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3602 | btrfs_device_get_total_bytes(device) - |
3603 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3604 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3605 | continue; |
3606 | ||
3607 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3608 | if (ret == -ENOSPC) |
3609 | break; | |
5a488b9d LB |
3610 | if (ret) { |
3611 | /* btrfs_shrink_device never returns ret > 0 */ | |
3612 | WARN_ON(ret > 0); | |
3613 | goto error; | |
3614 | } | |
ec44a35c | 3615 | |
a22285a6 | 3616 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3617 | if (IS_ERR(trans)) { |
3618 | ret = PTR_ERR(trans); | |
3619 | btrfs_info_in_rcu(fs_info, | |
3620 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3621 | rcu_str_deref(device->name), ret, | |
3622 | old_size, old_size - size_to_free); | |
3623 | goto error; | |
3624 | } | |
ec44a35c CM |
3625 | |
3626 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3627 | if (ret) { |
3a45bb20 | 3628 | btrfs_end_transaction(trans); |
5a488b9d LB |
3629 | /* btrfs_grow_device never returns ret > 0 */ |
3630 | WARN_ON(ret > 0); | |
3631 | btrfs_info_in_rcu(fs_info, | |
3632 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3633 | rcu_str_deref(device->name), ret, | |
3634 | old_size, old_size - size_to_free); | |
3635 | goto error; | |
3636 | } | |
ec44a35c | 3637 | |
3a45bb20 | 3638 | btrfs_end_transaction(trans); |
ec44a35c CM |
3639 | } |
3640 | ||
3641 | /* step two, relocate all the chunks */ | |
3642 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3643 | if (!path) { |
3644 | ret = -ENOMEM; | |
3645 | goto error; | |
3646 | } | |
19a39dce ID |
3647 | |
3648 | /* zero out stat counters */ | |
3649 | spin_lock(&fs_info->balance_lock); | |
3650 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3651 | spin_unlock(&fs_info->balance_lock); | |
3652 | again: | |
7d824b6f | 3653 | if (!counting) { |
12907fc7 DS |
3654 | /* |
3655 | * The single value limit and min/max limits use the same bytes | |
3656 | * in the | |
3657 | */ | |
7d824b6f DS |
3658 | bctl->data.limit = limit_data; |
3659 | bctl->meta.limit = limit_meta; | |
3660 | bctl->sys.limit = limit_sys; | |
3661 | } | |
ec44a35c CM |
3662 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3663 | key.offset = (u64)-1; | |
3664 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3665 | ||
d397712b | 3666 | while (1) { |
19a39dce | 3667 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3668 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3669 | ret = -ECANCELED; |
3670 | goto error; | |
3671 | } | |
3672 | ||
67c5e7d4 | 3673 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3674 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3675 | if (ret < 0) { |
3676 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3677 | goto error; |
67c5e7d4 | 3678 | } |
ec44a35c CM |
3679 | |
3680 | /* | |
3681 | * this shouldn't happen, it means the last relocate | |
3682 | * failed | |
3683 | */ | |
3684 | if (ret == 0) | |
c9e9f97b | 3685 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3686 | |
3687 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3688 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3689 | if (ret) { |
67c5e7d4 | 3690 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3691 | ret = 0; |
ec44a35c | 3692 | break; |
c9e9f97b | 3693 | } |
7d9eb12c | 3694 | |
f43ffb60 ID |
3695 | leaf = path->nodes[0]; |
3696 | slot = path->slots[0]; | |
3697 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3698 | |
67c5e7d4 FM |
3699 | if (found_key.objectid != key.objectid) { |
3700 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3701 | break; |
67c5e7d4 | 3702 | } |
7d9eb12c | 3703 | |
f43ffb60 | 3704 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3705 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3706 | |
19a39dce ID |
3707 | if (!counting) { |
3708 | spin_lock(&fs_info->balance_lock); | |
3709 | bctl->stat.considered++; | |
3710 | spin_unlock(&fs_info->balance_lock); | |
3711 | } | |
3712 | ||
2ff7e61e | 3713 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3714 | found_key.offset); |
2c9fe835 | 3715 | |
b3b4aa74 | 3716 | btrfs_release_path(path); |
67c5e7d4 FM |
3717 | if (!ret) { |
3718 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3719 | goto loop; |
67c5e7d4 | 3720 | } |
f43ffb60 | 3721 | |
19a39dce | 3722 | if (counting) { |
67c5e7d4 | 3723 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3724 | spin_lock(&fs_info->balance_lock); |
3725 | bctl->stat.expected++; | |
3726 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3727 | |
3728 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3729 | count_data++; | |
3730 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3731 | count_sys++; | |
3732 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3733 | count_meta++; | |
3734 | ||
3735 | goto loop; | |
3736 | } | |
3737 | ||
3738 | /* | |
3739 | * Apply limit_min filter, no need to check if the LIMITS | |
3740 | * filter is used, limit_min is 0 by default | |
3741 | */ | |
3742 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3743 | count_data < bctl->data.limit_min) | |
3744 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3745 | count_meta < bctl->meta.limit_min) | |
3746 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3747 | count_sys < bctl->sys.limit_min)) { | |
3748 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3749 | goto loop; |
3750 | } | |
3751 | ||
a6f93c71 LB |
3752 | if (!chunk_reserved) { |
3753 | /* | |
3754 | * We may be relocating the only data chunk we have, | |
3755 | * which could potentially end up with losing data's | |
3756 | * raid profile, so lets allocate an empty one in | |
3757 | * advance. | |
3758 | */ | |
3759 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3760 | found_key.offset); | |
2c9fe835 ZL |
3761 | if (ret < 0) { |
3762 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3763 | goto error; | |
a6f93c71 LB |
3764 | } else if (ret == 1) { |
3765 | chunk_reserved = 1; | |
2c9fe835 | 3766 | } |
2c9fe835 ZL |
3767 | } |
3768 | ||
5b4aacef | 3769 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3770 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3771 | if (ret && ret != -ENOSPC) |
3772 | goto error; | |
19a39dce | 3773 | if (ret == -ENOSPC) { |
c9e9f97b | 3774 | enospc_errors++; |
19a39dce ID |
3775 | } else { |
3776 | spin_lock(&fs_info->balance_lock); | |
3777 | bctl->stat.completed++; | |
3778 | spin_unlock(&fs_info->balance_lock); | |
3779 | } | |
f43ffb60 | 3780 | loop: |
795a3321 ID |
3781 | if (found_key.offset == 0) |
3782 | break; | |
ba1bf481 | 3783 | key.offset = found_key.offset - 1; |
ec44a35c | 3784 | } |
c9e9f97b | 3785 | |
19a39dce ID |
3786 | if (counting) { |
3787 | btrfs_release_path(path); | |
3788 | counting = false; | |
3789 | goto again; | |
3790 | } | |
ec44a35c CM |
3791 | error: |
3792 | btrfs_free_path(path); | |
c9e9f97b | 3793 | if (enospc_errors) { |
efe120a0 | 3794 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3795 | enospc_errors); |
c9e9f97b ID |
3796 | if (!ret) |
3797 | ret = -ENOSPC; | |
3798 | } | |
3799 | ||
ec44a35c CM |
3800 | return ret; |
3801 | } | |
3802 | ||
0c460c0d ID |
3803 | /** |
3804 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3805 | * @flags: profile to validate | |
3806 | * @extended: if true @flags is treated as an extended profile | |
3807 | */ | |
3808 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3809 | { | |
3810 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3811 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3812 | ||
3813 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3814 | ||
3815 | /* 1) check that all other bits are zeroed */ | |
3816 | if (flags & ~mask) | |
3817 | return 0; | |
3818 | ||
3819 | /* 2) see if profile is reduced */ | |
3820 | if (flags == 0) | |
3821 | return !extended; /* "0" is valid for usual profiles */ | |
3822 | ||
3823 | /* true if exactly one bit set */ | |
3824 | return (flags & (flags - 1)) == 0; | |
3825 | } | |
3826 | ||
837d5b6e ID |
3827 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3828 | { | |
a7e99c69 ID |
3829 | /* cancel requested || normal exit path */ |
3830 | return atomic_read(&fs_info->balance_cancel_req) || | |
3831 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3832 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3833 | } |
3834 | ||
c9e9f97b ID |
3835 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3836 | { | |
0940ebf6 ID |
3837 | int ret; |
3838 | ||
c9e9f97b | 3839 | unset_balance_control(fs_info); |
6bccf3ab | 3840 | ret = del_balance_item(fs_info); |
0f788c58 | 3841 | if (ret) |
34d97007 | 3842 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f | 3843 | |
171938e5 | 3844 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
c9e9f97b ID |
3845 | } |
3846 | ||
bdcd3c97 AM |
3847 | /* Non-zero return value signifies invalidity */ |
3848 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3849 | u64 allowed) | |
3850 | { | |
3851 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3852 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3853 | (bctl_arg->target & ~allowed))); | |
3854 | } | |
3855 | ||
c9e9f97b ID |
3856 | /* |
3857 | * Should be called with both balance and volume mutexes held | |
3858 | */ | |
3859 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3860 | struct btrfs_ioctl_balance_args *bargs) | |
3861 | { | |
3862 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
14506127 | 3863 | u64 meta_target, data_target; |
f43ffb60 | 3864 | u64 allowed; |
e4837f8f | 3865 | int mixed = 0; |
c9e9f97b | 3866 | int ret; |
8dabb742 | 3867 | u64 num_devices; |
de98ced9 | 3868 | unsigned seq; |
c9e9f97b | 3869 | |
837d5b6e | 3870 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3871 | atomic_read(&fs_info->balance_pause_req) || |
3872 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3873 | ret = -EINVAL; |
3874 | goto out; | |
3875 | } | |
3876 | ||
e4837f8f ID |
3877 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3878 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3879 | mixed = 1; | |
3880 | ||
f43ffb60 ID |
3881 | /* |
3882 | * In case of mixed groups both data and meta should be picked, | |
3883 | * and identical options should be given for both of them. | |
3884 | */ | |
e4837f8f ID |
3885 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3886 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3887 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3888 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3889 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3890 | btrfs_err(fs_info, |
3891 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3892 | ret = -EINVAL; |
3893 | goto out; | |
3894 | } | |
3895 | } | |
3896 | ||
8dabb742 | 3897 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3898 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3899 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3900 | BUG_ON(num_devices < 1); | |
3901 | num_devices--; | |
3902 | } | |
73beece9 | 3903 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
88be159c AH |
3904 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3905 | if (num_devices > 1) | |
e4d8ec0f | 3906 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3907 | if (num_devices > 2) |
3908 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3909 | if (num_devices > 3) | |
3910 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3911 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3912 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3913 | btrfs_err(fs_info, |
3914 | "unable to start balance with target data profile %llu", | |
3915 | bctl->data.target); | |
e4d8ec0f ID |
3916 | ret = -EINVAL; |
3917 | goto out; | |
3918 | } | |
bdcd3c97 | 3919 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3920 | btrfs_err(fs_info, |
5d163e0e JM |
3921 | "unable to start balance with target metadata profile %llu", |
3922 | bctl->meta.target); | |
e4d8ec0f ID |
3923 | ret = -EINVAL; |
3924 | goto out; | |
3925 | } | |
bdcd3c97 | 3926 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3927 | btrfs_err(fs_info, |
5d163e0e JM |
3928 | "unable to start balance with target system profile %llu", |
3929 | bctl->sys.target); | |
e4d8ec0f ID |
3930 | ret = -EINVAL; |
3931 | goto out; | |
3932 | } | |
3933 | ||
e4d8ec0f ID |
3934 | /* allow to reduce meta or sys integrity only if force set */ |
3935 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3936 | BTRFS_BLOCK_GROUP_RAID10 | |
3937 | BTRFS_BLOCK_GROUP_RAID5 | | |
3938 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3939 | do { |
3940 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3941 | ||
3942 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3943 | (fs_info->avail_system_alloc_bits & allowed) && | |
3944 | !(bctl->sys.target & allowed)) || | |
3945 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3946 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3947 | !(bctl->meta.target & allowed))) { | |
3948 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3949 | btrfs_info(fs_info, |
3950 | "force reducing metadata integrity"); | |
de98ced9 | 3951 | } else { |
5d163e0e JM |
3952 | btrfs_err(fs_info, |
3953 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3954 | ret = -EINVAL; |
3955 | goto out; | |
3956 | } | |
e4d8ec0f | 3957 | } |
de98ced9 | 3958 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3959 | |
14506127 AB |
3960 | /* if we're not converting, the target field is uninitialized */ |
3961 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3962 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3963 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3964 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3965 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3966 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 3967 | btrfs_warn(fs_info, |
5d163e0e | 3968 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
14506127 | 3969 | meta_target, data_target); |
ee592d07 ST |
3970 | } |
3971 | ||
6bccf3ab | 3972 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3973 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3974 | goto out; |
3975 | ||
59641015 ID |
3976 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3977 | BUG_ON(ret == -EEXIST); | |
3978 | set_balance_control(bctl); | |
3979 | } else { | |
3980 | BUG_ON(ret != -EEXIST); | |
3981 | spin_lock(&fs_info->balance_lock); | |
3982 | update_balance_args(bctl); | |
3983 | spin_unlock(&fs_info->balance_lock); | |
3984 | } | |
c9e9f97b | 3985 | |
837d5b6e | 3986 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3987 | mutex_unlock(&fs_info->balance_mutex); |
3988 | ||
3989 | ret = __btrfs_balance(fs_info); | |
3990 | ||
3991 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3992 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
3993 | |
3994 | if (bargs) { | |
3995 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3996 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3997 | } |
3998 | ||
3a01aa7a ID |
3999 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4000 | balance_need_close(fs_info)) { | |
4001 | __cancel_balance(fs_info); | |
4002 | } | |
4003 | ||
837d5b6e | 4004 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4005 | |
4006 | return ret; | |
4007 | out: | |
59641015 ID |
4008 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
4009 | __cancel_balance(fs_info); | |
ed0fb78f | 4010 | else { |
59641015 | 4011 | kfree(bctl); |
171938e5 | 4012 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
ed0fb78f | 4013 | } |
59641015 ID |
4014 | return ret; |
4015 | } | |
4016 | ||
4017 | static int balance_kthread(void *data) | |
4018 | { | |
2b6ba629 | 4019 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4020 | int ret = 0; |
59641015 ID |
4021 | |
4022 | mutex_lock(&fs_info->volume_mutex); | |
4023 | mutex_lock(&fs_info->balance_mutex); | |
4024 | ||
2b6ba629 | 4025 | if (fs_info->balance_ctl) { |
efe120a0 | 4026 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 4027 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 4028 | } |
59641015 ID |
4029 | |
4030 | mutex_unlock(&fs_info->balance_mutex); | |
4031 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 4032 | |
59641015 ID |
4033 | return ret; |
4034 | } | |
4035 | ||
2b6ba629 ID |
4036 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4037 | { | |
4038 | struct task_struct *tsk; | |
4039 | ||
4040 | spin_lock(&fs_info->balance_lock); | |
4041 | if (!fs_info->balance_ctl) { | |
4042 | spin_unlock(&fs_info->balance_lock); | |
4043 | return 0; | |
4044 | } | |
4045 | spin_unlock(&fs_info->balance_lock); | |
4046 | ||
3cdde224 | 4047 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 4048 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
4049 | return 0; |
4050 | } | |
4051 | ||
4052 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 4053 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4054 | } |
4055 | ||
68310a5e | 4056 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4057 | { |
59641015 ID |
4058 | struct btrfs_balance_control *bctl; |
4059 | struct btrfs_balance_item *item; | |
4060 | struct btrfs_disk_balance_args disk_bargs; | |
4061 | struct btrfs_path *path; | |
4062 | struct extent_buffer *leaf; | |
4063 | struct btrfs_key key; | |
4064 | int ret; | |
4065 | ||
4066 | path = btrfs_alloc_path(); | |
4067 | if (!path) | |
4068 | return -ENOMEM; | |
4069 | ||
59641015 | 4070 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4071 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4072 | key.offset = 0; |
4073 | ||
68310a5e | 4074 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4075 | if (ret < 0) |
68310a5e | 4076 | goto out; |
59641015 ID |
4077 | if (ret > 0) { /* ret = -ENOENT; */ |
4078 | ret = 0; | |
68310a5e ID |
4079 | goto out; |
4080 | } | |
4081 | ||
4082 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4083 | if (!bctl) { | |
4084 | ret = -ENOMEM; | |
4085 | goto out; | |
59641015 ID |
4086 | } |
4087 | ||
