]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 CM |
6 | #include <linux/sched.h> |
7 | #include <linux/bio.h> | |
5a0e3ad6 | 8 | #include <linux/slab.h> |
8a4b83cc | 9 | #include <linux/buffer_head.h> |
f2d8d74d | 10 | #include <linux/blkdev.h> |
442a4f63 | 11 | #include <linux/ratelimit.h> |
59641015 | 12 | #include <linux/kthread.h> |
53b381b3 | 13 | #include <linux/raid/pq.h> |
803b2f54 | 14 | #include <linux/semaphore.h> |
8da4b8c4 | 15 | #include <linux/uuid.h> |
f8e10cd3 | 16 | #include <linux/list_sort.h> |
0b86a832 CM |
17 | #include "ctree.h" |
18 | #include "extent_map.h" | |
19 | #include "disk-io.h" | |
20 | #include "transaction.h" | |
21 | #include "print-tree.h" | |
22 | #include "volumes.h" | |
53b381b3 | 23 | #include "raid56.h" |
8b712842 | 24 | #include "async-thread.h" |
21adbd5c | 25 | #include "check-integrity.h" |
606686ee | 26 | #include "rcu-string.h" |
3fed40cc | 27 | #include "math.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
0b86a832 | 30 | |
af902047 ZL |
31 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
32 | [BTRFS_RAID_RAID10] = { | |
33 | .sub_stripes = 2, | |
34 | .dev_stripes = 1, | |
35 | .devs_max = 0, /* 0 == as many as possible */ | |
36 | .devs_min = 4, | |
8789f4fe | 37 | .tolerated_failures = 1, |
af902047 ZL |
38 | .devs_increment = 2, |
39 | .ncopies = 2, | |
ed23467b | 40 | .raid_name = "raid10", |
41a6e891 | 41 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 42 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
43 | }, |
44 | [BTRFS_RAID_RAID1] = { | |
45 | .sub_stripes = 1, | |
46 | .dev_stripes = 1, | |
47 | .devs_max = 2, | |
48 | .devs_min = 2, | |
8789f4fe | 49 | .tolerated_failures = 1, |
af902047 ZL |
50 | .devs_increment = 2, |
51 | .ncopies = 2, | |
ed23467b | 52 | .raid_name = "raid1", |
41a6e891 | 53 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 54 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 ZL |
55 | }, |
56 | [BTRFS_RAID_DUP] = { | |
57 | .sub_stripes = 1, | |
58 | .dev_stripes = 2, | |
59 | .devs_max = 1, | |
60 | .devs_min = 1, | |
8789f4fe | 61 | .tolerated_failures = 0, |
af902047 ZL |
62 | .devs_increment = 1, |
63 | .ncopies = 2, | |
ed23467b | 64 | .raid_name = "dup", |
41a6e891 | 65 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 66 | .mindev_error = 0, |
af902047 ZL |
67 | }, |
68 | [BTRFS_RAID_RAID0] = { | |
69 | .sub_stripes = 1, | |
70 | .dev_stripes = 1, | |
71 | .devs_max = 0, | |
72 | .devs_min = 2, | |
8789f4fe | 73 | .tolerated_failures = 0, |
af902047 ZL |
74 | .devs_increment = 1, |
75 | .ncopies = 1, | |
ed23467b | 76 | .raid_name = "raid0", |
41a6e891 | 77 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 78 | .mindev_error = 0, |
af902047 ZL |
79 | }, |
80 | [BTRFS_RAID_SINGLE] = { | |
81 | .sub_stripes = 1, | |
82 | .dev_stripes = 1, | |
83 | .devs_max = 1, | |
84 | .devs_min = 1, | |
8789f4fe | 85 | .tolerated_failures = 0, |
af902047 ZL |
86 | .devs_increment = 1, |
87 | .ncopies = 1, | |
ed23467b | 88 | .raid_name = "single", |
41a6e891 | 89 | .bg_flag = 0, |
f9fbcaa2 | 90 | .mindev_error = 0, |
af902047 ZL |
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, | |
ed23467b | 100 | .raid_name = "raid5", |
41a6e891 | 101 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 102 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
103 | }, |
104 | [BTRFS_RAID_RAID6] = { | |
105 | .sub_stripes = 1, | |
106 | .dev_stripes = 1, | |
107 | .devs_max = 0, | |
108 | .devs_min = 3, | |
8789f4fe | 109 | .tolerated_failures = 2, |
af902047 ZL |
110 | .devs_increment = 1, |
111 | .ncopies = 3, | |
ed23467b | 112 | .raid_name = "raid6", |
41a6e891 | 113 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 114 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
115 | }, |
116 | }; | |
117 | ||
ed23467b AJ |
118 | const char *get_raid_name(enum btrfs_raid_types type) |
119 | { | |
120 | if (type >= BTRFS_NR_RAID_TYPES) | |
121 | return NULL; | |
122 | ||
123 | return btrfs_raid_array[type].raid_name; | |
124 | } | |
125 | ||
2b82032c | 126 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 127 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 128 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 129 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 130 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 131 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
132 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
133 | enum btrfs_map_op op, | |
134 | u64 logical, u64 *length, | |
135 | struct btrfs_bio **bbio_ret, | |
136 | int mirror_num, int need_raid_map); | |
2b82032c | 137 | |
9c6b1c4d DS |
138 | /* |
139 | * Device locking | |
140 | * ============== | |
141 | * | |
142 | * There are several mutexes that protect manipulation of devices and low-level | |
143 | * structures like chunks but not block groups, extents or files | |
144 | * | |
145 | * uuid_mutex (global lock) | |
146 | * ------------------------ | |
147 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
148 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
149 | * device) or requested by the device= mount option | |
150 | * | |
151 | * the mutex can be very coarse and can cover long-running operations | |
152 | * | |
153 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
154 | * seeding, structure cloning, openning/closing devices at mount/umount time | |
155 | * | |
156 | * global::fs_devs - add, remove, updates to the global list | |
157 | * | |
158 | * does not protect: manipulation of the fs_devices::devices list! | |
159 | * | |
160 | * btrfs_device::name - renames (write side), read is RCU | |
161 | * | |
162 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
163 | * ------------------------------------------------ | |
164 | * protects updates to fs_devices::devices, ie. adding and deleting | |
165 | * | |
166 | * simple list traversal with read-only actions can be done with RCU protection | |
167 | * | |
168 | * may be used to exclude some operations from running concurrently without any | |
169 | * modifications to the list (see write_all_supers) | |
170 | * | |
9c6b1c4d DS |
171 | * balance_mutex |
172 | * ------------- | |
173 | * protects balance structures (status, state) and context accessed from | |
174 | * several places (internally, ioctl) | |
175 | * | |
176 | * chunk_mutex | |
177 | * ----------- | |
178 | * protects chunks, adding or removing during allocation, trim or when a new | |
179 | * device is added/removed | |
180 | * | |
181 | * cleaner_mutex | |
182 | * ------------- | |
183 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
184 | * cleaning together with relocation or delayed iputs | |
185 | * | |
186 | * | |
187 | * Lock nesting | |
188 | * ============ | |
189 | * | |
190 | * uuid_mutex | |
191 | * volume_mutex | |
192 | * device_list_mutex | |
193 | * chunk_mutex | |
194 | * balance_mutex | |
89595e80 AJ |
195 | * |
196 | * | |
197 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
198 | * ====================================== | |
199 | * | |
200 | * Maintains the exclusivity of the following operations that apply to the | |
201 | * whole filesystem and cannot run in parallel. | |
202 | * | |
203 | * - Balance (*) | |
204 | * - Device add | |
205 | * - Device remove | |
206 | * - Device replace (*) | |
207 | * - Resize | |
208 | * | |
209 | * The device operations (as above) can be in one of the following states: | |
210 | * | |
211 | * - Running state | |
212 | * - Paused state | |
213 | * - Completed state | |
214 | * | |
215 | * Only device operations marked with (*) can go into the Paused state for the | |
216 | * following reasons: | |
217 | * | |
218 | * - ioctl (only Balance can be Paused through ioctl) | |
219 | * - filesystem remounted as read-only | |
220 | * - filesystem unmounted and mounted as read-only | |
221 | * - system power-cycle and filesystem mounted as read-only | |
222 | * - filesystem or device errors leading to forced read-only | |
223 | * | |
224 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
225 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
226 | * A device operation in Paused or Running state can be canceled or resumed | |
227 | * either by ioctl (Balance only) or when remounted as read-write. | |
228 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
229 | * completed. | |
9c6b1c4d DS |
230 | */ |
231 | ||
67a2c45e | 232 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 233 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
234 | struct list_head *btrfs_get_fs_uuids(void) |
235 | { | |
236 | return &fs_uuids; | |
237 | } | |
8a4b83cc | 238 | |
2dfeca9b DS |
239 | /* |
240 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
241 | * @fsid: if not NULL, copy the uuid to fs_devices::fsid | |
242 | * | |
243 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
244 | * The returned struct is not linked onto any lists and can be destroyed with | |
245 | * kfree() right away. | |
246 | */ | |
247 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
2208a378 ID |
248 | { |
249 | struct btrfs_fs_devices *fs_devs; | |
250 | ||
78f2c9e6 | 251 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
252 | if (!fs_devs) |
253 | return ERR_PTR(-ENOMEM); | |
254 | ||
255 | mutex_init(&fs_devs->device_list_mutex); | |
256 | ||
257 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 258 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 | 259 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
c4babc5e | 260 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
261 | if (fsid) |
262 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 ID |
263 | |
264 | return fs_devs; | |
265 | } | |
266 | ||
a425f9d4 | 267 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf DS |
268 | { |
269 | rcu_string_free(device->name); | |
270 | bio_put(device->flush_bio); | |
271 | kfree(device); | |
272 | } | |
273 | ||
e4404d6e YZ |
274 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
275 | { | |
276 | struct btrfs_device *device; | |
277 | WARN_ON(fs_devices->opened); | |
278 | while (!list_empty(&fs_devices->devices)) { | |
279 | device = list_entry(fs_devices->devices.next, | |
280 | struct btrfs_device, dev_list); | |
281 | list_del(&device->dev_list); | |
a425f9d4 | 282 | btrfs_free_device(device); |
e4404d6e YZ |
283 | } |
284 | kfree(fs_devices); | |
285 | } | |
286 | ||
b8b8ff59 LC |
287 | static void btrfs_kobject_uevent(struct block_device *bdev, |
288 | enum kobject_action action) | |
289 | { | |
290 | int ret; | |
291 | ||
292 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
293 | if (ret) | |
efe120a0 | 294 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
295 | action, |
296 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
297 | &disk_to_dev(bdev->bd_disk)->kobj); | |
298 | } | |
299 | ||
ffc5a379 | 300 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
301 | { |
302 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 303 | |
2b82032c YZ |
304 | while (!list_empty(&fs_uuids)) { |
305 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
306 | struct btrfs_fs_devices, fs_list); |
307 | list_del(&fs_devices->fs_list); | |
e4404d6e | 308 | free_fs_devices(fs_devices); |
8a4b83cc | 309 | } |
8a4b83cc CM |
310 | } |
311 | ||
48dae9cf DS |
312 | /* |
313 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
314 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 315 | * btrfs_free_device. |
48dae9cf | 316 | */ |
12bd2fc0 ID |
317 | static struct btrfs_device *__alloc_device(void) |
318 | { | |
319 | struct btrfs_device *dev; | |
320 | ||
78f2c9e6 | 321 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
322 | if (!dev) |
323 | return ERR_PTR(-ENOMEM); | |
324 | ||
e0ae9994 DS |
325 | /* |
326 | * Preallocate a bio that's always going to be used for flushing device | |
327 | * barriers and matches the device lifespan | |
328 | */ | |
329 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
330 | if (!dev->flush_bio) { | |
331 | kfree(dev); | |
332 | return ERR_PTR(-ENOMEM); | |
333 | } | |
e0ae9994 | 334 | |
12bd2fc0 ID |
335 | INIT_LIST_HEAD(&dev->dev_list); |
336 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 337 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
338 | |
339 | spin_lock_init(&dev->io_lock); | |
340 | ||
12bd2fc0 | 341 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 342 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 343 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 344 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 345 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
346 | |
347 | return dev; | |
348 | } | |
349 | ||
35c70103 DS |
350 | /* |
351 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
352 | * return NULL. | |
353 | * | |
354 | * If devid and uuid are both specified, the match must be exact, otherwise | |
355 | * only devid is used. | |
356 | */ | |
357 | static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices, | |
358 | u64 devid, const u8 *uuid) | |
8a4b83cc CM |
359 | { |
360 | struct btrfs_device *dev; | |
8a4b83cc | 361 | |
636d2c9d | 362 | list_for_each_entry(dev, &fs_devices->devices, dev_list) { |
a443755f | 363 | if (dev->devid == devid && |
8f18cf13 | 364 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 365 | return dev; |
a443755f | 366 | } |
8a4b83cc CM |
367 | } |
368 | return NULL; | |
369 | } | |
370 | ||
a1b32a59 | 371 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 372 | { |
8a4b83cc CM |
373 | struct btrfs_fs_devices *fs_devices; |
374 | ||
c4babc5e | 375 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
8a4b83cc CM |
376 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
377 | return fs_devices; | |
378 | } | |
379 | return NULL; | |
380 | } | |
381 | ||
beaf8ab3 SB |
382 | static int |
383 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
384 | int flush, struct block_device **bdev, | |
385 | struct buffer_head **bh) | |
386 | { | |
387 | int ret; | |
388 | ||
389 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
390 | ||
391 | if (IS_ERR(*bdev)) { | |
392 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
393 | goto error; |
394 | } | |
395 | ||
396 | if (flush) | |
397 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 398 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
399 | if (ret) { |
400 | blkdev_put(*bdev, flags); | |
401 | goto error; | |
402 | } | |
403 | invalidate_bdev(*bdev); | |
404 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
405 | if (IS_ERR(*bh)) { |
406 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
407 | blkdev_put(*bdev, flags); |
408 | goto error; | |
409 | } | |
410 | ||
411 | return 0; | |
412 | ||
413 | error: | |
414 | *bdev = NULL; | |
415 | *bh = NULL; | |
416 | return ret; | |
417 | } | |
418 | ||
ffbd517d CM |
419 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
420 | struct bio *head, struct bio *tail) | |
421 | { | |
422 | ||
423 | struct bio *old_head; | |
424 | ||
425 | old_head = pending_bios->head; | |
426 | pending_bios->head = head; | |
427 | if (pending_bios->tail) | |
428 | tail->bi_next = old_head; | |
429 | else | |
430 | pending_bios->tail = tail; | |
431 | } | |
432 | ||
8b712842 CM |
433 | /* |
434 | * we try to collect pending bios for a device so we don't get a large | |
435 | * number of procs sending bios down to the same device. This greatly | |
436 | * improves the schedulers ability to collect and merge the bios. | |
437 | * | |
438 | * But, it also turns into a long list of bios to process and that is sure | |
439 | * to eventually make the worker thread block. The solution here is to | |
440 | * make some progress and then put this work struct back at the end of | |
441 | * the list if the block device is congested. This way, multiple devices | |
442 | * can make progress from a single worker thread. | |
443 | */ | |
143bede5 | 444 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 445 | { |
0b246afa | 446 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
447 | struct bio *pending; |
448 | struct backing_dev_info *bdi; | |
ffbd517d | 449 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
450 | struct bio *tail; |
451 | struct bio *cur; | |
452 | int again = 0; | |
ffbd517d | 453 | unsigned long num_run; |
d644d8a1 | 454 | unsigned long batch_run = 0; |
b765ead5 | 455 | unsigned long last_waited = 0; |
d84275c9 | 456 | int force_reg = 0; |
0e588859 | 457 | int sync_pending = 0; |
211588ad CM |
458 | struct blk_plug plug; |
459 | ||
460 | /* | |
461 | * this function runs all the bios we've collected for | |
462 | * a particular device. We don't want to wander off to | |
463 | * another device without first sending all of these down. | |
464 | * So, setup a plug here and finish it off before we return | |
465 | */ | |
466 | blk_start_plug(&plug); | |
8b712842 | 467 | |
efa7c9f9 | 468 | bdi = device->bdev->bd_bdi; |
b64a2851 | 469 | |
8b712842 CM |
470 | loop: |
471 | spin_lock(&device->io_lock); | |
472 | ||
a6837051 | 473 | loop_lock: |
d84275c9 | 474 | num_run = 0; |
ffbd517d | 475 | |
8b712842 CM |
476 | /* take all the bios off the list at once and process them |
477 | * later on (without the lock held). But, remember the | |
478 | * tail and other pointers so the bios can be properly reinserted | |
479 | * into the list if we hit congestion | |
480 | */ | |
d84275c9 | 481 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 482 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
483 | force_reg = 1; |
484 | } else { | |
ffbd517d | 485 | pending_bios = &device->pending_bios; |
d84275c9 CM |
486 | force_reg = 0; |
487 | } | |
ffbd517d CM |
488 | |
489 | pending = pending_bios->head; | |
490 | tail = pending_bios->tail; | |
8b712842 | 491 | WARN_ON(pending && !tail); |
8b712842 CM |
492 | |
493 | /* | |
494 | * if pending was null this time around, no bios need processing | |
495 | * at all and we can stop. Otherwise it'll loop back up again | |
496 | * and do an additional check so no bios are missed. | |
497 | * | |
498 | * device->running_pending is used to synchronize with the | |
499 | * schedule_bio code. | |
500 | */ | |
ffbd517d CM |
501 | if (device->pending_sync_bios.head == NULL && |
502 | device->pending_bios.head == NULL) { | |
8b712842 CM |
503 | again = 0; |
504 | device->running_pending = 0; | |
ffbd517d CM |
505 | } else { |
506 | again = 1; | |
507 | device->running_pending = 1; | |
8b712842 | 508 | } |
ffbd517d CM |
509 | |
510 | pending_bios->head = NULL; | |
511 | pending_bios->tail = NULL; | |
512 | ||
8b712842 CM |
513 | spin_unlock(&device->io_lock); |
514 | ||
d397712b | 515 | while (pending) { |
ffbd517d CM |
516 | |
517 | rmb(); | |
d84275c9 CM |
518 | /* we want to work on both lists, but do more bios on the |
519 | * sync list than the regular list | |
520 | */ | |
521 | if ((num_run > 32 && | |
522 | pending_bios != &device->pending_sync_bios && | |
523 | device->pending_sync_bios.head) || | |
524 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
525 | device->pending_bios.head)) { | |
ffbd517d CM |
526 | spin_lock(&device->io_lock); |
527 | requeue_list(pending_bios, pending, tail); | |
528 | goto loop_lock; | |
529 | } | |
530 | ||
8b712842 CM |
531 | cur = pending; |
532 | pending = pending->bi_next; | |
533 | cur->bi_next = NULL; | |
b64a2851 | 534 | |
dac56212 | 535 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 536 | |
2ab1ba68 CM |
537 | /* |
538 | * if we're doing the sync list, record that our | |
539 | * plug has some sync requests on it | |
540 | * | |
541 | * If we're doing the regular list and there are | |
542 | * sync requests sitting around, unplug before | |
543 | * we add more | |
544 | */ | |
545 | if (pending_bios == &device->pending_sync_bios) { | |
546 | sync_pending = 1; | |
547 | } else if (sync_pending) { | |
548 | blk_finish_plug(&plug); | |
549 | blk_start_plug(&plug); | |
550 | sync_pending = 0; | |
551 | } | |
552 | ||
4e49ea4a | 553 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
554 | num_run++; |
555 | batch_run++; | |
853d8ec4 DS |
556 | |
557 | cond_resched(); | |
8b712842 CM |
558 | |
559 | /* | |
560 | * we made progress, there is more work to do and the bdi | |
561 | * is now congested. Back off and let other work structs | |
562 | * run instead | |
563 | */ | |
57fd5a5f | 564 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 565 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 566 | struct io_context *ioc; |
8b712842 | 567 | |
b765ead5 CM |
568 | ioc = current->io_context; |
569 | ||
570 | /* | |
571 | * the main goal here is that we don't want to | |
572 | * block if we're going to be able to submit | |
573 | * more requests without blocking. | |
574 | * | |
575 | * This code does two great things, it pokes into | |
576 | * the elevator code from a filesystem _and_ | |
577 | * it makes assumptions about how batching works. | |
578 | */ | |
579 | if (ioc && ioc->nr_batch_requests > 0 && | |
580 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
581 | (last_waited == 0 || | |
582 | ioc->last_waited == last_waited)) { | |
583 | /* | |
584 | * we want to go through our batch of | |
585 | * requests and stop. So, we copy out | |
586 | * the ioc->last_waited time and test | |
587 | * against it before looping | |
588 | */ | |
589 | last_waited = ioc->last_waited; | |
853d8ec4 | 590 | cond_resched(); |
b765ead5 CM |
591 | continue; |
592 | } | |
8b712842 | 593 | spin_lock(&device->io_lock); |
ffbd517d | 594 | requeue_list(pending_bios, pending, tail); |
a6837051 | 595 | device->running_pending = 1; |
8b712842 CM |
596 | |
597 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
598 | btrfs_queue_work(fs_info->submit_workers, |
599 | &device->work); | |
8b712842 CM |
600 | goto done; |
601 | } | |
602 | } | |
ffbd517d | 603 | |
51684082 CM |
604 | cond_resched(); |
605 | if (again) | |
606 | goto loop; | |
607 | ||
608 | spin_lock(&device->io_lock); | |
609 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
610 | goto loop_lock; | |
611 | spin_unlock(&device->io_lock); | |
612 | ||
8b712842 | 613 | done: |
211588ad | 614 | blk_finish_plug(&plug); |
8b712842 CM |
615 | } |
616 | ||
b2950863 | 617 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
618 | { |
619 | struct btrfs_device *device; | |
620 | ||
621 | device = container_of(work, struct btrfs_device, work); | |
622 | run_scheduled_bios(device); | |
623 | } | |
624 | ||
d8367db3 AJ |
625 | /* |
626 | * Search and remove all stale (devices which are not mounted) devices. | |
627 | * When both inputs are NULL, it will search and release all stale devices. | |
628 | * path: Optional. When provided will it release all unmounted devices | |
629 | * matching this path only. | |
630 | * skip_dev: Optional. Will skip this device when searching for the stale | |
631 | * devices. | |
632 | */ | |
633 | static void btrfs_free_stale_devices(const char *path, | |
fa6d2ae5 | 634 | struct btrfs_device *skip_device) |
4fde46f0 | 635 | { |
fa6d2ae5 AJ |
636 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
637 | struct btrfs_device *device, *tmp_device; | |
4fde46f0 | 638 | |
fa6d2ae5 | 639 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
7bcb8164 AJ |
640 | mutex_lock(&fs_devices->device_list_mutex); |
641 | if (fs_devices->opened) { | |
642 | mutex_unlock(&fs_devices->device_list_mutex); | |
4fde46f0 | 643 | continue; |
7bcb8164 | 644 | } |
4fde46f0 | 645 | |
fa6d2ae5 AJ |
646 | list_for_each_entry_safe(device, tmp_device, |
647 | &fs_devices->devices, dev_list) { | |
522f1b45 | 648 | int not_found = 0; |
4fde46f0 | 649 | |
fa6d2ae5 | 650 | if (skip_device && skip_device == device) |
d8367db3 | 651 | continue; |
fa6d2ae5 | 652 | if (path && !device->name) |
4fde46f0 AJ |
653 | continue; |
654 | ||
4fde46f0 | 655 | rcu_read_lock(); |
d8367db3 | 656 | if (path) |
fa6d2ae5 | 657 | not_found = strcmp(rcu_str_deref(device->name), |
d8367db3 | 658 | path); |
4fde46f0 | 659 | rcu_read_unlock(); |
38cf665d AJ |
660 | if (not_found) |
661 | continue; | |
4fde46f0 | 662 | |
4fde46f0 | 663 | /* delete the stale device */ |
7bcb8164 AJ |
664 | fs_devices->num_devices--; |
665 | list_del(&device->dev_list); | |
666 | btrfs_free_device(device); | |
667 | ||
668 | if (fs_devices->num_devices == 0) | |
fd649f10 | 669 | break; |
7bcb8164 AJ |
670 | } |
671 | mutex_unlock(&fs_devices->device_list_mutex); | |
672 | if (fs_devices->num_devices == 0) { | |
673 | btrfs_sysfs_remove_fsid(fs_devices); | |
674 | list_del(&fs_devices->fs_list); | |
675 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
676 | } |
677 | } | |
678 | } | |
679 | ||
0fb08bcc AJ |
680 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
681 | struct btrfs_device *device, fmode_t flags, | |
682 | void *holder) | |
683 | { | |
684 | struct request_queue *q; | |
685 | struct block_device *bdev; | |
686 | struct buffer_head *bh; | |
687 | struct btrfs_super_block *disk_super; | |
688 | u64 devid; | |
689 | int ret; | |
690 | ||
691 | if (device->bdev) | |
692 | return -EINVAL; | |
693 | if (!device->name) | |
694 | return -EINVAL; | |
695 | ||
696 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
697 | &bdev, &bh); | |
698 | if (ret) | |
699 | return ret; | |
700 | ||
701 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
702 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
703 | if (devid != device->devid) | |
704 | goto error_brelse; | |
705 | ||
706 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
707 | goto error_brelse; | |
708 | ||
709 | device->generation = btrfs_super_generation(disk_super); | |
710 | ||
711 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
ebbede42 | 712 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
713 | fs_devices->seeding = 1; |
714 | } else { | |
ebbede42 AJ |
715 | if (bdev_read_only(bdev)) |
716 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
717 | else | |
718 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
719 | } |
720 | ||
721 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
722 | if (!blk_queue_nonrot(q)) |
723 | fs_devices->rotating = 1; | |
724 | ||
725 | device->bdev = bdev; | |
e12c9621 | 726 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
727 | device->mode = flags; |
728 | ||
729 | fs_devices->open_devices++; | |
ebbede42 AJ |
730 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
731 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 732 | fs_devices->rw_devices++; |
b1b8e386 | 733 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
734 | } |
735 | brelse(bh); | |
736 | ||
737 | return 0; | |
738 | ||
739 | error_brelse: | |
740 | brelse(bh); | |
741 | blkdev_put(bdev, flags); | |
742 | ||
743 | return -EINVAL; | |
744 | } | |
745 | ||
60999ca4 DS |
746 | /* |
747 | * Add new device to list of registered devices | |
748 | * | |
749 | * Returns: | |
e124ece5 AJ |
750 | * device pointer which was just added or updated when successful |
751 | * error pointer when failed | |
60999ca4 | 752 | */ |
e124ece5 | 753 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
754 | struct btrfs_super_block *disk_super, |
755 | bool *new_device_added) | |
8a4b83cc CM |
756 | { |
757 | struct btrfs_device *device; | |
758 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 759 | struct rcu_string *name; |
8a4b83cc | 760 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 761 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
8a4b83cc CM |
762 | |
763 | fs_devices = find_fsid(disk_super->fsid); | |
764 | if (!fs_devices) { | |
2208a378 ID |
765 | fs_devices = alloc_fs_devices(disk_super->fsid); |
766 | if (IS_ERR(fs_devices)) | |
e124ece5 | 767 | return ERR_CAST(fs_devices); |
2208a378 | 768 | |
9c6d173e | 769 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 770 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 771 | |
8a4b83cc CM |
772 | device = NULL; |
773 | } else { | |
9c6d173e | 774 | mutex_lock(&fs_devices->device_list_mutex); |
35c70103 DS |
775 | device = find_device(fs_devices, devid, |
776 | disk_super->dev_item.uuid); | |
8a4b83cc | 777 | } |
443f24fe | 778 | |
8a4b83cc | 779 | if (!device) { |
9c6d173e AJ |
780 | if (fs_devices->opened) { |
781 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 782 | return ERR_PTR(-EBUSY); |
9c6d173e | 783 | } |
2b82032c | 784 | |
12bd2fc0 ID |
785 | device = btrfs_alloc_device(NULL, &devid, |
786 | disk_super->dev_item.uuid); | |
787 | if (IS_ERR(device)) { | |
9c6d173e | 788 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 789 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 790 | return device; |
8a4b83cc | 791 | } |
606686ee JB |
792 | |
793 | name = rcu_string_strdup(path, GFP_NOFS); | |
794 | if (!name) { | |
a425f9d4 | 795 | btrfs_free_device(device); |
9c6d173e | 796 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 797 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 798 | } |
606686ee | 799 | rcu_assign_pointer(device->name, name); |
90519d66 | 800 | |
1f78160c | 801 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 802 | fs_devices->num_devices++; |
e5e9a520 | 803 | |
2b82032c | 804 | device->fs_devices = fs_devices; |
4306a974 | 805 | *new_device_added = true; |
327f18cc AJ |
806 | |
807 | if (disk_super->label[0]) | |
808 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
809 | disk_super->label, devid, found_transid, path); | |
810 | else | |
811 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
812 | disk_super->fsid, devid, found_transid, path); | |
813 | ||
606686ee | 814 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