4088 | leaf = path->nodes[0]; | |
4089 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4090 | ||
68310a5e ID |
4091 | bctl->fs_info = fs_info; |
4092 | bctl->flags = btrfs_balance_flags(leaf, item); | |
4093 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4094 | |
4095 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4096 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4097 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4098 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4099 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4100 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4101 | ||
171938e5 | 4102 | WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)); |
ed0fb78f | 4103 | |
68310a5e ID |
4104 | mutex_lock(&fs_info->volume_mutex); |
4105 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 4106 | |
68310a5e ID |
4107 | set_balance_control(bctl); |
4108 | ||
4109 | mutex_unlock(&fs_info->balance_mutex); | |
4110 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4111 | out: |
4112 | btrfs_free_path(path); | |
ec44a35c CM |
4113 | return ret; |
4114 | } | |
4115 | ||
837d5b6e ID |
4116 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4117 | { | |
4118 | int ret = 0; | |
4119 | ||
4120 | mutex_lock(&fs_info->balance_mutex); | |
4121 | if (!fs_info->balance_ctl) { | |
4122 | mutex_unlock(&fs_info->balance_mutex); | |
4123 | return -ENOTCONN; | |
4124 | } | |
4125 | ||
4126 | if (atomic_read(&fs_info->balance_running)) { | |
4127 | atomic_inc(&fs_info->balance_pause_req); | |
4128 | mutex_unlock(&fs_info->balance_mutex); | |
4129 | ||
4130 | wait_event(fs_info->balance_wait_q, | |
4131 | atomic_read(&fs_info->balance_running) == 0); | |
4132 | ||
4133 | mutex_lock(&fs_info->balance_mutex); | |
4134 | /* we are good with balance_ctl ripped off from under us */ | |
4135 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4136 | atomic_dec(&fs_info->balance_pause_req); | |
4137 | } else { | |
4138 | ret = -ENOTCONN; | |
4139 | } | |
4140 | ||
4141 | mutex_unlock(&fs_info->balance_mutex); | |
4142 | return ret; | |
4143 | } | |
4144 | ||
a7e99c69 ID |
4145 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4146 | { | |
bc98a42c | 4147 | if (sb_rdonly(fs_info->sb)) |
e649e587 ID |
4148 | return -EROFS; |
4149 | ||
a7e99c69 ID |
4150 | mutex_lock(&fs_info->balance_mutex); |
4151 | if (!fs_info->balance_ctl) { | |
4152 | mutex_unlock(&fs_info->balance_mutex); | |
4153 | return -ENOTCONN; | |
4154 | } | |
4155 | ||
4156 | atomic_inc(&fs_info->balance_cancel_req); | |
4157 | /* | |
4158 | * if we are running just wait and return, balance item is | |
4159 | * deleted in btrfs_balance in this case | |
4160 | */ | |
4161 | if (atomic_read(&fs_info->balance_running)) { | |
4162 | mutex_unlock(&fs_info->balance_mutex); | |
4163 | wait_event(fs_info->balance_wait_q, | |
4164 | atomic_read(&fs_info->balance_running) == 0); | |
4165 | mutex_lock(&fs_info->balance_mutex); | |
4166 | } else { | |
4167 | /* __cancel_balance needs volume_mutex */ | |
4168 | mutex_unlock(&fs_info->balance_mutex); | |
4169 | mutex_lock(&fs_info->volume_mutex); | |
4170 | mutex_lock(&fs_info->balance_mutex); | |
4171 | ||
4172 | if (fs_info->balance_ctl) | |
4173 | __cancel_balance(fs_info); | |
4174 | ||
4175 | mutex_unlock(&fs_info->volume_mutex); | |
4176 | } | |
4177 | ||
4178 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4179 | atomic_dec(&fs_info->balance_cancel_req); | |
4180 | mutex_unlock(&fs_info->balance_mutex); | |
4181 | return 0; | |
4182 | } | |
4183 | ||
803b2f54 SB |
4184 | static int btrfs_uuid_scan_kthread(void *data) |
4185 | { | |
4186 | struct btrfs_fs_info *fs_info = data; | |
4187 | struct btrfs_root *root = fs_info->tree_root; | |
4188 | struct btrfs_key key; | |
803b2f54 SB |
4189 | struct btrfs_path *path = NULL; |
4190 | int ret = 0; | |
4191 | struct extent_buffer *eb; | |
4192 | int slot; | |
4193 | struct btrfs_root_item root_item; | |
4194 | u32 item_size; | |
f45388f3 | 4195 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4196 | |
4197 | path = btrfs_alloc_path(); | |
4198 | if (!path) { | |
4199 | ret = -ENOMEM; | |
4200 | goto out; | |
4201 | } | |
4202 | ||
4203 | key.objectid = 0; | |
4204 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4205 | key.offset = 0; | |
4206 | ||
803b2f54 | 4207 | while (1) { |
6174d3cb | 4208 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4209 | if (ret) { |
4210 | if (ret > 0) | |
4211 | ret = 0; | |
4212 | break; | |
4213 | } | |
4214 | ||
4215 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4216 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4217 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4218 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4219 | goto skip; | |
4220 | ||
4221 | eb = path->nodes[0]; | |
4222 | slot = path->slots[0]; | |
4223 | item_size = btrfs_item_size_nr(eb, slot); | |
4224 | if (item_size < sizeof(root_item)) | |
4225 | goto skip; | |
4226 | ||
803b2f54 SB |
4227 | read_extent_buffer(eb, &root_item, |
4228 | btrfs_item_ptr_offset(eb, slot), | |
4229 | (int)sizeof(root_item)); | |
4230 | if (btrfs_root_refs(&root_item) == 0) | |
4231 | goto skip; | |
f45388f3 FDBM |
4232 | |
4233 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4234 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4235 | if (trans) | |
4236 | goto update_tree; | |
4237 | ||
4238 | btrfs_release_path(path); | |
803b2f54 SB |
4239 | /* |
4240 | * 1 - subvol uuid item | |
4241 | * 1 - received_subvol uuid item | |
4242 | */ | |
4243 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4244 | if (IS_ERR(trans)) { | |
4245 | ret = PTR_ERR(trans); | |
4246 | break; | |
4247 | } | |
f45388f3 FDBM |
4248 | continue; |
4249 | } else { | |
4250 | goto skip; | |
4251 | } | |
4252 | update_tree: | |
4253 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
6bccf3ab | 4254 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4255 | root_item.uuid, |
4256 | BTRFS_UUID_KEY_SUBVOL, | |
4257 | key.objectid); | |
4258 | if (ret < 0) { | |
efe120a0 | 4259 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4260 | ret); |
803b2f54 SB |
4261 | break; |
4262 | } | |
4263 | } | |
4264 | ||
4265 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
6bccf3ab | 4266 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4267 | root_item.received_uuid, |
4268 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4269 | key.objectid); | |
4270 | if (ret < 0) { | |
efe120a0 | 4271 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4272 | ret); |
803b2f54 SB |
4273 | break; |
4274 | } | |
4275 | } | |
4276 | ||
f45388f3 | 4277 | skip: |
803b2f54 | 4278 | if (trans) { |
3a45bb20 | 4279 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4280 | trans = NULL; |
803b2f54 SB |
4281 | if (ret) |
4282 | break; | |
4283 | } | |
4284 | ||
803b2f54 SB |
4285 | btrfs_release_path(path); |
4286 | if (key.offset < (u64)-1) { | |
4287 | key.offset++; | |
4288 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4289 | key.offset = 0; | |
4290 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4291 | } else if (key.objectid < (u64)-1) { | |
4292 | key.offset = 0; | |
4293 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4294 | key.objectid++; | |
4295 | } else { | |
4296 | break; | |
4297 | } | |
4298 | cond_resched(); | |
4299 | } | |
4300 | ||
4301 | out: | |
4302 | btrfs_free_path(path); | |
f45388f3 | 4303 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4304 | btrfs_end_transaction(trans); |
803b2f54 | 4305 | if (ret) |
efe120a0 | 4306 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4307 | else |
afcdd129 | 4308 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4309 | up(&fs_info->uuid_tree_rescan_sem); |
4310 | return 0; | |
4311 | } | |
4312 | ||
70f80175 SB |
4313 | /* |
4314 | * Callback for btrfs_uuid_tree_iterate(). | |
4315 | * returns: | |
4316 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4317 | * < 0 if an error occurred. |
70f80175 SB |
4318 | * > 0 if the check failed, which means the caller shall remove the entry. |
4319 | */ | |
4320 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4321 | u8 *uuid, u8 type, u64 subid) | |
4322 | { | |
4323 | struct btrfs_key key; | |
4324 | int ret = 0; | |
4325 | struct btrfs_root *subvol_root; | |
4326 | ||
4327 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4328 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4329 | goto out; | |
4330 | ||
4331 | key.objectid = subid; | |
4332 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4333 | key.offset = (u64)-1; | |
4334 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4335 | if (IS_ERR(subvol_root)) { | |
4336 | ret = PTR_ERR(subvol_root); | |
4337 | if (ret == -ENOENT) | |
4338 | ret = 1; | |
4339 | goto out; | |
4340 | } | |
4341 | ||
4342 | switch (type) { | |
4343 | case BTRFS_UUID_KEY_SUBVOL: | |
4344 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4345 | ret = 1; | |
4346 | break; | |
4347 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4348 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4349 | BTRFS_UUID_SIZE)) | |
4350 | ret = 1; | |
4351 | break; | |
4352 | } | |
4353 | ||
4354 | out: | |
4355 | return ret; | |
4356 | } | |
4357 | ||
4358 | static int btrfs_uuid_rescan_kthread(void *data) | |
4359 | { | |
4360 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4361 | int ret; | |
4362 | ||
4363 | /* | |
4364 | * 1st step is to iterate through the existing UUID tree and | |
4365 | * to delete all entries that contain outdated data. | |
4366 | * 2nd step is to add all missing entries to the UUID tree. | |
4367 | */ | |
4368 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4369 | if (ret < 0) { | |
efe120a0 | 4370 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4371 | up(&fs_info->uuid_tree_rescan_sem); |
4372 | return ret; | |
4373 | } | |
4374 | return btrfs_uuid_scan_kthread(data); | |
4375 | } | |
4376 | ||
f7a81ea4 SB |
4377 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4378 | { | |
4379 | struct btrfs_trans_handle *trans; | |
4380 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4381 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4382 | struct task_struct *task; |
4383 | int ret; | |
f7a81ea4 SB |
4384 | |
4385 | /* | |
4386 | * 1 - root node | |
4387 | * 1 - root item | |
4388 | */ | |
4389 | trans = btrfs_start_transaction(tree_root, 2); | |
4390 | if (IS_ERR(trans)) | |
4391 | return PTR_ERR(trans); | |
4392 | ||
4393 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4394 | BTRFS_UUID_TREE_OBJECTID); | |
4395 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4396 | ret = PTR_ERR(uuid_root); |
66642832 | 4397 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4398 | btrfs_end_transaction(trans); |
6d13f549 | 4399 | return ret; |
f7a81ea4 SB |
4400 | } |
4401 | ||
4402 | fs_info->uuid_root = uuid_root; | |
4403 | ||
3a45bb20 | 4404 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4405 | if (ret) |
4406 | return ret; | |
4407 | ||
4408 | down(&fs_info->uuid_tree_rescan_sem); | |
4409 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4410 | if (IS_ERR(task)) { | |
70f80175 | 4411 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4412 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4413 | up(&fs_info->uuid_tree_rescan_sem); |
4414 | return PTR_ERR(task); | |
4415 | } | |
4416 | ||
4417 | return 0; | |
f7a81ea4 | 4418 | } |
803b2f54 | 4419 | |
70f80175 SB |
4420 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4421 | { | |
4422 | struct task_struct *task; | |
4423 | ||
4424 | down(&fs_info->uuid_tree_rescan_sem); | |
4425 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4426 | if (IS_ERR(task)) { | |
4427 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4428 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4429 | up(&fs_info->uuid_tree_rescan_sem); |
4430 | return PTR_ERR(task); | |
4431 | } | |
4432 | ||
4433 | return 0; | |
4434 | } | |
4435 | ||
8f18cf13 CM |
4436 | /* |
4437 | * shrinking a device means finding all of the device extents past | |
4438 | * the new size, and then following the back refs to the chunks. | |
4439 | * The chunk relocation code actually frees the device extent | |
4440 | */ | |
4441 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4442 | { | |
0b246afa JM |
4443 | struct btrfs_fs_info *fs_info = device->fs_info; |
4444 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4445 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4446 | struct btrfs_dev_extent *dev_extent = NULL; |
4447 | struct btrfs_path *path; | |
4448 | u64 length; | |
8f18cf13 CM |
4449 | u64 chunk_offset; |
4450 | int ret; | |
4451 | int slot; | |
ba1bf481 JB |
4452 | int failed = 0; |
4453 | bool retried = false; | |
53e489bc | 4454 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4455 | struct extent_buffer *l; |
4456 | struct btrfs_key key; | |
0b246afa | 4457 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4458 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4459 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4460 | u64 diff; |
4461 | ||
4462 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4463 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4464 | |
401e29c1 | 4465 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4466 | return -EINVAL; |
4467 | ||
8f18cf13 CM |
4468 | path = btrfs_alloc_path(); |
4469 | if (!path) | |
4470 | return -ENOMEM; | |
4471 | ||
e4058b54 | 4472 | path->reada = READA_FORWARD; |
8f18cf13 | 4473 | |
34441361 | 4474 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4475 | |
7cc8e58d | 4476 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4477 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4478 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4479 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4480 | } |
34441361 | 4481 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4482 | |
ba1bf481 | 4483 | again: |
8f18cf13 CM |
4484 | key.objectid = device->devid; |
4485 | key.offset = (u64)-1; | |
4486 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4487 | ||
213e64da | 4488 | do { |
0b246afa | 4489 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4490 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4491 | if (ret < 0) { |
0b246afa | 4492 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4493 | goto done; |
67c5e7d4 | 4494 | } |
8f18cf13 CM |
4495 | |
4496 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4497 | if (ret) |
0b246afa | 4498 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4499 | if (ret < 0) |
4500 | goto done; | |
4501 | if (ret) { | |
4502 | ret = 0; | |
b3b4aa74 | 4503 | btrfs_release_path(path); |
bf1fb512 | 4504 | break; |
8f18cf13 CM |
4505 | } |
4506 | ||
4507 | l = path->nodes[0]; | |
4508 | slot = path->slots[0]; | |
4509 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4510 | ||
ba1bf481 | 4511 | if (key.objectid != device->devid) { |
0b246afa | 4512 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4513 | btrfs_release_path(path); |
bf1fb512 | 4514 | break; |
ba1bf481 | 4515 | } |
8f18cf13 CM |
4516 | |
4517 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4518 | length = btrfs_dev_extent_length(l, dev_extent); | |
4519 | ||
ba1bf481 | 4520 | if (key.offset + length <= new_size) { |
0b246afa | 4521 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4522 | btrfs_release_path(path); |
d6397bae | 4523 | break; |
ba1bf481 | 4524 | } |
8f18cf13 | 4525 | |
8f18cf13 | 4526 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4527 | btrfs_release_path(path); |
8f18cf13 | 4528 | |
a6f93c71 LB |
4529 | /* |
4530 | * We may be relocating the only data chunk we have, | |
4531 | * which could potentially end up with losing data's | |
4532 | * raid profile, so lets allocate an empty one in | |
4533 | * advance. | |
4534 | */ | |
4535 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4536 | if (ret < 0) { | |
4537 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4538 | goto done; | |
4539 | } | |
4540 | ||
0b246afa JM |
4541 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4542 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4543 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4544 | goto done; |
ba1bf481 JB |
4545 | if (ret == -ENOSPC) |
4546 | failed++; | |
213e64da | 4547 | } while (key.offset-- > 0); |
ba1bf481 JB |
4548 | |
4549 | if (failed && !retried) { | |
4550 | failed = 0; | |
4551 | retried = true; | |
4552 | goto again; | |
4553 | } else if (failed && retried) { | |
4554 | ret = -ENOSPC; | |
ba1bf481 | 4555 | goto done; |
8f18cf13 CM |
4556 | } |
4557 | ||
d6397bae | 4558 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4559 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4560 | if (IS_ERR(trans)) { |
4561 | ret = PTR_ERR(trans); | |
4562 | goto done; | |
4563 | } | |
4564 | ||
34441361 | 4565 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4566 | |
4567 | /* | |
4568 | * We checked in the above loop all device extents that were already in | |
4569 | * the device tree. However before we have updated the device's | |
4570 | * total_bytes to the new size, we might have had chunk allocations that | |
4571 | * have not complete yet (new block groups attached to transaction | |
4572 | * handles), and therefore their device extents were not yet in the | |
4573 | * device tree and we missed them in the loop above. So if we have any | |
4574 | * pending chunk using a device extent that overlaps the device range | |
4575 | * that we can not use anymore, commit the current transaction and | |
4576 | * repeat the search on the device tree - this way we guarantee we will | |
4577 | * not have chunks using device extents that end beyond 'new_size'. | |
4578 | */ | |
4579 | if (!checked_pending_chunks) { | |
4580 | u64 start = new_size; | |
4581 | u64 len = old_size - new_size; | |
4582 | ||
499f377f JM |
4583 | if (contains_pending_extent(trans->transaction, device, |
4584 | &start, len)) { | |
34441361 | 4585 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4586 | checked_pending_chunks = true; |