815 | /* |
816 | * When FS is already mounted. | |
817 | * 1. If you are here and if the device->name is NULL that | |
818 | * means this device was missing at time of FS mount. | |
819 | * 2. If you are here and if the device->name is different | |
820 | * from 'path' that means either | |
821 | * a. The same device disappeared and reappeared with | |
822 | * different name. or | |
823 | * b. The missing-disk-which-was-replaced, has | |
824 | * reappeared now. | |
825 | * | |
826 | * We must allow 1 and 2a above. But 2b would be a spurious | |
827 | * and unintentional. | |
828 | * | |
829 | * Further in case of 1 and 2a above, the disk at 'path' | |
830 | * would have missed some transaction when it was away and | |
831 | * in case of 2a the stale bdev has to be updated as well. | |
832 | * 2b must not be allowed at all time. | |
833 | */ | |
834 | ||
835 | /* | |
0f23ae74 CM |
836 | * For now, we do allow update to btrfs_fs_device through the |
837 | * btrfs dev scan cli after FS has been mounted. We're still | |
838 | * tracking a problem where systems fail mount by subvolume id | |
839 | * when we reject replacement on a mounted FS. | |
b96de000 | 840 | */ |
0f23ae74 | 841 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
842 | /* |
843 | * That is if the FS is _not_ mounted and if you | |
844 | * are here, that means there is more than one | |
845 | * disk with same uuid and devid.We keep the one | |
846 | * with larger generation number or the last-in if | |
847 | * generation are equal. | |
848 | */ | |
9c6d173e | 849 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 850 | return ERR_PTR(-EEXIST); |
77bdae4d | 851 | } |
b96de000 | 852 | |
606686ee | 853 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
854 | if (!name) { |
855 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 856 | return ERR_PTR(-ENOMEM); |
9c6d173e | 857 | } |
606686ee JB |
858 | rcu_string_free(device->name); |
859 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 860 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 861 | fs_devices->missing_devices--; |
e6e674bd | 862 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 863 | } |
8a4b83cc CM |
864 | } |
865 | ||
77bdae4d AJ |
866 | /* |
867 | * Unmount does not free the btrfs_device struct but would zero | |
868 | * generation along with most of the other members. So just update | |
869 | * it back. We need it to pick the disk with largest generation | |
870 | * (as above). | |
871 | */ | |
872 | if (!fs_devices->opened) | |
873 | device->generation = found_transid; | |
874 | ||
f2788d2f AJ |
875 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
876 | ||
9c6d173e | 877 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 878 | return device; |
8a4b83cc CM |
879 | } |
880 | ||
e4404d6e YZ |
881 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
882 | { | |
883 | struct btrfs_fs_devices *fs_devices; | |
884 | struct btrfs_device *device; | |
885 | struct btrfs_device *orig_dev; | |
886 | ||
2208a378 ID |
887 | fs_devices = alloc_fs_devices(orig->fsid); |
888 | if (IS_ERR(fs_devices)) | |
889 | return fs_devices; | |
e4404d6e | 890 | |
adbbb863 | 891 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 892 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 893 | |
46224705 | 894 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 895 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
896 | struct rcu_string *name; |
897 | ||
12bd2fc0 ID |
898 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
899 | orig_dev->uuid); | |
900 | if (IS_ERR(device)) | |
e4404d6e YZ |
901 | goto error; |
902 | ||
606686ee JB |
903 | /* |
904 | * This is ok to do without rcu read locked because we hold the | |
905 | * uuid mutex so nothing we touch in here is going to disappear. | |
906 | */ | |
e755f780 | 907 | if (orig_dev->name) { |
78f2c9e6 DS |
908 | name = rcu_string_strdup(orig_dev->name->str, |
909 | GFP_KERNEL); | |
e755f780 | 910 | if (!name) { |
a425f9d4 | 911 | btrfs_free_device(device); |
e755f780 AJ |
912 | goto error; |
913 | } | |
914 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 915 | } |
e4404d6e | 916 | |
e4404d6e YZ |
917 | list_add(&device->dev_list, &fs_devices->devices); |
918 | device->fs_devices = fs_devices; | |
919 | fs_devices->num_devices++; | |
920 | } | |
adbbb863 | 921 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
922 | return fs_devices; |
923 | error: | |
adbbb863 | 924 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
925 | free_fs_devices(fs_devices); |
926 | return ERR_PTR(-ENOMEM); | |
927 | } | |
928 | ||
9b99b115 AJ |
929 | /* |
930 | * After we have read the system tree and know devids belonging to | |
931 | * this filesystem, remove the device which does not belong there. | |
932 | */ | |
933 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 934 | { |
c6e30871 | 935 | struct btrfs_device *device, *next; |
443f24fe | 936 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 937 | |
dfe25020 CM |
938 | mutex_lock(&uuid_mutex); |
939 | again: | |
46224705 | 940 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 941 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
942 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
943 | &device->dev_state)) { | |
401e29c1 AJ |
944 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
945 | &device->dev_state) && | |
946 | (!latest_dev || | |
947 | device->generation > latest_dev->generation)) { | |
443f24fe | 948 | latest_dev = device; |
a6b0d5c8 | 949 | } |
2b82032c | 950 | continue; |
a6b0d5c8 | 951 | } |
2b82032c | 952 | |
8dabb742 SB |
953 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
954 | /* | |
955 | * In the first step, keep the device which has | |
956 | * the correct fsid and the devid that is used | |
957 | * for the dev_replace procedure. | |
958 | * In the second step, the dev_replace state is | |
959 | * read from the device tree and it is known | |
960 | * whether the procedure is really active or | |
961 | * not, which means whether this device is | |
962 | * used or whether it should be removed. | |
963 | */ | |
401e29c1 AJ |
964 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
965 | &device->dev_state)) { | |
8dabb742 SB |
966 | continue; |
967 | } | |
968 | } | |
2b82032c | 969 | if (device->bdev) { |
d4d77629 | 970 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
971 | device->bdev = NULL; |
972 | fs_devices->open_devices--; | |
973 | } | |
ebbede42 | 974 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 975 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 976 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
977 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
978 | &device->dev_state)) | |
8dabb742 | 979 | fs_devices->rw_devices--; |
2b82032c | 980 | } |
e4404d6e YZ |
981 | list_del_init(&device->dev_list); |
982 | fs_devices->num_devices--; | |
a425f9d4 | 983 | btrfs_free_device(device); |
dfe25020 | 984 | } |
2b82032c YZ |
985 | |
986 | if (fs_devices->seed) { | |
987 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
988 | goto again; |
989 | } | |
990 | ||
443f24fe | 991 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 992 | |
dfe25020 | 993 | mutex_unlock(&uuid_mutex); |
dfe25020 | 994 | } |
a0af469b | 995 | |
f06c5965 | 996 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
997 | { |
998 | struct btrfs_device *device; | |
999 | ||
9f5316c1 | 1000 | device = container_of(head, struct btrfs_device, rcu); |
a425f9d4 | 1001 | btrfs_free_device(device); |
1f78160c XG |
1002 | } |
1003 | ||
14238819 AJ |
1004 | static void btrfs_close_bdev(struct btrfs_device *device) |
1005 | { | |
08ffcae8 DS |
1006 | if (!device->bdev) |
1007 | return; | |
1008 | ||
ebbede42 | 1009 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1010 | sync_blockdev(device->bdev); |
1011 | invalidate_bdev(device->bdev); | |
1012 | } | |
1013 | ||
08ffcae8 | 1014 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1015 | } |
1016 | ||
959b1c04 | 1017 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1018 | { |
1019 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1020 | struct btrfs_device *new_device; | |
1021 | struct rcu_string *name; | |
1022 | ||
1023 | if (device->bdev) | |
1024 | fs_devices->open_devices--; | |
1025 | ||
ebbede42 | 1026 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1027 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1028 | list_del_init(&device->dev_alloc_list); | |
1029 | fs_devices->rw_devices--; | |
1030 | } | |
1031 | ||
e6e674bd | 1032 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1033 | fs_devices->missing_devices--; |
1034 | ||
959b1c04 NB |
1035 | btrfs_close_bdev(device); |
1036 | ||
f448341a AJ |
1037 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1038 | device->uuid); | |
1039 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1040 | ||
1041 | /* Safe because we are under uuid_mutex */ | |
1042 | if (device->name) { | |
1043 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1044 | BUG_ON(!name); /* -ENOMEM */ | |
1045 | rcu_assign_pointer(new_device->name, name); | |
1046 | } | |
1047 | ||
1048 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1049 | new_device->fs_devices = device->fs_devices; | |
959b1c04 NB |
1050 | |
1051 | call_rcu(&device->rcu, free_device_rcu); | |
f448341a AJ |
1052 | } |
1053 | ||
0226e0eb | 1054 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1055 | { |
2037a093 | 1056 | struct btrfs_device *device, *tmp; |
e4404d6e | 1057 | |
2b82032c YZ |
1058 | if (--fs_devices->opened > 0) |
1059 | return 0; | |
8a4b83cc | 1060 | |
c9513edb | 1061 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1062 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1063 | btrfs_close_one_device(device); |
8a4b83cc | 1064 | } |
c9513edb XG |
1065 | mutex_unlock(&fs_devices->device_list_mutex); |
1066 | ||
e4404d6e YZ |
1067 | WARN_ON(fs_devices->open_devices); |
1068 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1069 | fs_devices->opened = 0; |
1070 | fs_devices->seeding = 0; | |
2b82032c | 1071 | |
8a4b83cc CM |
1072 | return 0; |
1073 | } | |
1074 | ||
2b82032c YZ |
1075 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1076 | { | |
e4404d6e | 1077 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1078 | int ret; |
1079 | ||
1080 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1081 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1082 | if (!fs_devices->opened) { |
1083 | seed_devices = fs_devices->seed; | |
1084 | fs_devices->seed = NULL; | |
1085 | } | |
2b82032c | 1086 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1087 | |
1088 | while (seed_devices) { | |
1089 | fs_devices = seed_devices; | |
1090 | seed_devices = fs_devices->seed; | |
0226e0eb | 1091 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1092 | free_fs_devices(fs_devices); |
1093 | } | |
2b82032c YZ |
1094 | return ret; |
1095 | } | |
1096 | ||
897fb573 | 1097 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1098 | fmode_t flags, void *holder) |
8a4b83cc | 1099 | { |
8a4b83cc | 1100 | struct btrfs_device *device; |
443f24fe | 1101 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1102 | int ret = 0; |
8a4b83cc | 1103 | |
d4d77629 TH |
1104 | flags |= FMODE_EXCL; |
1105 | ||
f117e290 | 1106 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1107 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1108 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1109 | continue; |
a0af469b | 1110 | |
9f050db4 AJ |
1111 | if (!latest_dev || |
1112 | device->generation > latest_dev->generation) | |
1113 | latest_dev = device; | |
8a4b83cc | 1114 | } |
a0af469b | 1115 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1116 | ret = -EINVAL; |
a0af469b CM |
1117 | goto out; |
1118 | } | |
2b82032c | 1119 | fs_devices->opened = 1; |
443f24fe | 1120 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1121 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1122 | out: |
2b82032c YZ |
1123 | return ret; |
1124 | } | |
1125 | ||
f8e10cd3 AJ |
1126 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1127 | { | |
1128 | struct btrfs_device *dev1, *dev2; | |
1129 | ||
1130 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1131 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1132 | ||
1133 | if (dev1->devid < dev2->devid) | |
1134 | return -1; | |
1135 | else if (dev1->devid > dev2->devid) | |
1136 | return 1; | |
1137 | return 0; | |
1138 | } | |
1139 | ||
2b82032c | 1140 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1141 | fmode_t flags, void *holder) |
2b82032c YZ |
1142 | { |
1143 | int ret; | |
1144 | ||
f5194e34 DS |
1145 | lockdep_assert_held(&uuid_mutex); |
1146 | ||
542c5908 | 1147 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1148 | if (fs_devices->opened) { |
e4404d6e YZ |
1149 | fs_devices->opened++; |
1150 | ret = 0; | |
2b82032c | 1151 | } else { |
f8e10cd3 | 1152 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1153 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1154 | } |
542c5908 AJ |
1155 | mutex_unlock(&fs_devices->device_list_mutex); |
1156 | ||
8a4b83cc CM |
1157 | return ret; |
1158 | } | |
1159 | ||
c9162bdf | 1160 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1161 | { |
1162 | kunmap(page); | |
1163 | put_page(page); | |
1164 | } | |
1165 | ||
c9162bdf OS |
1166 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1167 | struct page **page, | |
1168 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1169 | { |
1170 | void *p; | |
1171 | pgoff_t index; | |
1172 | ||
1173 | /* make sure our super fits in the device */ | |
1174 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1175 | return 1; | |
1176 | ||
1177 | /* make sure our super fits in the page */ | |
1178 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1179 | return 1; | |
1180 | ||
1181 | /* make sure our super doesn't straddle pages on disk */ | |
1182 | index = bytenr >> PAGE_SHIFT; | |
1183 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1184 | return 1; | |
1185 | ||
1186 | /* pull in the page with our super */ | |
1187 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1188 | index, GFP_KERNEL); | |
1189 | ||
1190 | if (IS_ERR_OR_NULL(*page)) | |
1191 | return 1; | |
1192 | ||
1193 | p = kmap(*page); | |
1194 | ||
1195 | /* align our pointer to the offset of the super block */ | |
1196 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1197 | ||
1198 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1199 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1200 | btrfs_release_disk_super(*page); | |
1201 | return 1; | |
1202 | } | |
1203 | ||
1204 | if ((*disk_super)->label[0] && | |
1205 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1206 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1207 | ||
1208 | return 0; | |
1209 | } | |
1210 | ||
6f60cbd3 DS |
1211 | /* |
1212 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1213 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1214 | * is read via pagecache | |
1215 | */ | |
36350e95 GJ |
1216 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1217 | void *holder) | |
8a4b83cc CM |
1218 | { |
1219 | struct btrfs_super_block *disk_super; | |
4306a974 | 1220 | bool new_device_added = false; |
36350e95 | 1221 | struct btrfs_device *device = NULL; |
8a4b83cc | 1222 | struct block_device *bdev; |
6f60cbd3 | 1223 | struct page *page; |
6f60cbd3 | 1224 | u64 bytenr; |
8a4b83cc | 1225 | |
899f9307 DS |
1226 | lockdep_assert_held(&uuid_mutex); |
1227 | ||
6f60cbd3 DS |
1228 | /* |
1229 | * we would like to check all the supers, but that would make | |
1230 | * a btrfs mount succeed after a mkfs from a different FS. | |
1231 | * So, we need to add a special mount option to scan for | |
1232 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1233 | */ | |
1234 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1235 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1236 | |
1237 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1238 | if (IS_ERR(bdev)) |
36350e95 | 1239 | return ERR_CAST(bdev); |
6f60cbd3 | 1240 | |
05a5c55d | 1241 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1242 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1243 | goto error_bdev_put; |
05a5c55d | 1244 | } |
6f60cbd3 | 1245 | |
4306a974 | 1246 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1247 | if (!IS_ERR(device)) { |
4306a974 AJ |
1248 | if (new_device_added) |
1249 | btrfs_free_stale_devices(path, device); | |
1250 | } | |
6f60cbd3 | 1251 | |
6cf86a00 | 1252 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1253 | |
1254 | error_bdev_put: | |
d4d77629 | 1255 | blkdev_put(bdev, flags); |
b6ed73bc | 1256 | |
36350e95 | 1257 | return device; |
8a4b83cc | 1258 | } |
0b86a832 | 1259 | |
499f377f | 1260 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1261 | struct btrfs_device *device, |
1262 | u64 *start, u64 len) | |
1263 | { | |
fb456252 | 1264 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1265 | struct extent_map *em; |
499f377f | 1266 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1267 | int ret = 0; |
1b984508 | 1268 | u64 physical_start = *start; |
6df9a95e | 1269 | |
499f377f JM |
1270 | if (transaction) |
1271 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1272 | again: |
1273 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1274 | struct map_lookup *map; |
1275 | int i; | |
1276 | ||
95617d69 | 1277 | map = em->map_lookup; |
6df9a95e | 1278 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1279 | u64 end; |
1280 | ||
6df9a95e JB |
1281 | if (map->stripes[i].dev != device) |
1282 | continue; | |
1b984508 | 1283 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1284 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1285 | physical_start) |
6df9a95e | 1286 | continue; |
c152b63e FM |
1287 | /* |
1288 | * Make sure that while processing the pinned list we do | |
1289 | * not override our *start with a lower value, because | |
1290 | * we can have pinned chunks that fall within this | |
1291 | * device hole and that have lower physical addresses | |
1292 | * than the pending chunks we processed before. If we | |
1293 | * do not take this special care we can end up getting | |
1294 | * 2 pending chunks that start at the same physical | |
1295 | * device offsets because the end offset of a pinned | |
1296 | * chunk can be equal to the start offset of some | |
1297 | * pending chunk. | |
1298 | */ | |
1299 | end = map->stripes[i].physical + em->orig_block_len; | |
1300 | if (end > *start) { | |
1301 | *start = end; | |
1302 | ret = 1; | |
1303 | } | |
6df9a95e JB |
1304 | } |
1305 | } | |
499f377f JM |
1306 | if (search_list != &fs_info->pinned_chunks) { |
1307 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1308 | goto again; |
1309 | } | |
6df9a95e JB |
1310 | |
1311 | return ret; | |
1312 | } | |
1313 | ||
1314 | ||
0b86a832 | 1315 | /* |
499f377f JM |
1316 | * find_free_dev_extent_start - find free space in the specified device |
1317 | * @device: the device which we search the free space in | |
1318 | * @num_bytes: the size of the free space that we need | |
1319 | * @search_start: the position from which to begin the search | |
1320 | * @start: store the start of the free space. | |
1321 | * @len: the size of the free space. that we find, or the size | |
1322 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1323 | * |
0b86a832 CM |
1324 | * this uses a pretty simple search, the expectation is that it is |
1325 | * called very infrequently and that a given device has a small number | |
1326 | * of extents | |
7bfc837d MX |
1327 | * |
1328 | * @start is used to store the start of the free space if we find. But if we | |
1329 | * don't find suitable free space, it will be used to store the start position | |
1330 | * of the max free space. | |
1331 | * | |
1332 | * @len is used to store the size of the free space that we find. | |
1333 | * But if we don't find suitable free space, it is used to store the size of | |
1334 | * the max free space. | |
0b86a832 | 1335 | */ |
499f377f JM |
1336 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1337 | struct btrfs_device *device, u64 num_bytes, | |
1338 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1339 | { |
0b246afa JM |
1340 | struct btrfs_fs_info *fs_info = device->fs_info; |
1341 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1342 | struct btrfs_key key; |
7bfc837d | 1343 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1344 | struct btrfs_path *path; |
7bfc837d MX |
1345 | u64 hole_size; |
1346 | u64 max_hole_start; | |
1347 | u64 max_hole_size; | |
1348 | u64 extent_end; | |
0b86a832 CM |
1349 | u64 search_end = device->total_bytes; |
1350 | int ret; | |
7bfc837d | 1351 | int slot; |
0b86a832 | 1352 | struct extent_buffer *l; |
8cdc7c5b FM |
1353 | |
1354 | /* | |
1355 | * We don't want to overwrite the superblock on the drive nor any area | |
1356 | * used by the boot loader (grub for example), so we make sure to start | |
1357 | * at an offset of at least 1MB. | |
1358 | */ | |
0d0c71b3 | 1359 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1360 | |
6df9a95e JB |
1361 | path = btrfs_alloc_path(); |
1362 | if (!path) | |
1363 | return -ENOMEM; | |
f2ab7618 | 1364 | |
7bfc837d MX |
1365 | max_hole_start = search_start; |
1366 | max_hole_size = 0; | |
1367 | ||
f2ab7618 | 1368 | again: |
401e29c1 AJ |
1369 | if (search_start >= search_end || |
1370 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1371 | ret = -ENOSPC; |
6df9a95e | 1372 | goto out; |
7bfc837d MX |
1373 | } |
1374 | ||
e4058b54 | 1375 | path->reada = READA_FORWARD; |
6df9a95e JB |
1376 | path->search_commit_root = 1; |
1377 | path->skip_locking = 1; | |
7bfc837d | 1378 | |
0b86a832 CM |
1379 | key.objectid = device->devid; |
1380 | key.offset = search_start; | |
1381 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1382 | |
125ccb0a | 1383 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1384 | if (ret < 0) |
7bfc837d | 1385 | goto out; |
1fcbac58 YZ |
1386 | if (ret > 0) { |
1387 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1388 | if (ret < 0) | |
7bfc837d | 1389 | goto out; |
1fcbac58 | 1390 | } |
7bfc837d | 1391 | |
0b86a832 CM |
1392 | while (1) { |
1393 | l = path->nodes[0]; | |
1394 | slot = path->slots[0]; | |
1395 | if (slot >= btrfs_header_nritems(l)) { | |
1396 | ret = btrfs_next_leaf(root, path); | |
1397 | if (ret == 0) | |
1398 | continue; | |
1399 | if (ret < 0) | |
7bfc837d MX |
1400 | goto out; |
1401 | ||
1402 | break; | |
0b86a832 CM |
1403 | } |
1404 | btrfs_item_key_to_cpu(l, &key, slot); | |
1405 | ||
1406 | if (key.objectid < device->devid) | |
1407 | goto next; | |
1408 | ||
1409 | if (key.objectid > device->devid) | |
7bfc837d | 1410 | break; |
0b86a832 | 1411 | |
962a298f | 1412 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1413 | goto next; |
9779b72f | 1414 | |
7bfc837d MX |
1415 | if (key.offset > search_start) { |
1416 | hole_size = key.offset - search_start; | |
9779b72f | 1417 | |
6df9a95e JB |
1418 | /* |
1419 | * Have to check before we set max_hole_start, otherwise | |
1420 | * we could end up sending back this offset anyway. | |
1421 | */ | |
499f377f | 1422 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1423 | &search_start, |
1b984508 FL |
1424 | hole_size)) { |
1425 | if (key.offset >= search_start) { | |
1426 | hole_size = key.offset - search_start; | |
1427 | } else { | |
1428 | WARN_ON_ONCE(1); | |
1429 | hole_size = 0; | |
1430 | } | |
1431 | } | |
6df9a95e | 1432 | |
7bfc837d MX |
1433 | if (hole_size > max_hole_size) { |
1434 | max_hole_start = search_start; | |
1435 | max_hole_size = hole_size; | |
1436 | } | |
9779b72f | 1437 | |
7bfc837d MX |
1438 | /* |
1439 | * If this free space is greater than which we need, | |
1440 | * it must be the max free space that we have found | |
1441 | * until now, so max_hole_start must point to the start | |
1442 | * of this free space and the length of this free space | |
1443 | * is stored in max_hole_size. Thus, we return | |
1444 | * max_hole_start and max_hole_size and go back to the | |
1445 | * caller. | |
1446 | */ | |
1447 | if (hole_size >= num_bytes) { | |
1448 | ret = 0; | |
1449 | goto out; | |
0b86a832 CM |
1450 | } |
1451 | } | |
0b86a832 | 1452 | |
0b86a832 | 1453 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1454 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1455 | dev_extent); | |
1456 | if (extent_end > search_start) | |
1457 | search_start = extent_end; | |
0b86a832 CM |
1458 | next: |
1459 | path->slots[0]++; | |
1460 | cond_resched(); | |
1461 | } | |
0b86a832 | 1462 | |
38c01b96 | 1463 | /* |
1464 | * At this point, search_start should be the end of | |
1465 | * allocated dev extents, and when shrinking the device, | |
1466 | * search_end may be smaller than search_start. | |
1467 | */ | |
f2ab7618 | 1468 | if (search_end > search_start) { |
38c01b96 | 1469 | hole_size = search_end - search_start; |
1470 | ||
499f377f | 1471 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1472 | hole_size)) { |
1473 | btrfs_release_path(path); | |
1474 | goto again; | |
1475 | } | |
0b86a832 | 1476 | |
f2ab7618 ZL |
1477 | if (hole_size > max_hole_size) { |
1478 | max_hole_start = search_start; | |
1479 | max_hole_size = hole_size; | |
1480 | } | |
6df9a95e JB |
1481 | } |
1482 | ||
7bfc837d | 1483 | /* See above. */ |
f2ab7618 | 1484 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1485 | ret = -ENOSPC; |
1486 | else | |
1487 | ret = 0; | |
1488 | ||
1489 | out: | |
2b82032c | 1490 | btrfs_free_path(path); |
7bfc837d | 1491 | *start = max_hole_start; |
b2117a39 | 1492 | if (len) |
7bfc837d | 1493 | *len = max_hole_size; |
0b86a832 CM |
1494 | return ret; |
1495 | } | |
1496 | ||
499f377f JM |
1497 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1498 | struct btrfs_device *device, u64 num_bytes, | |
1499 | u64 *start, u64 *len) | |
1500 | { | |
499f377f | 1501 | /* FIXME use last free of some kind */ |
499f377f | 1502 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1503 | num_bytes, 0, start, len); |
499f377f JM |
1504 | } |
1505 | ||
b2950863 | 1506 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1507 | struct btrfs_device *device, |
2196d6e8 | 1508 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1509 | { |
0b246afa JM |
1510 | struct btrfs_fs_info *fs_info = device->fs_info; |
1511 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1512 | int ret; |
1513 | struct btrfs_path *path; | |
8f18cf13 | 1514 | struct btrfs_key key; |
a061fc8d CM |
1515 | struct btrfs_key found_key; |
1516 | struct extent_buffer *leaf = NULL; | |
1517 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1518 | |
1519 | path = btrfs_alloc_path(); | |
1520 | if (!path) | |
1521 | return -ENOMEM; | |
1522 | ||
1523 | key.objectid = device->devid; | |
1524 | key.offset = start; | |
1525 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1526 | again: |
8f18cf13 | 1527 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1528 | if (ret > 0) { |
1529 | ret = btrfs_previous_item(root, path, key.objectid, | |
1530 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1531 | if (ret) |
1532 | goto out; | |
a061fc8d CM |
1533 | leaf = path->nodes[0]; |
1534 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1535 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1536 | struct btrfs_dev_extent); | |
1537 | BUG_ON(found_key.offset > start || found_key.offset + | |
1538 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1539 | key = found_key; |
1540 | btrfs_release_path(path); | |
1541 | goto again; | |
a061fc8d CM |
1542 | } else if (ret == 0) { |
1543 | leaf = path->nodes[0]; | |
1544 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1545 | struct btrfs_dev_extent); | |
79787eaa | 1546 | } else { |
0b246afa | 1547 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1548 | goto out; |
a061fc8d | 1549 | } |
8f18cf13 | 1550 | |
2196d6e8 MX |
1551 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1552 | ||
8f18cf13 | 1553 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1554 | if (ret) { |
0b246afa JM |
1555 | btrfs_handle_fs_error(fs_info, ret, |
1556 | "Failed to remove dev extent item"); | |
13212b54 | 1557 | } else { |
3204d33c | 1558 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1559 | } |
b0b802d7 | 1560 | out: |
8f18cf13 CM |
1561 | btrfs_free_path(path); |
1562 | return ret; | |
1563 | } | |
1564 | ||
48a3b636 ES |
1565 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1566 | struct btrfs_device *device, | |
48a3b636 | 1567 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1568 | { |
1569 | int ret; | |
1570 | struct btrfs_path *path; | |
0b246afa JM |
1571 | struct btrfs_fs_info *fs_info = device->fs_info; |
1572 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1573 | struct btrfs_dev_extent *extent; |
1574 | struct extent_buffer *leaf; | |
1575 | struct btrfs_key key; | |
1576 | ||
e12c9621 | 1577 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1578 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1579 | path = btrfs_alloc_path(); |
1580 | if (!path) | |
1581 | return -ENOMEM; | |
1582 | ||
0b86a832 | 1583 | key.objectid = device->devid; |
2b82032c | 1584 | key.offset = start; |
0b86a832 CM |
1585 | key.type = BTRFS_DEV_EXTENT_KEY; |
1586 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1587 | sizeof(*extent)); | |
2cdcecbc MF |
1588 | if (ret) |
1589 | goto out; | |
0b86a832 CM |
1590 | |
1591 | leaf = path->nodes[0]; | |
1592 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1593 | struct btrfs_dev_extent); | |
b5d9071c NB |
1594 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1595 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1596 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1597 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1598 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1599 | ||
0b86a832 CM |
1600 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1601 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1602 | out: |
0b86a832 CM |
1603 | btrfs_free_path(path); |
1604 | return ret; | |
1605 | } | |
1606 | ||
6df9a95e | 1607 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1608 | { |
6df9a95e JB |
1609 | struct extent_map_tree *em_tree; |
1610 | struct extent_map *em; | |
1611 | struct rb_node *n; | |
1612 | u64 ret = 0; | |
0b86a832 | 1613 | |
6df9a95e JB |
1614 | em_tree = &fs_info->mapping_tree.map_tree; |
1615 | read_lock(&em_tree->lock); | |
1616 | n = rb_last(&em_tree->map); | |
1617 | if (n) { | |
1618 | em = rb_entry(n, struct extent_map, rb_node); | |
1619 | ret = em->start + em->len; | |
0b86a832 | 1620 | } |
6df9a95e JB |
1621 | read_unlock(&em_tree->lock); |
1622 | ||
0b86a832 CM |
1623 | return ret; |
1624 | } | |
1625 | ||
53f10659 ID |
1626 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1627 | u64 *devid_ret) | |
0b86a832 CM |
1628 | { |
1629 | int ret; | |
1630 | struct btrfs_key key; | |
1631 | struct btrfs_key found_key; | |
2b82032c YZ |
1632 | struct btrfs_path *path; |
1633 | ||
2b82032c YZ |
1634 | path = btrfs_alloc_path(); |
1635 | if (!path) | |
1636 | return -ENOMEM; | |
0b86a832 CM |
1637 | |
1638 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1639 | key.type = BTRFS_DEV_ITEM_KEY; | |
1640 | key.offset = (u64)-1; | |
1641 | ||
53f10659 | 1642 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1643 | if (ret < 0) |
1644 | goto error; | |
1645 | ||
79787eaa | 1646 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1647 | |
53f10659 ID |
1648 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1649 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1650 | BTRFS_DEV_ITEM_KEY); |
1651 | if (ret) { | |
53f10659 | 1652 | *devid_ret = 1; |
0b86a832 CM |
1653 | } else { |
1654 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1655 | path->slots[0]); | |
53f10659 | 1656 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1657 | } |
1658 | ret = 0; | |
1659 | error: | |
2b82032c | 1660 | btrfs_free_path(path); |
0b86a832 CM |
1661 | return ret; |
1662 | } | |
1663 | ||
1664 | /* | |
1665 | * the device information is stored in the chunk root | |
1666 | * the btrfs_device struct should be fully filled in | |
1667 | */ | |
c74a0b02 | 1668 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1669 | struct btrfs_device *device) |
0b86a832 CM |
1670 | { |
1671 | int ret; | |
1672 | struct btrfs_path *path; | |
1673 | struct btrfs_dev_item *dev_item; | |
1674 | struct extent_buffer *leaf; | |
1675 | struct btrfs_key key; | |
1676 | unsigned long ptr; | |
0b86a832 | 1677 | |
0b86a832 CM |
1678 | path = btrfs_alloc_path(); |
1679 | if (!path) | |
1680 | return -ENOMEM; | |
1681 | ||
0b86a832 CM |
1682 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1683 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1684 | key.offset = device->devid; |
0b86a832 | 1685 | |
8e87e856 NB |
1686 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1687 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1688 | if (ret) |
1689 | goto out; | |
1690 | ||
1691 | leaf = path->nodes[0]; | |
1692 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1693 | ||
1694 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1695 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1696 | btrfs_set_device_type(leaf, dev_item, device->type); |
1697 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1698 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1699 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1700 | btrfs_set_device_total_bytes(leaf, dev_item, |
1701 | btrfs_device_get_disk_total_bytes(device)); | |
1702 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1703 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1704 | btrfs_set_device_group(leaf, dev_item, 0); |
1705 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1706 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1707 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1708 | |
410ba3a2 | 1709 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1710 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1711 | ptr = btrfs_device_fsid(dev_item); |
8e87e856 | 1712 | write_extent_buffer(leaf, trans->fs_info->fsid, ptr, BTRFS_FSID_SIZE); |
0b86a832 | 1713 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1714 | |
2b82032c | 1715 | ret = 0; |
0b86a832 CM |
1716 | out: |
1717 | btrfs_free_path(path); | |
1718 | return ret; | |
1719 | } | |
8f18cf13 | 1720 | |
5a1972bd QW |
1721 | /* |
1722 | * Function to update ctime/mtime for a given device path. | |
1723 | * Mainly used for ctime/mtime based probe like libblkid. | |
1724 | */ | |
da353f6b | 1725 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1726 | { |
1727 | struct file *filp; | |
1728 | ||
1729 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1730 | if (IS_ERR(filp)) |
5a1972bd QW |
1731 | return; |
1732 | file_update_time(filp); | |
1733 | filp_close(filp, NULL); | |
5a1972bd QW |
1734 | } |
1735 | ||
5b4aacef | 1736 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1737 | struct btrfs_device *device) |
1738 | { | |
5b4aacef | 1739 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1740 | int ret; |
1741 | struct btrfs_path *path; | |
a061fc8d | 1742 | struct btrfs_key key; |
a061fc8d CM |
1743 | struct btrfs_trans_handle *trans; |
1744 | ||
a061fc8d CM |
1745 | path = btrfs_alloc_path(); |
1746 | if (!path) | |
1747 | return -ENOMEM; | |
1748 | ||
a22285a6 | 1749 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1750 | if (IS_ERR(trans)) { |
1751 | btrfs_free_path(path); | |
1752 | return PTR_ERR(trans); | |
1753 | } | |
a061fc8d CM |
1754 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1755 | key.type = BTRFS_DEV_ITEM_KEY; | |
1756 | key.offset = device->devid; | |
1757 | ||
1758 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1759 | if (ret) { |
1760 | if (ret > 0) | |
1761 | ret = -ENOENT; | |
1762 | btrfs_abort_transaction(trans, ret); | |
1763 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1764 | goto out; |
1765 | } | |
1766 | ||
1767 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1768 | if (ret) { |
1769 | btrfs_abort_transaction(trans, ret); | |
1770 | btrfs_end_transaction(trans); | |
1771 | } | |
1772 | ||
a061fc8d CM |
1773 | out: |
1774 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1775 | if (!ret) |
1776 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1777 | return ret; |
1778 | } | |
1779 | ||
3cc31a0d DS |
1780 | /* |
1781 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1782 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1783 | * replace. | |
1784 | */ | |
1785 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1786 | u64 num_devices) | |
a061fc8d | 1787 | { |
a061fc8d | 1788 | u64 all_avail; |
de98ced9 | 1789 | unsigned seq; |
418775a2 | 1790 | int i; |
a061fc8d | 1791 | |
de98ced9 | 1792 | do { |
bd45ffbc | 1793 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1794 | |
bd45ffbc AJ |
1795 | all_avail = fs_info->avail_data_alloc_bits | |
1796 | fs_info->avail_system_alloc_bits | | |
1797 | fs_info->avail_metadata_alloc_bits; | |
1798 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1799 | |
418775a2 | 1800 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1801 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1802 | continue; |
a061fc8d | 1803 | |
418775a2 | 1804 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 1805 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 1806 | |
418775a2 DS |
1807 | if (ret) |
1808 | return ret; | |
1809 | } | |
53b381b3 DW |
1810 | } |
1811 | ||
bd45ffbc | 1812 | return 0; |
f1fa7f26 AJ |
1813 | } |
1814 | ||
c9162bdf OS |
1815 | static struct btrfs_device * btrfs_find_next_active_device( |
1816 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1817 | { |
2b82032c | 1818 | struct btrfs_device *next_device; |
88acff64 AJ |
1819 | |
1820 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1821 | if (next_device != device && | |
e6e674bd AJ |
1822 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1823 | && next_device->bdev) | |
88acff64 AJ |
1824 | return next_device; |
1825 | } | |
1826 | ||
1827 | return NULL; | |
1828 | } | |
1829 | ||
1830 | /* | |
1831 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1832 | * and replace it with the provided or the next active device, in the context | |
1833 | * where this function called, there should be always be another device (or | |
1834 | * this_dev) which is active. | |
1835 | */ | |
d6507cf1 NB |
1836 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
1837 | struct btrfs_device *this_dev) | |
88acff64 | 1838 | { |
d6507cf1 | 1839 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
1840 | struct btrfs_device *next_device; |
1841 | ||
1842 | if (this_dev) | |
1843 | next_device = this_dev; | |
1844 | else | |
1845 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1846 | device); | |
1847 | ASSERT(next_device); | |
1848 | ||
1849 | if (fs_info->sb->s_bdev && | |
1850 | (fs_info->sb->s_bdev == device->bdev)) | |
1851 | fs_info->sb->s_bdev = next_device->bdev; | |
1852 | ||
1853 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1854 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1855 | } | |
1856 | ||
da353f6b DS |
1857 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1858 | u64 devid) | |
f1fa7f26 AJ |
1859 | { |
1860 | struct btrfs_device *device; | |
1f78160c | 1861 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 1862 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 1863 | u64 num_devices; |
a061fc8d CM |
1864 | int ret = 0; |
1865 | ||
a061fc8d CM |
1866 | mutex_lock(&uuid_mutex); |
1867 | ||
b5185197 | 1868 | num_devices = fs_devices->num_devices; |
7e79cb86 | 1869 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
0b246afa | 1870 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
8dabb742 SB |
1871 | WARN_ON(num_devices < 1); |
1872 | num_devices--; | |
1873 | } | |
7e79cb86 | 1874 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
8dabb742 | 1875 | |
0b246afa | 1876 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1877 | if (ret) |
a061fc8d | 1878 | goto out; |
a061fc8d | 1879 | |
2ff7e61e JM |
1880 | ret = btrfs_find_device_by_devspec(fs_info, devid, device_path, |
1881 | &device); | |
24fc572f | 1882 | if (ret) |
53b381b3 | 1883 | goto out; |
dfe25020 | 1884 | |
401e29c1 | 1885 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1886 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1887 | goto out; |
63a212ab SB |
1888 | } |
1889 | ||
ebbede42 AJ |
1890 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1891 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1892 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1893 | goto out; |
2b82032c YZ |
1894 | } |
1895 | ||
ebbede42 | 1896 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1897 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1898 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1899 | device->fs_devices->rw_devices--; |
34441361 | 1900 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 1901 | } |
a061fc8d | 1902 | |
d7901554 | 1903 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1904 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1905 | mutex_lock(&uuid_mutex); |
a061fc8d | 1906 | if (ret) |
9b3517e9 | 1907 | goto error_undo; |
a061fc8d | 1908 | |
63a212ab SB |
1909 | /* |
1910 | * TODO: the superblock still includes this device in its num_devices | |
1911 | * counter although write_all_supers() is not locked out. This | |
1912 | * could give a filesystem state which requires a degraded mount. | |
1913 | */ | |
0b246afa | 1914 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1915 | if (ret) |
9b3517e9 | 1916 | goto error_undo; |
a061fc8d | 1917 | |
e12c9621 | 1918 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 1919 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1920 | |
1921 | /* | |
1922 | * the device list mutex makes sure that we don't change | |
1923 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1924 | * the device supers. Whoever is writing all supers, should |
1925 | * lock the device list mutex before getting the number of | |
1926 | * devices in the super block (super_copy). Conversely, | |
1927 | * whoever updates the number of devices in the super block | |
1928 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1929 | */ |
1f78160c | 1930 | |
41a52a0f AJ |
1931 | /* |
1932 | * In normal cases the cur_devices == fs_devices. But in case | |
1933 | * of deleting a seed device, the cur_devices should point to | |
1934 | * its own fs_devices listed under the fs_devices->seed. | |
1935 | */ | |
1f78160c | 1936 | cur_devices = device->fs_devices; |
b5185197 | 1937 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 1938 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1939 | |
41a52a0f AJ |
1940 | cur_devices->num_devices--; |
1941 | cur_devices->total_devices--; | |
b4993e64 AJ |
1942 | /* Update total_devices of the parent fs_devices if it's seed */ |
1943 | if (cur_devices != fs_devices) | |
1944 | fs_devices->total_devices--; | |
2b82032c | 1945 | |
e6e674bd | 1946 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 1947 | cur_devices->missing_devices--; |
cd02dca5 | 1948 | |
d6507cf1 | 1949 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 1950 | |
0bfaa9c5 | 1951 | if (device->bdev) { |
41a52a0f | 1952 | cur_devices->open_devices--; |
0bfaa9c5 | 1953 | /* remove sysfs entry */ |
b5185197 | 1954 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 1955 | } |
99994cde | 1956 | |
0b246afa JM |
1957 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
1958 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 1959 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 1960 | |
cea67ab9 JM |
1961 | /* |
1962 | * at this point, the device is zero sized and detached from | |
1963 | * the devices list. All that's left is to zero out the old | |
1964 | * supers and free the device. | |
1965 | */ | |
ebbede42 | 1966 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
1967 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
1968 | ||
1969 | btrfs_close_bdev(device); | |
f06c5965 | 1970 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 1971 | |
1f78160c | 1972 | if (cur_devices->open_devices == 0) { |
e4404d6e | 1973 | while (fs_devices) { |
8321cf25 RS |
1974 | if (fs_devices->seed == cur_devices) { |
1975 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1976 | break; |
8321cf25 | 1977 | } |
e4404d6e | 1978 | fs_devices = fs_devices->seed; |
2b82032c | 1979 | } |
1f78160c | 1980 | cur_devices->seed = NULL; |
0226e0eb | 1981 | close_fs_devices(cur_devices); |
1f78160c | 1982 | free_fs_devices(cur_devices); |
2b82032c YZ |
1983 | } |
1984 | ||
a061fc8d CM |
1985 | out: |
1986 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1987 | return ret; |
24fc572f | 1988 | |
9b3517e9 | 1989 | error_undo: |
ebbede42 | 1990 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1991 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 1992 | list_add(&device->dev_alloc_list, |
b5185197 | 1993 | &fs_devices->alloc_list); |
c3929c36 | 1994 | device->fs_devices->rw_devices++; |
34441361 | 1995 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 1996 | } |
24fc572f | 1997 | goto out; |
a061fc8d CM |
1998 | } |
1999 | ||
68a9db5f | 2000 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2001 | { |
d51908ce AJ |
2002 | struct btrfs_fs_devices *fs_devices; |
2003 | ||
68a9db5f | 2004 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2005 | |
25e8e911 AJ |
2006 | /* |
2007 | * in case of fs with no seed, srcdev->fs_devices will point | |
2008 | * to fs_devices of fs_info. However when the dev being replaced is | |
2009 | * a seed dev it will point to the seed's local fs_devices. In short | |
2010 | * srcdev will have its correct fs_devices in both the cases. | |
2011 | */ | |
2012 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2013 | |
e93c89c1 | 2014 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2015 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2016 | fs_devices->num_devices--; |
e6e674bd | 2017 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2018 | fs_devices->missing_devices--; |
e93c89c1 | 2019 | |
ebbede42 | 2020 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2021 | fs_devices->rw_devices--; |
1357272f | 2022 | |
82372bc8 | 2023 | if (srcdev->bdev) |
d51908ce | 2024 | fs_devices->open_devices--; |
084b6e7c QW |
2025 | } |
2026 | ||
2027 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2028 | struct btrfs_device *srcdev) | |
2029 | { | |
2030 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2031 | |
ebbede42 | 2032 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2033 | /* zero out the old super if it is writable */ |
2034 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2035 | } | |
14238819 AJ |
2036 | |
2037 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2038 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2039 | |
94d5f0c2 AJ |
2040 | /* if this is no devs we rather delete the fs_devices */ |
2041 | if (!fs_devices->num_devices) { | |
2042 | struct btrfs_fs_devices *tmp_fs_devices; | |
2043 | ||
6dd38f81 AJ |
2044 | /* |
2045 | * On a mounted FS, num_devices can't be zero unless it's a | |
2046 | * seed. In case of a seed device being replaced, the replace | |
2047 | * target added to the sprout FS, so there will be no more | |
2048 | * device left under the seed FS. | |
2049 | */ | |
2050 | ASSERT(fs_devices->seeding); | |
2051 | ||
94d5f0c2 AJ |
2052 | tmp_fs_devices = fs_info->fs_devices; |
2053 | while (tmp_fs_devices) { | |
2054 | if (tmp_fs_devices->seed == fs_devices) { | |
2055 | tmp_fs_devices->seed = fs_devices->seed; | |
2056 | break; | |
2057 | } | |
2058 | tmp_fs_devices = tmp_fs_devices->seed; | |
2059 | } | |
2060 | fs_devices->seed = NULL; | |
0226e0eb | 2061 | close_fs_devices(fs_devices); |
8bef8401 | 2062 | free_fs_devices(fs_devices); |
94d5f0c2 | 2063 | } |
e93c89c1 SB |
2064 | } |
2065 | ||
4f5ad7bd | 2066 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2067 | { |
4f5ad7bd | 2068 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2069 | |
e93c89c1 | 2070 | WARN_ON(!tgtdev); |
d9a071f0 | 2071 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2072 | |
d9a071f0 | 2073 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2074 | |
779bf3fe | 2075 | if (tgtdev->bdev) |
d9a071f0 | 2076 | fs_devices->open_devices--; |
779bf3fe | 2077 | |
d9a071f0 | 2078 | fs_devices->num_devices--; |
e93c89c1 | 2079 | |
d6507cf1 | 2080 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2081 | |
e93c89c1 | 2082 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2083 | |
d9a071f0 | 2084 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2085 | |
2086 | /* | |
2087 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2088 | * may lead to a call to btrfs_show_devname() which will try | |
2089 | * to hold device_list_mutex. And here this device | |
2090 | * is already out of device list, so we don't have to hold | |
2091 | * the device_list_mutex lock. | |
2092 | */ | |
2093 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2094 | |
2095 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2096 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2097 | } |
2098 | ||
2ff7e61e | 2099 | static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2100 | const char *device_path, |
48a3b636 | 2101 | struct btrfs_device **device) |
7ba15b7d SB |
2102 | { |
2103 | int ret = 0; | |
2104 | struct btrfs_super_block *disk_super; | |
2105 | u64 devid; | |
2106 | u8 *dev_uuid; | |
2107 | struct block_device *bdev; | |
2108 | struct buffer_head *bh; | |
2109 | ||
2110 | *device = NULL; | |
2111 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2112 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2113 | if (ret) |
2114 | return ret; | |
2115 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2116 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2117 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2118 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2119 | brelse(bh); |
2120 | if (!*device) | |
2121 | ret = -ENOENT; | |
2122 | blkdev_put(bdev, FMODE_READ); | |
2123 | return ret; | |
2124 | } | |
2125 | ||
2ff7e61e | 2126 | int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info, |
da353f6b | 2127 | const char *device_path, |
7ba15b7d SB |
2128 | struct btrfs_device **device) |
2129 | { | |
2130 | *device = NULL; | |
2131 | if (strcmp(device_path, "missing") == 0) { | |
2132 | struct list_head *devices; | |
2133 | struct btrfs_device *tmp; | |
2134 | ||
0b246afa | 2135 | devices = &fs_info->fs_devices->devices; |
7ba15b7d | 2136 | list_for_each_entry(tmp, devices, dev_list) { |
e12c9621 AJ |
2137 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2138 | &tmp->dev_state) && !tmp->bdev) { | |
7ba15b7d SB |
2139 | *device = tmp; |
2140 | break; | |
2141 | } | |
2142 | } | |
2143 | ||
d74a6259 AJ |
2144 | if (!*device) |
2145 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2146 | |
2147 | return 0; | |
2148 | } else { | |
2ff7e61e | 2149 | return btrfs_find_device_by_path(fs_info, device_path, device); |
7ba15b7d SB |
2150 | } |
2151 | } | |
2152 | ||
5c5c0df0 DS |
2153 | /* |
2154 | * Lookup a device given by device id, or the path if the id is 0. | |
2155 | */ | |
2ff7e61e | 2156 | int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid, |
da353f6b DS |
2157 | const char *devpath, |
2158 | struct btrfs_device **device) | |
24e0474b AJ |
2159 | { |
2160 | int ret; | |
2161 | ||
5c5c0df0 | 2162 | if (devid) { |
24e0474b | 2163 | ret = 0; |
0b246afa | 2164 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2165 | if (!*device) |
2166 | ret = -ENOENT; | |
2167 | } else { | |
5c5c0df0 | 2168 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2169 | return -EINVAL; |
2170 | ||
2ff7e61e | 2171 | ret = btrfs_find_device_missing_or_by_path(fs_info, devpath, |
24e0474b AJ |
2172 | device); |
2173 | } | |
2174 | return ret; | |
2175 | } | |
2176 | ||
2b82032c YZ |
2177 | /* |
2178 | * does all the dirty work required for changing file system's UUID. | |
2179 | */ | |
2ff7e61e | 2180 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2181 | { |
0b246afa | 2182 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2183 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2184 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2185 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2186 | struct btrfs_device *device; |
2187 | u64 super_flags; | |
2188 | ||
a32bf9a3 | 2189 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2190 | if (!fs_devices->seeding) |
2b82032c YZ |
2191 | return -EINVAL; |
2192 | ||
2dfeca9b | 2193 | seed_devices = alloc_fs_devices(NULL); |
2208a378 ID |
2194 | if (IS_ERR(seed_devices)) |
2195 | return PTR_ERR(seed_devices); | |
2b82032c | 2196 | |
e4404d6e YZ |
2197 | old_devices = clone_fs_devices(fs_devices); |
2198 | if (IS_ERR(old_devices)) { | |
2199 | kfree(seed_devices); | |
2200 | return PTR_ERR(old_devices); | |
2b82032c | 2201 | } |
e4404d6e | 2202 | |
c4babc5e | 2203 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2204 | |
e4404d6e YZ |
2205 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2206 | seed_devices->opened = 1; | |
2207 | INIT_LIST_HEAD(&seed_devices->devices); | |
2208 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2209 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2210 | |
321a4bf7 | 2211 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2212 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2213 | synchronize_rcu); | |
2196d6e8 MX |
2214 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2215 | device->fs_devices = seed_devices; | |
c9513edb | 2216 | |
34441361 | 2217 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2218 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2219 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2220 | |
2b82032c YZ |
2221 | fs_devices->seeding = 0; |
2222 | fs_devices->num_devices = 0; | |
2223 | fs_devices->open_devices = 0; | |
69611ac8 | 2224 | fs_devices->missing_devices = 0; |
69611ac8 | 2225 | fs_devices->rotating = 0; |
e4404d6e | 2226 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2227 | |
2228 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2229 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2230 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2231 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2232 | |
2b82032c YZ |
2233 | super_flags = btrfs_super_flags(disk_super) & |
2234 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2235 | btrfs_set_super_flags(disk_super, super_flags); | |
2236 | ||
2237 | return 0; | |
2238 | } | |
2239 | ||
2240 | /* | |
01327610 | 2241 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2242 | */ |
2243 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2244 | struct btrfs_fs_info *fs_info) |
2b82032c | 2245 | { |
5b4aacef | 2246 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2247 | struct btrfs_path *path; |
2248 | struct extent_buffer *leaf; | |
2249 | struct btrfs_dev_item *dev_item; | |
2250 | struct btrfs_device *device; | |
2251 | struct btrfs_key key; | |
44880fdc | 2252 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2253 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2254 | u64 devid; | |
2255 | int ret; | |
2256 | ||
2257 | path = btrfs_alloc_path(); | |
2258 | if (!path) | |
2259 | return -ENOMEM; | |
2260 | ||
2b82032c YZ |
2261 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2262 | key.offset = 0; | |
2263 | key.type = BTRFS_DEV_ITEM_KEY; | |
2264 | ||
2265 | while (1) { | |
2266 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2267 | if (ret < 0) | |
2268 | goto error; | |
2269 | ||
2270 | leaf = path->nodes[0]; | |
2271 | next_slot: | |
2272 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2273 | ret = btrfs_next_leaf(root, path); | |
2274 | if (ret > 0) | |
2275 | break; | |
2276 | if (ret < 0) | |
2277 | goto error; | |
2278 | leaf = path->nodes[0]; | |
2279 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2280 | btrfs_release_path(path); |
2b82032c YZ |
2281 | continue; |
2282 | } | |
2283 | ||
2284 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2285 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2286 | key.type != BTRFS_DEV_ITEM_KEY) | |
2287 | break; | |
2288 | ||
2289 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2290 | struct btrfs_dev_item); | |
2291 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2292 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2293 | BTRFS_UUID_SIZE); |
1473b24e | 2294 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2295 | BTRFS_FSID_SIZE); |
0b246afa | 2296 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2297 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2298 | |
2299 | if (device->fs_devices->seeding) { | |
2300 | btrfs_set_device_generation(leaf, dev_item, | |
2301 | device->generation); | |
2302 | btrfs_mark_buffer_dirty(leaf); | |
2303 | } | |
2304 | ||
2305 | path->slots[0]++; | |
2306 | goto next_slot; | |
2307 | } | |
2308 | ret = 0; | |
2309 | error: | |
2310 | btrfs_free_path(path); | |
2311 | return ret; | |
2312 | } | |
2313 | ||
da353f6b | 2314 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2315 | { |
5112febb | 2316 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2317 | struct request_queue *q; |
788f20eb CM |
2318 | struct btrfs_trans_handle *trans; |
2319 | struct btrfs_device *device; | |
2320 | struct block_device *bdev; | |
0b246afa | 2321 | struct super_block *sb = fs_info->sb; |
606686ee | 2322 | struct rcu_string *name; |
5da54bc1 | 2323 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
3c1dbdf5 | 2324 | u64 tmp; |
2b82032c | 2325 | int seeding_dev = 0; |
788f20eb | 2326 | int ret = 0; |
7132a262 | 2327 | bool unlocked = false; |
788f20eb | 2328 | |
5da54bc1 | 2329 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2330 | return -EROFS; |
788f20eb | 2331 | |
a5d16333 | 2332 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2333 | fs_info->bdev_holder); |
7f59203a JB |
2334 | if (IS_ERR(bdev)) |
2335 | return PTR_ERR(bdev); | |
a2135011 | 2336 | |
5da54bc1 | 2337 | if (fs_devices->seeding) { |
2b82032c YZ |
2338 | seeding_dev = 1; |
2339 | down_write(&sb->s_umount); | |
2340 | mutex_lock(&uuid_mutex); | |
2341 | } | |
2342 | ||
8c8bee1d | 2343 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2344 | |
5da54bc1 | 2345 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2346 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2347 | if (device->bdev == bdev) { |
2348 | ret = -EEXIST; | |
d25628bd | 2349 | mutex_unlock( |
5da54bc1 | 2350 | &fs_devices->device_list_mutex); |
2b82032c | 2351 | goto error; |
788f20eb CM |
2352 | } |
2353 | } | |
5da54bc1 | 2354 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2355 | |
0b246afa | 2356 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2357 | if (IS_ERR(device)) { |