4587 | failed = 0; | |
4588 | retried = false; | |
3a45bb20 | 4589 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4590 | if (ret) |
4591 | goto done; | |
4592 | goto again; | |
4593 | } | |
4594 | } | |
4595 | ||
7cc8e58d | 4596 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4597 | if (list_empty(&device->resized_list)) |
4598 | list_add_tail(&device->resized_list, | |
0b246afa | 4599 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4600 | |
d6397bae | 4601 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4602 | btrfs_set_super_total_bytes(super_copy, |
4603 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4604 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4605 | |
4606 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4607 | ret = btrfs_update_device(trans, device); | |
3a45bb20 | 4608 | btrfs_end_transaction(trans); |
8f18cf13 CM |
4609 | done: |
4610 | btrfs_free_path(path); | |
53e489bc | 4611 | if (ret) { |
34441361 | 4612 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4613 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4614 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4615 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4616 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4617 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4618 | } |
8f18cf13 CM |
4619 | return ret; |
4620 | } | |
4621 | ||
2ff7e61e | 4622 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4623 | struct btrfs_key *key, |
4624 | struct btrfs_chunk *chunk, int item_size) | |
4625 | { | |
0b246afa | 4626 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4627 | struct btrfs_disk_key disk_key; |
4628 | u32 array_size; | |
4629 | u8 *ptr; | |
4630 | ||
34441361 | 4631 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4632 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4633 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4634 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4635 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4636 | return -EFBIG; |
fe48a5c0 | 4637 | } |
0b86a832 CM |
4638 | |
4639 | ptr = super_copy->sys_chunk_array + array_size; | |
4640 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4641 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4642 | ptr += sizeof(disk_key); | |
4643 | memcpy(ptr, chunk, item_size); | |
4644 | item_size += sizeof(disk_key); | |
4645 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4646 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4647 | |
0b86a832 CM |
4648 | return 0; |
4649 | } | |
4650 | ||
73c5de00 AJ |
4651 | /* |
4652 | * sort the devices in descending order by max_avail, total_avail | |
4653 | */ | |
4654 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4655 | { |
73c5de00 AJ |
4656 | const struct btrfs_device_info *di_a = a; |
4657 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4658 | |
73c5de00 | 4659 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4660 | return -1; |
73c5de00 | 4661 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4662 | return 1; |
73c5de00 AJ |
4663 | if (di_a->total_avail > di_b->total_avail) |
4664 | return -1; | |
4665 | if (di_a->total_avail < di_b->total_avail) | |
4666 | return 1; | |
4667 | return 0; | |
b2117a39 | 4668 | } |
0b86a832 | 4669 | |
53b381b3 DW |
4670 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4671 | { | |
ffe2d203 | 4672 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4673 | return; |
4674 | ||
ceda0864 | 4675 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4676 | } |
4677 | ||
062d4d1f | 4678 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
4679 | - sizeof(struct btrfs_chunk)) \ |
4680 | / sizeof(struct btrfs_stripe) + 1) | |
4681 | ||
4682 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4683 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4684 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4685 | / sizeof(struct btrfs_stripe) + 1) | |
4686 | ||
73c5de00 | 4687 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4688 | u64 start, u64 type) |
b2117a39 | 4689 | { |
2ff7e61e | 4690 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4691 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4692 | struct btrfs_device *device; |
73c5de00 AJ |
4693 | struct map_lookup *map = NULL; |
4694 | struct extent_map_tree *em_tree; | |
4695 | struct extent_map *em; | |
4696 | struct btrfs_device_info *devices_info = NULL; | |
4697 | u64 total_avail; | |
4698 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4699 | int data_stripes; /* number of stripes that count for |
4700 | block group size */ | |
73c5de00 AJ |
4701 | int sub_stripes; /* sub_stripes info for map */ |
4702 | int dev_stripes; /* stripes per dev */ | |
4703 | int devs_max; /* max devs to use */ | |
4704 | int devs_min; /* min devs needed */ | |
4705 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4706 | int ncopies; /* how many copies to data has */ | |
4707 | int ret; | |
4708 | u64 max_stripe_size; | |
4709 | u64 max_chunk_size; | |
4710 | u64 stripe_size; | |
4711 | u64 num_bytes; | |
4712 | int ndevs; | |
4713 | int i; | |
4714 | int j; | |
31e50229 | 4715 | int index; |
593060d7 | 4716 | |
0c460c0d | 4717 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4718 | |
4117f207 QW |
4719 | if (list_empty(&fs_devices->alloc_list)) { |
4720 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4721 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4722 | return -ENOSPC; |
4117f207 | 4723 | } |
b2117a39 | 4724 | |
31e50229 | 4725 | index = __get_raid_index(type); |
73c5de00 | 4726 | |
31e50229 LB |
4727 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4728 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4729 | devs_max = btrfs_raid_array[index].devs_max; | |
4730 | devs_min = btrfs_raid_array[index].devs_min; | |
4731 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4732 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4733 | |
9b3f68b9 | 4734 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4735 | max_stripe_size = SZ_1G; |
73c5de00 | 4736 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 | 4737 | if (!devs_max) |
062d4d1f | 4738 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 4739 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4740 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4741 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4742 | max_stripe_size = SZ_1G; | |
1100373f | 4743 | else |
ee22184b | 4744 | max_stripe_size = SZ_256M; |
73c5de00 | 4745 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 4746 | if (!devs_max) |
062d4d1f | 4747 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 4748 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4749 | max_stripe_size = SZ_32M; |
73c5de00 | 4750 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4751 | if (!devs_max) |
4752 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4753 | } else { |
351fd353 | 4754 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4755 | type); |
4756 | BUG_ON(1); | |
9b3f68b9 CM |
4757 | } |
4758 | ||
2b82032c YZ |
4759 | /* we don't want a chunk larger than 10% of writeable space */ |
4760 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4761 | max_chunk_size); | |
9b3f68b9 | 4762 | |
31e818fe | 4763 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4764 | GFP_NOFS); |
4765 | if (!devices_info) | |
4766 | return -ENOMEM; | |
0cad8a11 | 4767 | |
9f680ce0 | 4768 | /* |
73c5de00 AJ |
4769 | * in the first pass through the devices list, we gather information |
4770 | * about the available holes on each device. | |
9f680ce0 | 4771 | */ |
73c5de00 | 4772 | ndevs = 0; |
ebcc9301 | 4773 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4774 | u64 max_avail; |
4775 | u64 dev_offset; | |
b2117a39 | 4776 | |
ebbede42 | 4777 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4778 | WARN(1, KERN_ERR |
efe120a0 | 4779 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4780 | continue; |
4781 | } | |
b2117a39 | 4782 | |
e12c9621 AJ |
4783 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4784 | &device->dev_state) || | |
401e29c1 | 4785 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4786 | continue; |
b2117a39 | 4787 | |
73c5de00 AJ |
4788 | if (device->total_bytes > device->bytes_used) |
4789 | total_avail = device->total_bytes - device->bytes_used; | |
4790 | else | |
4791 | total_avail = 0; | |
38c01b96 | 4792 | |
4793 | /* If there is no space on this device, skip it. */ | |
4794 | if (total_avail == 0) | |
4795 | continue; | |
b2117a39 | 4796 | |
6df9a95e | 4797 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4798 | max_stripe_size * dev_stripes, |
4799 | &dev_offset, &max_avail); | |
4800 | if (ret && ret != -ENOSPC) | |
4801 | goto error; | |
b2117a39 | 4802 | |
73c5de00 AJ |
4803 | if (ret == 0) |
4804 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4805 | |
4117f207 QW |
4806 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4807 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4808 | btrfs_debug(info, | |
4809 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4810 | __func__, device->devid, max_avail, | |
4811 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4812 | continue; |
4117f207 | 4813 | } |
b2117a39 | 4814 | |
063d006f ES |
4815 | if (ndevs == fs_devices->rw_devices) { |
4816 | WARN(1, "%s: found more than %llu devices\n", | |
4817 | __func__, fs_devices->rw_devices); | |
4818 | break; | |
4819 | } | |
73c5de00 AJ |
4820 | devices_info[ndevs].dev_offset = dev_offset; |
4821 | devices_info[ndevs].max_avail = max_avail; | |
4822 | devices_info[ndevs].total_avail = total_avail; | |
4823 | devices_info[ndevs].dev = device; | |
4824 | ++ndevs; | |
4825 | } | |
b2117a39 | 4826 | |
73c5de00 AJ |
4827 | /* |
4828 | * now sort the devices by hole size / available space | |
4829 | */ | |
4830 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4831 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4832 | |
73c5de00 | 4833 | /* round down to number of usable stripes */ |
e5600fd6 | 4834 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4835 | |
73c5de00 AJ |
4836 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4837 | ret = -ENOSPC; | |
4117f207 QW |
4838 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4839 | btrfs_debug(info, | |
4840 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
4841 | __func__, ndevs, min(devs_min, | |
4842 | devs_increment * sub_stripes)); | |
4843 | } | |
73c5de00 | 4844 | goto error; |
b2117a39 | 4845 | } |
9f680ce0 | 4846 | |
f148ef4d NB |
4847 | ndevs = min(ndevs, devs_max); |
4848 | ||
73c5de00 | 4849 | /* |
92e222df HK |
4850 | * The primary goal is to maximize the number of stripes, so use as |
4851 | * many devices as possible, even if the stripes are not maximum sized. | |
4852 | * | |
4853 | * The DUP profile stores more than one stripe per device, the | |
4854 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4855 | */ |
92e222df | 4856 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4857 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4858 | |
53b381b3 DW |
4859 | /* |
4860 | * this will have to be fixed for RAID1 and RAID10 over | |
4861 | * more drives | |
4862 | */ | |
4863 | data_stripes = num_stripes / ncopies; | |
4864 | ||
500ceed8 | 4865 | if (type & BTRFS_BLOCK_GROUP_RAID5) |
53b381b3 | 4866 | data_stripes = num_stripes - 1; |
500ceed8 NB |
4867 | |
4868 | if (type & BTRFS_BLOCK_GROUP_RAID6) | |
53b381b3 | 4869 | data_stripes = num_stripes - 2; |
86db2578 CM |
4870 | |
4871 | /* | |
4872 | * Use the number of data stripes to figure out how big this chunk | |
4873 | * is really going to be in terms of logical address space, | |
4874 | * and compare that answer with the max chunk size | |
4875 | */ | |
4876 | if (stripe_size * data_stripes > max_chunk_size) { | |
4877 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4878 | |
4879 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4880 | |
4881 | /* bump the answer up to a 16MB boundary */ | |
4882 | stripe_size = (stripe_size + mask) & ~mask; | |
4883 | ||
4884 | /* but don't go higher than the limits we found | |
4885 | * while searching for free extents | |
4886 | */ | |
4887 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4888 | stripe_size = devices_info[ndevs-1].max_avail; | |
4889 | } | |
4890 | ||
37db63a4 | 4891 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 4892 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4893 | |
4894 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4895 | if (!map) { | |
4896 | ret = -ENOMEM; | |
4897 | goto error; | |
4898 | } | |
4899 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4900 | |
73c5de00 AJ |
4901 | for (i = 0; i < ndevs; ++i) { |
4902 | for (j = 0; j < dev_stripes; ++j) { | |
4903 | int s = i * dev_stripes + j; | |
4904 | map->stripes[s].dev = devices_info[i].dev; | |
4905 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4906 | j * stripe_size; | |
6324fbf3 | 4907 | } |
6324fbf3 | 4908 | } |
500ceed8 NB |
4909 | map->stripe_len = BTRFS_STRIPE_LEN; |
4910 | map->io_align = BTRFS_STRIPE_LEN; | |
4911 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4912 | map->type = type; |
2b82032c | 4913 | map->sub_stripes = sub_stripes; |
0b86a832 | 4914 | |
53b381b3 | 4915 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4916 | |
6bccf3ab | 4917 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4918 | |
172ddd60 | 4919 | em = alloc_extent_map(); |
2b82032c | 4920 | if (!em) { |
298a8f9c | 4921 | kfree(map); |
b2117a39 MX |
4922 | ret = -ENOMEM; |
4923 | goto error; | |
593060d7 | 4924 | } |
298a8f9c | 4925 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4926 | em->map_lookup = map; |
2b82032c | 4927 | em->start = start; |
73c5de00 | 4928 | em->len = num_bytes; |
2b82032c YZ |
4929 | em->block_start = 0; |
4930 | em->block_len = em->len; | |
6df9a95e | 4931 | em->orig_block_len = stripe_size; |
593060d7 | 4932 | |
0b246afa | 4933 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4934 | write_lock(&em_tree->lock); |
09a2a8f9 | 4935 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 4936 | if (ret) { |
1efb72a3 | 4937 | write_unlock(&em_tree->lock); |
0f5d42b2 | 4938 | free_extent_map(em); |
1dd4602f | 4939 | goto error; |
0f5d42b2 | 4940 | } |
0b86a832 | 4941 | |
1efb72a3 NB |
4942 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
4943 | refcount_inc(&em->refs); | |
4944 | write_unlock(&em_tree->lock); | |
4945 | ||
0174484d | 4946 | ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes); |
6df9a95e JB |
4947 | if (ret) |
4948 | goto error_del_extent; | |
2b82032c | 4949 | |
7cc8e58d MX |
4950 | for (i = 0; i < map->num_stripes; i++) { |
4951 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4952 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4953 | } | |
43530c46 | 4954 | |
a5ed45f8 | 4955 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4956 | |
0f5d42b2 | 4957 | free_extent_map(em); |
0b246afa | 4958 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4959 | |
b2117a39 | 4960 | kfree(devices_info); |
2b82032c | 4961 | return 0; |
b2117a39 | 4962 | |
6df9a95e | 4963 | error_del_extent: |
0f5d42b2 JB |
4964 | write_lock(&em_tree->lock); |
4965 | remove_extent_mapping(em_tree, em); | |
4966 | write_unlock(&em_tree->lock); | |
4967 | ||
4968 | /* One for our allocation */ | |
4969 | free_extent_map(em); | |
4970 | /* One for the tree reference */ | |
4971 | free_extent_map(em); | |
495e64f4 FM |
4972 | /* One for the pending_chunks list reference */ |
4973 | free_extent_map(em); | |
b2117a39 | 4974 | error: |
b2117a39 MX |
4975 | kfree(devices_info); |
4976 | return ret; | |
2b82032c YZ |
4977 | } |
4978 | ||
6df9a95e | 4979 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
6bccf3ab | 4980 | struct btrfs_fs_info *fs_info, |
6df9a95e | 4981 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4982 | { |
6bccf3ab JM |
4983 | struct btrfs_root *extent_root = fs_info->extent_root; |
4984 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4985 | struct btrfs_key key; |
2b82032c YZ |
4986 | struct btrfs_device *device; |
4987 | struct btrfs_chunk *chunk; | |
4988 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4989 | struct extent_map *em; |
4990 | struct map_lookup *map; | |
4991 | size_t item_size; | |
4992 | u64 dev_offset; | |
4993 | u64 stripe_size; | |
4994 | int i = 0; | |
140e639f | 4995 | int ret = 0; |
2b82032c | 4996 | |
592d92ee LB |
4997 | em = get_chunk_map(fs_info, chunk_offset, chunk_size); |
4998 | if (IS_ERR(em)) | |
4999 | return PTR_ERR(em); | |
6df9a95e | 5000 | |
95617d69 | 5001 | map = em->map_lookup; |
6df9a95e JB |
5002 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5003 | stripe_size = em->orig_block_len; | |
5004 | ||
2b82032c | 5005 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5006 | if (!chunk) { |
5007 | ret = -ENOMEM; | |
5008 | goto out; | |
5009 | } | |
5010 | ||
50460e37 FM |
5011 | /* |
5012 | * Take the device list mutex to prevent races with the final phase of | |
5013 | * a device replace operation that replaces the device object associated | |
5014 | * with the map's stripes, because the device object's id can change | |
5015 | * at any time during that final phase of the device replace operation | |
5016 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5017 | */ | |
0b246afa | 5018 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5019 | for (i = 0; i < map->num_stripes; i++) { |
5020 | device = map->stripes[i].dev; | |
5021 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5022 | |
0b86a832 | 5023 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5024 | if (ret) |
50460e37 | 5025 | break; |
b5d9071c NB |
5026 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5027 | dev_offset, stripe_size); | |
6df9a95e | 5028 | if (ret) |
50460e37 FM |
5029 | break; |
5030 | } | |
5031 | if (ret) { | |
0b246afa | 5032 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5033 | goto out; |
2b82032c YZ |
5034 | } |
5035 | ||
2b82032c | 5036 | stripe = &chunk->stripe; |
6df9a95e JB |
5037 | for (i = 0; i < map->num_stripes; i++) { |
5038 | device = map->stripes[i].dev; | |
5039 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5040 | |
e17cade2 CM |
5041 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5042 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5043 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5044 | stripe++; |
0b86a832 | 5045 | } |
0b246afa | 5046 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5047 | |
2b82032c | 5048 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5049 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5050 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5051 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5052 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5053 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5054 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5055 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5056 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5057 | |
2b82032c YZ |
5058 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5059 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5060 | key.offset = chunk_offset; | |
0b86a832 | 5061 | |
2b82032c | 5062 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5063 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5064 | /* | |
5065 | * TODO: Cleanup of inserted chunk root in case of | |
5066 | * failure. | |
5067 | */ | |
2ff7e61e | 5068 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5069 | } |
1abe9b8a | 5070 | |
6df9a95e | 5071 | out: |
0b86a832 | 5072 | kfree(chunk); |
6df9a95e | 5073 | free_extent_map(em); |
4ed1d16e | 5074 | return ret; |
2b82032c | 5075 | } |
0b86a832 | 5076 | |
2b82032c YZ |
5077 | /* |
5078 | * Chunk allocation falls into two parts. The first part does works | |
5079 | * that make the new allocated chunk useable, but not do any operation | |
5080 | * that modifies the chunk tree. The second part does the works that | |
5081 | * require modifying the chunk tree. This division is important for the | |
5082 | * bootstrap process of adding storage to a seed btrfs. | |
5083 | */ | |
5084 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 5085 | struct btrfs_fs_info *fs_info, u64 type) |
2b82032c YZ |
5086 | { |
5087 | u64 chunk_offset; | |
2b82032c | 5088 | |
0b246afa JM |
5089 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
5090 | chunk_offset = find_next_chunk(fs_info); | |
72b468c8 | 5091 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5092 | } |
5093 | ||
d397712b | 5094 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5095 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5096 | { |
5097 | u64 chunk_offset; | |
5098 | u64 sys_chunk_offset; | |
2b82032c | 5099 | u64 alloc_profile; |
2b82032c YZ |
5100 | int ret; |
5101 | ||
6df9a95e | 5102 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5103 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5104 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5105 | if (ret) |
5106 | return ret; | |
2b82032c | 5107 | |
0b246afa | 5108 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5109 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5110 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5111 | return ret; |
2b82032c YZ |
5112 | } |
5113 | ||
d20983b4 MX |
5114 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5115 | { | |
5116 | int max_errors; | |
5117 | ||
5118 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5119 | BTRFS_BLOCK_GROUP_RAID10 | | |
5120 | BTRFS_BLOCK_GROUP_RAID5 | | |
5121 | BTRFS_BLOCK_GROUP_DUP)) { | |
5122 | max_errors = 1; | |
5123 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5124 | max_errors = 2; | |
5125 | } else { | |
5126 | max_errors = 0; | |
005d6427 | 5127 | } |
2b82032c | 5128 | |
d20983b4 | 5129 | return max_errors; |
2b82032c YZ |
5130 | } |
5131 | ||
2ff7e61e | 5132 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5133 | { |
5134 | struct extent_map *em; | |
5135 | struct map_lookup *map; | |
2b82032c | 5136 | int readonly = 0; |
d20983b4 | 5137 | int miss_ndevs = 0; |
2b82032c YZ |
5138 | int i; |
5139 | ||
592d92ee LB |
5140 | em = get_chunk_map(fs_info, chunk_offset, 1); |
5141 | if (IS_ERR(em)) | |
2b82032c YZ |
5142 | return 1; |
5143 | ||
95617d69 | 5144 | map = em->map_lookup; |
2b82032c | 5145 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5146 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5147 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5148 | miss_ndevs++; |
5149 | continue; | |
5150 | } | |
ebbede42 AJ |
5151 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5152 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5153 | readonly = 1; |
d20983b4 | 5154 | goto end; |
2b82032c YZ |
5155 | } |
5156 | } | |
d20983b4 MX |
5157 | |
5158 | /* | |
5159 | * If the number of missing devices is larger than max errors, | |
5160 | * we can not write the data into that chunk successfully, so | |
5161 | * set it readonly. | |
5162 | */ | |
5163 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5164 | readonly = 1; | |
5165 | end: | |
0b86a832 | 5166 | free_extent_map(em); |
2b82032c | 5167 | return readonly; |
0b86a832 CM |
5168 | } |
5169 | ||
5170 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5171 | { | |
a8067e02 | 5172 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5173 | } |
5174 | ||
5175 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5176 | { | |
5177 | struct extent_map *em; | |
5178 | ||
d397712b | 5179 | while (1) { |
890871be | 5180 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5181 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5182 | if (em) | |
5183 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5184 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5185 | if (!em) |
5186 | break; | |
0b86a832 CM |
5187 | /* once for us */ |
5188 | free_extent_map(em); | |
5189 | /* once for the tree */ | |
5190 | free_extent_map(em); | |
5191 | } | |
5192 | } | |
5193 | ||
5d964051 | 5194 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5195 | { |
5196 | struct extent_map *em; | |
5197 | struct map_lookup *map; | |
f188591e CM |
5198 | int ret; |
5199 | ||
592d92ee LB |
5200 | em = get_chunk_map(fs_info, logical, len); |
5201 | if (IS_ERR(em)) | |
5202 | /* | |
5203 | * We could return errors for these cases, but that could get | |
5204 | * ugly and we'd probably do the same thing which is just not do | |
5205 | * anything else and exit, so return 1 so the callers don't try | |
5206 | * to use other copies. | |
5207 | */ | |
fb7669b5 | 5208 | return 1; |
fb7669b5 | 5209 | |
95617d69 | 5210 | map = em->map_lookup; |
f188591e CM |
5211 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5212 | ret = map->num_stripes; | |
321aecc6 CM |
5213 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5214 | ret = map->sub_stripes; | |
53b381b3 DW |
5215 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5216 | ret = 2; | |
5217 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5218 | /* |
5219 | * There could be two corrupted data stripes, we need | |
5220 | * to loop retry in order to rebuild the correct data. | |
5221 | * | |
5222 | * Fail a stripe at a time on every retry except the | |
5223 | * stripe under reconstruction. | |
5224 | */ | |
5225 | ret = map->num_stripes; | |
f188591e CM |
5226 | else |
5227 | ret = 1; | |
5228 | free_extent_map(em); | |
ad6d620e | 5229 | |
73beece9 | 5230 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
6fad823f LB |
5231 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5232 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5233 | ret++; |
73beece9 | 5234 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5235 | |
f188591e CM |
5236 | return ret; |
5237 | } | |
5238 | ||
2ff7e61e | 5239 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5240 | u64 logical) |
5241 | { | |
5242 | struct extent_map *em; | |
5243 | struct map_lookup *map; | |
0b246afa | 5244 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5245 | |
592d92ee | 5246 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5247 | |
69f03f13 NB |
5248 | if (!WARN_ON(IS_ERR(em))) { |
5249 | map = em->map_lookup; | |
5250 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5251 | len = map->stripe_len * nr_data_stripes(map); | |
5252 | free_extent_map(em); | |
5253 | } | |
53b381b3 DW |
5254 | return len; |
5255 | } | |
5256 | ||
e4ff5fb5 | 5257 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5258 | { |
5259 | struct extent_map *em; | |
5260 | struct map_lookup *map; | |
53b381b3 DW |
5261 | int ret = 0; |
5262 | ||
592d92ee | 5263 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5264 | |
69f03f13 NB |
5265 | if(!WARN_ON(IS_ERR(em))) { |
5266 | map = em->map_lookup; | |
5267 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5268 | ret = 1; | |
5269 | free_extent_map(em); | |
5270 | } | |
53b381b3 DW |
5271 | return ret; |
5272 | } | |
5273 | ||
30d9861f SB |
5274 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5275 | struct map_lookup *map, int first, int num, | |
5276 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5277 | { |
5278 | int i; | |
30d9861f SB |
5279 | int tolerance; |
5280 | struct btrfs_device *srcdev; | |
5281 | ||
5282 | if (dev_replace_is_ongoing && | |
5283 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5284 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5285 | srcdev = fs_info->dev_replace.srcdev; | |
5286 | else | |
5287 | srcdev = NULL; | |
5288 | ||
5289 | /* | |
5290 | * try to avoid the drive that is the source drive for a | |
5291 | * dev-replace procedure, only choose it if no other non-missing | |
5292 | * mirror is available | |
5293 | */ | |
5294 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5295 | if (map->stripes[optimal].dev->bdev && | |
5296 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5297 | return optimal; | |
5298 | for (i = first; i < first + num; i++) { | |
5299 | if (map->stripes[i].dev->bdev && | |
5300 | (tolerance || map->stripes[i].dev != srcdev)) | |
5301 | return i; | |
5302 | } | |
dfe25020 | 5303 | } |
30d9861f | 5304 | |
dfe25020 CM |
5305 | /* we couldn't find one that doesn't fail. Just return something |
5306 | * and the io error handling code will clean up eventually | |
5307 | */ | |
5308 | return optimal; | |
5309 | } | |
5310 | ||
53b381b3 DW |
5311 | static inline int parity_smaller(u64 a, u64 b) |
5312 | { | |
5313 | return a > b; | |
5314 | } | |
5315 | ||
5316 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5317 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5318 | { |
5319 | struct btrfs_bio_stripe s; | |
5320 | int i; | |
5321 | u64 l; | |
5322 | int again = 1; | |
5323 | ||
5324 | while (again) { | |
5325 | again = 0; | |
cc7539ed | 5326 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5327 | if (parity_smaller(bbio->raid_map[i], |
5328 | bbio->raid_map[i+1])) { | |
53b381b3 | 5329 | s = bbio->stripes[i]; |
8e5cfb55 | 5330 | l = bbio->raid_map[i]; |
53b381b3 | 5331 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5332 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5333 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5334 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5335 | |
53b381b3 DW |
5336 | again = 1; |
5337 | } | |
5338 | } | |
5339 | } | |
5340 | } | |
5341 | ||
6e9606d2 ZL |
5342 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5343 | { | |
5344 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5345 | /* the size of the btrfs_bio */ |
6e9606d2 | 5346 | sizeof(struct btrfs_bio) + |
e57cf21e | 5347 | /* plus the variable array for the stripes */ |
6e9606d2 | 5348 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5349 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5350 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5351 | /* |
5352 | * plus the raid_map, which includes both the tgt dev | |
5353 | * and the stripes | |
5354 | */ | |
5355 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5356 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5357 | |
5358 | atomic_set(&bbio->error, 0); | |
140475ae | 5359 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5360 | |
5361 | return bbio; | |
5362 | } | |
5363 | ||
5364 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5365 | { | |
140475ae ER |
5366 | WARN_ON(!refcount_read(&bbio->refs)); |
5367 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5368 | } |
5369 | ||
5370 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5371 | { | |
5372 | if (!bbio) | |
5373 | return; | |
140475ae | 5374 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5375 | kfree(bbio); |
5376 | } | |
5377 | ||
0b3d4cd3 LB |
5378 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5379 | /* | |
5380 | * Please note that, discard won't be sent to target device of device | |
5381 | * replace. | |
5382 | */ | |
5383 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5384 | u64 logical, u64 length, | |
5385 | struct btrfs_bio **bbio_ret) | |
5386 | { | |
5387 | struct extent_map *em; | |
5388 | struct map_lookup *map; | |
5389 | struct btrfs_bio *bbio; | |
5390 | u64 offset; | |
5391 | u64 stripe_nr; | |
5392 | u64 stripe_nr_end; | |
5393 | u64 stripe_end_offset; | |
5394 | u64 stripe_cnt; | |
5395 | u64 stripe_len; | |
5396 | u64 stripe_offset; | |
5397 | u64 num_stripes; | |
5398 | u32 stripe_index; | |
5399 | u32 factor = 0; | |
5400 | u32 sub_stripes = 0; | |
5401 | u64 stripes_per_dev = 0; | |
5402 | u32 remaining_stripes = 0; | |
5403 | u32 last_stripe = 0; | |
5404 | int ret = 0; | |
5405 | int i; | |
5406 | ||
5407 | /* discard always return a bbio */ | |
5408 | ASSERT(bbio_ret); | |
5409 | ||
5410 | em = get_chunk_map(fs_info, logical, length); | |
5411 | if (IS_ERR(em)) | |
5412 | return PTR_ERR(em); | |
5413 | ||
5414 | map = em->map_lookup; | |
5415 | /* we don't discard raid56 yet */ | |
5416 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5417 | ret = -EOPNOTSUPP; | |
5418 | goto out; | |
5419 | } | |
5420 | ||
5421 | offset = logical - em->start; | |
5422 | length = min_t(u64, em->len - offset, length); | |
5423 | ||
5424 | stripe_len = map->stripe_len; | |
5425 | /* | |
5426 | * stripe_nr counts the total number of stripes we have to stride | |
5427 | * to get to this block | |
5428 | */ | |
5429 | stripe_nr = div64_u64(offset, stripe_len); | |
5430 | ||
5431 | /* stripe_offset is the offset of this block in its stripe */ | |
5432 | stripe_offset = offset - stripe_nr * stripe_len; | |
5433 | ||
5434 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5435 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5436 | stripe_cnt = stripe_nr_end - stripe_nr; |
5437 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5438 | (offset + length); | |
5439 | /* | |
5440 | * after this, stripe_nr is the number of stripes on this | |
5441 | * device we have to walk to find the data, and stripe_index is | |
5442 | * the number of our device in the stripe array | |
5443 | */ | |
5444 | num_stripes = 1; | |
5445 | stripe_index = 0; | |
5446 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5447 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5448 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5449 | sub_stripes = 1; | |
5450 | else | |
5451 | sub_stripes = map->sub_stripes; | |
5452 | ||
5453 | factor = map->num_stripes / sub_stripes; | |
5454 | num_stripes = min_t(u64, map->num_stripes, | |
5455 | sub_stripes * stripe_cnt); | |
5456 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5457 | stripe_index *= sub_stripes; | |
5458 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5459 | &remaining_stripes); | |
5460 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5461 | last_stripe *= sub_stripes; | |
5462 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5463 | BTRFS_BLOCK_GROUP_DUP)) { | |
5464 | num_stripes = map->num_stripes; | |
5465 | } else { | |
5466 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5467 | &stripe_index); | |
5468 | } | |
5469 | ||
5470 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5471 | if (!bbio) { | |
5472 | ret = -ENOMEM; | |
5473 | goto out; | |
5474 | } | |
5475 | ||
5476 | for (i = 0; i < num_stripes; i++) { | |
5477 | bbio->stripes[i].physical = | |
5478 | map->stripes[stripe_index].physical + | |
5479 | stripe_offset + stripe_nr * map->stripe_len; | |
5480 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5481 | ||
5482 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5483 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5484 | bbio->stripes[i].length = stripes_per_dev * | |
5485 | map->stripe_len; | |
5486 | ||
5487 | if (i / sub_stripes < remaining_stripes) | |
5488 | bbio->stripes[i].length += | |
5489 | map->stripe_len; | |
5490 | ||
5491 | /* | |
5492 | * Special for the first stripe and | |
5493 | * the last stripe: | |
5494 | * | |
5495 | * |-------|...|-------| | |
5496 | * |----------| | |
5497 | * off end_off | |
5498 | */ | |
5499 | if (i < sub_stripes) | |
5500 | bbio->stripes[i].length -= | |
5501 | stripe_offset; | |
5502 | ||
5503 | if (stripe_index >= last_stripe && | |
5504 | stripe_index <= (last_stripe + | |
5505 | sub_stripes - 1)) | |
5506 | bbio->stripes[i].length -= | |
5507 | stripe_end_offset; | |
5508 | ||
5509 | if (i == sub_stripes - 1) | |
5510 | stripe_offset = 0; | |
5511 | } else { | |
5512 | bbio->stripes[i].length = length; | |
5513 | } | |
5514 | ||
5515 | stripe_index++; | |
5516 | if (stripe_index == map->num_stripes) { | |
5517 | stripe_index = 0; | |
5518 | stripe_nr++; | |
5519 | } | |
5520 | } | |
5521 | ||
5522 | *bbio_ret = bbio; | |
5523 | bbio->map_type = map->type; | |
5524 | bbio->num_stripes = num_stripes; | |
5525 | out: | |
5526 | free_extent_map(em); | |
5527 | return ret; | |
5528 | } | |
5529 | ||
5ab56090 LB |
5530 | /* |
5531 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5532 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5533 | * left cursor can also be read from the target drive. | |
5534 | * | |
5535 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5536 | * array of stripes. | |