788f20eb | 2358 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2359 | ret = PTR_ERR(device); |
2b82032c | 2360 | goto error; |
788f20eb CM |
2361 | } |
2362 | ||
78f2c9e6 | 2363 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2364 | if (!name) { |
2b82032c | 2365 | ret = -ENOMEM; |
5c4cf6c9 | 2366 | goto error_free_device; |
788f20eb | 2367 | } |
606686ee | 2368 | rcu_assign_pointer(device->name, name); |
2b82032c | 2369 | |
a22285a6 | 2370 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2371 | if (IS_ERR(trans)) { |
98d5dc13 | 2372 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2373 | goto error_free_device; |
98d5dc13 TI |
2374 | } |
2375 | ||
d5e2003c | 2376 | q = bdev_get_queue(bdev); |
ebbede42 | 2377 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2378 | device->generation = trans->transid; |
0b246afa JM |
2379 | device->io_width = fs_info->sectorsize; |
2380 | device->io_align = fs_info->sectorsize; | |
2381 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2382 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2383 | fs_info->sectorsize); | |
2cc3c559 | 2384 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2385 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2386 | device->fs_info = fs_info; |
788f20eb | 2387 | device->bdev = bdev; |
e12c9621 | 2388 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2389 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2390 | device->mode = FMODE_EXCL; |
27087f37 | 2391 | device->dev_stats_valid = 1; |
9f6d2510 | 2392 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2393 | |
2b82032c | 2394 | if (seeding_dev) { |
1751e8a6 | 2395 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2396 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2397 | if (ret) { |
2398 | btrfs_abort_transaction(trans, ret); | |
2399 | goto error_trans; | |
2400 | } | |
2b82032c | 2401 | } |
788f20eb | 2402 | |
5da54bc1 | 2403 | device->fs_devices = fs_devices; |
e5e9a520 | 2404 | |
5da54bc1 | 2405 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2406 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2407 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2408 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2409 | fs_devices->num_devices++; | |
2410 | fs_devices->open_devices++; | |
2411 | fs_devices->rw_devices++; | |
2412 | fs_devices->total_devices++; | |
2413 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2414 | |
a5ed45f8 | 2415 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2416 | |
e884f4f0 | 2417 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2418 | fs_devices->rotating = 1; |
c289811c | 2419 | |
0b246afa JM |
2420 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2421 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
7dfb8be1 | 2422 | round_down(tmp + device->total_bytes, fs_info->sectorsize)); |
788f20eb | 2423 | |
0b246afa JM |
2424 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2425 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2426 | |
2427 | /* add sysfs device entry */ | |
5da54bc1 | 2428 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2429 | |
2196d6e8 MX |
2430 | /* |
2431 | * we've got more storage, clear any full flags on the space | |
2432 | * infos | |
2433 | */ | |
0b246afa | 2434 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2435 | |
34441361 | 2436 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2437 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2438 | |
2b82032c | 2439 | if (seeding_dev) { |
34441361 | 2440 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2441 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2442 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2443 | if (ret) { |
66642832 | 2444 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2445 | goto error_sysfs; |
005d6427 | 2446 | } |
2196d6e8 MX |
2447 | } |
2448 | ||
8e87e856 | 2449 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2450 | if (ret) { |
66642832 | 2451 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2452 | goto error_sysfs; |
2196d6e8 MX |
2453 | } |
2454 | ||
2455 | if (seeding_dev) { | |
2456 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2457 | ||
0b246afa | 2458 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2459 | if (ret) { |
66642832 | 2460 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2461 | goto error_sysfs; |
005d6427 | 2462 | } |
b2373f25 AJ |
2463 | |
2464 | /* Sprouting would change fsid of the mounted root, | |
2465 | * so rename the fsid on the sysfs | |
2466 | */ | |
2467 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa | 2468 | fs_info->fsid); |
5da54bc1 | 2469 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2470 | btrfs_warn(fs_info, |
2471 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2472 | } |
2473 | ||
3a45bb20 | 2474 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2475 | |
2b82032c YZ |
2476 | if (seeding_dev) { |
2477 | mutex_unlock(&uuid_mutex); | |
2478 | up_write(&sb->s_umount); | |
7132a262 | 2479 | unlocked = true; |
788f20eb | 2480 | |
79787eaa JM |
2481 | if (ret) /* transaction commit */ |
2482 | return ret; | |
2483 | ||
2ff7e61e | 2484 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2485 | if (ret < 0) |
0b246afa | 2486 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2487 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2488 | trans = btrfs_attach_transaction(root); |
2489 | if (IS_ERR(trans)) { | |
2490 | if (PTR_ERR(trans) == -ENOENT) | |
2491 | return 0; | |
7132a262 AJ |
2492 | ret = PTR_ERR(trans); |
2493 | trans = NULL; | |
2494 | goto error_sysfs; | |
671415b7 | 2495 | } |
3a45bb20 | 2496 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2497 | } |
c9e9f97b | 2498 | |
5a1972bd QW |
2499 | /* Update ctime/mtime for libblkid */ |
2500 | update_dev_time(device_path); | |
2b82032c | 2501 | return ret; |
79787eaa | 2502 | |
d31c32f6 | 2503 | error_sysfs: |
5da54bc1 | 2504 | btrfs_sysfs_rm_device_link(fs_devices, device); |
79787eaa | 2505 | error_trans: |
0af2c4bf | 2506 | if (seeding_dev) |
1751e8a6 | 2507 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2508 | if (trans) |
2509 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2510 | error_free_device: |
a425f9d4 | 2511 | btrfs_free_device(device); |
2b82032c | 2512 | error: |
e525fd89 | 2513 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2514 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2515 | mutex_unlock(&uuid_mutex); |
2516 | up_write(&sb->s_umount); | |
2517 | } | |
c9e9f97b | 2518 | return ret; |
788f20eb CM |
2519 | } |
2520 | ||
d397712b CM |
2521 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2522 | struct btrfs_device *device) | |
0b86a832 CM |
2523 | { |
2524 | int ret; | |
2525 | struct btrfs_path *path; | |
0b246afa | 2526 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2527 | struct btrfs_dev_item *dev_item; |
2528 | struct extent_buffer *leaf; | |
2529 | struct btrfs_key key; | |
2530 | ||
0b86a832 CM |
2531 | path = btrfs_alloc_path(); |
2532 | if (!path) | |
2533 | return -ENOMEM; | |
2534 | ||
2535 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2536 | key.type = BTRFS_DEV_ITEM_KEY; | |
2537 | key.offset = device->devid; | |
2538 | ||
2539 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2540 | if (ret < 0) | |
2541 | goto out; | |
2542 | ||
2543 | if (ret > 0) { | |
2544 | ret = -ENOENT; | |
2545 | goto out; | |
2546 | } | |
2547 | ||
2548 | leaf = path->nodes[0]; | |
2549 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2550 | ||
2551 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2552 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2553 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2554 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2555 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2556 | btrfs_set_device_total_bytes(leaf, dev_item, |
2557 | btrfs_device_get_disk_total_bytes(device)); | |
2558 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2559 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2560 | btrfs_mark_buffer_dirty(leaf); |
2561 | ||
2562 | out: | |
2563 | btrfs_free_path(path); | |
2564 | return ret; | |
2565 | } | |
2566 | ||
2196d6e8 | 2567 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2568 | struct btrfs_device *device, u64 new_size) |
2569 | { | |
0b246afa JM |
2570 | struct btrfs_fs_info *fs_info = device->fs_info; |
2571 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2572 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2573 | u64 old_total; |
2574 | u64 diff; | |
8f18cf13 | 2575 | |
ebbede42 | 2576 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2577 | return -EACCES; |
2196d6e8 | 2578 | |
7dfb8be1 NB |
2579 | new_size = round_down(new_size, fs_info->sectorsize); |
2580 | ||
34441361 | 2581 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2582 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2583 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2584 | |
63a212ab | 2585 | if (new_size <= device->total_bytes || |
401e29c1 | 2586 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2587 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2588 | return -EINVAL; |
2196d6e8 | 2589 | } |
2b82032c | 2590 | |
0b246afa | 2591 | fs_devices = fs_info->fs_devices; |
2b82032c | 2592 | |
7dfb8be1 NB |
2593 | btrfs_set_super_total_bytes(super_copy, |
2594 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2595 | device->fs_devices->total_rw_bytes += diff; |
2596 | ||
7cc8e58d MX |
2597 | btrfs_device_set_total_bytes(device, new_size); |
2598 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2599 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2600 | if (list_empty(&device->resized_list)) |
2601 | list_add_tail(&device->resized_list, | |
2602 | &fs_devices->resized_devices); | |
34441361 | 2603 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2604 | |
8f18cf13 CM |
2605 | return btrfs_update_device(trans, device); |
2606 | } | |
2607 | ||
f4208794 | 2608 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2609 | { |
f4208794 | 2610 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2611 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2612 | int ret; |
2613 | struct btrfs_path *path; | |
2614 | struct btrfs_key key; | |
2615 | ||
8f18cf13 CM |
2616 | path = btrfs_alloc_path(); |
2617 | if (!path) | |
2618 | return -ENOMEM; | |
2619 | ||
408fbf19 | 2620 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2621 | key.offset = chunk_offset; |
2622 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2623 | ||
2624 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2625 | if (ret < 0) |
2626 | goto out; | |
2627 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2628 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2629 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2630 | ret = -ENOENT; |
2631 | goto out; | |
2632 | } | |
8f18cf13 CM |
2633 | |
2634 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2635 | if (ret < 0) |
0b246afa JM |
2636 | btrfs_handle_fs_error(fs_info, ret, |
2637 | "Failed to delete chunk item."); | |
79787eaa | 2638 | out: |
8f18cf13 | 2639 | btrfs_free_path(path); |
65a246c5 | 2640 | return ret; |
8f18cf13 CM |
2641 | } |
2642 | ||
408fbf19 | 2643 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2644 | { |
0b246afa | 2645 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2646 | struct btrfs_disk_key *disk_key; |
2647 | struct btrfs_chunk *chunk; | |
2648 | u8 *ptr; | |
2649 | int ret = 0; | |
2650 | u32 num_stripes; | |
2651 | u32 array_size; | |
2652 | u32 len = 0; | |
2653 | u32 cur; | |
2654 | struct btrfs_key key; | |
2655 | ||
34441361 | 2656 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2657 | array_size = btrfs_super_sys_array_size(super_copy); |
2658 | ||
2659 | ptr = super_copy->sys_chunk_array; | |
2660 | cur = 0; | |
2661 | ||
2662 | while (cur < array_size) { | |
2663 | disk_key = (struct btrfs_disk_key *)ptr; | |
2664 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2665 | ||
2666 | len = sizeof(*disk_key); | |
2667 | ||
2668 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2669 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2670 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2671 | len += btrfs_chunk_item_size(num_stripes); | |
2672 | } else { | |
2673 | ret = -EIO; | |
2674 | break; | |
2675 | } | |
408fbf19 | 2676 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2677 | key.offset == chunk_offset) { |
2678 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2679 | array_size -= len; | |
2680 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2681 | } else { | |
2682 | ptr += len; | |
2683 | cur += len; | |
2684 | } | |
2685 | } | |
34441361 | 2686 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2687 | return ret; |
2688 | } | |
2689 | ||
592d92ee LB |
2690 | static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info, |
2691 | u64 logical, u64 length) | |
2692 | { | |
2693 | struct extent_map_tree *em_tree; | |
2694 | struct extent_map *em; | |
2695 | ||
2696 | em_tree = &fs_info->mapping_tree.map_tree; | |
2697 | read_lock(&em_tree->lock); | |
2698 | em = lookup_extent_mapping(em_tree, logical, length); | |
2699 | read_unlock(&em_tree->lock); | |
2700 | ||
2701 | if (!em) { | |
2702 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2703 | logical, length); | |
2704 | return ERR_PTR(-EINVAL); | |
2705 | } | |
2706 | ||
2707 | if (em->start > logical || em->start + em->len < logical) { | |
2708 | btrfs_crit(fs_info, | |
2709 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2710 | logical, length, em->start, em->start + em->len); | |
2711 | free_extent_map(em); | |
2712 | return ERR_PTR(-EINVAL); | |
2713 | } | |
2714 | ||
2715 | /* callers are responsible for dropping em's ref. */ | |
2716 | return em; | |
2717 | } | |
2718 | ||
97aff912 | 2719 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2720 | { |
97aff912 | 2721 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2722 | struct extent_map *em; |
2723 | struct map_lookup *map; | |
2196d6e8 | 2724 | u64 dev_extent_len = 0; |
47ab2a6c | 2725 | int i, ret = 0; |
0b246afa | 2726 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2727 | |
592d92ee LB |
2728 | em = get_chunk_map(fs_info, chunk_offset, 1); |
2729 | if (IS_ERR(em)) { | |
47ab2a6c JB |
2730 | /* |
2731 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2732 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2733 | * do anything we still error out. |
2734 | */ | |
2735 | ASSERT(0); | |
592d92ee | 2736 | return PTR_ERR(em); |
47ab2a6c | 2737 | } |
95617d69 | 2738 | map = em->map_lookup; |
34441361 | 2739 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2740 | check_system_chunk(trans, map->type); |
34441361 | 2741 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2742 | |
57ba4cb8 FM |
2743 | /* |
2744 | * Take the device list mutex to prevent races with the final phase of | |
2745 | * a device replace operation that replaces the device object associated | |
2746 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2747 | */ | |
2748 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2749 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2750 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2751 | ret = btrfs_free_dev_extent(trans, device, |
2752 | map->stripes[i].physical, | |
2753 | &dev_extent_len); | |
47ab2a6c | 2754 | if (ret) { |
57ba4cb8 | 2755 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2756 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2757 | goto out; |
2758 | } | |
a061fc8d | 2759 | |
2196d6e8 | 2760 | if (device->bytes_used > 0) { |
34441361 | 2761 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2762 | btrfs_device_set_bytes_used(device, |
2763 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2764 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2765 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2766 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2767 | } |
a061fc8d | 2768 | |
dfe25020 CM |
2769 | if (map->stripes[i].dev) { |
2770 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2771 | if (ret) { |
57ba4cb8 | 2772 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2773 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2774 | goto out; |
2775 | } | |
dfe25020 | 2776 | } |
8f18cf13 | 2777 | } |
57ba4cb8 FM |
2778 | mutex_unlock(&fs_devices->device_list_mutex); |
2779 | ||
f4208794 | 2780 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 2781 | if (ret) { |
66642832 | 2782 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2783 | goto out; |
2784 | } | |
8f18cf13 | 2785 | |
6bccf3ab | 2786 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2787 | |
8f18cf13 | 2788 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2789 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2790 | if (ret) { |
66642832 | 2791 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2792 | goto out; |
2793 | } | |
8f18cf13 CM |
2794 | } |
2795 | ||
5a98ec01 | 2796 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 2797 | if (ret) { |
66642832 | 2798 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2799 | goto out; |
2800 | } | |
2b82032c | 2801 | |
47ab2a6c | 2802 | out: |
2b82032c YZ |
2803 | /* once for us */ |
2804 | free_extent_map(em); | |
47ab2a6c JB |
2805 | return ret; |
2806 | } | |
2b82032c | 2807 | |
5b4aacef | 2808 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2809 | { |
5b4aacef | 2810 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2811 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2812 | int ret; |
2b82032c | 2813 | |
67c5e7d4 FM |
2814 | /* |
2815 | * Prevent races with automatic removal of unused block groups. | |
2816 | * After we relocate and before we remove the chunk with offset | |
2817 | * chunk_offset, automatic removal of the block group can kick in, | |
2818 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2819 | * | |
2820 | * Make sure to acquire this mutex before doing a tree search (dev | |
2821 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2822 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2823 | * we release the path used to search the chunk/dev tree and before | |
2824 | * the current task acquires this mutex and calls us. | |
2825 | */ | |
a32bf9a3 | 2826 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 2827 | |
0b246afa | 2828 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2829 | if (ret) |
2830 | return -ENOSPC; | |
2831 | ||
2832 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 2833 | btrfs_scrub_pause(fs_info); |
0b246afa | 2834 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2835 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2836 | if (ret) |
2837 | return ret; | |
2838 | ||
75cb379d JM |
2839 | /* |
2840 | * We add the kobjects here (and after forcing data chunk creation) | |
2841 | * since relocation is the only place we'll create chunks of a new | |
2842 | * type at runtime. The only place where we'll remove the last | |
2843 | * chunk of a type is the call immediately below this one. Even | |
2844 | * so, we're protected against races with the cleaner thread since | |
2845 | * we're covered by the delete_unused_bgs_mutex. | |
2846 | */ | |
2847 | btrfs_add_raid_kobjects(fs_info); | |
2848 | ||
19c4d2f9 CM |
2849 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2850 | chunk_offset); | |
2851 | if (IS_ERR(trans)) { | |
2852 | ret = PTR_ERR(trans); | |
2853 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2854 | return ret; | |
2855 | } | |
2856 | ||
47ab2a6c | 2857 | /* |
19c4d2f9 CM |
2858 | * step two, delete the device extents and the |
2859 | * chunk tree entries | |
47ab2a6c | 2860 | */ |
97aff912 | 2861 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 2862 | btrfs_end_transaction(trans); |
19c4d2f9 | 2863 | return ret; |
2b82032c YZ |
2864 | } |
2865 | ||
2ff7e61e | 2866 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2867 | { |
0b246afa | 2868 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2869 | struct btrfs_path *path; |
2870 | struct extent_buffer *leaf; | |
2871 | struct btrfs_chunk *chunk; | |
2872 | struct btrfs_key key; | |
2873 | struct btrfs_key found_key; | |
2b82032c | 2874 | u64 chunk_type; |
ba1bf481 JB |
2875 | bool retried = false; |
2876 | int failed = 0; | |
2b82032c YZ |
2877 | int ret; |
2878 | ||
2879 | path = btrfs_alloc_path(); | |
2880 | if (!path) | |
2881 | return -ENOMEM; | |
2882 | ||
ba1bf481 | 2883 | again: |
2b82032c YZ |
2884 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2885 | key.offset = (u64)-1; | |
2886 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2887 | ||
2888 | while (1) { | |
0b246afa | 2889 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2890 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2891 | if (ret < 0) { |
0b246afa | 2892 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2893 | goto error; |
67c5e7d4 | 2894 | } |
79787eaa | 2895 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2896 | |
2897 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2898 | key.type); | |
67c5e7d4 | 2899 | if (ret) |
0b246afa | 2900 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
2901 | if (ret < 0) |
2902 | goto error; | |
2903 | if (ret > 0) | |
2904 | break; | |
1a40e23b | 2905 | |
2b82032c YZ |
2906 | leaf = path->nodes[0]; |
2907 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2908 | |
2b82032c YZ |
2909 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2910 | struct btrfs_chunk); | |
2911 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2912 | btrfs_release_path(path); |
8f18cf13 | 2913 | |
2b82032c | 2914 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 2915 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
2916 | if (ret == -ENOSPC) |
2917 | failed++; | |
14586651 HS |
2918 | else |
2919 | BUG_ON(ret); | |
2b82032c | 2920 | } |
0b246afa | 2921 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 2922 | |
2b82032c YZ |
2923 | if (found_key.offset == 0) |
2924 | break; | |
2925 | key.offset = found_key.offset - 1; | |
2926 | } | |
2927 | ret = 0; | |
ba1bf481 JB |
2928 | if (failed && !retried) { |
2929 | failed = 0; | |
2930 | retried = true; | |
2931 | goto again; | |
fae7f21c | 2932 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2933 | ret = -ENOSPC; |
2934 | } | |
2b82032c YZ |
2935 | error: |
2936 | btrfs_free_path(path); | |
2937 | return ret; | |
8f18cf13 CM |
2938 | } |
2939 | ||
a6f93c71 LB |
2940 | /* |
2941 | * return 1 : allocate a data chunk successfully, | |
2942 | * return <0: errors during allocating a data chunk, | |
2943 | * return 0 : no need to allocate a data chunk. | |
2944 | */ | |
2945 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
2946 | u64 chunk_offset) | |
2947 | { | |
2948 | struct btrfs_block_group_cache *cache; | |
2949 | u64 bytes_used; | |
2950 | u64 chunk_type; | |
2951 | ||
2952 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2953 | ASSERT(cache); | |
2954 | chunk_type = cache->flags; | |
2955 | btrfs_put_block_group(cache); | |
2956 | ||
2957 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
2958 | spin_lock(&fs_info->data_sinfo->lock); | |
2959 | bytes_used = fs_info->data_sinfo->bytes_used; | |
2960 | spin_unlock(&fs_info->data_sinfo->lock); | |
2961 | ||
2962 | if (!bytes_used) { | |
2963 | struct btrfs_trans_handle *trans; | |
2964 | int ret; | |
2965 | ||
2966 | trans = btrfs_join_transaction(fs_info->tree_root); | |
2967 | if (IS_ERR(trans)) | |
2968 | return PTR_ERR(trans); | |
2969 | ||
43a7e99d | 2970 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
2971 | BTRFS_BLOCK_GROUP_DATA); |
2972 | btrfs_end_transaction(trans); | |
2973 | if (ret < 0) | |
2974 | return ret; | |
2975 | ||
75cb379d JM |
2976 | btrfs_add_raid_kobjects(fs_info); |
2977 | ||
a6f93c71 LB |
2978 | return 1; |
2979 | } | |
2980 | } | |
2981 | return 0; | |
2982 | } | |
2983 | ||
6bccf3ab | 2984 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
2985 | struct btrfs_balance_control *bctl) |
2986 | { | |
6bccf3ab | 2987 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
2988 | struct btrfs_trans_handle *trans; |
2989 | struct btrfs_balance_item *item; | |
2990 | struct btrfs_disk_balance_args disk_bargs; | |
2991 | struct btrfs_path *path; | |
2992 | struct extent_buffer *leaf; | |
2993 | struct btrfs_key key; | |
2994 | int ret, err; | |
2995 | ||
2996 | path = btrfs_alloc_path(); | |
2997 | if (!path) | |
2998 | return -ENOMEM; | |
2999 | ||
3000 | trans = btrfs_start_transaction(root, 0); | |
3001 | if (IS_ERR(trans)) { | |
3002 | btrfs_free_path(path); | |
3003 | return PTR_ERR(trans); | |
3004 | } | |
3005 | ||
3006 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3007 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3008 | key.offset = 0; |
3009 | ||
3010 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3011 | sizeof(*item)); | |
3012 | if (ret) | |
3013 | goto out; | |
3014 | ||
3015 | leaf = path->nodes[0]; | |
3016 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3017 | ||
b159fa28 | 3018 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3019 | |
3020 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3021 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3022 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3023 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3024 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3025 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3026 | ||
3027 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3028 | ||
3029 | btrfs_mark_buffer_dirty(leaf); | |
3030 | out: | |
3031 | btrfs_free_path(path); | |
3a45bb20 | 3032 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3033 | if (err && !ret) |
3034 | ret = err; | |
3035 | return ret; | |
3036 | } | |
3037 | ||
6bccf3ab | 3038 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3039 | { |
6bccf3ab | 3040 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3041 | struct btrfs_trans_handle *trans; |
3042 | struct btrfs_path *path; | |
3043 | struct btrfs_key key; | |
3044 | int ret, err; | |
3045 | ||
3046 | path = btrfs_alloc_path(); | |
3047 | if (!path) | |
3048 | return -ENOMEM; | |
3049 | ||
3050 | trans = btrfs_start_transaction(root, 0); | |
3051 | if (IS_ERR(trans)) { | |
3052 | btrfs_free_path(path); | |
3053 | return PTR_ERR(trans); | |
3054 | } | |
3055 | ||
3056 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3057 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3058 | key.offset = 0; |
3059 | ||
3060 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3061 | if (ret < 0) | |
3062 | goto out; | |
3063 | if (ret > 0) { | |
3064 | ret = -ENOENT; | |
3065 | goto out; | |
3066 | } | |
3067 | ||
3068 | ret = btrfs_del_item(trans, root, path); | |
3069 | out: | |
3070 | btrfs_free_path(path); | |
3a45bb20 | 3071 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3072 | if (err && !ret) |
3073 | ret = err; | |
3074 | return ret; | |
3075 | } | |
3076 | ||
59641015 ID |
3077 | /* |
3078 | * This is a heuristic used to reduce the number of chunks balanced on | |
3079 | * resume after balance was interrupted. | |
3080 | */ | |
3081 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3082 | { | |
3083 | /* | |
3084 | * Turn on soft mode for chunk types that were being converted. | |
3085 | */ | |
3086 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3087 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3088 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3089 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3090 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3091 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3092 | ||
3093 | /* | |
3094 | * Turn on usage filter if is not already used. The idea is | |
3095 | * that chunks that we have already balanced should be | |
3096 | * reasonably full. Don't do it for chunks that are being | |
3097 | * converted - that will keep us from relocating unconverted | |
3098 | * (albeit full) chunks. | |
3099 | */ | |
3100 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3101 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3102 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3103 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3104 | bctl->data.usage = 90; | |
3105 | } | |
3106 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3107 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3108 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3109 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3110 | bctl->sys.usage = 90; | |
3111 | } | |
3112 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3113 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3114 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3115 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3116 | bctl->meta.usage = 90; | |
3117 | } | |
3118 | } | |
3119 | ||
149196a2 DS |
3120 | /* |
3121 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3122 | */ | |
3123 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3124 | { |
3125 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3126 | int ret; |
c9e9f97b ID |
3127 | |
3128 | BUG_ON(!fs_info->balance_ctl); | |
3129 | ||
3130 | spin_lock(&fs_info->balance_lock); | |
3131 | fs_info->balance_ctl = NULL; | |
3132 | spin_unlock(&fs_info->balance_lock); | |
3133 | ||
3134 | kfree(bctl); | |
149196a2 DS |
3135 | ret = del_balance_item(fs_info); |
3136 | if (ret) | |
3137 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3138 | } |
3139 | ||
ed25e9b2 ID |
3140 | /* |
3141 | * Balance filters. Return 1 if chunk should be filtered out | |
3142 | * (should not be balanced). | |
3143 | */ | |
899c81ea | 3144 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3145 | struct btrfs_balance_args *bargs) |
3146 | { | |
899c81ea ID |
3147 | chunk_type = chunk_to_extended(chunk_type) & |
3148 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3149 | |
899c81ea | 3150 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3151 | return 0; |
3152 | ||
3153 | return 1; | |
3154 | } | |
3155 | ||
dba72cb3 | 3156 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3157 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3158 | { |
3159 | struct btrfs_block_group_cache *cache; | |
3160 | u64 chunk_used; | |
3161 | u64 user_thresh_min; | |
3162 | u64 user_thresh_max; | |
3163 | int ret = 1; | |
3164 | ||
3165 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3166 | chunk_used = btrfs_block_group_used(&cache->item); | |
3167 | ||
3168 | if (bargs->usage_min == 0) | |
3169 | user_thresh_min = 0; | |
3170 | else | |
3171 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3172 | bargs->usage_min); | |
3173 | ||
3174 | if (bargs->usage_max == 0) | |
3175 | user_thresh_max = 1; | |
3176 | else if (bargs->usage_max > 100) | |
3177 | user_thresh_max = cache->key.offset; | |
3178 | else | |
3179 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3180 | bargs->usage_max); | |
3181 | ||
3182 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3183 | ret = 0; | |
3184 | ||
3185 | btrfs_put_block_group(cache); | |
3186 | return ret; | |
3187 | } | |
3188 | ||
dba72cb3 | 3189 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3190 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3191 | { |
3192 | struct btrfs_block_group_cache *cache; | |
3193 | u64 chunk_used, user_thresh; | |
3194 | int ret = 1; | |
3195 | ||
3196 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3197 | chunk_used = btrfs_block_group_used(&cache->item); | |
3198 | ||
bc309467 | 3199 | if (bargs->usage_min == 0) |
3e39cea6 | 3200 | user_thresh = 1; |
a105bb88 ID |
3201 | else if (bargs->usage > 100) |
3202 | user_thresh = cache->key.offset; | |
3203 | else | |
3204 | user_thresh = div_factor_fine(cache->key.offset, | |
3205 | bargs->usage); | |
3206 | ||
5ce5b3c0 ID |
3207 | if (chunk_used < user_thresh) |
3208 | ret = 0; | |
3209 | ||
3210 | btrfs_put_block_group(cache); | |
3211 | return ret; | |
3212 | } | |
3213 | ||
409d404b ID |
3214 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3215 | struct btrfs_chunk *chunk, | |
3216 | struct btrfs_balance_args *bargs) | |
3217 | { | |
3218 | struct btrfs_stripe *stripe; | |
3219 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3220 | int i; | |
3221 | ||
3222 | for (i = 0; i < num_stripes; i++) { | |
3223 | stripe = btrfs_stripe_nr(chunk, i); | |
3224 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3225 | return 0; | |
3226 | } | |
3227 | ||
3228 | return 1; | |
3229 | } | |
3230 | ||
94e60d5a ID |
3231 | /* [pstart, pend) */ |
3232 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3233 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3234 | struct btrfs_balance_args *bargs) |
3235 | { | |
3236 | struct btrfs_stripe *stripe; | |
3237 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3238 | u64 stripe_offset; | |
3239 | u64 stripe_length; | |
3240 | int factor; | |
3241 | int i; | |
3242 | ||
3243 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3244 | return 0; | |
3245 | ||
3246 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3247 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3248 | factor = num_stripes / 2; | |
3249 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3250 | factor = num_stripes - 1; | |
3251 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3252 | factor = num_stripes - 2; | |
3253 | } else { | |
3254 | factor = num_stripes; | |
3255 | } | |
94e60d5a ID |
3256 | |
3257 | for (i = 0; i < num_stripes; i++) { | |
3258 | stripe = btrfs_stripe_nr(chunk, i); | |
3259 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3260 | continue; | |
3261 | ||
3262 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3263 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3264 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3265 | |
3266 | if (stripe_offset < bargs->pend && | |
3267 | stripe_offset + stripe_length > bargs->pstart) | |
3268 | return 0; | |
3269 | } | |
3270 | ||
3271 | return 1; | |
3272 | } | |
3273 | ||
ea67176a ID |
3274 | /* [vstart, vend) */ |
3275 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3276 | struct btrfs_chunk *chunk, | |
3277 | u64 chunk_offset, | |
3278 | struct btrfs_balance_args *bargs) | |
3279 | { | |
3280 | if (chunk_offset < bargs->vend && | |
3281 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3282 | /* at least part of the chunk is inside this vrange */ | |
3283 | return 0; | |
3284 | ||
3285 | return 1; | |
3286 | } | |
3287 | ||
dee32d0a GAP |
3288 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3289 | struct btrfs_chunk *chunk, | |
3290 | struct btrfs_balance_args *bargs) | |
3291 | { | |
3292 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3293 | ||
3294 | if (bargs->stripes_min <= num_stripes | |
3295 | && num_stripes <= bargs->stripes_max) | |
3296 | return 0; | |
3297 | ||
3298 | return 1; | |
3299 | } | |
3300 | ||
899c81ea | 3301 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3302 | struct btrfs_balance_args *bargs) |
3303 | { | |
3304 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3305 | return 0; | |
3306 | ||
899c81ea ID |
3307 | chunk_type = chunk_to_extended(chunk_type) & |
3308 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3309 | |
899c81ea | 3310 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3311 | return 1; |
3312 | ||
3313 | return 0; | |
3314 | } | |
3315 | ||
2ff7e61e | 3316 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3317 | struct extent_buffer *leaf, |
3318 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3319 | { | |
0b246afa | 3320 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3321 | struct btrfs_balance_args *bargs = NULL; |
3322 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3323 | ||
3324 | /* type filter */ | |
3325 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3326 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3327 | return 0; | |
3328 | } | |
3329 | ||
3330 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3331 | bargs = &bctl->data; | |
3332 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3333 | bargs = &bctl->sys; | |
3334 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3335 | bargs = &bctl->meta; | |
3336 | ||
ed25e9b2 ID |
3337 | /* profiles filter */ |
3338 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3339 | chunk_profiles_filter(chunk_type, bargs)) { | |
3340 | return 0; | |
5ce5b3c0 ID |
3341 | } |
3342 | ||
3343 | /* usage filter */ | |
3344 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3345 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3346 | return 0; |
bc309467 | 3347 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3348 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3349 | return 0; |
409d404b ID |
3350 | } |
3351 | ||
3352 | /* devid filter */ | |
3353 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3354 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3355 | return 0; | |
94e60d5a ID |
3356 | } |
3357 | ||
3358 | /* drange filter, makes sense only with devid filter */ | |
3359 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3360 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3361 | return 0; |
ea67176a ID |
3362 | } |
3363 | ||
3364 | /* vrange filter */ | |
3365 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3366 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3367 | return 0; | |
ed25e9b2 ID |
3368 | } |
3369 | ||
dee32d0a GAP |
3370 | /* stripes filter */ |
3371 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3372 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3373 | return 0; | |
3374 | } | |
3375 | ||
cfa4c961 ID |
3376 | /* soft profile changing mode */ |
3377 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3378 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3379 | return 0; | |
3380 | } | |
3381 | ||
7d824b6f DS |
3382 | /* |
3383 | * limited by count, must be the last filter | |
3384 | */ | |
3385 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3386 | if (bargs->limit == 0) | |
3387 | return 0; | |
3388 | else | |
3389 | bargs->limit--; | |
12907fc7 DS |
3390 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3391 | /* | |
3392 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3393 | * determined here because we do not have the global information |
12907fc7 DS |
3394 | * about the count of all chunks that satisfy the filters. |
3395 | */ | |
3396 | if (bargs->limit_max == 0) | |
3397 | return 0; | |
3398 | else | |
3399 | bargs->limit_max--; | |
7d824b6f DS |
3400 | } |
3401 | ||
f43ffb60 ID |
3402 | return 1; |
3403 | } | |
3404 | ||
c9e9f97b | 3405 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3406 | { |
19a39dce | 3407 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3408 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3409 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3410 | struct list_head *devices; | |
ec44a35c CM |
3411 | struct btrfs_device *device; |
3412 | u64 old_size; | |
3413 | u64 size_to_free; | |
12907fc7 | 3414 | u64 chunk_type; |
f43ffb60 | 3415 | struct btrfs_chunk *chunk; |
5a488b9d | 3416 | struct btrfs_path *path = NULL; |
ec44a35c | 3417 | struct btrfs_key key; |
ec44a35c | 3418 | struct btrfs_key found_key; |
c9e9f97b | 3419 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3420 | struct extent_buffer *leaf; |
3421 | int slot; | |
c9e9f97b ID |
3422 | int ret; |
3423 | int enospc_errors = 0; | |
19a39dce | 3424 | bool counting = true; |
12907fc7 | 3425 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3426 | u64 limit_data = bctl->data.limit; |
3427 | u64 limit_meta = bctl->meta.limit; | |
3428 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3429 | u32 count_data = 0; |
3430 | u32 count_meta = 0; | |
3431 | u32 count_sys = 0; | |
2c9fe835 | 3432 | int chunk_reserved = 0; |
ec44a35c | 3433 | |
ec44a35c | 3434 | /* step one make some room on all the devices */ |
c9e9f97b | 3435 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3436 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3437 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3438 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3439 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3440 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3441 | btrfs_device_get_total_bytes(device) - |
3442 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3443 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3444 | continue; |
3445 | ||
3446 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3447 | if (ret == -ENOSPC) |
3448 | break; | |
5a488b9d LB |
3449 | if (ret) { |
3450 | /* btrfs_shrink_device never returns ret > 0 */ | |
3451 | WARN_ON(ret > 0); | |
3452 | goto error; | |
3453 | } | |
ec44a35c | 3454 | |
a22285a6 | 3455 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3456 | if (IS_ERR(trans)) { |
3457 | ret = PTR_ERR(trans); | |
3458 | btrfs_info_in_rcu(fs_info, | |
3459 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3460 | rcu_str_deref(device->name), ret, | |
3461 | old_size, old_size - size_to_free); | |
3462 | goto error; | |
3463 | } | |
ec44a35c CM |
3464 | |
3465 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3466 | if (ret) { |
3a45bb20 | 3467 | btrfs_end_transaction(trans); |
5a488b9d LB |
3468 | /* btrfs_grow_device never returns ret > 0 */ |
3469 | WARN_ON(ret > 0); | |
3470 | btrfs_info_in_rcu(fs_info, | |
3471 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3472 | rcu_str_deref(device->name), ret, | |
3473 | old_size, old_size - size_to_free); | |
3474 | goto error; | |
3475 | } | |
ec44a35c | 3476 | |
3a45bb20 | 3477 | btrfs_end_transaction(trans); |
ec44a35c CM |
3478 | } |
3479 | ||
3480 | /* step two, relocate all the chunks */ | |
3481 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3482 | if (!path) { |
3483 | ret = -ENOMEM; | |
3484 | goto error; | |
3485 | } | |
19a39dce ID |
3486 | |
3487 | /* zero out stat counters */ | |
3488 | spin_lock(&fs_info->balance_lock); | |
3489 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3490 | spin_unlock(&fs_info->balance_lock); | |
3491 | again: | |
7d824b6f | 3492 | if (!counting) { |
12907fc7 DS |
3493 | /* |
3494 | * The single value limit and min/max limits use the same bytes | |
3495 | * in the | |
3496 | */ | |
7d824b6f DS |
3497 | bctl->data.limit = limit_data; |
3498 | bctl->meta.limit = limit_meta; | |
3499 | bctl->sys.limit = limit_sys; | |
3500 | } | |
ec44a35c CM |
3501 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3502 | key.offset = (u64)-1; | |
3503 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3504 | ||
d397712b | 3505 | while (1) { |
19a39dce | 3506 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3507 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3508 | ret = -ECANCELED; |
3509 | goto error; | |
3510 | } | |
3511 | ||
67c5e7d4 | 3512 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3513 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3514 | if (ret < 0) { |
3515 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3516 | goto error; |
67c5e7d4 | 3517 | } |
ec44a35c CM |
3518 | |
3519 | /* | |
3520 | * this shouldn't happen, it means the last relocate | |
3521 | * failed | |
3522 | */ | |
3523 | if (ret == 0) | |
c9e9f97b | 3524 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3525 | |
3526 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3527 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3528 | if (ret) { |
67c5e7d4 | 3529 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3530 | ret = 0; |
ec44a35c | 3531 | break; |
c9e9f97b | 3532 | } |
7d9eb12c | 3533 | |
f43ffb60 ID |
3534 | leaf = path->nodes[0]; |
3535 | slot = path->slots[0]; | |
3536 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3537 | |
67c5e7d4 FM |
3538 | if (found_key.objectid != key.objectid) { |
3539 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3540 | break; |
67c5e7d4 | 3541 | } |
7d9eb12c | 3542 | |
f43ffb60 | 3543 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3544 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3545 | |
19a39dce ID |
3546 | if (!counting) { |
3547 | spin_lock(&fs_info->balance_lock); | |
3548 | bctl->stat.considered++; | |
3549 | spin_unlock(&fs_info->balance_lock); | |
3550 | } | |
3551 | ||
2ff7e61e | 3552 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3553 | found_key.offset); |
2c9fe835 | 3554 | |
b3b4aa74 | 3555 | btrfs_release_path(path); |
67c5e7d4 FM |
3556 | if (!ret) { |
3557 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3558 | goto loop; |
67c5e7d4 | 3559 | } |
f43ffb60 | 3560 | |
19a39dce | 3561 | if (counting) { |
67c5e7d4 | 3562 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3563 | spin_lock(&fs_info->balance_lock); |
3564 | bctl->stat.expected++; | |
3565 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3566 | |
3567 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3568 | count_data++; | |
3569 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3570 | count_sys++; | |
3571 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3572 | count_meta++; | |
3573 | ||
3574 | goto loop; | |
3575 | } | |
3576 | ||
3577 | /* | |
3578 | * Apply limit_min filter, no need to check if the LIMITS | |
3579 | * filter is used, limit_min is 0 by default | |
3580 | */ | |
3581 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3582 | count_data < bctl->data.limit_min) | |
3583 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3584 | count_meta < bctl->meta.limit_min) | |
3585 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3586 | count_sys < bctl->sys.limit_min)) { | |
3587 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3588 | goto loop; |
3589 | } | |
3590 | ||
a6f93c71 LB |
3591 | if (!chunk_reserved) { |
3592 | /* | |
3593 | * We may be relocating the only data chunk we have, | |
3594 | * which could potentially end up with losing data's | |
3595 | * raid profile, so lets allocate an empty one in | |
3596 | * advance. | |
3597 | */ | |
3598 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3599 | found_key.offset); | |
2c9fe835 ZL |
3600 | if (ret < 0) { |
3601 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3602 | goto error; | |
a6f93c71 LB |
3603 | } else if (ret == 1) { |
3604 | chunk_reserved = 1; | |
2c9fe835 | 3605 | } |
2c9fe835 ZL |
3606 | } |
3607 | ||
5b4aacef | 3608 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3609 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3610 | if (ret && ret != -ENOSPC) |
3611 | goto error; | |
19a39dce | 3612 | if (ret == -ENOSPC) { |
c9e9f97b | 3613 | enospc_errors++; |
19a39dce ID |
3614 | } else { |
3615 | spin_lock(&fs_info->balance_lock); | |
3616 | bctl->stat.completed++; | |
3617 | spin_unlock(&fs_info->balance_lock); | |
3618 | } | |
f43ffb60 | 3619 | loop: |
795a3321 ID |
3620 | if (found_key.offset == 0) |
3621 | break; | |
ba1bf481 | 3622 | key.offset = found_key.offset - 1; |
ec44a35c | 3623 | } |
c9e9f97b | 3624 | |
19a39dce ID |
3625 | if (counting) { |
3626 | btrfs_release_path(path); | |
3627 | counting = false; | |
3628 | goto again; | |
3629 | } | |
ec44a35c CM |
3630 | error: |
3631 | btrfs_free_path(path); | |
c9e9f97b | 3632 | if (enospc_errors) { |
efe120a0 | 3633 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3634 | enospc_errors); |
c9e9f97b ID |
3635 | if (!ret) |
3636 | ret = -ENOSPC; | |
3637 | } | |
3638 | ||
ec44a35c CM |
3639 | return ret; |
3640 | } | |
3641 | ||
0c460c0d ID |
3642 | /** |
3643 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3644 | * @flags: profile to validate | |
3645 | * @extended: if true @flags is treated as an extended profile | |
3646 | */ | |
3647 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3648 | { | |
3649 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3650 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3651 | ||
3652 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3653 | ||
3654 | /* 1) check that all other bits are zeroed */ | |
3655 | if (flags & ~mask) | |
3656 | return 0; | |
3657 | ||
3658 | /* 2) see if profile is reduced */ | |
3659 | if (flags == 0) | |
3660 | return !extended; /* "0" is valid for usual profiles */ | |
3661 | ||
3662 | /* true if exactly one bit set */ | |
3663 | return (flags & (flags - 1)) == 0; | |
3664 | } | |
3665 | ||
837d5b6e ID |
3666 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3667 | { | |
a7e99c69 ID |
3668 | /* cancel requested || normal exit path */ |
3669 | return atomic_read(&fs_info->balance_cancel_req) || | |
3670 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3671 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3672 | } |
3673 | ||
bdcd3c97 AM |
3674 | /* Non-zero return value signifies invalidity */ |
3675 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3676 | u64 allowed) | |
3677 | { | |
3678 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3679 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3680 | (bctl_arg->target & ~allowed))); | |
3681 | } | |
3682 | ||
c9e9f97b | 3683 | /* |
dccdb07b | 3684 | * Should be called with balance mutexe held |
c9e9f97b | 3685 | */ |
6fcf6e2b DS |
3686 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
3687 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
3688 | struct btrfs_ioctl_balance_args *bargs) |
3689 | { | |
14506127 | 3690 | u64 meta_target, data_target; |
f43ffb60 | 3691 | u64 allowed; |
e4837f8f | 3692 | int mixed = 0; |
c9e9f97b | 3693 | int ret; |
8dabb742 | 3694 | u64 num_devices; |
de98ced9 | 3695 | unsigned seq; |
c9e9f97b | 3696 | |
837d5b6e | 3697 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3698 | atomic_read(&fs_info->balance_pause_req) || |
3699 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3700 | ret = -EINVAL; |
3701 | goto out; | |
3702 | } | |
3703 | ||
e4837f8f ID |
3704 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3705 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3706 | mixed = 1; | |
3707 | ||
f43ffb60 ID |
3708 | /* |
3709 | * In case of mixed groups both data and meta should be picked, | |
3710 | * and identical options should be given for both of them. | |
3711 | */ | |
e4837f8f ID |
3712 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3713 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3714 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3715 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3716 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 3717 | btrfs_err(fs_info, |
6dac13f8 | 3718 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
3719 | ret = -EINVAL; |
3720 | goto out; | |
3721 | } | |
3722 | } | |
3723 | ||
8dabb742 | 3724 | num_devices = fs_info->fs_devices->num_devices; |
7e79cb86 | 3725 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
8dabb742 SB |
3726 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3727 | BUG_ON(num_devices < 1); | |
3728 | num_devices--; | |
3729 | } | |
7e79cb86 | 3730 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
88be159c AH |
3731 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3732 | if (num_devices > 1) | |
e4d8ec0f | 3733 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3734 | if (num_devices > 2) |
3735 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3736 | if (num_devices > 3) | |
3737 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3738 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3739 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
3740 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
3741 | ||
5d163e0e | 3742 | btrfs_err(fs_info, |
6dac13f8 AJ |
3743 | "balance: invalid convert data profile %s", |
3744 | get_raid_name(index)); | |
e4d8ec0f ID |
3745 | ret = -EINVAL; |
3746 | goto out; | |
3747 | } | |
bdcd3c97 | 3748 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
3749 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
3750 | ||
efe120a0 | 3751 | btrfs_err(fs_info, |
6dac13f8 AJ |
3752 | "balance: invalid convert metadata profile %s", |
3753 | get_raid_name(index)); | |
e4d8ec0f ID |
3754 | ret = -EINVAL; |
3755 | goto out; | |
3756 | } | |
bdcd3c97 | 3757 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
3758 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
3759 | ||
efe120a0 | 3760 | btrfs_err(fs_info, |
6dac13f8 AJ |
3761 | "balance: invalid convert system profile %s", |
3762 | get_raid_name(index)); | |
e4d8ec0f ID |
3763 | ret = -EINVAL; |
3764 | goto out; | |
3765 | } | |
3766 | ||
e4d8ec0f ID |
3767 | /* allow to reduce meta or sys integrity only if force set */ |
3768 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3769 | BTRFS_BLOCK_GROUP_RAID10 | |
3770 | BTRFS_BLOCK_GROUP_RAID5 | | |
3771 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3772 | do { |
3773 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3774 | ||
3775 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3776 | (fs_info->avail_system_alloc_bits & allowed) && | |
3777 | !(bctl->sys.target & allowed)) || | |
3778 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3779 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3780 | !(bctl->meta.target & allowed))) { | |
3781 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e | 3782 | btrfs_info(fs_info, |
6dac13f8 | 3783 | "balance: force reducing metadata integrity"); |
de98ced9 | 3784 | } else { |
5d163e0e | 3785 | btrfs_err(fs_info, |
6dac13f8 | 3786 | "balance: reduces metadata integrity, use --force if you want this"); |
de98ced9 MX |
3787 | ret = -EINVAL; |
3788 | goto out; | |
3789 | } | |
e4d8ec0f | 3790 | } |
de98ced9 | 3791 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3792 | |
14506127 AB |
3793 | /* if we're not converting, the target field is uninitialized */ |
3794 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3795 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3796 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3797 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3798 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3799 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
3800 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
3801 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
3802 | ||
ee592d07 | 3803 | btrfs_warn(fs_info, |
6dac13f8 AJ |
3804 | "balance: metadata profile %s has lower redundancy than data profile %s", |
3805 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
3806 | } |
3807 | ||
6bccf3ab | 3808 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3809 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3810 | goto out; |
3811 | ||
59641015 ID |
3812 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3813 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
3814 | BUG_ON(fs_info->balance_ctl); |
3815 | spin_lock(&fs_info->balance_lock); | |
3816 | fs_info->balance_ctl = bctl; | |
3817 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
3818 | } else { |
3819 | BUG_ON(ret != -EEXIST); | |
3820 | spin_lock(&fs_info->balance_lock); | |
3821 | update_balance_args(bctl); | |
3822 | spin_unlock(&fs_info->balance_lock); | |
3823 | } | |
c9e9f97b | 3824 | |
3009a62f DS |
3825 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
3826 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
c9e9f97b ID |
3827 | mutex_unlock(&fs_info->balance_mutex); |
3828 | ||
3829 | ret = __btrfs_balance(fs_info); | |
3830 | ||
3831 | mutex_lock(&fs_info->balance_mutex); | |
3009a62f | 3832 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
3833 | |
3834 | if (bargs) { | |
3835 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 3836 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
3837 | } |
3838 | ||
3a01aa7a ID |
3839 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3840 | balance_need_close(fs_info)) { | |
149196a2 | 3841 | reset_balance_state(fs_info); |
a17c95df | 3842 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
3843 | } |
3844 | ||
837d5b6e | 3845 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3846 | |
3847 | return ret; | |
3848 | out: | |
59641015 | 3849 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 3850 | reset_balance_state(fs_info); |
a17c95df | 3851 | else |
59641015 | 3852 | kfree(bctl); |
a17c95df DS |
3853 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3854 | ||
59641015 ID |
3855 | return ret; |
3856 | } | |
3857 | ||
3858 | static int balance_kthread(void *data) | |
3859 | { | |
2b6ba629 | 3860 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3861 | int ret = 0; |
59641015 | 3862 | |
59641015 | 3863 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 3864 | if (fs_info->balance_ctl) { |
6dac13f8 | 3865 | btrfs_info(fs_info, "balance: resuming"); |
6fcf6e2b | 3866 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
9555c6c1 | 3867 | } |
59641015 | 3868 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 3869 | |
59641015 ID |
3870 | return ret; |
3871 | } | |
3872 | ||
2b6ba629 ID |
3873 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3874 | { | |
3875 | struct task_struct *tsk; | |
3876 | ||
1354e1a1 | 3877 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 3878 | if (!fs_info->balance_ctl) { |
1354e1a1 | 3879 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
3880 | return 0; |
3881 | } | |
1354e1a1 | 3882 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 3883 | |
3cdde224 | 3884 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 3885 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
3886 | return 0; |
3887 | } | |
3888 | ||
02ee654d AJ |
3889 | /* |
3890 | * A ro->rw remount sequence should continue with the paused balance | |
3891 | * regardless of who pauses it, system or the user as of now, so set | |
3892 | * the resume flag. | |
3893 | */ | |
3894 | spin_lock(&fs_info->balance_lock); | |
3895 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
3896 | spin_unlock(&fs_info->balance_lock); | |
3897 | ||
2b6ba629 | 3898 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 3899 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3900 | } |
3901 | ||
68310a5e | 3902 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3903 | { |
59641015 ID |
3904 | struct btrfs_balance_control *bctl; |
3905 | struct btrfs_balance_item *item; | |
3906 | struct btrfs_disk_balance_args disk_bargs; | |
3907 | struct btrfs_path *path; | |
3908 | struct extent_buffer *leaf; | |
3909 | struct btrfs_key key; | |
3910 | int ret; | |
3911 | ||
3912 | path = btrfs_alloc_path(); | |
3913 | if (!path) | |
3914 | return -ENOMEM; | |
3915 | ||
59641015 | 3916 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3917 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
3918 | key.offset = 0; |
3919 | ||
68310a5e | 3920 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3921 | if (ret < 0) |
68310a5e | 3922 | goto out; |
59641015 ID |
3923 | if (ret > 0) { /* ret = -ENOENT; */ |
3924 | ret = 0; | |
68310a5e ID |
3925 | goto out; |
3926 | } | |
3927 | ||
3928 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3929 | if (!bctl) { | |