5537 | * For READ, it also needs to be supported using the same mirror number. | |
5538 | * | |
5539 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5540 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5541 | * case. | |
5542 | */ | |
5543 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5544 | u64 logical, u64 length, | |
5545 | u64 srcdev_devid, int *mirror_num, | |
5546 | u64 *physical) | |
5547 | { | |
5548 | struct btrfs_bio *bbio = NULL; | |
5549 | int num_stripes; | |
5550 | int index_srcdev = 0; | |
5551 | int found = 0; | |
5552 | u64 physical_of_found = 0; | |
5553 | int i; | |
5554 | int ret = 0; | |
5555 | ||
5556 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5557 | logical, &length, &bbio, 0, 0); | |
5558 | if (ret) { | |
5559 | ASSERT(bbio == NULL); | |
5560 | return ret; | |
5561 | } | |
5562 | ||
5563 | num_stripes = bbio->num_stripes; | |
5564 | if (*mirror_num > num_stripes) { | |
5565 | /* | |
5566 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5567 | * that means that the requested area is not left of the left | |
5568 | * cursor | |
5569 | */ | |
5570 | btrfs_put_bbio(bbio); | |
5571 | return -EIO; | |
5572 | } | |
5573 | ||
5574 | /* | |
5575 | * process the rest of the function using the mirror_num of the source | |
5576 | * drive. Therefore look it up first. At the end, patch the device | |
5577 | * pointer to the one of the target drive. | |
5578 | */ | |
5579 | for (i = 0; i < num_stripes; i++) { | |
5580 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5581 | continue; | |
5582 | ||
5583 | /* | |
5584 | * In case of DUP, in order to keep it simple, only add the | |
5585 | * mirror with the lowest physical address | |
5586 | */ | |
5587 | if (found && | |
5588 | physical_of_found <= bbio->stripes[i].physical) | |
5589 | continue; | |
5590 | ||
5591 | index_srcdev = i; | |
5592 | found = 1; | |
5593 | physical_of_found = bbio->stripes[i].physical; | |
5594 | } | |
5595 | ||
5596 | btrfs_put_bbio(bbio); | |
5597 | ||
5598 | ASSERT(found); | |
5599 | if (!found) | |
5600 | return -EIO; | |
5601 | ||
5602 | *mirror_num = index_srcdev + 1; | |
5603 | *physical = physical_of_found; | |
5604 | return ret; | |
5605 | } | |
5606 | ||
73c0f228 LB |
5607 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5608 | struct btrfs_bio **bbio_ret, | |
5609 | struct btrfs_dev_replace *dev_replace, | |
5610 | int *num_stripes_ret, int *max_errors_ret) | |
5611 | { | |
5612 | struct btrfs_bio *bbio = *bbio_ret; | |
5613 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5614 | int tgtdev_indexes = 0; | |
5615 | int num_stripes = *num_stripes_ret; | |
5616 | int max_errors = *max_errors_ret; | |
5617 | int i; | |
5618 | ||
5619 | if (op == BTRFS_MAP_WRITE) { | |
5620 | int index_where_to_add; | |
5621 | ||
5622 | /* | |
5623 | * duplicate the write operations while the dev replace | |
5624 | * procedure is running. Since the copying of the old disk to | |
5625 | * the new disk takes place at run time while the filesystem is | |
5626 | * mounted writable, the regular write operations to the old | |
5627 | * disk have to be duplicated to go to the new disk as well. | |
5628 | * | |
5629 | * Note that device->missing is handled by the caller, and that | |
5630 | * the write to the old disk is already set up in the stripes | |
5631 | * array. | |
5632 | */ | |
5633 | index_where_to_add = num_stripes; | |
5634 | for (i = 0; i < num_stripes; i++) { | |
5635 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5636 | /* write to new disk, too */ | |
5637 | struct btrfs_bio_stripe *new = | |
5638 | bbio->stripes + index_where_to_add; | |
5639 | struct btrfs_bio_stripe *old = | |
5640 | bbio->stripes + i; | |
5641 | ||
5642 | new->physical = old->physical; | |
5643 | new->length = old->length; | |
5644 | new->dev = dev_replace->tgtdev; | |
5645 | bbio->tgtdev_map[i] = index_where_to_add; | |
5646 | index_where_to_add++; | |
5647 | max_errors++; | |
5648 | tgtdev_indexes++; | |
5649 | } | |
5650 | } | |
5651 | num_stripes = index_where_to_add; | |
5652 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5653 | int index_srcdev = 0; | |
5654 | int found = 0; | |
5655 | u64 physical_of_found = 0; | |
5656 | ||
5657 | /* | |
5658 | * During the dev-replace procedure, the target drive can also | |
5659 | * be used to read data in case it is needed to repair a corrupt | |
5660 | * block elsewhere. This is possible if the requested area is | |
5661 | * left of the left cursor. In this area, the target drive is a | |
5662 | * full copy of the source drive. | |
5663 | */ | |
5664 | for (i = 0; i < num_stripes; i++) { | |
5665 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5666 | /* | |
5667 | * In case of DUP, in order to keep it simple, | |
5668 | * only add the mirror with the lowest physical | |
5669 | * address | |
5670 | */ | |
5671 | if (found && | |
5672 | physical_of_found <= | |
5673 | bbio->stripes[i].physical) | |
5674 | continue; | |
5675 | index_srcdev = i; | |
5676 | found = 1; | |
5677 | physical_of_found = bbio->stripes[i].physical; | |
5678 | } | |
5679 | } | |
5680 | if (found) { | |
5681 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5682 | bbio->stripes + num_stripes; | |
5683 | ||
5684 | tgtdev_stripe->physical = physical_of_found; | |
5685 | tgtdev_stripe->length = | |
5686 | bbio->stripes[index_srcdev].length; | |
5687 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5688 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5689 | ||
5690 | tgtdev_indexes++; | |
5691 | num_stripes++; | |
5692 | } | |
5693 | } | |
5694 | ||
5695 | *num_stripes_ret = num_stripes; | |
5696 | *max_errors_ret = max_errors; | |
5697 | bbio->num_tgtdevs = tgtdev_indexes; | |
5698 | *bbio_ret = bbio; | |
5699 | } | |
5700 | ||
2b19a1fe LB |
5701 | static bool need_full_stripe(enum btrfs_map_op op) |
5702 | { | |
5703 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5704 | } | |
5705 | ||
cf8cddd3 CH |
5706 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5707 | enum btrfs_map_op op, | |
f2d8d74d | 5708 | u64 logical, u64 *length, |
a1d3c478 | 5709 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5710 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5711 | { |
5712 | struct extent_map *em; | |
5713 | struct map_lookup *map; | |
0b86a832 | 5714 | u64 offset; |
593060d7 CM |
5715 | u64 stripe_offset; |
5716 | u64 stripe_nr; | |
53b381b3 | 5717 | u64 stripe_len; |
9d644a62 | 5718 | u32 stripe_index; |
cea9e445 | 5719 | int i; |
de11cc12 | 5720 | int ret = 0; |
f2d8d74d | 5721 | int num_stripes; |
a236aed1 | 5722 | int max_errors = 0; |
2c8cdd6e | 5723 | int tgtdev_indexes = 0; |
a1d3c478 | 5724 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5725 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5726 | int dev_replace_is_ongoing = 0; | |
5727 | int num_alloc_stripes; | |
ad6d620e SB |
5728 | int patch_the_first_stripe_for_dev_replace = 0; |
5729 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5730 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5731 | |
0b3d4cd3 LB |
5732 | if (op == BTRFS_MAP_DISCARD) |
5733 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5734 | *length, bbio_ret); | |
5735 | ||
592d92ee LB |
5736 | em = get_chunk_map(fs_info, logical, *length); |
5737 | if (IS_ERR(em)) | |
5738 | return PTR_ERR(em); | |
0b86a832 | 5739 | |
95617d69 | 5740 | map = em->map_lookup; |
0b86a832 | 5741 | offset = logical - em->start; |
593060d7 | 5742 | |
53b381b3 | 5743 | stripe_len = map->stripe_len; |
593060d7 CM |
5744 | stripe_nr = offset; |
5745 | /* | |
5746 | * stripe_nr counts the total number of stripes we have to stride | |
5747 | * to get to this block | |
5748 | */ | |
47c5713f | 5749 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5750 | |
53b381b3 | 5751 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5752 | if (offset < stripe_offset) { |
5d163e0e JM |
5753 | btrfs_crit(fs_info, |
5754 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5755 | stripe_offset, offset, em->start, logical, |
5756 | stripe_len); | |
5757 | free_extent_map(em); | |
5758 | return -EINVAL; | |
5759 | } | |
593060d7 CM |
5760 | |
5761 | /* stripe_offset is the offset of this block in its stripe*/ | |
5762 | stripe_offset = offset - stripe_offset; | |
5763 | ||
53b381b3 | 5764 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5765 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5766 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5767 | raid56_full_stripe_start = offset; | |
5768 | ||
5769 | /* allow a write of a full stripe, but make sure we don't | |
5770 | * allow straddling of stripes | |
5771 | */ | |
47c5713f DS |
5772 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5773 | full_stripe_len); | |
53b381b3 DW |
5774 | raid56_full_stripe_start *= full_stripe_len; |
5775 | } | |
5776 | ||
0b3d4cd3 | 5777 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5778 | u64 max_len; |
5779 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5780 | For other RAID types and for RAID[56] reads, just allow a single | |
5781 | stripe (on a single disk). */ | |
ffe2d203 | 5782 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5783 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5784 | max_len = stripe_len * nr_data_stripes(map) - |
5785 | (offset - raid56_full_stripe_start); | |
5786 | } else { | |
5787 | /* we limit the length of each bio to what fits in a stripe */ | |
5788 | max_len = stripe_len - stripe_offset; | |
5789 | } | |
5790 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5791 | } else { |
5792 | *length = em->len - offset; | |
5793 | } | |
f2d8d74d | 5794 | |
53b381b3 DW |
5795 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5796 | it cares about is the length */ | |
a1d3c478 | 5797 | if (!bbio_ret) |
cea9e445 CM |
5798 | goto out; |
5799 | ||
73beece9 | 5800 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5801 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5802 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5803 | btrfs_dev_replace_unlock(dev_replace, 0); |
5804 | else | |
5805 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5806 | |
ad6d620e | 5807 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5808 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5809 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5810 | dev_replace->srcdev->devid, | |
5811 | &mirror_num, | |
5812 | &physical_to_patch_in_first_stripe); | |
5813 | if (ret) | |
ad6d620e | 5814 | goto out; |
5ab56090 LB |
5815 | else |
5816 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5817 | } else if (mirror_num > map->num_stripes) { |
5818 | mirror_num = 0; | |
5819 | } | |
5820 | ||
f2d8d74d | 5821 | num_stripes = 1; |
cea9e445 | 5822 | stripe_index = 0; |
fce3bb9a | 5823 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5824 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5825 | &stripe_index); | |
de483734 | 5826 | if (!need_full_stripe(op)) |
28e1cc7d | 5827 | mirror_num = 1; |
fce3bb9a | 5828 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 5829 | if (need_full_stripe(op)) |
f2d8d74d | 5830 | num_stripes = map->num_stripes; |
2fff734f | 5831 | else if (mirror_num) |
f188591e | 5832 | stripe_index = mirror_num - 1; |
dfe25020 | 5833 | else { |
30d9861f | 5834 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5835 | map->num_stripes, |
30d9861f SB |
5836 | current->pid % map->num_stripes, |
5837 | dev_replace_is_ongoing); | |
a1d3c478 | 5838 | mirror_num = stripe_index + 1; |
dfe25020 | 5839 | } |
2fff734f | 5840 | |
611f0e00 | 5841 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5842 | if (need_full_stripe(op)) { |
f2d8d74d | 5843 | num_stripes = map->num_stripes; |
a1d3c478 | 5844 | } else if (mirror_num) { |
f188591e | 5845 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5846 | } else { |
5847 | mirror_num = 1; | |
5848 | } | |
2fff734f | 5849 | |
321aecc6 | 5850 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5851 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5852 | |
47c5713f | 5853 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5854 | stripe_index *= map->sub_stripes; |
5855 | ||
de483734 | 5856 | if (need_full_stripe(op)) |
f2d8d74d | 5857 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5858 | else if (mirror_num) |
5859 | stripe_index += mirror_num - 1; | |
dfe25020 | 5860 | else { |
3e74317a | 5861 | int old_stripe_index = stripe_index; |
30d9861f SB |
5862 | stripe_index = find_live_mirror(fs_info, map, |
5863 | stripe_index, | |
dfe25020 | 5864 | map->sub_stripes, stripe_index + |
30d9861f SB |
5865 | current->pid % map->sub_stripes, |
5866 | dev_replace_is_ongoing); | |
3e74317a | 5867 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5868 | } |
53b381b3 | 5869 | |
ffe2d203 | 5870 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5871 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5872 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5873 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5874 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5875 | |
5876 | /* RAID[56] write or recovery. Return all stripes */ | |
5877 | num_stripes = map->num_stripes; | |
5878 | max_errors = nr_parity_stripes(map); | |
5879 | ||
53b381b3 DW |
5880 | *length = map->stripe_len; |
5881 | stripe_index = 0; | |
5882 | stripe_offset = 0; | |
5883 | } else { | |
5884 | /* | |
5885 | * Mirror #0 or #1 means the original data block. | |
5886 | * Mirror #2 is RAID5 parity block. | |
5887 | * Mirror #3 is RAID6 Q block. | |
5888 | */ | |
47c5713f DS |
5889 | stripe_nr = div_u64_rem(stripe_nr, |
5890 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5891 | if (mirror_num > 1) |
5892 | stripe_index = nr_data_stripes(map) + | |
5893 | mirror_num - 2; | |
5894 | ||
5895 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5896 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5897 | &stripe_index); | |
de483734 | 5898 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 5899 | mirror_num = 1; |
53b381b3 | 5900 | } |
8790d502 CM |
5901 | } else { |
5902 | /* | |
47c5713f DS |
5903 | * after this, stripe_nr is the number of stripes on this |
5904 | * device we have to walk to find the data, and stripe_index is | |
5905 | * the number of our device in the stripe array | |
8790d502 | 5906 | */ |
47c5713f DS |
5907 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5908 | &stripe_index); | |
a1d3c478 | 5909 | mirror_num = stripe_index + 1; |
8790d502 | 5910 | } |
e042d1ec | 5911 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5912 | btrfs_crit(fs_info, |
5913 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5914 | stripe_index, map->num_stripes); |
5915 | ret = -EINVAL; | |
5916 | goto out; | |
5917 | } | |
cea9e445 | 5918 | |
472262f3 | 5919 | num_alloc_stripes = num_stripes; |
6fad823f | 5920 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5921 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5922 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5923 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5924 | num_alloc_stripes++; |
2c8cdd6e | 5925 | tgtdev_indexes = num_stripes; |
ad6d620e | 5926 | } |
2c8cdd6e | 5927 | |
6e9606d2 | 5928 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5929 | if (!bbio) { |
5930 | ret = -ENOMEM; | |
5931 | goto out; | |
5932 | } | |
6fad823f | 5933 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5934 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5935 | |
8e5cfb55 | 5936 | /* build raid_map */ |
2b19a1fe LB |
5937 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5938 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5939 | u64 tmp; |
9d644a62 | 5940 | unsigned rot; |
8e5cfb55 ZL |
5941 | |
5942 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5943 | sizeof(struct btrfs_bio_stripe) * | |
5944 | num_alloc_stripes + | |
5945 | sizeof(int) * tgtdev_indexes); | |
5946 | ||
5947 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5948 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5949 | |
5950 | /* Fill in the logical address of each stripe */ | |
5951 | tmp = stripe_nr * nr_data_stripes(map); | |
5952 | for (i = 0; i < nr_data_stripes(map); i++) | |
5953 | bbio->raid_map[(i+rot) % num_stripes] = | |
5954 | em->start + (tmp + i) * map->stripe_len; | |
5955 | ||
5956 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5957 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5958 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5959 | RAID6_Q_STRIPE; | |
5960 | } | |
5961 | ||
b89203f7 | 5962 | |
0b3d4cd3 LB |
5963 | for (i = 0; i < num_stripes; i++) { |
5964 | bbio->stripes[i].physical = | |
5965 | map->stripes[stripe_index].physical + | |
5966 | stripe_offset + | |
5967 | stripe_nr * map->stripe_len; | |
5968 | bbio->stripes[i].dev = | |
5969 | map->stripes[stripe_index].dev; | |
5970 | stripe_index++; | |
593060d7 | 5971 | } |
de11cc12 | 5972 | |
2b19a1fe | 5973 | if (need_full_stripe(op)) |
d20983b4 | 5974 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5975 | |
8e5cfb55 ZL |
5976 | if (bbio->raid_map) |
5977 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5978 | |
73c0f228 | 5979 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5980 | need_full_stripe(op)) { |
73c0f228 LB |
5981 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5982 | &max_errors); | |
472262f3 SB |
5983 | } |
5984 | ||
de11cc12 | 5985 | *bbio_ret = bbio; |
10f11900 | 5986 | bbio->map_type = map->type; |
de11cc12 LZ |
5987 | bbio->num_stripes = num_stripes; |
5988 | bbio->max_errors = max_errors; | |
5989 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5990 | |
5991 | /* | |
5992 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5993 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5994 | * available as a mirror | |
5995 | */ | |
5996 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5997 | WARN_ON(num_stripes > 1); | |
5998 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5999 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6000 | bbio->mirror_num = map->num_stripes + 1; | |
6001 | } | |