3930 | ret = -ENOMEM; | |
3931 | goto out; | |
59641015 ID |
3932 | } |
3933 | ||
3934 | leaf = path->nodes[0]; | |
3935 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3936 | ||
68310a5e ID |
3937 | bctl->flags = btrfs_balance_flags(leaf, item); |
3938 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3939 | |
3940 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3941 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3942 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3943 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3944 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3945 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3946 | ||
eee95e3f DS |
3947 | /* |
3948 | * This should never happen, as the paused balance state is recovered | |
3949 | * during mount without any chance of other exclusive ops to collide. | |
3950 | * | |
3951 | * This gives the exclusive op status to balance and keeps in paused | |
3952 | * state until user intervention (cancel or umount). If the ownership | |
3953 | * cannot be assigned, show a message but do not fail. The balance | |
3954 | * is in a paused state and must have fs_info::balance_ctl properly | |
3955 | * set up. | |
3956 | */ | |
3957 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
3958 | btrfs_warn(fs_info, | |
6dac13f8 | 3959 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 3960 | |
68310a5e | 3961 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
3962 | BUG_ON(fs_info->balance_ctl); |
3963 | spin_lock(&fs_info->balance_lock); | |
3964 | fs_info->balance_ctl = bctl; | |
3965 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 3966 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
3967 | out: |
3968 | btrfs_free_path(path); | |
ec44a35c CM |
3969 | return ret; |
3970 | } | |
3971 | ||
837d5b6e ID |
3972 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3973 | { | |
3974 | int ret = 0; | |
3975 | ||
3976 | mutex_lock(&fs_info->balance_mutex); | |
3977 | if (!fs_info->balance_ctl) { | |
3978 | mutex_unlock(&fs_info->balance_mutex); | |
3979 | return -ENOTCONN; | |
3980 | } | |
3981 | ||
3009a62f | 3982 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
3983 | atomic_inc(&fs_info->balance_pause_req); |
3984 | mutex_unlock(&fs_info->balance_mutex); | |
3985 | ||
3986 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 3987 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
3988 | |
3989 | mutex_lock(&fs_info->balance_mutex); | |
3990 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 3991 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
3992 | atomic_dec(&fs_info->balance_pause_req); |
3993 | } else { | |
3994 | ret = -ENOTCONN; | |
3995 | } | |
3996 | ||
3997 | mutex_unlock(&fs_info->balance_mutex); | |
3998 | return ret; | |
3999 | } | |
4000 | ||
a7e99c69 ID |
4001 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4002 | { | |
4003 | mutex_lock(&fs_info->balance_mutex); | |
4004 | if (!fs_info->balance_ctl) { | |
4005 | mutex_unlock(&fs_info->balance_mutex); | |
4006 | return -ENOTCONN; | |
4007 | } | |
4008 | ||
cf7d20f4 DS |
4009 | /* |
4010 | * A paused balance with the item stored on disk can be resumed at | |
4011 | * mount time if the mount is read-write. Otherwise it's still paused | |
4012 | * and we must not allow cancelling as it deletes the item. | |
4013 | */ | |
4014 | if (sb_rdonly(fs_info->sb)) { | |
4015 | mutex_unlock(&fs_info->balance_mutex); | |
4016 | return -EROFS; | |
4017 | } | |
4018 | ||
a7e99c69 ID |
4019 | atomic_inc(&fs_info->balance_cancel_req); |
4020 | /* | |
4021 | * if we are running just wait and return, balance item is | |
4022 | * deleted in btrfs_balance in this case | |
4023 | */ | |
3009a62f | 4024 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4025 | mutex_unlock(&fs_info->balance_mutex); |
4026 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4027 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4028 | mutex_lock(&fs_info->balance_mutex); |
4029 | } else { | |
a7e99c69 | 4030 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4031 | /* |
4032 | * Lock released to allow other waiters to continue, we'll | |
4033 | * reexamine the status again. | |
4034 | */ | |
a7e99c69 ID |
4035 | mutex_lock(&fs_info->balance_mutex); |
4036 | ||
a17c95df | 4037 | if (fs_info->balance_ctl) { |
149196a2 | 4038 | reset_balance_state(fs_info); |
a17c95df | 4039 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4040 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4041 | } |
a7e99c69 ID |
4042 | } |
4043 | ||
3009a62f DS |
4044 | BUG_ON(fs_info->balance_ctl || |
4045 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4046 | atomic_dec(&fs_info->balance_cancel_req); |
4047 | mutex_unlock(&fs_info->balance_mutex); | |
4048 | return 0; | |
4049 | } | |
4050 | ||
803b2f54 SB |
4051 | static int btrfs_uuid_scan_kthread(void *data) |
4052 | { | |
4053 | struct btrfs_fs_info *fs_info = data; | |
4054 | struct btrfs_root *root = fs_info->tree_root; | |
4055 | struct btrfs_key key; | |
803b2f54 SB |
4056 | struct btrfs_path *path = NULL; |
4057 | int ret = 0; | |
4058 | struct extent_buffer *eb; | |
4059 | int slot; | |
4060 | struct btrfs_root_item root_item; | |
4061 | u32 item_size; | |
f45388f3 | 4062 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4063 | |
4064 | path = btrfs_alloc_path(); | |
4065 | if (!path) { | |
4066 | ret = -ENOMEM; | |
4067 | goto out; | |
4068 | } | |
4069 | ||
4070 | key.objectid = 0; | |
4071 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4072 | key.offset = 0; | |
4073 | ||
803b2f54 | 4074 | while (1) { |
7c829b72 AJ |
4075 | ret = btrfs_search_forward(root, &key, path, |
4076 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4077 | if (ret) { |
4078 | if (ret > 0) | |
4079 | ret = 0; | |
4080 | break; | |
4081 | } | |
4082 | ||
4083 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4084 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4085 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4086 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4087 | goto skip; | |
4088 | ||
4089 | eb = path->nodes[0]; | |
4090 | slot = path->slots[0]; | |
4091 | item_size = btrfs_item_size_nr(eb, slot); | |
4092 | if (item_size < sizeof(root_item)) | |
4093 | goto skip; | |
4094 | ||
803b2f54 SB |
4095 | read_extent_buffer(eb, &root_item, |
4096 | btrfs_item_ptr_offset(eb, slot), | |
4097 | (int)sizeof(root_item)); | |
4098 | if (btrfs_root_refs(&root_item) == 0) | |
4099 | goto skip; | |
f45388f3 FDBM |
4100 | |
4101 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4102 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4103 | if (trans) | |
4104 | goto update_tree; | |
4105 | ||
4106 | btrfs_release_path(path); | |
803b2f54 SB |
4107 | /* |
4108 | * 1 - subvol uuid item | |
4109 | * 1 - received_subvol uuid item | |
4110 | */ | |
4111 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4112 | if (IS_ERR(trans)) { | |
4113 | ret = PTR_ERR(trans); | |
4114 | break; | |
4115 | } | |
f45388f3 FDBM |
4116 | continue; |
4117 | } else { | |
4118 | goto skip; | |
4119 | } | |
4120 | update_tree: | |
4121 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4122 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4123 | BTRFS_UUID_KEY_SUBVOL, |
4124 | key.objectid); | |
4125 | if (ret < 0) { | |
efe120a0 | 4126 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4127 | ret); |
803b2f54 SB |
4128 | break; |
4129 | } | |
4130 | } | |
4131 | ||
4132 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4133 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4134 | root_item.received_uuid, |
4135 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4136 | key.objectid); | |
4137 | if (ret < 0) { | |
efe120a0 | 4138 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4139 | ret); |
803b2f54 SB |
4140 | break; |
4141 | } | |
4142 | } | |
4143 | ||
f45388f3 | 4144 | skip: |
803b2f54 | 4145 | if (trans) { |
3a45bb20 | 4146 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4147 | trans = NULL; |
803b2f54 SB |
4148 | if (ret) |
4149 | break; | |
4150 | } | |
4151 | ||
803b2f54 SB |
4152 | btrfs_release_path(path); |
4153 | if (key.offset < (u64)-1) { | |
4154 | key.offset++; | |
4155 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4156 | key.offset = 0; | |
4157 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4158 | } else if (key.objectid < (u64)-1) { | |
4159 | key.offset = 0; | |
4160 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4161 | key.objectid++; | |
4162 | } else { | |
4163 | break; | |
4164 | } | |
4165 | cond_resched(); | |
4166 | } | |
4167 | ||
4168 | out: | |
4169 | btrfs_free_path(path); | |
f45388f3 | 4170 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4171 | btrfs_end_transaction(trans); |
803b2f54 | 4172 | if (ret) |
efe120a0 | 4173 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4174 | else |
afcdd129 | 4175 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4176 | up(&fs_info->uuid_tree_rescan_sem); |
4177 | return 0; | |
4178 | } | |
4179 | ||
70f80175 SB |
4180 | /* |
4181 | * Callback for btrfs_uuid_tree_iterate(). | |
4182 | * returns: | |
4183 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4184 | * < 0 if an error occurred. |
70f80175 SB |
4185 | * > 0 if the check failed, which means the caller shall remove the entry. |
4186 | */ | |
4187 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4188 | u8 *uuid, u8 type, u64 subid) | |
4189 | { | |
4190 | struct btrfs_key key; | |
4191 | int ret = 0; | |
4192 | struct btrfs_root *subvol_root; | |
4193 | ||
4194 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4195 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4196 | goto out; | |
4197 | ||
4198 | key.objectid = subid; | |
4199 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4200 | key.offset = (u64)-1; | |
4201 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4202 | if (IS_ERR(subvol_root)) { | |
4203 | ret = PTR_ERR(subvol_root); | |
4204 | if (ret == -ENOENT) | |
4205 | ret = 1; | |
4206 | goto out; | |
4207 | } | |
4208 | ||
4209 | switch (type) { | |
4210 | case BTRFS_UUID_KEY_SUBVOL: | |
4211 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4212 | ret = 1; | |
4213 | break; | |
4214 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4215 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4216 | BTRFS_UUID_SIZE)) | |
4217 | ret = 1; | |
4218 | break; | |
4219 | } | |
4220 | ||
4221 | out: | |
4222 | return ret; | |
4223 | } | |
4224 | ||
4225 | static int btrfs_uuid_rescan_kthread(void *data) | |
4226 | { | |
4227 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4228 | int ret; | |
4229 | ||
4230 | /* | |
4231 | * 1st step is to iterate through the existing UUID tree and | |
4232 | * to delete all entries that contain outdated data. | |
4233 | * 2nd step is to add all missing entries to the UUID tree. | |
4234 | */ | |
4235 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4236 | if (ret < 0) { | |
efe120a0 | 4237 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4238 | up(&fs_info->uuid_tree_rescan_sem); |
4239 | return ret; | |
4240 | } | |
4241 | return btrfs_uuid_scan_kthread(data); | |
4242 | } | |
4243 | ||
f7a81ea4 SB |
4244 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4245 | { | |
4246 | struct btrfs_trans_handle *trans; | |
4247 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4248 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4249 | struct task_struct *task; |
4250 | int ret; | |
f7a81ea4 SB |
4251 | |
4252 | /* | |
4253 | * 1 - root node | |
4254 | * 1 - root item | |
4255 | */ | |
4256 | trans = btrfs_start_transaction(tree_root, 2); | |
4257 | if (IS_ERR(trans)) | |
4258 | return PTR_ERR(trans); | |
4259 | ||
4260 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4261 | BTRFS_UUID_TREE_OBJECTID); | |
4262 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4263 | ret = PTR_ERR(uuid_root); |
66642832 | 4264 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4265 | btrfs_end_transaction(trans); |
6d13f549 | 4266 | return ret; |
f7a81ea4 SB |
4267 | } |
4268 | ||
4269 | fs_info->uuid_root = uuid_root; | |
4270 | ||
3a45bb20 | 4271 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4272 | if (ret) |
4273 | return ret; | |
4274 | ||
4275 | down(&fs_info->uuid_tree_rescan_sem); | |
4276 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4277 | if (IS_ERR(task)) { | |
70f80175 | 4278 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4279 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4280 | up(&fs_info->uuid_tree_rescan_sem); |
4281 | return PTR_ERR(task); | |
4282 | } | |
4283 | ||
4284 | return 0; | |
f7a81ea4 | 4285 | } |
803b2f54 | 4286 | |
70f80175 SB |
4287 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4288 | { | |
4289 | struct task_struct *task; | |
4290 | ||
4291 | down(&fs_info->uuid_tree_rescan_sem); | |
4292 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4293 | if (IS_ERR(task)) { | |
4294 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4295 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4296 | up(&fs_info->uuid_tree_rescan_sem); |
4297 | return PTR_ERR(task); | |
4298 | } | |
4299 | ||
4300 | return 0; | |
4301 | } | |
4302 | ||
8f18cf13 CM |
4303 | /* |
4304 | * shrinking a device means finding all of the device extents past | |
4305 | * the new size, and then following the back refs to the chunks. | |
4306 | * The chunk relocation code actually frees the device extent | |
4307 | */ | |
4308 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4309 | { | |
0b246afa JM |
4310 | struct btrfs_fs_info *fs_info = device->fs_info; |
4311 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4312 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4313 | struct btrfs_dev_extent *dev_extent = NULL; |
4314 | struct btrfs_path *path; | |
4315 | u64 length; | |
8f18cf13 CM |
4316 | u64 chunk_offset; |
4317 | int ret; | |
4318 | int slot; | |
ba1bf481 JB |
4319 | int failed = 0; |
4320 | bool retried = false; | |
53e489bc | 4321 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4322 | struct extent_buffer *l; |
4323 | struct btrfs_key key; | |
0b246afa | 4324 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4325 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4326 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4327 | u64 diff; |
4328 | ||
4329 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4330 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4331 | |
401e29c1 | 4332 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4333 | return -EINVAL; |
4334 | ||
8f18cf13 CM |
4335 | path = btrfs_alloc_path(); |
4336 | if (!path) | |
4337 | return -ENOMEM; | |
4338 | ||
0338dff6 | 4339 | path->reada = READA_BACK; |
8f18cf13 | 4340 | |
34441361 | 4341 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4342 | |
7cc8e58d | 4343 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4344 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4345 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4346 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4347 | } |
34441361 | 4348 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4349 | |
ba1bf481 | 4350 | again: |
8f18cf13 CM |
4351 | key.objectid = device->devid; |
4352 | key.offset = (u64)-1; | |
4353 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4354 | ||
213e64da | 4355 | do { |
0b246afa | 4356 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4357 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4358 | if (ret < 0) { |
0b246afa | 4359 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4360 | goto done; |
67c5e7d4 | 4361 | } |
8f18cf13 CM |
4362 | |
4363 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4364 | if (ret) |
0b246afa | 4365 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4366 | if (ret < 0) |
4367 | goto done; | |
4368 | if (ret) { | |
4369 | ret = 0; | |
b3b4aa74 | 4370 | btrfs_release_path(path); |
bf1fb512 | 4371 | break; |
8f18cf13 CM |
4372 | } |
4373 | ||
4374 | l = path->nodes[0]; | |
4375 | slot = path->slots[0]; | |
4376 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4377 | ||
ba1bf481 | 4378 | if (key.objectid != device->devid) { |
0b246afa | 4379 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4380 | btrfs_release_path(path); |
bf1fb512 | 4381 | break; |
ba1bf481 | 4382 | } |
8f18cf13 CM |
4383 | |
4384 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4385 | length = btrfs_dev_extent_length(l, dev_extent); | |
4386 | ||
ba1bf481 | 4387 | if (key.offset + length <= new_size) { |
0b246afa | 4388 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4389 | btrfs_release_path(path); |
d6397bae | 4390 | break; |
ba1bf481 | 4391 | } |
8f18cf13 | 4392 | |
8f18cf13 | 4393 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4394 | btrfs_release_path(path); |
8f18cf13 | 4395 | |
a6f93c71 LB |
4396 | /* |
4397 | * We may be relocating the only data chunk we have, | |
4398 | * which could potentially end up with losing data's | |
4399 | * raid profile, so lets allocate an empty one in | |
4400 | * advance. | |
4401 | */ | |
4402 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4403 | if (ret < 0) { | |
4404 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4405 | goto done; | |
4406 | } | |
4407 | ||
0b246afa JM |
4408 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4409 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4410 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4411 | goto done; |
ba1bf481 JB |
4412 | if (ret == -ENOSPC) |
4413 | failed++; | |
213e64da | 4414 | } while (key.offset-- > 0); |
ba1bf481 JB |
4415 | |
4416 | if (failed && !retried) { | |
4417 | failed = 0; | |
4418 | retried = true; | |
4419 | goto again; | |
4420 | } else if (failed && retried) { | |
4421 | ret = -ENOSPC; | |
ba1bf481 | 4422 | goto done; |
8f18cf13 CM |
4423 | } |
4424 | ||
d6397bae | 4425 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4426 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4427 | if (IS_ERR(trans)) { |
4428 | ret = PTR_ERR(trans); | |
4429 | goto done; | |
4430 | } | |
4431 | ||
34441361 | 4432 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4433 | |
4434 | /* | |
4435 | * We checked in the above loop all device extents that were already in | |
4436 | * the device tree. However before we have updated the device's | |
4437 | * total_bytes to the new size, we might have had chunk allocations that | |
4438 | * have not complete yet (new block groups attached to transaction | |
4439 | * handles), and therefore their device extents were not yet in the | |
4440 | * device tree and we missed them in the loop above. So if we have any | |
4441 | * pending chunk using a device extent that overlaps the device range | |
4442 | * that we can not use anymore, commit the current transaction and | |
4443 | * repeat the search on the device tree - this way we guarantee we will | |
4444 | * not have chunks using device extents that end beyond 'new_size'. | |
4445 | */ | |
4446 | if (!checked_pending_chunks) { | |
4447 | u64 start = new_size; | |
4448 | u64 len = old_size - new_size; | |
4449 | ||
499f377f JM |
4450 | if (contains_pending_extent(trans->transaction, device, |
4451 | &start, len)) { | |
34441361 | 4452 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4453 | checked_pending_chunks = true; |
4454 | failed = 0; | |
4455 | retried = false; | |
3a45bb20 | 4456 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4457 | if (ret) |
4458 | goto done; | |
4459 | goto again; | |
4460 | } | |
4461 | } | |
4462 | ||
7cc8e58d | 4463 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4464 | if (list_empty(&device->resized_list)) |
4465 | list_add_tail(&device->resized_list, | |
0b246afa | 4466 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4467 | |
d6397bae | 4468 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4469 | btrfs_set_super_total_bytes(super_copy, |
4470 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4471 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4472 | |
4473 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4474 | ret = btrfs_update_device(trans, device); | |
3a45bb20 | 4475 | btrfs_end_transaction(trans); |
8f18cf13 CM |
4476 | done: |
4477 | btrfs_free_path(path); | |
53e489bc | 4478 | if (ret) { |
34441361 | 4479 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4480 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4481 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4482 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4483 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4484 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4485 | } |
8f18cf13 CM |
4486 | return ret; |
4487 | } | |
4488 | ||
2ff7e61e | 4489 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4490 | struct btrfs_key *key, |
4491 | struct btrfs_chunk *chunk, int item_size) | |
4492 | { | |
0b246afa | 4493 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4494 | struct btrfs_disk_key disk_key; |
4495 | u32 array_size; | |
4496 | u8 *ptr; | |
4497 | ||
34441361 | 4498 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4499 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4500 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4501 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4502 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4503 | return -EFBIG; |
fe48a5c0 | 4504 | } |
0b86a832 CM |
4505 | |
4506 | ptr = super_copy->sys_chunk_array + array_size; | |
4507 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4508 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4509 | ptr += sizeof(disk_key); | |
4510 | memcpy(ptr, chunk, item_size); | |
4511 | item_size += sizeof(disk_key); | |
4512 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4513 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4514 | |
0b86a832 CM |
4515 | return 0; |
4516 | } | |
4517 | ||
73c5de00 AJ |
4518 | /* |
4519 | * sort the devices in descending order by max_avail, total_avail | |
4520 | */ | |
4521 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4522 | { |
73c5de00 AJ |
4523 | const struct btrfs_device_info *di_a = a; |
4524 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4525 | |
73c5de00 | 4526 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4527 | return -1; |
73c5de00 | 4528 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4529 | return 1; |
73c5de00 AJ |
4530 | if (di_a->total_avail > di_b->total_avail) |
4531 | return -1; | |
4532 | if (di_a->total_avail < di_b->total_avail) | |
4533 | return 1; | |
4534 | return 0; | |
b2117a39 | 4535 | } |
0b86a832 | 4536 | |
53b381b3 DW |
4537 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4538 | { | |
ffe2d203 | 4539 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4540 | return; |
4541 | ||
ceda0864 | 4542 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4543 | } |
4544 | ||
062d4d1f | 4545 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
4546 | - sizeof(struct btrfs_chunk)) \ |
4547 | / sizeof(struct btrfs_stripe) + 1) | |
4548 | ||
4549 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4550 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4551 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4552 | / sizeof(struct btrfs_stripe) + 1) | |
4553 | ||
73c5de00 | 4554 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4555 | u64 start, u64 type) |
b2117a39 | 4556 | { |
2ff7e61e | 4557 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4558 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4559 | struct btrfs_device *device; |
73c5de00 AJ |
4560 | struct map_lookup *map = NULL; |
4561 | struct extent_map_tree *em_tree; | |
4562 | struct extent_map *em; | |
4563 | struct btrfs_device_info *devices_info = NULL; | |
4564 | u64 total_avail; | |
4565 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4566 | int data_stripes; /* number of stripes that count for |
4567 | block group size */ | |
73c5de00 AJ |
4568 | int sub_stripes; /* sub_stripes info for map */ |
4569 | int dev_stripes; /* stripes per dev */ | |
4570 | int devs_max; /* max devs to use */ | |
4571 | int devs_min; /* min devs needed */ | |
4572 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4573 | int ncopies; /* how many copies to data has */ | |
4574 | int ret; | |
4575 | u64 max_stripe_size; | |
4576 | u64 max_chunk_size; | |
4577 | u64 stripe_size; | |
4578 | u64 num_bytes; | |
4579 | int ndevs; | |
4580 | int i; | |
4581 | int j; | |
31e50229 | 4582 | int index; |
593060d7 | 4583 | |
0c460c0d | 4584 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4585 | |
4117f207 QW |
4586 | if (list_empty(&fs_devices->alloc_list)) { |
4587 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4588 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4589 | return -ENOSPC; |
4117f207 | 4590 | } |
b2117a39 | 4591 | |
3e72ee88 | 4592 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4593 | |
31e50229 LB |
4594 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4595 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4596 | devs_max = btrfs_raid_array[index].devs_max; | |
4597 | devs_min = btrfs_raid_array[index].devs_min; | |
4598 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4599 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4600 | |
9b3f68b9 | 4601 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4602 | max_stripe_size = SZ_1G; |
fce466ea | 4603 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 4604 | if (!devs_max) |
062d4d1f | 4605 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 4606 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4607 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4608 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4609 | max_stripe_size = SZ_1G; | |
1100373f | 4610 | else |
ee22184b | 4611 | max_stripe_size = SZ_256M; |
73c5de00 | 4612 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 4613 | if (!devs_max) |
062d4d1f | 4614 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 4615 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4616 | max_stripe_size = SZ_32M; |
73c5de00 | 4617 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4618 | if (!devs_max) |
4619 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4620 | } else { |
351fd353 | 4621 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4622 | type); |
4623 | BUG_ON(1); | |
9b3f68b9 CM |
4624 | } |
4625 | ||
2b82032c YZ |
4626 | /* we don't want a chunk larger than 10% of writeable space */ |
4627 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4628 | max_chunk_size); | |
9b3f68b9 | 4629 | |
31e818fe | 4630 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4631 | GFP_NOFS); |
4632 | if (!devices_info) | |
4633 | return -ENOMEM; | |
0cad8a11 | 4634 | |
9f680ce0 | 4635 | /* |
73c5de00 AJ |
4636 | * in the first pass through the devices list, we gather information |
4637 | * about the available holes on each device. | |
9f680ce0 | 4638 | */ |
73c5de00 | 4639 | ndevs = 0; |
ebcc9301 | 4640 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4641 | u64 max_avail; |
4642 | u64 dev_offset; | |
b2117a39 | 4643 | |
ebbede42 | 4644 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4645 | WARN(1, KERN_ERR |
efe120a0 | 4646 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4647 | continue; |
4648 | } | |
b2117a39 | 4649 | |
e12c9621 AJ |
4650 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4651 | &device->dev_state) || | |
401e29c1 | 4652 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4653 | continue; |
b2117a39 | 4654 | |
73c5de00 AJ |
4655 | if (device->total_bytes > device->bytes_used) |
4656 | total_avail = device->total_bytes - device->bytes_used; | |
4657 | else | |
4658 | total_avail = 0; | |
38c01b96 | 4659 | |
4660 | /* If there is no space on this device, skip it. */ | |
4661 | if (total_avail == 0) | |
4662 | continue; | |
b2117a39 | 4663 | |
6df9a95e | 4664 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4665 | max_stripe_size * dev_stripes, |
4666 | &dev_offset, &max_avail); | |
4667 | if (ret && ret != -ENOSPC) | |
4668 | goto error; | |
b2117a39 | 4669 | |
73c5de00 AJ |
4670 | if (ret == 0) |
4671 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4672 | |
4117f207 QW |
4673 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4674 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4675 | btrfs_debug(info, | |
4676 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4677 | __func__, device->devid, max_avail, | |
4678 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4679 | continue; |
4117f207 | 4680 | } |
b2117a39 | 4681 | |
063d006f ES |
4682 | if (ndevs == fs_devices->rw_devices) { |
4683 | WARN(1, "%s: found more than %llu devices\n", | |
4684 | __func__, fs_devices->rw_devices); | |
4685 | break; | |
4686 | } | |
73c5de00 AJ |
4687 | devices_info[ndevs].dev_offset = dev_offset; |
4688 | devices_info[ndevs].max_avail = max_avail; | |
4689 | devices_info[ndevs].total_avail = total_avail; | |
4690 | devices_info[ndevs].dev = device; | |
4691 | ++ndevs; | |
4692 | } | |
b2117a39 | 4693 | |
73c5de00 AJ |
4694 | /* |
4695 | * now sort the devices by hole size / available space | |
4696 | */ | |
4697 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4698 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4699 | |
73c5de00 | 4700 | /* round down to number of usable stripes */ |
e5600fd6 | 4701 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4702 | |
ba89b802 | 4703 | if (ndevs < devs_min) { |
73c5de00 | 4704 | ret = -ENOSPC; |
4117f207 QW |
4705 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4706 | btrfs_debug(info, | |
4707 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 4708 | __func__, ndevs, devs_min); |
4117f207 | 4709 | } |
73c5de00 | 4710 | goto error; |
b2117a39 | 4711 | } |
9f680ce0 | 4712 | |
f148ef4d NB |
4713 | ndevs = min(ndevs, devs_max); |
4714 | ||
73c5de00 | 4715 | /* |
92e222df HK |
4716 | * The primary goal is to maximize the number of stripes, so use as |
4717 | * many devices as possible, even if the stripes are not maximum sized. | |
4718 | * | |
4719 | * The DUP profile stores more than one stripe per device, the | |
4720 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4721 | */ |
92e222df | 4722 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4723 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4724 | |
53b381b3 DW |
4725 | /* |
4726 | * this will have to be fixed for RAID1 and RAID10 over | |
4727 | * more drives | |
4728 | */ | |
4729 | data_stripes = num_stripes / ncopies; | |
4730 | ||
500ceed8 | 4731 | if (type & BTRFS_BLOCK_GROUP_RAID5) |
53b381b3 | 4732 | data_stripes = num_stripes - 1; |
500ceed8 NB |
4733 | |
4734 | if (type & BTRFS_BLOCK_GROUP_RAID6) | |
53b381b3 | 4735 | data_stripes = num_stripes - 2; |
86db2578 CM |
4736 | |
4737 | /* | |
4738 | * Use the number of data stripes to figure out how big this chunk | |
4739 | * is really going to be in terms of logical address space, | |
4740 | * and compare that answer with the max chunk size | |
4741 | */ | |
4742 | if (stripe_size * data_stripes > max_chunk_size) { | |
b8b93add | 4743 | stripe_size = div_u64(max_chunk_size, data_stripes); |
86db2578 CM |
4744 | |
4745 | /* bump the answer up to a 16MB boundary */ | |
793ff2c8 | 4746 | stripe_size = round_up(stripe_size, SZ_16M); |
86db2578 | 4747 | |
793ff2c8 QW |
4748 | /* |
4749 | * But don't go higher than the limits we found while searching | |
4750 | * for free extents | |
86db2578 | 4751 | */ |
793ff2c8 QW |
4752 | stripe_size = min(devices_info[ndevs - 1].max_avail, |
4753 | stripe_size); | |
86db2578 CM |
4754 | } |
4755 | ||
37db63a4 | 4756 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 4757 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4758 | |
4759 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4760 | if (!map) { | |
4761 | ret = -ENOMEM; | |
4762 | goto error; | |
4763 | } | |
4764 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4765 | |
73c5de00 AJ |
4766 | for (i = 0; i < ndevs; ++i) { |
4767 | for (j = 0; j < dev_stripes; ++j) { | |
4768 | int s = i * dev_stripes + j; | |
4769 | map->stripes[s].dev = devices_info[i].dev; | |
4770 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4771 | j * stripe_size; | |
6324fbf3 | 4772 | } |
6324fbf3 | 4773 | } |
500ceed8 NB |
4774 | map->stripe_len = BTRFS_STRIPE_LEN; |
4775 | map->io_align = BTRFS_STRIPE_LEN; | |
4776 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4777 | map->type = type; |
2b82032c | 4778 | map->sub_stripes = sub_stripes; |
0b86a832 | 4779 | |
53b381b3 | 4780 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4781 | |
6bccf3ab | 4782 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4783 | |
172ddd60 | 4784 | em = alloc_extent_map(); |
2b82032c | 4785 | if (!em) { |
298a8f9c | 4786 | kfree(map); |
b2117a39 MX |
4787 | ret = -ENOMEM; |
4788 | goto error; | |
593060d7 | 4789 | } |
298a8f9c | 4790 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4791 | em->map_lookup = map; |
2b82032c | 4792 | em->start = start; |
73c5de00 | 4793 | em->len = num_bytes; |
2b82032c YZ |
4794 | em->block_start = 0; |
4795 | em->block_len = em->len; | |
6df9a95e | 4796 | em->orig_block_len = stripe_size; |
593060d7 | 4797 | |
0b246afa | 4798 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4799 | write_lock(&em_tree->lock); |
09a2a8f9 | 4800 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 4801 | if (ret) { |
1efb72a3 | 4802 | write_unlock(&em_tree->lock); |
0f5d42b2 | 4803 | free_extent_map(em); |
1dd4602f | 4804 | goto error; |
0f5d42b2 | 4805 | } |
0b86a832 | 4806 | |
1efb72a3 NB |
4807 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
4808 | refcount_inc(&em->refs); | |
4809 | write_unlock(&em_tree->lock); | |
4810 | ||
e7e02096 | 4811 | ret = btrfs_make_block_group(trans, 0, type, start, num_bytes); |
6df9a95e JB |
4812 | if (ret) |
4813 | goto error_del_extent; | |
2b82032c | 4814 | |
7cc8e58d MX |
4815 | for (i = 0; i < map->num_stripes; i++) { |
4816 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4817 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4818 | } | |
43530c46 | 4819 | |
a5ed45f8 | 4820 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4821 | |
0f5d42b2 | 4822 | free_extent_map(em); |
0b246afa | 4823 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4824 | |
b2117a39 | 4825 | kfree(devices_info); |
2b82032c | 4826 | return 0; |
b2117a39 | 4827 | |
6df9a95e | 4828 | error_del_extent: |
0f5d42b2 JB |
4829 | write_lock(&em_tree->lock); |
4830 | remove_extent_mapping(em_tree, em); | |
4831 | write_unlock(&em_tree->lock); | |
4832 | ||
4833 | /* One for our allocation */ | |
4834 | free_extent_map(em); | |
4835 | /* One for the tree reference */ | |
4836 | free_extent_map(em); | |
495e64f4 FM |
4837 | /* One for the pending_chunks list reference */ |
4838 | free_extent_map(em); | |
b2117a39 | 4839 | error: |
b2117a39 MX |
4840 | kfree(devices_info); |
4841 | return ret; | |
2b82032c YZ |
4842 | } |
4843 | ||
6df9a95e | 4844 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 4845 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4846 | { |
97aff912 | 4847 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
4848 | struct btrfs_root *extent_root = fs_info->extent_root; |
4849 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4850 | struct btrfs_key key; |
2b82032c YZ |
4851 | struct btrfs_device *device; |
4852 | struct btrfs_chunk *chunk; | |
4853 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4854 | struct extent_map *em; |
4855 | struct map_lookup *map; | |
4856 | size_t item_size; | |
4857 | u64 dev_offset; | |
4858 | u64 stripe_size; | |
4859 | int i = 0; | |
140e639f | 4860 | int ret = 0; |
2b82032c | 4861 | |
592d92ee LB |
4862 | em = get_chunk_map(fs_info, chunk_offset, chunk_size); |
4863 | if (IS_ERR(em)) | |
4864 | return PTR_ERR(em); | |
6df9a95e | 4865 | |
95617d69 | 4866 | map = em->map_lookup; |
6df9a95e JB |
4867 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4868 | stripe_size = em->orig_block_len; | |
4869 | ||
2b82032c | 4870 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4871 | if (!chunk) { |
4872 | ret = -ENOMEM; | |
4873 | goto out; | |
4874 | } | |
4875 | ||
50460e37 FM |
4876 | /* |
4877 | * Take the device list mutex to prevent races with the final phase of | |
4878 | * a device replace operation that replaces the device object associated | |
4879 | * with the map's stripes, because the device object's id can change | |
4880 | * at any time during that final phase of the device replace operation | |
4881 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4882 | */ | |
0b246afa | 4883 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
4884 | for (i = 0; i < map->num_stripes; i++) { |
4885 | device = map->stripes[i].dev; | |
4886 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4887 | |
0b86a832 | 4888 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4889 | if (ret) |
50460e37 | 4890 | break; |
b5d9071c NB |
4891 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
4892 | dev_offset, stripe_size); | |
6df9a95e | 4893 | if (ret) |
50460e37 FM |
4894 | break; |
4895 | } | |
4896 | if (ret) { | |
0b246afa | 4897 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 4898 | goto out; |
2b82032c YZ |
4899 | } |
4900 | ||
2b82032c | 4901 | stripe = &chunk->stripe; |
6df9a95e JB |
4902 | for (i = 0; i < map->num_stripes; i++) { |
4903 | device = map->stripes[i].dev; | |
4904 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4905 | |
e17cade2 CM |
4906 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4907 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4908 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4909 | stripe++; |
0b86a832 | 4910 | } |
0b246afa | 4911 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4912 | |
2b82032c | 4913 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4914 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4915 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4916 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4917 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4918 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4919 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 4920 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 4921 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4922 | |
2b82032c YZ |
4923 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4924 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4925 | key.offset = chunk_offset; | |
0b86a832 | 4926 | |
2b82032c | 4927 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4928 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4929 | /* | |
4930 | * TODO: Cleanup of inserted chunk root in case of | |
4931 | * failure. | |
4932 | */ | |
2ff7e61e | 4933 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 4934 | } |
1abe9b8a | 4935 | |
6df9a95e | 4936 | out: |
0b86a832 | 4937 | kfree(chunk); |
6df9a95e | 4938 | free_extent_map(em); |
4ed1d16e | 4939 | return ret; |
2b82032c | 4940 | } |
0b86a832 | 4941 | |
2b82032c YZ |
4942 | /* |
4943 | * Chunk allocation falls into two parts. The first part does works | |
4944 | * that make the new allocated chunk useable, but not do any operation | |
4945 | * that modifies the chunk tree. The second part does the works that | |
4946 | * require modifying the chunk tree. This division is important for the | |
4947 | * bootstrap process of adding storage to a seed btrfs. | |
4948 | */ | |
c216b203 | 4949 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
4950 | { |
4951 | u64 chunk_offset; | |
2b82032c | 4952 | |
c216b203 NB |
4953 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
4954 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 4955 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
4956 | } |
4957 | ||
d397712b | 4958 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 4959 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
4960 | { |
4961 | u64 chunk_offset; | |
4962 | u64 sys_chunk_offset; | |
2b82032c | 4963 | u64 alloc_profile; |
2b82032c YZ |
4964 | int ret; |
4965 | ||
6df9a95e | 4966 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 4967 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 4968 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
4969 | if (ret) |
4970 | return ret; | |
2b82032c | 4971 | |
0b246afa | 4972 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 4973 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 4974 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 4975 | return ret; |
2b82032c YZ |
4976 | } |
4977 | ||
d20983b4 MX |
4978 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
4979 | { | |
4980 | int max_errors; | |
4981 | ||
4982 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4983 | BTRFS_BLOCK_GROUP_RAID10 | | |
4984 | BTRFS_BLOCK_GROUP_RAID5 | | |
4985 | BTRFS_BLOCK_GROUP_DUP)) { | |
4986 | max_errors = 1; | |
4987 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4988 | max_errors = 2; | |
4989 | } else { | |
4990 | max_errors = 0; | |
005d6427 | 4991 | } |
2b82032c | 4992 | |
d20983b4 | 4993 | return max_errors; |
2b82032c YZ |
4994 | } |
4995 | ||
2ff7e61e | 4996 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
4997 | { |
4998 | struct extent_map *em; | |
4999 | struct map_lookup *map; | |
2b82032c | 5000 | int readonly = 0; |
d20983b4 | 5001 | int miss_ndevs = 0; |
2b82032c YZ |
5002 | int i; |
5003 | ||
592d92ee LB |
5004 | em = get_chunk_map(fs_info, chunk_offset, 1); |
5005 | if (IS_ERR(em)) | |
2b82032c YZ |
5006 | return 1; |
5007 | ||
95617d69 | 5008 | map = em->map_lookup; |
2b82032c | 5009 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5010 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5011 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5012 | miss_ndevs++; |
5013 | continue; | |
5014 | } | |
ebbede42 AJ |
5015 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5016 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5017 | readonly = 1; |
d20983b4 | 5018 | goto end; |
2b82032c YZ |
5019 | } |
5020 | } | |
d20983b4 MX |
5021 | |
5022 | /* | |
5023 | * If the number of missing devices is larger than max errors, | |
5024 | * we can not write the data into that chunk successfully, so | |
5025 | * set it readonly. | |
5026 | */ | |
5027 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5028 | readonly = 1; | |
5029 | end: | |
0b86a832 | 5030 | free_extent_map(em); |
2b82032c | 5031 | return readonly; |
0b86a832 CM |
5032 | } |
5033 | ||
5034 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5035 | { | |
a8067e02 | 5036 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5037 | } |
5038 | ||
5039 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5040 | { | |
5041 | struct extent_map *em; | |
5042 | ||
d397712b | 5043 | while (1) { |
890871be | 5044 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5045 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5046 | if (em) | |
5047 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5048 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5049 | if (!em) |
5050 | break; | |
0b86a832 CM |
5051 | /* once for us */ |
5052 | free_extent_map(em); | |
5053 | /* once for the tree */ | |
5054 | free_extent_map(em); | |
5055 | } | |
5056 | } | |
5057 | ||
5d964051 | 5058 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5059 | { |
5060 | struct extent_map *em; | |
5061 | struct map_lookup *map; | |
f188591e CM |
5062 | int ret; |
5063 | ||
592d92ee LB |
5064 | em = get_chunk_map(fs_info, logical, len); |
5065 | if (IS_ERR(em)) | |
5066 | /* | |
5067 | * We could return errors for these cases, but that could get | |
5068 | * ugly and we'd probably do the same thing which is just not do | |
5069 | * anything else and exit, so return 1 so the callers don't try | |
5070 | * to use other copies. | |
5071 | */ | |
fb7669b5 | 5072 | return 1; |
fb7669b5 | 5073 | |
95617d69 | 5074 | map = em->map_lookup; |
f188591e CM |
5075 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5076 | ret = map->num_stripes; | |
321aecc6 CM |
5077 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5078 | ret = map->sub_stripes; | |
53b381b3 DW |
5079 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5080 | ret = 2; | |
5081 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5082 | /* |
5083 | * There could be two corrupted data stripes, we need | |
5084 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5085 | * |
8810f751 LB |
5086 | * Fail a stripe at a time on every retry except the |
5087 | * stripe under reconstruction. | |
5088 | */ | |
5089 | ret = map->num_stripes; | |
f188591e CM |
5090 | else |
5091 | ret = 1; | |
5092 | free_extent_map(em); | |
ad6d620e | 5093 | |
7e79cb86 | 5094 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
6fad823f LB |
5095 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5096 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5097 | ret++; |
7e79cb86 | 5098 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
ad6d620e | 5099 | |
f188591e CM |
5100 | return ret; |
5101 | } | |
5102 | ||
2ff7e61e | 5103 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5104 | u64 logical) |
5105 | { | |
5106 | struct extent_map *em; | |
5107 | struct map_lookup *map; | |
0b246afa | 5108 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5109 | |
592d92ee | 5110 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5111 | |
69f03f13 NB |
5112 | if (!WARN_ON(IS_ERR(em))) { |
5113 | map = em->map_lookup; | |
5114 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5115 | len = map->stripe_len * nr_data_stripes(map); | |
5116 | free_extent_map(em); | |
5117 | } | |
53b381b3 DW |
5118 | return len; |
5119 | } | |
5120 | ||
e4ff5fb5 | 5121 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5122 | { |
5123 | struct extent_map *em; | |
5124 | struct map_lookup *map; | |
53b381b3 DW |
5125 | int ret = 0; |
5126 | ||
592d92ee | 5127 | em = get_chunk_map(fs_info, logical, len); |
53b381b3 | 5128 | |
69f03f13 NB |
5129 | if(!WARN_ON(IS_ERR(em))) { |
5130 | map = em->map_lookup; | |
5131 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5132 | ret = 1; | |
5133 | free_extent_map(em); | |
5134 | } | |
53b381b3 DW |
5135 | return ret; |
5136 | } | |
5137 | ||
30d9861f | 5138 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5139 | struct map_lookup *map, int first, |
8ba0ae78 | 5140 | int dev_replace_is_ongoing) |
dfe25020 CM |
5141 | { |
5142 | int i; | |
99f92a7c | 5143 | int num_stripes; |
8ba0ae78 | 5144 | int preferred_mirror; |
30d9861f SB |
5145 | int tolerance; |
5146 | struct btrfs_device *srcdev; | |
5147 | ||
99f92a7c AJ |
5148 | ASSERT((map->type & |
5149 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5150 | ||
5151 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5152 | num_stripes = map->sub_stripes; | |
5153 | else | |
5154 | num_stripes = map->num_stripes; | |
5155 | ||
8ba0ae78 AJ |
5156 | preferred_mirror = first + current->pid % num_stripes; |
5157 | ||
30d9861f SB |
5158 | if (dev_replace_is_ongoing && |
5159 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5160 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5161 | srcdev = fs_info->dev_replace.srcdev; | |
5162 | else | |
5163 | srcdev = NULL; | |
5164 | ||
5165 | /* | |
5166 | * try to avoid the drive that is the source drive for a | |
5167 | * dev-replace procedure, only choose it if no other non-missing | |
5168 | * mirror is available | |
5169 | */ | |
5170 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5171 | if (map->stripes[preferred_mirror].dev->bdev && |
5172 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5173 | return preferred_mirror; | |
99f92a7c | 5174 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5175 | if (map->stripes[i].dev->bdev && |
5176 | (tolerance || map->stripes[i].dev != srcdev)) | |
5177 | return i; | |
5178 | } | |
dfe25020 | 5179 | } |
30d9861f | 5180 | |
dfe25020 CM |
5181 | /* we couldn't find one that doesn't fail. Just return something |
5182 | * and the io error handling code will clean up eventually | |
5183 | */ | |
8ba0ae78 | 5184 | return preferred_mirror; |
dfe25020 CM |
5185 | } |
5186 | ||
53b381b3 DW |
5187 | static inline int parity_smaller(u64 a, u64 b) |
5188 | { | |
5189 | return a > b; | |
5190 | } | |
5191 | ||
5192 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5193 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5194 | { |
5195 | struct btrfs_bio_stripe s; | |
5196 | int i; | |
5197 | u64 l; | |
5198 | int again = 1; | |
5199 | ||
5200 | while (again) { | |
5201 | again = 0; | |
cc7539ed | 5202 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5203 | if (parity_smaller(bbio->raid_map[i], |
5204 | bbio->raid_map[i+1])) { | |
53b381b3 | 5205 | s = bbio->stripes[i]; |
8e5cfb55 | 5206 | l = bbio->raid_map[i]; |
53b381b3 | 5207 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5208 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5209 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5210 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5211 | |
53b381b3 DW |
5212 | again = 1; |
5213 | } | |
5214 | } | |
5215 | } | |
5216 | } | |
5217 | ||
6e9606d2 ZL |
5218 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5219 | { | |
5220 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5221 | /* the size of the btrfs_bio */ |
6e9606d2 | 5222 | sizeof(struct btrfs_bio) + |
e57cf21e | 5223 | /* plus the variable array for the stripes */ |
6e9606d2 | 5224 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5225 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5226 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5227 | /* |
5228 | * plus the raid_map, which includes both the tgt dev | |
5229 | * and the stripes | |
5230 | */ | |
5231 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5232 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5233 | |
5234 | atomic_set(&bbio->error, 0); | |
140475ae | 5235 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5236 | |
5237 | return bbio; | |
5238 | } | |
5239 | ||
5240 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5241 | { | |
140475ae ER |
5242 | WARN_ON(!refcount_read(&bbio->refs)); |
5243 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5244 | } |
5245 | ||
5246 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5247 | { | |
5248 | if (!bbio) | |
5249 | return; | |
140475ae | 5250 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5251 | kfree(bbio); |
5252 | } | |
5253 | ||
0b3d4cd3 LB |
5254 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5255 | /* | |
5256 | * Please note that, discard won't be sent to target device of device | |
5257 | * replace. | |
5258 | */ | |
5259 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5260 | u64 logical, u64 length, | |
5261 | struct btrfs_bio **bbio_ret) | |
5262 | { | |
5263 | struct extent_map *em; | |
5264 | struct map_lookup *map; | |
5265 | struct btrfs_bio *bbio; | |
5266 | u64 offset; | |
5267 | u64 stripe_nr; | |
5268 | u64 stripe_nr_end; | |
5269 | u64 stripe_end_offset; | |
5270 | u64 stripe_cnt; | |
5271 | u64 stripe_len; | |
5272 | u64 stripe_offset; | |
5273 | u64 num_stripes; | |
5274 | u32 stripe_index; | |
5275 | u32 factor = 0; | |
5276 | u32 sub_stripes = 0; | |
5277 | u64 stripes_per_dev = 0; | |
5278 | u32 remaining_stripes = 0; | |
5279 | u32 last_stripe = 0; | |
5280 | int ret = 0; | |
5281 | int i; | |
5282 | ||
5283 | /* discard always return a bbio */ | |
5284 | ASSERT(bbio_ret); | |
5285 | ||
5286 | em = get_chunk_map(fs_info, logical, length); | |
5287 | if (IS_ERR(em)) | |
5288 | return PTR_ERR(em); | |
5289 | ||
5290 | map = em->map_lookup; | |
5291 | /* we don't discard raid56 yet */ | |
5292 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5293 | ret = -EOPNOTSUPP; | |
5294 | goto out; | |
5295 | } | |
5296 | ||
5297 | offset = logical - em->start; | |
5298 | length = min_t(u64, em->len - offset, length); | |
5299 | ||
5300 | stripe_len = map->stripe_len; | |
5301 | /* | |
5302 | * stripe_nr counts the total number of stripes we have to stride | |
5303 | * to get to this block | |
5304 | */ | |
5305 | stripe_nr = div64_u64(offset, stripe_len); | |
5306 | ||
5307 | /* stripe_offset is the offset of this block in its stripe */ | |
5308 | stripe_offset = offset - stripe_nr * stripe_len; | |
5309 | ||
5310 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5311 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5312 | stripe_cnt = stripe_nr_end - stripe_nr; |
5313 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5314 | (offset + length); | |
5315 | /* | |
5316 | * after this, stripe_nr is the number of stripes on this | |
5317 | * device we have to walk to find the data, and stripe_index is | |
5318 | * the number of our device in the stripe array | |
5319 | */ | |
5320 | num_stripes = 1; | |
5321 | stripe_index = 0; | |
5322 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5323 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5324 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5325 | sub_stripes = 1; | |
5326 | else | |
5327 | sub_stripes = map->sub_stripes; | |
5328 | ||
5329 | factor = map->num_stripes / sub_stripes; | |
5330 | num_stripes = min_t(u64, map->num_stripes, | |
5331 | sub_stripes * stripe_cnt); | |
5332 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5333 | stripe_index *= sub_stripes; | |
5334 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5335 | &remaining_stripes); | |
5336 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5337 | last_stripe *= sub_stripes; | |
5338 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5339 | BTRFS_BLOCK_GROUP_DUP)) { | |
5340 | num_stripes = map->num_stripes; | |
5341 | } else { | |
5342 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5343 | &stripe_index); | |
5344 | } | |
5345 | ||
5346 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5347 | if (!bbio) { | |
5348 | ret = -ENOMEM; | |
5349 | goto out; | |
5350 | } | |
5351 | ||
5352 | for (i = 0; i < num_stripes; i++) { | |
5353 | bbio->stripes[i].physical = | |
5354 | map->stripes[stripe_index].physical + | |
5355 | stripe_offset + stripe_nr * map->stripe_len; | |
5356 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5357 | ||
5358 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5359 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5360 | bbio->stripes[i].length = stripes_per_dev * | |
5361 | map->stripe_len; | |
5362 | ||
5363 | if (i / sub_stripes < remaining_stripes) | |
5364 | bbio->stripes[i].length += | |
5365 | map->stripe_len; | |
5366 | ||
5367 | /* | |
5368 | * Special for the first stripe and | |
5369 | * the last stripe: | |
5370 | * | |
5371 | * |-------|...|-------| | |
5372 | * |----------| | |
5373 | * off end_off | |
5374 | */ | |
5375 | if (i < sub_stripes) | |
5376 | bbio->stripes[i].length -= | |
5377 | stripe_offset; | |
5378 | ||
5379 | if (stripe_index >= last_stripe && | |
5380 | stripe_index <= (last_stripe + | |
5381 | sub_stripes - 1)) | |
5382 | bbio->stripes[i].length -= | |
5383 | stripe_end_offset; | |
5384 | ||
5385 | if (i == sub_stripes - 1) | |
5386 | stripe_offset = 0; | |
5387 | } else { | |
5388 | bbio->stripes[i].length = length; | |
5389 | } | |
5390 | ||
5391 | stripe_index++; | |
5392 | if (stripe_index == map->num_stripes) { | |
5393 | stripe_index = 0; | |
5394 | stripe_nr++; | |
5395 | } | |
5396 | } | |
5397 | ||
5398 | *bbio_ret = bbio; | |
5399 | bbio->map_type = map->type; | |
5400 | bbio->num_stripes = num_stripes; | |
5401 | out: | |
5402 | free_extent_map(em); | |
5403 | return ret; | |
5404 | } | |
5405 | ||
5ab56090 LB |
5406 | /* |
5407 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5408 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5409 | * left cursor can also be read from the target drive. | |
5410 | * | |
5411 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5412 | * array of stripes. | |
5413 | * For READ, it also needs to be supported using the same mirror number. | |
5414 | * | |
5415 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5416 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5417 | * case. | |
5418 | */ | |
5419 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5420 | u64 logical, u64 length, | |
5421 | u64 srcdev_devid, int *mirror_num, | |
5422 | u64 *physical) | |
5423 | { | |
5424 | struct btrfs_bio *bbio = NULL; | |
5425 | int num_stripes; | |
5426 | int index_srcdev = 0; | |
5427 | int found = 0; | |
5428 | u64 physical_of_found = 0; | |
5429 | int i; | |
5430 | int ret = 0; | |
5431 | ||
5432 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5433 | logical, &length, &bbio, 0, 0); | |
5434 | if (ret) { | |
5435 | ASSERT(bbio == NULL); | |
5436 | return ret; | |
5437 | } | |
5438 | ||
5439 | num_stripes = bbio->num_stripes; | |
5440 | if (*mirror_num > num_stripes) { | |
5441 | /* | |
5442 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5443 | * that means that the requested area is not left of the left | |
5444 | * cursor | |
5445 | */ | |
5446 | btrfs_put_bbio(bbio); | |
5447 | return -EIO; | |
5448 | } | |
5449 | ||
5450 | /* | |
5451 | * process the rest of the function using the mirror_num of the source | |
5452 | * drive. Therefore look it up first. At the end, patch the device | |
5453 | * pointer to the one of the target drive. | |
5454 | */ | |
5455 | for (i = 0; i < num_stripes; i++) { | |
5456 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5457 | continue; | |
5458 | ||
5459 | /* | |
5460 | * In case of DUP, in order to keep it simple, only add the | |
5461 | * mirror with the lowest physical address | |
5462 | */ | |
5463 | if (found && | |
5464 | physical_of_found <= bbio->stripes[i].physical) | |
5465 | continue; | |
5466 | ||
5467 | index_srcdev = i; | |
5468 | found = 1; | |
5469 | physical_of_found = bbio->stripes[i].physical; | |
5470 | } | |
5471 | ||
5472 | btrfs_put_bbio(bbio); | |
5473 | ||
5474 | ASSERT(found); | |
5475 | if (!found) | |
5476 | return -EIO; | |
5477 | ||
5478 | *mirror_num = index_srcdev + 1; | |
5479 | *physical = physical_of_found; | |
5480 | return ret; | |
5481 | } | |
5482 | ||
73c0f228 LB |
5483 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5484 | struct btrfs_bio **bbio_ret, | |
5485 | struct btrfs_dev_replace *dev_replace, | |
5486 | int *num_stripes_ret, int *max_errors_ret) | |
5487 | { | |
5488 | struct btrfs_bio *bbio = *bbio_ret; | |
5489 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5490 | int tgtdev_indexes = 0; | |
5491 | int num_stripes = *num_stripes_ret; | |
5492 | int max_errors = *max_errors_ret; | |
5493 | int i; | |
5494 | ||
5495 | if (op == BTRFS_MAP_WRITE) { | |
5496 | int index_where_to_add; | |
5497 | ||
5498 | /* | |
5499 | * duplicate the write operations while the dev replace | |
5500 | * procedure is running. Since the copying of the old disk to | |
5501 | * the new disk takes place at run time while the filesystem is | |