cea9e445 | 6002 | out: |
73beece9 LB |
6003 | if (dev_replace_is_ongoing) { |
6004 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
6005 | btrfs_dev_replace_unlock(dev_replace, 0); | |
6006 | } | |
0b86a832 | 6007 | free_extent_map(em); |
de11cc12 | 6008 | return ret; |
0b86a832 CM |
6009 | } |
6010 | ||
cf8cddd3 | 6011 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6012 | u64 logical, u64 *length, |
a1d3c478 | 6013 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6014 | { |
b3d3fa51 | 6015 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6016 | mirror_num, 0); |
f2d8d74d CM |
6017 | } |
6018 | ||
af8e2d1d | 6019 | /* For Scrub/replace */ |
cf8cddd3 | 6020 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6021 | u64 logical, u64 *length, |
825ad4c9 | 6022 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6023 | { |
825ad4c9 | 6024 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6025 | } |
6026 | ||
ab8d0fc4 | 6027 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
6028 | u64 chunk_start, u64 physical, u64 devid, |
6029 | u64 **logical, int *naddrs, int *stripe_len) | |
6030 | { | |
a512bbf8 YZ |
6031 | struct extent_map *em; |
6032 | struct map_lookup *map; | |
6033 | u64 *buf; | |
6034 | u64 bytenr; | |
6035 | u64 length; | |
6036 | u64 stripe_nr; | |
53b381b3 | 6037 | u64 rmap_len; |
a512bbf8 YZ |
6038 | int i, j, nr = 0; |
6039 | ||
592d92ee LB |
6040 | em = get_chunk_map(fs_info, chunk_start, 1); |
6041 | if (IS_ERR(em)) | |
835d974f | 6042 | return -EIO; |
835d974f | 6043 | |
95617d69 | 6044 | map = em->map_lookup; |
a512bbf8 | 6045 | length = em->len; |
53b381b3 DW |
6046 | rmap_len = map->stripe_len; |
6047 | ||
a512bbf8 | 6048 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6049 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6050 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6051 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6052 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6053 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6054 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6055 | } | |
a512bbf8 | 6056 | |
31e818fe | 6057 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6058 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6059 | |
6060 | for (i = 0; i < map->num_stripes; i++) { | |
6061 | if (devid && map->stripes[i].dev->devid != devid) | |
6062 | continue; | |
6063 | if (map->stripes[i].physical > physical || | |
6064 | map->stripes[i].physical + length <= physical) | |
6065 | continue; | |
6066 | ||
6067 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6068 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6069 | |
6070 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6071 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6072 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6073 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6074 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6075 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6076 | * Alternatively, just use rmap_len below instead of | |
6077 | * map->stripe_len */ | |
6078 | ||
6079 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6080 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6081 | for (j = 0; j < nr; j++) { |
6082 | if (buf[j] == bytenr) | |
6083 | break; | |
6084 | } | |
934d375b CM |
6085 | if (j == nr) { |
6086 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6087 | buf[nr++] = bytenr; |
934d375b | 6088 | } |
a512bbf8 YZ |
6089 | } |
6090 | ||
a512bbf8 YZ |
6091 | *logical = buf; |
6092 | *naddrs = nr; | |
53b381b3 | 6093 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6094 | |
6095 | free_extent_map(em); | |
6096 | return 0; | |
f2d8d74d CM |
6097 | } |
6098 | ||
4246a0b6 | 6099 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6100 | { |
326e1dbb MS |
6101 | bio->bi_private = bbio->private; |
6102 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6103 | bio_endio(bio); |
326e1dbb | 6104 | |
6e9606d2 | 6105 | btrfs_put_bbio(bbio); |
8408c716 MX |
6106 | } |
6107 | ||
4246a0b6 | 6108 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6109 | { |
9be3395b | 6110 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6111 | int is_orig_bio = 0; |
8790d502 | 6112 | |
4e4cbee9 | 6113 | if (bio->bi_status) { |
a1d3c478 | 6114 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6115 | if (bio->bi_status == BLK_STS_IOERR || |
6116 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6117 | unsigned int stripe_index = |
9be3395b | 6118 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6119 | struct btrfs_device *dev; |
442a4f63 SB |
6120 | |
6121 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6122 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6123 | if (dev->bdev) { |
37226b21 | 6124 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6125 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6126 | BTRFS_DEV_STAT_WRITE_ERRS); |
6127 | else | |
1cb34c8e | 6128 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6129 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6130 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6131 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6132 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6133 | } |
442a4f63 SB |
6134 | } |
6135 | } | |
8790d502 | 6136 | |
a1d3c478 | 6137 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6138 | is_orig_bio = 1; |
6139 | ||
c404e0dc MX |
6140 | btrfs_bio_counter_dec(bbio->fs_info); |
6141 | ||
a1d3c478 | 6142 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6143 | if (!is_orig_bio) { |
6144 | bio_put(bio); | |
a1d3c478 | 6145 | bio = bbio->orig_bio; |
7d2b4daa | 6146 | } |
c7b22bb1 | 6147 | |
9be3395b | 6148 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6149 | /* only send an error to the higher layers if it is |
53b381b3 | 6150 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6151 | */ |
a1d3c478 | 6152 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6153 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6154 | } else { |
1259ab75 CM |
6155 | /* |
6156 | * this bio is actually up to date, we didn't | |
6157 | * go over the max number of errors | |
6158 | */ | |
2dbe0c77 | 6159 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6160 | } |
c55f1396 | 6161 | |
4246a0b6 | 6162 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6163 | } else if (!is_orig_bio) { |
8790d502 CM |
6164 | bio_put(bio); |
6165 | } | |
8790d502 CM |
6166 | } |
6167 | ||
8b712842 CM |
6168 | /* |
6169 | * see run_scheduled_bios for a description of why bios are collected for | |
6170 | * async submit. | |
6171 | * | |
6172 | * This will add one bio to the pending list for a device and make sure | |
6173 | * the work struct is scheduled. | |
6174 | */ | |
2ff7e61e | 6175 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6176 | struct bio *bio) |
8b712842 | 6177 | { |
0b246afa | 6178 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6179 | int should_queue = 1; |
ffbd517d | 6180 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6181 | |
e6e674bd AJ |
6182 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) || |
6183 | !device->bdev) { | |
4246a0b6 | 6184 | bio_io_error(bio); |
53b381b3 DW |
6185 | return; |
6186 | } | |
6187 | ||
8b712842 | 6188 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6189 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6190 | btrfsic_submit_bio(bio); |
143bede5 | 6191 | return; |
8b712842 CM |
6192 | } |
6193 | ||
492bb6de | 6194 | WARN_ON(bio->bi_next); |
8b712842 | 6195 | bio->bi_next = NULL; |
8b712842 CM |
6196 | |
6197 | spin_lock(&device->io_lock); | |
67f055c7 | 6198 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6199 | pending_bios = &device->pending_sync_bios; |
6200 | else | |
6201 | pending_bios = &device->pending_bios; | |
8b712842 | 6202 | |
ffbd517d CM |
6203 | if (pending_bios->tail) |
6204 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6205 | |
ffbd517d CM |
6206 | pending_bios->tail = bio; |
6207 | if (!pending_bios->head) | |
6208 | pending_bios->head = bio; | |
8b712842 CM |
6209 | if (device->running_pending) |
6210 | should_queue = 0; | |
6211 | ||
6212 | spin_unlock(&device->io_lock); | |
6213 | ||
6214 | if (should_queue) | |
0b246afa | 6215 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6216 | } |
6217 | ||
2ff7e61e JM |
6218 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6219 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6220 | { |
6221 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6222 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6223 | |
6224 | bio->bi_private = bbio; | |
9be3395b | 6225 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6226 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6227 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6228 | #ifdef DEBUG |
6229 | { | |
6230 | struct rcu_string *name; | |
6231 | ||
6232 | rcu_read_lock(); | |
6233 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6234 | btrfs_debug(fs_info, |
6235 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6236 | bio_op(bio), bio->bi_opf, | |
6237 | (u64)bio->bi_iter.bi_sector, | |
6238 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6239 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6240 | rcu_read_unlock(); |
6241 | } | |
6242 | #endif | |
74d46992 | 6243 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6244 | |
2ff7e61e | 6245 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6246 | |
de1ee92a | 6247 | if (async) |
2ff7e61e | 6248 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6249 | else |
4e49ea4a | 6250 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6251 | } |
6252 | ||
de1ee92a JB |
6253 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6254 | { | |
6255 | atomic_inc(&bbio->error); | |
6256 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6257 | /* Should be the original bio. */ |
8408c716 MX |
6258 | WARN_ON(bio != bbio->orig_bio); |
6259 | ||
9be3395b | 6260 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6261 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6262 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6263 | bio->bi_status = BLK_STS_IOERR; | |
6264 | else | |
6265 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6266 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6267 | } |
6268 | } | |
6269 | ||
58efbc9f OS |
6270 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6271 | int mirror_num, int async_submit) | |
0b86a832 | 6272 | { |
0b86a832 | 6273 | struct btrfs_device *dev; |
8790d502 | 6274 | struct bio *first_bio = bio; |
4f024f37 | 6275 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6276 | u64 length = 0; |
6277 | u64 map_length; | |
0b86a832 | 6278 | int ret; |
08da757d ZL |
6279 | int dev_nr; |
6280 | int total_devs; | |
a1d3c478 | 6281 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6282 | |
4f024f37 | 6283 | length = bio->bi_iter.bi_size; |
0b86a832 | 6284 | map_length = length; |
cea9e445 | 6285 | |
0b246afa | 6286 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6287 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6288 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6289 | if (ret) { |
0b246afa | 6290 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6291 | return errno_to_blk_status(ret); |
c404e0dc | 6292 | } |
cea9e445 | 6293 | |
a1d3c478 | 6294 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6295 | bbio->orig_bio = first_bio; |
6296 | bbio->private = first_bio->bi_private; | |
6297 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6298 | bbio->fs_info = fs_info; |
53b381b3 DW |
6299 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6300 | ||
ad1ba2a0 | 6301 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6302 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6303 | /* In this case, map_length has been set to the length of |
6304 | a single stripe; not the whole write */ | |
37226b21 | 6305 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6306 | ret = raid56_parity_write(fs_info, bio, bbio, |
6307 | map_length); | |
53b381b3 | 6308 | } else { |
2ff7e61e JM |
6309 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6310 | map_length, mirror_num, 1); | |
53b381b3 | 6311 | } |
4245215d | 6312 | |
0b246afa | 6313 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6314 | return errno_to_blk_status(ret); |
53b381b3 DW |
6315 | } |
6316 | ||
cea9e445 | 6317 | if (map_length < length) { |
0b246afa | 6318 | btrfs_crit(fs_info, |
5d163e0e JM |
6319 | "mapping failed logical %llu bio len %llu len %llu", |
6320 | logical, length, map_length); | |
cea9e445 CM |
6321 | BUG(); |
6322 | } | |
a1d3c478 | 6323 | |
08da757d | 6324 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6325 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6326 | if (!dev || !dev->bdev || |
ebbede42 AJ |
6327 | (bio_op(first_bio) == REQ_OP_WRITE && |
6328 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6329 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6330 | continue; |
6331 | } | |
6332 | ||
3aa8e074 | 6333 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6334 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6335 | else |
a1d3c478 | 6336 | bio = first_bio; |
de1ee92a | 6337 | |
2ff7e61e JM |
6338 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6339 | dev_nr, async_submit); | |
8790d502 | 6340 | } |
0b246afa | 6341 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6342 | return BLK_STS_OK; |
0b86a832 CM |
6343 | } |
6344 | ||
aa1b8cd4 | 6345 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6346 | u8 *uuid, u8 *fsid) |
0b86a832 | 6347 | { |
2b82032c YZ |
6348 | struct btrfs_device *device; |
6349 | struct btrfs_fs_devices *cur_devices; | |
6350 | ||
aa1b8cd4 | 6351 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6352 | while (cur_devices) { |
6353 | if (!fsid || | |
44880fdc | 6354 | !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6355 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6356 | if (device) |
6357 | return device; | |
6358 | } | |
6359 | cur_devices = cur_devices->seed; | |
6360 | } | |
6361 | return NULL; | |
0b86a832 CM |
6362 | } |
6363 | ||
2ff7e61e | 6364 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6365 | u64 devid, u8 *dev_uuid) |
6366 | { | |
6367 | struct btrfs_device *device; | |
dfe25020 | 6368 | |
12bd2fc0 ID |
6369 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6370 | if (IS_ERR(device)) | |
adfb69af | 6371 | return device; |
12bd2fc0 ID |
6372 | |
6373 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6374 | device->fs_devices = fs_devices; |
dfe25020 | 6375 | fs_devices->num_devices++; |
12bd2fc0 | 6376 | |
e6e674bd | 6377 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6378 | fs_devices->missing_devices++; |
12bd2fc0 | 6379 | |
dfe25020 CM |
6380 | return device; |
6381 | } | |
6382 | ||
12bd2fc0 ID |
6383 | /** |
6384 | * btrfs_alloc_device - allocate struct btrfs_device | |
6385 | * @fs_info: used only for generating a new devid, can be NULL if | |
6386 | * devid is provided (i.e. @devid != NULL). | |
6387 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6388 | * is generated. | |
6389 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6390 | * is generated. | |
6391 | * | |
6392 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf DS |
6393 | * on error. Returned struct is not linked onto any lists and must be |
6394 | * destroyed with free_device. | |
12bd2fc0 ID |
6395 | */ |
6396 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6397 | const u64 *devid, | |
6398 | const u8 *uuid) | |
6399 | { | |
6400 | struct btrfs_device *dev; | |
6401 | u64 tmp; | |
6402 | ||
fae7f21c | 6403 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6404 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6405 | |
6406 | dev = __alloc_device(); | |
6407 | if (IS_ERR(dev)) | |
6408 | return dev; | |
6409 | ||
6410 | if (devid) | |
6411 | tmp = *devid; | |
6412 | else { | |
6413 | int ret; | |
6414 | ||
6415 | ret = find_next_devid(fs_info, &tmp); | |
6416 | if (ret) { | |
55de4803 | 6417 | free_device(dev); |
12bd2fc0 ID |
6418 | return ERR_PTR(ret); |
6419 | } | |
6420 | } | |
6421 | dev->devid = tmp; | |
6422 | ||
6423 | if (uuid) | |
6424 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6425 | else | |
6426 | generate_random_uuid(dev->uuid); | |
6427 | ||
9e0af237 LB |
6428 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6429 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6430 | |
6431 | return dev; | |
6432 | } | |
6433 | ||
e06cd3dd | 6434 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6435 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6436 | struct extent_buffer *leaf, |
6437 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6438 | { |
0b86a832 | 6439 | u64 length; |
f04b772b | 6440 | u64 stripe_len; |
e06cd3dd LB |
6441 | u16 num_stripes; |
6442 | u16 sub_stripes; | |
6443 | u64 type; | |
0b86a832 | 6444 | |
e17cade2 | 6445 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6446 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6447 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6448 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6449 | type = btrfs_chunk_type(leaf, chunk); | |
6450 | ||
f04b772b | 6451 | if (!num_stripes) { |
0b246afa | 6452 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6453 | num_stripes); |
6454 | return -EIO; | |
6455 | } | |
0b246afa JM |
6456 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6457 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6458 | return -EIO; |
6459 | } | |
0b246afa JM |
6460 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6461 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6462 | btrfs_chunk_sector_size(leaf, chunk)); |
6463 | return -EIO; | |
6464 | } | |
0b246afa JM |
6465 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6466 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6467 | return -EIO; |
6468 | } | |
3d8da678 | 6469 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6470 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6471 | stripe_len); |
6472 | return -EIO; | |
6473 | } | |
6474 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6475 | type) { |
0b246afa | 6476 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6477 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6478 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6479 | btrfs_chunk_type(leaf, chunk)); | |
6480 | return -EIO; | |
6481 | } | |
e06cd3dd LB |
6482 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6483 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6484 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6485 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6486 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6487 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6488 | num_stripes != 1)) { | |
0b246afa | 6489 | btrfs_err(fs_info, |
e06cd3dd LB |
6490 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6491 | num_stripes, sub_stripes, | |
6492 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6493 | return -EIO; | |
6494 | } | |
6495 | ||
6496 | return 0; | |
6497 | } | |
6498 | ||
5a2b8e60 | 6499 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6500 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6501 | { |
2b902dfc AJ |
6502 | if (error) |
6503 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6504 | devid, uuid); | |
6505 | else | |
6506 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6507 | devid, uuid); | |
5a2b8e60 AJ |
6508 | } |
6509 | ||
2ff7e61e | 6510 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6511 | struct extent_buffer *leaf, |
6512 | struct btrfs_chunk *chunk) | |
6513 | { | |
0b246afa | 6514 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6515 | struct map_lookup *map; |
6516 | struct extent_map *em; | |
6517 | u64 logical; | |
6518 | u64 length; | |
e06cd3dd LB |
6519 | u64 devid; |
6520 | u8 uuid[BTRFS_UUID_SIZE]; | |
6521 | int num_stripes; | |
6522 | int ret; | |
6523 | int i; | |
6524 | ||
6525 | logical = key->offset; | |
6526 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6527 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6528 | ||
2ff7e61e | 6529 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6530 | if (ret) |
6531 | return ret; | |
a061fc8d | 6532 | |
890871be | 6533 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6534 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6535 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6536 | |
6537 | /* already mapped? */ | |
6538 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6539 | free_extent_map(em); | |
0b86a832 CM |
6540 | return 0; |
6541 | } else if (em) { | |
6542 | free_extent_map(em); | |
6543 | } | |
0b86a832 | 6544 | |
172ddd60 | 6545 | em = alloc_extent_map(); |
0b86a832 CM |
6546 | if (!em) |
6547 | return -ENOMEM; | |
593060d7 | 6548 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6549 | if (!map) { |
6550 | free_extent_map(em); | |
6551 | return -ENOMEM; | |
6552 | } | |
6553 | ||
298a8f9c | 6554 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6555 | em->map_lookup = map; |
0b86a832 CM |
6556 | em->start = logical; |
6557 | em->len = length; | |
70c8a91c | 6558 | em->orig_start = 0; |
0b86a832 | 6559 | em->block_start = 0; |
c8b97818 | 6560 | em->block_len = em->len; |
0b86a832 | 6561 | |
593060d7 CM |
6562 | map->num_stripes = num_stripes; |
6563 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6564 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6565 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6566 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6567 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6568 | for (i = 0; i < num_stripes; i++) { |
6569 | map->stripes[i].physical = | |
6570 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6571 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6572 | read_extent_buffer(leaf, uuid, (unsigned long) |
6573 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6574 | BTRFS_UUID_SIZE); | |
0b246afa | 6575 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6576 | uuid, NULL); |
3cdde224 | 6577 | if (!map->stripes[i].dev && |
0b246afa | 6578 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6579 | free_extent_map(em); |
2b902dfc | 6580 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6581 | return -ENOENT; |
593060d7 | 6582 | } |
dfe25020 CM |
6583 | if (!map->stripes[i].dev) { |
6584 | map->stripes[i].dev = | |
2ff7e61e JM |
6585 | add_missing_dev(fs_info->fs_devices, devid, |
6586 | uuid); | |
adfb69af | 6587 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6588 | free_extent_map(em); |
adfb69af AJ |
6589 | btrfs_err(fs_info, |
6590 | "failed to init missing dev %llu: %ld", | |
6591 | devid, PTR_ERR(map->stripes[i].dev)); | |
6592 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6593 | } |
2b902dfc | 6594 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6595 | } |
e12c9621 AJ |
6596 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6597 | &(map->stripes[i].dev->dev_state)); | |
6598 | ||
0b86a832 CM |
6599 | } |
6600 | ||
890871be | 6601 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6602 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6603 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6604 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6605 | free_extent_map(em); |
6606 | ||
6607 | return 0; | |
6608 | } | |
6609 | ||
143bede5 | 6610 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6611 | struct btrfs_dev_item *dev_item, |
6612 | struct btrfs_device *device) | |
6613 | { | |
6614 | unsigned long ptr; | |
0b86a832 CM |
6615 | |
6616 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6617 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6618 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6619 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6620 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6621 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6622 | device->type = btrfs_device_type(leaf, dev_item); |
6623 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6624 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6625 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6626 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6627 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6628 | |
410ba3a2 | 6629 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6630 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6631 | } |
6632 | ||
2ff7e61e | 6633 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6634 | u8 *fsid) |
2b82032c YZ |
6635 | { |
6636 | struct btrfs_fs_devices *fs_devices; | |
6637 | int ret; | |
6638 | ||
b367e47f | 6639 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2dfeca9b | 6640 | ASSERT(fsid); |
2b82032c | 6641 | |
0b246afa | 6642 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6643 | while (fs_devices) { |
44880fdc | 6644 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6645 | return fs_devices; |
6646 | ||
2b82032c YZ |
6647 | fs_devices = fs_devices->seed; |
6648 | } | |
6649 | ||
6650 | fs_devices = find_fsid(fsid); | |
6651 | if (!fs_devices) { | |
0b246afa | 6652 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6653 | return ERR_PTR(-ENOENT); |
6654 | ||
6655 | fs_devices = alloc_fs_devices(fsid); | |
6656 | if (IS_ERR(fs_devices)) | |
6657 | return fs_devices; | |
6658 | ||
6659 | fs_devices->seeding = 1; | |
6660 | fs_devices->opened = 1; | |
6661 | return fs_devices; | |
2b82032c | 6662 | } |
e4404d6e YZ |
6663 | |
6664 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6665 | if (IS_ERR(fs_devices)) |
6666 | return fs_devices; | |
2b82032c | 6667 | |
97288f2c | 6668 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
0b246afa | 6669 | fs_info->bdev_holder); |
48d28232 JL |
6670 | if (ret) { |
6671 | free_fs_devices(fs_devices); | |
5f375835 | 6672 | fs_devices = ERR_PTR(ret); |
2b82032c | 6673 | goto out; |
48d28232 | 6674 | } |
2b82032c YZ |
6675 | |
6676 | if (!fs_devices->seeding) { | |
6677 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6678 | free_fs_devices(fs_devices); |
5f375835 | 6679 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6680 | goto out; |
6681 | } | |
6682 | ||
0b246afa JM |
6683 | fs_devices->seed = fs_info->fs_devices->seed; |
6684 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6685 | out: |
5f375835 | 6686 | return fs_devices; |
2b82032c YZ |
6687 | } |
6688 | ||
2ff7e61e | 6689 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6690 | struct extent_buffer *leaf, |
6691 | struct btrfs_dev_item *dev_item) | |
6692 | { | |
0b246afa | 6693 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6694 | struct btrfs_device *device; |
6695 | u64 devid; | |
6696 | int ret; | |
44880fdc | 6697 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6698 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6699 | ||
0b86a832 | 6700 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6701 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6702 | BTRFS_UUID_SIZE); |
1473b24e | 6703 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6704 | BTRFS_FSID_SIZE); |
2b82032c | 6705 | |
44880fdc | 6706 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6707 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6708 | if (IS_ERR(fs_devices)) |
6709 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6710 | } |
6711 | ||
0b246afa | 6712 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6713 | if (!device) { |
c5502451 | 6714 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6715 | btrfs_report_missing_device(fs_info, devid, |
6716 | dev_uuid, true); | |
45dbdbc9 | 6717 | return -ENOENT; |
c5502451 | 6718 | } |
2b82032c | 6719 | |
2ff7e61e | 6720 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6721 | if (IS_ERR(device)) { |
6722 | btrfs_err(fs_info, | |
6723 | "failed to add missing dev %llu: %ld", | |
6724 | devid, PTR_ERR(device)); | |
6725 | return PTR_ERR(device); | |
6726 | } | |
2b902dfc | 6727 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6728 | } else { |
c5502451 | 6729 | if (!device->bdev) { |
2b902dfc AJ |
6730 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6731 | btrfs_report_missing_device(fs_info, | |
6732 | devid, dev_uuid, true); | |
45dbdbc9 | 6733 | return -ENOENT; |
2b902dfc AJ |
6734 | } |
6735 | btrfs_report_missing_device(fs_info, devid, | |
6736 | dev_uuid, false); | |
c5502451 | 6737 | } |
5f375835 | 6738 | |
e6e674bd AJ |
6739 | if (!device->bdev && |
6740 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6741 | /* |
6742 | * this happens when a device that was properly setup | |
6743 | * in the device info lists suddenly goes bad. | |
6744 | * device->bdev is NULL, and so we have to set | |
6745 | * device->missing to one here | |
6746 | */ | |
5f375835 | 6747 | device->fs_devices->missing_devices++; |
e6e674bd | 6748 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6749 | } |
5f375835 MX |
6750 | |
6751 | /* Move the device to its own fs_devices */ | |
6752 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6753 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6754 | &device->dev_state)); | |
5f375835 MX |
6755 | |
6756 | list_move(&device->dev_list, &fs_devices->devices); | |
6757 | device->fs_devices->num_devices--; | |
6758 | fs_devices->num_devices++; | |
6759 | ||
6760 | device->fs_devices->missing_devices--; | |
6761 | fs_devices->missing_devices++; | |
6762 | ||
6763 | device->fs_devices = fs_devices; | |
6764 | } | |
2b82032c YZ |
6765 | } |
6766 | ||
0b246afa | 6767 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6768 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6769 | if (device->generation != |
6770 | btrfs_device_generation(leaf, dev_item)) | |
6771 | return -EINVAL; | |
6324fbf3 | 6772 | } |
0b86a832 CM |
6773 | |
6774 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6775 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6776 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6777 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6778 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6779 | atomic64_add(device->total_bytes - device->bytes_used, |
6780 | &fs_info->free_chunk_space); | |
2bf64758 | 6781 | } |
0b86a832 | 6782 | ret = 0; |
0b86a832 CM |
6783 | return ret; |
6784 | } | |
6785 | ||
6bccf3ab | 6786 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6787 | { |
6bccf3ab | 6788 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6789 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6790 | struct extent_buffer *sb; |
0b86a832 | 6791 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6792 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6793 | u8 *array_ptr; |
6794 | unsigned long sb_array_offset; | |
84eed90f | 6795 | int ret = 0; |
0b86a832 CM |
6796 | u32 num_stripes; |
6797 | u32 array_size; | |
6798 | u32 len = 0; | |
1ffb22cf | 6799 | u32 cur_offset; |
e06cd3dd | 6800 | u64 type; |
84eed90f | 6801 | struct btrfs_key key; |
0b86a832 | 6802 | |
0b246afa | 6803 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6804 | /* |
6805 | * This will create extent buffer of nodesize, superblock size is | |
6806 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6807 | * overallocate but we can keep it as-is, only the first page is used. | |
6808 | */ | |
2ff7e61e | 6809 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6810 | if (IS_ERR(sb)) |
6811 | return PTR_ERR(sb); | |
4db8c528 | 6812 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6813 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6814 | /* |
01327610 | 6815 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6816 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6817 | * pages up-to-date when the page is larger: extent does not cover the |
6818 | * whole page and consequently check_page_uptodate does not find all | |
6819 | * the page's extents up-to-date (the hole beyond sb), | |
6820 | * write_extent_buffer then triggers a WARN_ON. | |
6821 | * | |
6822 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6823 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6824 | * to silence the warning eg. on PowerPC 64. | |
6825 | */ | |
09cbfeaf | 6826 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6827 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6828 | |
a061fc8d | 6829 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6830 | array_size = btrfs_super_sys_array_size(super_copy); |
6831 | ||
1ffb22cf DS |
6832 | array_ptr = super_copy->sys_chunk_array; |
6833 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6834 | cur_offset = 0; | |
0b86a832 | 6835 | |
1ffb22cf DS |
6836 | while (cur_offset < array_size) { |
6837 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6838 | len = sizeof(*disk_key); |
6839 | if (cur_offset + len > array_size) | |
6840 | goto out_short_read; | |
6841 | ||
0b86a832 CM |
6842 | btrfs_disk_key_to_cpu(&key, disk_key); |
6843 | ||
1ffb22cf DS |
6844 | array_ptr += len; |
6845 | sb_array_offset += len; | |
6846 | cur_offset += len; | |
0b86a832 | 6847 | |
0d81ba5d | 6848 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6849 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6850 | /* |
6851 | * At least one btrfs_chunk with one stripe must be | |
6852 | * present, exact stripe count check comes afterwards | |
6853 | */ | |
6854 | len = btrfs_chunk_item_size(1); | |
6855 | if (cur_offset + len > array_size) | |
6856 | goto out_short_read; | |
6857 | ||
6858 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6859 | if (!num_stripes) { |
ab8d0fc4 JM |
6860 | btrfs_err(fs_info, |
6861 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6862 | num_stripes, cur_offset); |
6863 | ret = -EIO; | |
6864 | break; | |
6865 | } | |
6866 | ||
e06cd3dd LB |
6867 | type = btrfs_chunk_type(sb, chunk); |
6868 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6869 | btrfs_err(fs_info, |
e06cd3dd LB |
6870 | "invalid chunk type %llu in sys_array at offset %u", |
6871 | type, cur_offset); | |
6872 | ret = -EIO; | |
6873 | break; | |
6874 | } | |
6875 | ||
e3540eab DS |
6876 | len = btrfs_chunk_item_size(num_stripes); |
6877 | if (cur_offset + len > array_size) | |
6878 | goto out_short_read; | |
6879 | ||
2ff7e61e | 6880 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6881 | if (ret) |
6882 | break; | |
0b86a832 | 6883 | } else { |
ab8d0fc4 JM |
6884 | btrfs_err(fs_info, |
6885 | "unexpected item type %u in sys_array at offset %u", | |
6886 | (u32)key.type, cur_offset); | |
84eed90f CM |
6887 | ret = -EIO; |
6888 | break; | |
0b86a832 | 6889 | } |
1ffb22cf DS |
6890 | array_ptr += len; |
6891 | sb_array_offset += len; | |
6892 | cur_offset += len; | |
0b86a832 | 6893 | } |
d865177a | 6894 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6895 | free_extent_buffer_stale(sb); |
84eed90f | 6896 | return ret; |
e3540eab DS |
6897 | |
6898 | out_short_read: | |
ab8d0fc4 | 6899 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6900 | len, cur_offset); |
d865177a | 6901 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6902 | free_extent_buffer_stale(sb); |
e3540eab | 6903 | return -EIO; |
0b86a832 CM |
6904 | } |
6905 | ||
21634a19 QW |
6906 | /* |
6907 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6908 | * | |
6528b99d AJ |
6909 | * If the @failing_dev is specified, it's accounted as missing. |
6910 | * | |
21634a19 QW |
6911 | * Return true if all chunks meet the minimal RW mount requirements. |
6912 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6913 | */ | |
6528b99d AJ |
6914 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6915 | struct btrfs_device *failing_dev) | |
21634a19 QW |
6916 | { |
6917 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
6918 | struct extent_map *em; | |
6919 | u64 next_start = 0; | |
6920 | bool ret = true; | |
6921 | ||
6922 | read_lock(&map_tree->map_tree.lock); | |
6923 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
6924 | read_unlock(&map_tree->map_tree.lock); | |
6925 | /* No chunk at all? Return false anyway */ | |
6926 | if (!em) { | |
6927 | ret = false; | |
6928 | goto out; | |
6929 | } | |
6930 | while (em) { | |
6931 | struct map_lookup *map; | |
6932 | int missing = 0; | |
6933 | int max_tolerated; | |
6934 | int i; | |
6935 | ||
6936 | map = em->map_lookup; | |
6937 | max_tolerated = | |
6938 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6939 | map->type); | |
6940 | for (i = 0; i < map->num_stripes; i++) { | |
6941 | struct btrfs_device *dev = map->stripes[i].dev; | |
6942 | ||
e6e674bd AJ |
6943 | if (!dev || !dev->bdev || |
6944 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6945 | dev->last_flush_error) |
6946 | missing++; | |
6528b99d AJ |
6947 | else if (failing_dev && failing_dev == dev) |
6948 | missing++; | |
21634a19 QW |
6949 | } |
6950 | if (missing > max_tolerated) { | |
6528b99d AJ |
6951 | if (!failing_dev) |
6952 | btrfs_warn(fs_info, | |
21634a19 QW |
6953 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
6954 | em->start, missing, max_tolerated); | |
6955 | free_extent_map(em); | |
6956 | ret = false; | |
6957 | goto out; | |
6958 | } | |
6959 | next_start = extent_map_end(em); | |
6960 | free_extent_map(em); | |
6961 | ||
6962 | read_lock(&map_tree->map_tree.lock); | |
6963 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
6964 | (u64)(-1) - next_start); | |
6965 | read_unlock(&map_tree->map_tree.lock); | |
6966 | } | |
6967 | out: | |
6968 | return ret; | |
6969 | } | |
6970 | ||
5b4aacef | 6971 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6972 | { |
5b4aacef | 6973 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6974 | struct btrfs_path *path; |
6975 | struct extent_buffer *leaf; | |
6976 | struct btrfs_key key; | |
6977 | struct btrfs_key found_key; | |
6978 | int ret; | |
6979 | int slot; | |
99e3ecfc | 6980 | u64 total_dev = 0; |
0b86a832 | 6981 | |
0b86a832 CM |
6982 | path = btrfs_alloc_path(); |
6983 | if (!path) | |
6984 | return -ENOMEM; | |
6985 | ||
b367e47f | 6986 | mutex_lock(&uuid_mutex); |
34441361 | 6987 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6988 | |
395927a9 FDBM |
6989 | /* |
6990 | * Read all device items, and then all the chunk items. All | |
6991 | * device items are found before any chunk item (their object id | |
6992 | * is smaller than the lowest possible object id for a chunk | |
6993 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6994 | */ |
6995 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6996 | key.offset = 0; | |
6997 | key.type = 0; | |
0b86a832 | 6998 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6999 | if (ret < 0) |
7000 | goto error; | |
d397712b | 7001 | while (1) { |
0b86a832 CM |
7002 | leaf = path->nodes[0]; |
7003 | slot = path->slots[0]; | |
7004 | if (slot >= btrfs_header_nritems(leaf)) { | |
7005 | ret = btrfs_next_leaf(root, path); | |
7006 | if (ret == 0) | |
7007 | continue; | |
7008 | if (ret < 0) | |
7009 | goto error; | |
7010 | break; | |
7011 | } | |
7012 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7013 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7014 | struct btrfs_dev_item *dev_item; | |
7015 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7016 | struct btrfs_dev_item); |
2ff7e61e | 7017 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
7018 | if (ret) |
7019 | goto error; | |
99e3ecfc | 7020 | total_dev++; |
0b86a832 CM |
7021 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7022 | struct btrfs_chunk *chunk; | |
7023 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7024 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7025 | if (ret) |
7026 | goto error; | |
0b86a832 CM |
7027 | } |
7028 | path->slots[0]++; | |
7029 | } | |
99e3ecfc LB |
7030 | |
7031 | /* | |
7032 | * After loading chunk tree, we've got all device information, | |
7033 | * do another round of validation checks. | |
7034 | */ | |
0b246afa JM |
7035 | if (total_dev != fs_info->fs_devices->total_devices) { |
7036 | btrfs_err(fs_info, | |
99e3ecfc | 7037 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7038 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7039 | total_dev); |
7040 | ret = -EINVAL; | |
7041 | goto error; | |
7042 | } | |
0b246afa JM |
7043 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7044 | fs_info->fs_devices->total_rw_bytes) { | |
7045 | btrfs_err(fs_info, | |
99e3ecfc | 7046 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7047 | btrfs_super_total_bytes(fs_info->super_copy), |
7048 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7049 | ret = -EINVAL; |
7050 | goto error; | |
7051 | } | |
0b86a832 CM |
7052 | ret = 0; |
7053 | error: | |
34441361 | 7054 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7055 | mutex_unlock(&uuid_mutex); |
7056 | ||
2b82032c | 7057 | btrfs_free_path(path); |
0b86a832 CM |
7058 | return ret; |
7059 | } | |
442a4f63 | 7060 | |
cb517eab MX |
7061 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7062 | { | |
7063 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7064 | struct btrfs_device *device; | |
7065 | ||
29cc83f6 LB |
7066 | while (fs_devices) { |
7067 | mutex_lock(&fs_devices->device_list_mutex); | |
7068 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7069 | device->fs_info = fs_info; |
29cc83f6 LB |
7070 | mutex_unlock(&fs_devices->device_list_mutex); |
7071 | ||
7072 | fs_devices = fs_devices->seed; | |
7073 | } | |
cb517eab MX |
7074 | } |
7075 | ||
733f4fbb SB |
7076 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7077 | { | |
7078 | int i; | |
7079 | ||
7080 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7081 | btrfs_dev_stat_reset(dev, i); | |
7082 | } | |
7083 | ||
7084 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7085 | { | |
7086 | struct btrfs_key key; | |
7087 | struct btrfs_key found_key; | |
7088 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7089 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7090 | struct extent_buffer *eb; | |
7091 | int slot; | |
7092 | int ret = 0; | |
7093 | struct btrfs_device *device; | |
7094 | struct btrfs_path *path = NULL; | |
7095 | int i; | |
7096 | ||
7097 | path = btrfs_alloc_path(); | |
7098 | if (!path) { | |
7099 | ret = -ENOMEM; | |
7100 | goto out; | |
7101 | } | |
7102 | ||
7103 | mutex_lock(&fs_devices->device_list_mutex); | |
7104 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7105 | int item_size; | |
7106 | struct btrfs_dev_stats_item *ptr; | |
7107 | ||
242e2956 DS |
7108 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7109 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7110 | key.offset = device->devid; |
7111 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7112 | if (ret) { | |
733f4fbb SB |
7113 | __btrfs_reset_dev_stats(device); |
7114 | device->dev_stats_valid = 1; | |
7115 | btrfs_release_path(path); | |
7116 | continue; | |
7117 | } | |
7118 | slot = path->slots[0]; | |
7119 | eb = path->nodes[0]; | |
7120 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7121 | item_size = btrfs_item_size_nr(eb, slot); | |
7122 | ||
7123 | ptr = btrfs_item_ptr(eb, slot, | |
7124 | struct btrfs_dev_stats_item); | |
7125 | ||
7126 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7127 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7128 | btrfs_dev_stat_set(device, i, | |
7129 | btrfs_dev_stats_value(eb, ptr, i)); | |
7130 | else | |
7131 | btrfs_dev_stat_reset(device, i); | |
7132 | } | |
7133 | ||
7134 | device->dev_stats_valid = 1; | |
7135 | btrfs_dev_stat_print_on_load(device); | |
7136 | btrfs_release_path(path); | |
7137 | } | |
7138 | mutex_unlock(&fs_devices->device_list_mutex); | |
7139 | ||
7140 | out: | |
7141 | btrfs_free_path(path); | |
7142 | return ret < 0 ? ret : 0; | |
7143 | } | |
7144 | ||
7145 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6bccf3ab | 7146 | struct btrfs_fs_info *fs_info, |
733f4fbb SB |
7147 | struct btrfs_device *device) |
7148 | { | |
6bccf3ab | 7149 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7150 | struct btrfs_path *path; |
7151 | struct btrfs_key key; | |
7152 | struct extent_buffer *eb; | |
7153 | struct btrfs_dev_stats_item *ptr; | |
7154 | int ret; | |
7155 | int i; | |
7156 | ||
242e2956 DS |
7157 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7158 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7159 | key.offset = device->devid; |
7160 | ||
7161 | path = btrfs_alloc_path(); | |
fa252992 DS |
7162 | if (!path) |
7163 | return -ENOMEM; | |
733f4fbb SB |
7164 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7165 | if (ret < 0) { | |
0b246afa | 7166 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7167 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7168 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7169 | goto out; |
7170 | } | |
7171 | ||
7172 | if (ret == 0 && | |
7173 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7174 | /* need to delete old one and insert a new one */ | |
7175 | ret = btrfs_del_item(trans, dev_root, path); | |
7176 | if (ret != 0) { | |
0b246afa | 7177 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7178 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7179 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7180 | goto out; |
7181 | } | |
7182 | ret = 1; | |
7183 | } | |
7184 | ||
7185 | if (ret == 1) { | |
7186 | /* need to insert a new item */ | |
7187 | btrfs_release_path(path); | |
7188 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7189 | &key, sizeof(*ptr)); | |
7190 | if (ret < 0) { | |
0b246afa | 7191 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7192 | "insert dev_stats item for device %s failed %d", |
7193 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7194 | goto out; |
7195 | } | |
7196 | } | |
7197 | ||
7198 | eb = path->nodes[0]; | |
7199 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7200 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7201 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7202 | btrfs_dev_stat_read(device, i)); | |
7203 | btrfs_mark_buffer_dirty(eb); | |
7204 | ||
7205 | out: | |
7206 | btrfs_free_path(path); | |
7207 | return ret; | |
7208 | } | |
7209 | ||
7210 | /* | |
7211 | * called from commit_transaction. Writes all changed device stats to disk. | |
7212 | */ | |
7213 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7214 | struct btrfs_fs_info *fs_info) | |
7215 | { | |
733f4fbb SB |
7216 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7217 | struct btrfs_device *device; | |
addc3fa7 | 7218 | int stats_cnt; |
733f4fbb SB |
7219 | int ret = 0; |
7220 | ||
7221 | mutex_lock(&fs_devices->device_list_mutex); | |
7222 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7223 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7224 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7225 | continue; |
7226 | ||
9deae968 NB |
7227 | |
7228 | /* | |
7229 | * There is a LOAD-LOAD control dependency between the value of | |
7230 | * dev_stats_ccnt and updating the on-disk values which requires | |
7231 | * reading the in-memory counters. Such control dependencies | |
7232 | * require explicit read memory barriers. | |
7233 | * | |
7234 | * This memory barriers pairs with smp_mb__before_atomic in | |
7235 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7236 | * barrier implied by atomic_xchg in | |
7237 | * btrfs_dev_stats_read_and_reset | |
7238 | */ | |
7239 | smp_rmb(); | |
7240 | ||
6bccf3ab | 7241 | ret = update_dev_stat_item(trans, fs_info, device); |
733f4fbb | 7242 | if (!ret) |
addc3fa7 | 7243 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7244 | } |
7245 | mutex_unlock(&fs_devices->device_list_mutex); | |
7246 | ||
7247 | return ret; | |
7248 | } | |
7249 | ||
442a4f63 SB |
7250 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7251 | { | |
7252 | btrfs_dev_stat_inc(dev, index); | |
7253 | btrfs_dev_stat_print_on_error(dev); | |
7254 | } | |
7255 | ||
48a3b636 | 7256 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7257 | { |
733f4fbb SB |
7258 | if (!dev->dev_stats_valid) |
7259 | return; | |
fb456252 | 7260 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7261 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7262 | rcu_str_deref(dev->name), |
442a4f63 SB |
7263 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7264 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7265 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7266 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7267 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7268 | } |
c11d2c23 | 7269 | |
733f4fbb SB |
7270 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7271 | { | |
a98cdb85 SB |
7272 | int i; |
7273 | ||
7274 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7275 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7276 | break; | |
7277 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7278 | return; /* all values == 0, suppress message */ | |
7279 | ||
fb456252 | 7280 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7281 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7282 | rcu_str_deref(dev->name), |
733f4fbb SB |
7283 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7284 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7285 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7286 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7287 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7288 | } | |
7289 | ||
2ff7e61e | 7290 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7291 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7292 | { |
7293 | struct btrfs_device *dev; | |
0b246afa | 7294 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7295 | int i; |
7296 | ||
7297 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7298 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7299 | mutex_unlock(&fs_devices->device_list_mutex); |
7300 | ||
7301 | if (!dev) { | |
0b246afa | 7302 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7303 | return -ENODEV; |
733f4fbb | 7304 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7305 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7306 | return -ENODEV; |
b27f7c0c | 7307 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7308 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7309 | if (stats->nr_items > i) | |
7310 | stats->values[i] = | |
7311 | btrfs_dev_stat_read_and_reset(dev, i); | |
7312 | else | |
7313 | btrfs_dev_stat_reset(dev, i); | |
7314 | } | |
7315 | } else { | |
7316 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7317 | if (stats->nr_items > i) | |
7318 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7319 | } | |
7320 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7321 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7322 | return 0; | |
7323 | } | |
a8a6dab7 | 7324 | |
da353f6b | 7325 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7326 | { |
7327 | struct buffer_head *bh; | |
7328 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7329 | int copy_num; |
a8a6dab7 | 7330 | |
12b1c263 AJ |
7331 | if (!bdev) |
7332 | return; | |
a8a6dab7 | 7333 | |
12b1c263 AJ |
7334 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7335 | copy_num++) { | |
a8a6dab7 | 7336 | |
12b1c263 AJ |
7337 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7338 | continue; | |
7339 | ||
7340 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7341 | ||
7342 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7343 | set_buffer_dirty(bh); | |
7344 | sync_dirty_buffer(bh); | |
7345 | brelse(bh); | |
7346 | } | |
7347 | ||
7348 | /* Notify udev that device has changed */ | |
7349 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7350 | ||
7351 | /* Update ctime/mtime for device path for libblkid */ | |
7352 | update_dev_time(device_path); | |
a8a6dab7 | 7353 | } |
935e5cc9 MX |
7354 | |
7355 | /* | |
7356 | * Update the size of all devices, which is used for writing out the | |
7357 | * super blocks. | |
7358 | */ | |
7359 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7360 | { | |
7361 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7362 | struct btrfs_device *curr, *next; | |
7363 | ||
7364 | if (list_empty(&fs_devices->resized_devices)) | |
7365 | return; | |
7366 | ||
7367 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7368 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7369 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7370 | resized_list) { | |
7371 | list_del_init(&curr->resized_list); | |
7372 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7373 | } | |
34441361 | 7374 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7375 | mutex_unlock(&fs_devices->device_list_mutex); |
7376 | } | |
ce7213c7 MX |
7377 | |
7378 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7379 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7380 | { |
e9b919b1 | 7381 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7382 | struct extent_map *em; |
7383 | struct map_lookup *map; | |
7384 | struct btrfs_device *dev; | |
7385 | int i; | |
7386 | ||
e9b919b1 | 7387 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7388 | return; |
7389 | ||
7390 | /* In order to kick the device replace finish process */ | |
34441361 | 7391 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7392 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7393 | map = em->map_lookup; |
ce7213c7 MX |
7394 | |
7395 | for (i = 0; i < map->num_stripes; i++) { | |
7396 | dev = map->stripes[i].dev; | |
7397 | dev->commit_bytes_used = dev->bytes_used; | |
7398 | } | |
7399 | } | |
34441361 | 7400 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7401 | } |
5a13f430 AJ |
7402 | |
7403 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7404 | { | |
7405 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7406 | while (fs_devices) { | |
7407 | fs_devices->fs_info = fs_info; | |
7408 | fs_devices = fs_devices->seed; | |
7409 | } | |
7410 | } | |
7411 | ||
7412 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7413 | { | |
7414 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7415 | while (fs_devices) { | |
7416 | fs_devices->fs_info = NULL; | |
7417 | fs_devices = fs_devices->seed; | |
7418 | } | |
7419 | } |