5502 | * mounted writable, the regular write operations to the old | |
5503 | * disk have to be duplicated to go to the new disk as well. | |
5504 | * | |
5505 | * Note that device->missing is handled by the caller, and that | |
5506 | * the write to the old disk is already set up in the stripes | |
5507 | * array. | |
5508 | */ | |
5509 | index_where_to_add = num_stripes; | |
5510 | for (i = 0; i < num_stripes; i++) { | |
5511 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5512 | /* write to new disk, too */ | |
5513 | struct btrfs_bio_stripe *new = | |
5514 | bbio->stripes + index_where_to_add; | |
5515 | struct btrfs_bio_stripe *old = | |
5516 | bbio->stripes + i; | |
5517 | ||
5518 | new->physical = old->physical; | |
5519 | new->length = old->length; | |
5520 | new->dev = dev_replace->tgtdev; | |
5521 | bbio->tgtdev_map[i] = index_where_to_add; | |
5522 | index_where_to_add++; | |
5523 | max_errors++; | |
5524 | tgtdev_indexes++; | |
5525 | } | |
5526 | } | |
5527 | num_stripes = index_where_to_add; | |
5528 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5529 | int index_srcdev = 0; | |
5530 | int found = 0; | |
5531 | u64 physical_of_found = 0; | |
5532 | ||
5533 | /* | |
5534 | * During the dev-replace procedure, the target drive can also | |
5535 | * be used to read data in case it is needed to repair a corrupt | |
5536 | * block elsewhere. This is possible if the requested area is | |
5537 | * left of the left cursor. In this area, the target drive is a | |
5538 | * full copy of the source drive. | |
5539 | */ | |
5540 | for (i = 0; i < num_stripes; i++) { | |
5541 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5542 | /* | |
5543 | * In case of DUP, in order to keep it simple, | |
5544 | * only add the mirror with the lowest physical | |
5545 | * address | |
5546 | */ | |
5547 | if (found && | |
5548 | physical_of_found <= | |
5549 | bbio->stripes[i].physical) | |
5550 | continue; | |
5551 | index_srcdev = i; | |
5552 | found = 1; | |
5553 | physical_of_found = bbio->stripes[i].physical; | |
5554 | } | |
5555 | } | |
5556 | if (found) { | |
5557 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5558 | bbio->stripes + num_stripes; | |
5559 | ||
5560 | tgtdev_stripe->physical = physical_of_found; | |
5561 | tgtdev_stripe->length = | |
5562 | bbio->stripes[index_srcdev].length; | |
5563 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5564 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5565 | ||
5566 | tgtdev_indexes++; | |
5567 | num_stripes++; | |
5568 | } | |
5569 | } | |
5570 | ||
5571 | *num_stripes_ret = num_stripes; | |
5572 | *max_errors_ret = max_errors; | |
5573 | bbio->num_tgtdevs = tgtdev_indexes; | |
5574 | *bbio_ret = bbio; | |
5575 | } | |
5576 | ||
2b19a1fe LB |
5577 | static bool need_full_stripe(enum btrfs_map_op op) |
5578 | { | |
5579 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5580 | } | |
5581 | ||
cf8cddd3 CH |
5582 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5583 | enum btrfs_map_op op, | |
f2d8d74d | 5584 | u64 logical, u64 *length, |
a1d3c478 | 5585 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5586 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5587 | { |
5588 | struct extent_map *em; | |
5589 | struct map_lookup *map; | |
0b86a832 | 5590 | u64 offset; |
593060d7 CM |
5591 | u64 stripe_offset; |
5592 | u64 stripe_nr; | |
53b381b3 | 5593 | u64 stripe_len; |
9d644a62 | 5594 | u32 stripe_index; |
cea9e445 | 5595 | int i; |
de11cc12 | 5596 | int ret = 0; |
f2d8d74d | 5597 | int num_stripes; |
a236aed1 | 5598 | int max_errors = 0; |
2c8cdd6e | 5599 | int tgtdev_indexes = 0; |
a1d3c478 | 5600 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5601 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5602 | int dev_replace_is_ongoing = 0; | |
5603 | int num_alloc_stripes; | |
ad6d620e SB |
5604 | int patch_the_first_stripe_for_dev_replace = 0; |
5605 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5606 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5607 | |
0b3d4cd3 LB |
5608 | if (op == BTRFS_MAP_DISCARD) |
5609 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5610 | *length, bbio_ret); | |
5611 | ||
592d92ee LB |
5612 | em = get_chunk_map(fs_info, logical, *length); |
5613 | if (IS_ERR(em)) | |
5614 | return PTR_ERR(em); | |
0b86a832 | 5615 | |
95617d69 | 5616 | map = em->map_lookup; |
0b86a832 | 5617 | offset = logical - em->start; |
593060d7 | 5618 | |
53b381b3 | 5619 | stripe_len = map->stripe_len; |
593060d7 CM |
5620 | stripe_nr = offset; |
5621 | /* | |
5622 | * stripe_nr counts the total number of stripes we have to stride | |
5623 | * to get to this block | |
5624 | */ | |
47c5713f | 5625 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5626 | |
53b381b3 | 5627 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5628 | if (offset < stripe_offset) { |
5d163e0e JM |
5629 | btrfs_crit(fs_info, |
5630 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5631 | stripe_offset, offset, em->start, logical, |
5632 | stripe_len); | |
5633 | free_extent_map(em); | |
5634 | return -EINVAL; | |
5635 | } | |
593060d7 CM |
5636 | |
5637 | /* stripe_offset is the offset of this block in its stripe*/ | |
5638 | stripe_offset = offset - stripe_offset; | |
5639 | ||
53b381b3 | 5640 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5641 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5642 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5643 | raid56_full_stripe_start = offset; | |
5644 | ||
5645 | /* allow a write of a full stripe, but make sure we don't | |
5646 | * allow straddling of stripes | |
5647 | */ | |
47c5713f DS |
5648 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5649 | full_stripe_len); | |
53b381b3 DW |
5650 | raid56_full_stripe_start *= full_stripe_len; |
5651 | } | |
5652 | ||
0b3d4cd3 | 5653 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5654 | u64 max_len; |
5655 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5656 | For other RAID types and for RAID[56] reads, just allow a single | |
5657 | stripe (on a single disk). */ | |
ffe2d203 | 5658 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5659 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5660 | max_len = stripe_len * nr_data_stripes(map) - |
5661 | (offset - raid56_full_stripe_start); | |
5662 | } else { | |
5663 | /* we limit the length of each bio to what fits in a stripe */ | |
5664 | max_len = stripe_len - stripe_offset; | |
5665 | } | |
5666 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5667 | } else { |
5668 | *length = em->len - offset; | |
5669 | } | |
f2d8d74d | 5670 | |
53b381b3 DW |
5671 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5672 | it cares about is the length */ | |
a1d3c478 | 5673 | if (!bbio_ret) |
cea9e445 CM |
5674 | goto out; |
5675 | ||
7e79cb86 | 5676 | btrfs_dev_replace_read_lock(dev_replace); |
472262f3 SB |
5677 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5678 | if (!dev_replace_is_ongoing) | |
7e79cb86 | 5679 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 LB |
5680 | else |
5681 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5682 | |
ad6d620e | 5683 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5684 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5685 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5686 | dev_replace->srcdev->devid, | |
5687 | &mirror_num, | |
5688 | &physical_to_patch_in_first_stripe); | |
5689 | if (ret) | |
ad6d620e | 5690 | goto out; |
5ab56090 LB |
5691 | else |
5692 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5693 | } else if (mirror_num > map->num_stripes) { |
5694 | mirror_num = 0; | |
5695 | } | |
5696 | ||
f2d8d74d | 5697 | num_stripes = 1; |
cea9e445 | 5698 | stripe_index = 0; |
fce3bb9a | 5699 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5700 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5701 | &stripe_index); | |
de483734 | 5702 | if (!need_full_stripe(op)) |
28e1cc7d | 5703 | mirror_num = 1; |
fce3bb9a | 5704 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 5705 | if (need_full_stripe(op)) |
f2d8d74d | 5706 | num_stripes = map->num_stripes; |
2fff734f | 5707 | else if (mirror_num) |
f188591e | 5708 | stripe_index = mirror_num - 1; |
dfe25020 | 5709 | else { |
30d9861f | 5710 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 5711 | dev_replace_is_ongoing); |
a1d3c478 | 5712 | mirror_num = stripe_index + 1; |
dfe25020 | 5713 | } |
2fff734f | 5714 | |
611f0e00 | 5715 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5716 | if (need_full_stripe(op)) { |
f2d8d74d | 5717 | num_stripes = map->num_stripes; |
a1d3c478 | 5718 | } else if (mirror_num) { |
f188591e | 5719 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5720 | } else { |
5721 | mirror_num = 1; | |
5722 | } | |
2fff734f | 5723 | |
321aecc6 | 5724 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5725 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5726 | |
47c5713f | 5727 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5728 | stripe_index *= map->sub_stripes; |
5729 | ||
de483734 | 5730 | if (need_full_stripe(op)) |
f2d8d74d | 5731 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5732 | else if (mirror_num) |
5733 | stripe_index += mirror_num - 1; | |
dfe25020 | 5734 | else { |
3e74317a | 5735 | int old_stripe_index = stripe_index; |
30d9861f SB |
5736 | stripe_index = find_live_mirror(fs_info, map, |
5737 | stripe_index, | |
30d9861f | 5738 | dev_replace_is_ongoing); |
3e74317a | 5739 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5740 | } |
53b381b3 | 5741 | |
ffe2d203 | 5742 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5743 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5744 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5745 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5746 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5747 | |
5748 | /* RAID[56] write or recovery. Return all stripes */ | |
5749 | num_stripes = map->num_stripes; | |
5750 | max_errors = nr_parity_stripes(map); | |
5751 | ||
53b381b3 DW |
5752 | *length = map->stripe_len; |
5753 | stripe_index = 0; | |
5754 | stripe_offset = 0; | |
5755 | } else { | |
5756 | /* | |
5757 | * Mirror #0 or #1 means the original data block. | |
5758 | * Mirror #2 is RAID5 parity block. | |
5759 | * Mirror #3 is RAID6 Q block. | |
5760 | */ | |
47c5713f DS |
5761 | stripe_nr = div_u64_rem(stripe_nr, |
5762 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5763 | if (mirror_num > 1) |
5764 | stripe_index = nr_data_stripes(map) + | |
5765 | mirror_num - 2; | |
5766 | ||
5767 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5768 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5769 | &stripe_index); | |
de483734 | 5770 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 5771 | mirror_num = 1; |
53b381b3 | 5772 | } |
8790d502 CM |
5773 | } else { |
5774 | /* | |
47c5713f DS |
5775 | * after this, stripe_nr is the number of stripes on this |
5776 | * device we have to walk to find the data, and stripe_index is | |
5777 | * the number of our device in the stripe array | |
8790d502 | 5778 | */ |
47c5713f DS |
5779 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5780 | &stripe_index); | |
a1d3c478 | 5781 | mirror_num = stripe_index + 1; |
8790d502 | 5782 | } |
e042d1ec | 5783 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5784 | btrfs_crit(fs_info, |
5785 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5786 | stripe_index, map->num_stripes); |
5787 | ret = -EINVAL; | |
5788 | goto out; | |
5789 | } | |
cea9e445 | 5790 | |
472262f3 | 5791 | num_alloc_stripes = num_stripes; |
6fad823f | 5792 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5793 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5794 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5795 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5796 | num_alloc_stripes++; |
2c8cdd6e | 5797 | tgtdev_indexes = num_stripes; |
ad6d620e | 5798 | } |
2c8cdd6e | 5799 | |
6e9606d2 | 5800 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5801 | if (!bbio) { |
5802 | ret = -ENOMEM; | |
5803 | goto out; | |
5804 | } | |
6fad823f | 5805 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5806 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5807 | |
8e5cfb55 | 5808 | /* build raid_map */ |
2b19a1fe LB |
5809 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5810 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5811 | u64 tmp; |
9d644a62 | 5812 | unsigned rot; |
8e5cfb55 ZL |
5813 | |
5814 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5815 | sizeof(struct btrfs_bio_stripe) * | |
5816 | num_alloc_stripes + | |
5817 | sizeof(int) * tgtdev_indexes); | |
5818 | ||
5819 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5820 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5821 | |
5822 | /* Fill in the logical address of each stripe */ | |
5823 | tmp = stripe_nr * nr_data_stripes(map); | |
5824 | for (i = 0; i < nr_data_stripes(map); i++) | |
5825 | bbio->raid_map[(i+rot) % num_stripes] = | |
5826 | em->start + (tmp + i) * map->stripe_len; | |
5827 | ||
5828 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5829 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5830 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5831 | RAID6_Q_STRIPE; | |
5832 | } | |
5833 | ||
b89203f7 | 5834 | |
0b3d4cd3 LB |
5835 | for (i = 0; i < num_stripes; i++) { |
5836 | bbio->stripes[i].physical = | |
5837 | map->stripes[stripe_index].physical + | |
5838 | stripe_offset + | |
5839 | stripe_nr * map->stripe_len; | |
5840 | bbio->stripes[i].dev = | |
5841 | map->stripes[stripe_index].dev; | |
5842 | stripe_index++; | |
593060d7 | 5843 | } |
de11cc12 | 5844 | |
2b19a1fe | 5845 | if (need_full_stripe(op)) |
d20983b4 | 5846 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5847 | |
8e5cfb55 ZL |
5848 | if (bbio->raid_map) |
5849 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5850 | |
73c0f228 | 5851 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5852 | need_full_stripe(op)) { |
73c0f228 LB |
5853 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5854 | &max_errors); | |
472262f3 SB |
5855 | } |
5856 | ||
de11cc12 | 5857 | *bbio_ret = bbio; |
10f11900 | 5858 | bbio->map_type = map->type; |
de11cc12 LZ |
5859 | bbio->num_stripes = num_stripes; |
5860 | bbio->max_errors = max_errors; | |
5861 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5862 | |
5863 | /* | |
5864 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5865 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5866 | * available as a mirror | |
5867 | */ | |
5868 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5869 | WARN_ON(num_stripes > 1); | |
5870 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5871 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5872 | bbio->mirror_num = map->num_stripes + 1; | |
5873 | } | |
cea9e445 | 5874 | out: |
73beece9 LB |
5875 | if (dev_replace_is_ongoing) { |
5876 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
7e79cb86 | 5877 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 | 5878 | } |
0b86a832 | 5879 | free_extent_map(em); |
de11cc12 | 5880 | return ret; |
0b86a832 CM |
5881 | } |
5882 | ||
cf8cddd3 | 5883 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 5884 | u64 logical, u64 *length, |
a1d3c478 | 5885 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5886 | { |
b3d3fa51 | 5887 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5888 | mirror_num, 0); |
f2d8d74d CM |
5889 | } |
5890 | ||
af8e2d1d | 5891 | /* For Scrub/replace */ |
cf8cddd3 | 5892 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 5893 | u64 logical, u64 *length, |
825ad4c9 | 5894 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 5895 | { |
825ad4c9 | 5896 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
5897 | } |
5898 | ||
63a9c7b9 NB |
5899 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
5900 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 5901 | { |
a512bbf8 YZ |
5902 | struct extent_map *em; |
5903 | struct map_lookup *map; | |
5904 | u64 *buf; | |
5905 | u64 bytenr; | |
5906 | u64 length; | |
5907 | u64 stripe_nr; | |
53b381b3 | 5908 | u64 rmap_len; |
a512bbf8 YZ |
5909 | int i, j, nr = 0; |
5910 | ||
592d92ee LB |
5911 | em = get_chunk_map(fs_info, chunk_start, 1); |
5912 | if (IS_ERR(em)) | |
835d974f | 5913 | return -EIO; |
835d974f | 5914 | |
95617d69 | 5915 | map = em->map_lookup; |
a512bbf8 | 5916 | length = em->len; |
53b381b3 DW |
5917 | rmap_len = map->stripe_len; |
5918 | ||
a512bbf8 | 5919 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5920 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5921 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5922 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5923 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5924 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5925 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5926 | } | |
a512bbf8 | 5927 | |
31e818fe | 5928 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5929 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5930 | |
5931 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
5932 | if (map->stripes[i].physical > physical || |
5933 | map->stripes[i].physical + length <= physical) | |
5934 | continue; | |
5935 | ||
5936 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 5937 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5938 | |
5939 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5940 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5941 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5942 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5943 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5944 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5945 | * Alternatively, just use rmap_len below instead of | |
5946 | * map->stripe_len */ | |
5947 | ||
5948 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5949 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5950 | for (j = 0; j < nr; j++) { |
5951 | if (buf[j] == bytenr) | |
5952 | break; | |
5953 | } | |
934d375b CM |
5954 | if (j == nr) { |
5955 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5956 | buf[nr++] = bytenr; |
934d375b | 5957 | } |
a512bbf8 YZ |
5958 | } |
5959 | ||
a512bbf8 YZ |
5960 | *logical = buf; |
5961 | *naddrs = nr; | |
53b381b3 | 5962 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5963 | |
5964 | free_extent_map(em); | |
5965 | return 0; | |
f2d8d74d CM |
5966 | } |
5967 | ||
4246a0b6 | 5968 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 5969 | { |
326e1dbb MS |
5970 | bio->bi_private = bbio->private; |
5971 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 5972 | bio_endio(bio); |
326e1dbb | 5973 | |
6e9606d2 | 5974 | btrfs_put_bbio(bbio); |
8408c716 MX |
5975 | } |
5976 | ||
4246a0b6 | 5977 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 5978 | { |
9be3395b | 5979 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 5980 | int is_orig_bio = 0; |
8790d502 | 5981 | |
4e4cbee9 | 5982 | if (bio->bi_status) { |
a1d3c478 | 5983 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
5984 | if (bio->bi_status == BLK_STS_IOERR || |
5985 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 5986 | unsigned int stripe_index = |
9be3395b | 5987 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 5988 | struct btrfs_device *dev; |
442a4f63 SB |
5989 | |
5990 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5991 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 5992 | if (dev->bdev) { |
37226b21 | 5993 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 5994 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
5995 | BTRFS_DEV_STAT_WRITE_ERRS); |
5996 | else | |
1cb34c8e | 5997 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 5998 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 5999 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6000 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6001 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6002 | } |
442a4f63 SB |
6003 | } |
6004 | } | |
8790d502 | 6005 | |
a1d3c478 | 6006 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6007 | is_orig_bio = 1; |
6008 | ||
c404e0dc MX |
6009 | btrfs_bio_counter_dec(bbio->fs_info); |
6010 | ||
a1d3c478 | 6011 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6012 | if (!is_orig_bio) { |
6013 | bio_put(bio); | |
a1d3c478 | 6014 | bio = bbio->orig_bio; |
7d2b4daa | 6015 | } |
c7b22bb1 | 6016 | |
9be3395b | 6017 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6018 | /* only send an error to the higher layers if it is |
53b381b3 | 6019 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6020 | */ |
a1d3c478 | 6021 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6022 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6023 | } else { |
1259ab75 CM |
6024 | /* |
6025 | * this bio is actually up to date, we didn't | |
6026 | * go over the max number of errors | |
6027 | */ | |
2dbe0c77 | 6028 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6029 | } |
c55f1396 | 6030 | |
4246a0b6 | 6031 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6032 | } else if (!is_orig_bio) { |
8790d502 CM |
6033 | bio_put(bio); |
6034 | } | |
8790d502 CM |
6035 | } |
6036 | ||
8b712842 CM |
6037 | /* |
6038 | * see run_scheduled_bios for a description of why bios are collected for | |
6039 | * async submit. | |
6040 | * | |
6041 | * This will add one bio to the pending list for a device and make sure | |
6042 | * the work struct is scheduled. | |
6043 | */ | |
2ff7e61e | 6044 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6045 | struct bio *bio) |
8b712842 | 6046 | { |
0b246afa | 6047 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6048 | int should_queue = 1; |
ffbd517d | 6049 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6050 | |
e6e674bd AJ |
6051 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) || |
6052 | !device->bdev) { | |
4246a0b6 | 6053 | bio_io_error(bio); |
53b381b3 DW |
6054 | return; |
6055 | } | |
6056 | ||
8b712842 | 6057 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6058 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6059 | btrfsic_submit_bio(bio); |
143bede5 | 6060 | return; |
8b712842 CM |
6061 | } |
6062 | ||
492bb6de | 6063 | WARN_ON(bio->bi_next); |
8b712842 | 6064 | bio->bi_next = NULL; |
8b712842 CM |
6065 | |
6066 | spin_lock(&device->io_lock); | |
67f055c7 | 6067 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6068 | pending_bios = &device->pending_sync_bios; |
6069 | else | |
6070 | pending_bios = &device->pending_bios; | |
8b712842 | 6071 | |
ffbd517d CM |
6072 | if (pending_bios->tail) |
6073 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6074 | |
ffbd517d CM |
6075 | pending_bios->tail = bio; |
6076 | if (!pending_bios->head) | |
6077 | pending_bios->head = bio; | |
8b712842 CM |
6078 | if (device->running_pending) |
6079 | should_queue = 0; | |
6080 | ||
6081 | spin_unlock(&device->io_lock); | |
6082 | ||
6083 | if (should_queue) | |
0b246afa | 6084 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6085 | } |
6086 | ||
2ff7e61e JM |
6087 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6088 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6089 | { |
6090 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6091 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6092 | |
6093 | bio->bi_private = bbio; | |
9be3395b | 6094 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6095 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6096 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6097 | btrfs_debug_in_rcu(fs_info, |
6098 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6099 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6100 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6101 | bio->bi_iter.bi_size); | |
74d46992 | 6102 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6103 | |
2ff7e61e | 6104 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6105 | |
de1ee92a | 6106 | if (async) |
2ff7e61e | 6107 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6108 | else |
4e49ea4a | 6109 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6110 | } |
6111 | ||
de1ee92a JB |
6112 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6113 | { | |
6114 | atomic_inc(&bbio->error); | |
6115 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6116 | /* Should be the original bio. */ |
8408c716 MX |
6117 | WARN_ON(bio != bbio->orig_bio); |
6118 | ||
9be3395b | 6119 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6120 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6121 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6122 | bio->bi_status = BLK_STS_IOERR; | |
6123 | else | |
6124 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6125 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6126 | } |
6127 | } | |
6128 | ||
58efbc9f OS |
6129 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6130 | int mirror_num, int async_submit) | |
0b86a832 | 6131 | { |
0b86a832 | 6132 | struct btrfs_device *dev; |
8790d502 | 6133 | struct bio *first_bio = bio; |
4f024f37 | 6134 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6135 | u64 length = 0; |
6136 | u64 map_length; | |
0b86a832 | 6137 | int ret; |
08da757d ZL |
6138 | int dev_nr; |
6139 | int total_devs; | |
a1d3c478 | 6140 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6141 | |
4f024f37 | 6142 | length = bio->bi_iter.bi_size; |
0b86a832 | 6143 | map_length = length; |
cea9e445 | 6144 | |
0b246afa | 6145 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6146 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6147 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6148 | if (ret) { |
0b246afa | 6149 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6150 | return errno_to_blk_status(ret); |
c404e0dc | 6151 | } |
cea9e445 | 6152 | |
a1d3c478 | 6153 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6154 | bbio->orig_bio = first_bio; |
6155 | bbio->private = first_bio->bi_private; | |
6156 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6157 | bbio->fs_info = fs_info; |
53b381b3 DW |
6158 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6159 | ||
ad1ba2a0 | 6160 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6161 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6162 | /* In this case, map_length has been set to the length of |
6163 | a single stripe; not the whole write */ | |
37226b21 | 6164 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6165 | ret = raid56_parity_write(fs_info, bio, bbio, |
6166 | map_length); | |
53b381b3 | 6167 | } else { |
2ff7e61e JM |
6168 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6169 | map_length, mirror_num, 1); | |
53b381b3 | 6170 | } |
4245215d | 6171 | |
0b246afa | 6172 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6173 | return errno_to_blk_status(ret); |
53b381b3 DW |
6174 | } |
6175 | ||
cea9e445 | 6176 | if (map_length < length) { |
0b246afa | 6177 | btrfs_crit(fs_info, |
5d163e0e JM |
6178 | "mapping failed logical %llu bio len %llu len %llu", |
6179 | logical, length, map_length); | |
cea9e445 CM |
6180 | BUG(); |
6181 | } | |
a1d3c478 | 6182 | |
08da757d | 6183 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6184 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6185 | if (!dev || !dev->bdev || |
ebbede42 AJ |
6186 | (bio_op(first_bio) == REQ_OP_WRITE && |
6187 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6188 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6189 | continue; |
6190 | } | |
6191 | ||
3aa8e074 | 6192 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6193 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6194 | else |
a1d3c478 | 6195 | bio = first_bio; |
de1ee92a | 6196 | |
2ff7e61e JM |
6197 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6198 | dev_nr, async_submit); | |
8790d502 | 6199 | } |
0b246afa | 6200 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6201 | return BLK_STS_OK; |
0b86a832 CM |
6202 | } |
6203 | ||
aa1b8cd4 | 6204 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6205 | u8 *uuid, u8 *fsid) |
0b86a832 | 6206 | { |
2b82032c YZ |
6207 | struct btrfs_device *device; |
6208 | struct btrfs_fs_devices *cur_devices; | |
6209 | ||
aa1b8cd4 | 6210 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6211 | while (cur_devices) { |
6212 | if (!fsid || | |
44880fdc | 6213 | !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6214 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6215 | if (device) |
6216 | return device; | |
6217 | } | |
6218 | cur_devices = cur_devices->seed; | |
6219 | } | |
6220 | return NULL; | |
0b86a832 CM |
6221 | } |
6222 | ||
2ff7e61e | 6223 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6224 | u64 devid, u8 *dev_uuid) |
6225 | { | |
6226 | struct btrfs_device *device; | |
dfe25020 | 6227 | |
12bd2fc0 ID |
6228 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6229 | if (IS_ERR(device)) | |
adfb69af | 6230 | return device; |
12bd2fc0 ID |
6231 | |
6232 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6233 | device->fs_devices = fs_devices; |
dfe25020 | 6234 | fs_devices->num_devices++; |
12bd2fc0 | 6235 | |
e6e674bd | 6236 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6237 | fs_devices->missing_devices++; |
12bd2fc0 | 6238 | |
dfe25020 CM |
6239 | return device; |
6240 | } | |
6241 | ||
12bd2fc0 ID |
6242 | /** |
6243 | * btrfs_alloc_device - allocate struct btrfs_device | |
6244 | * @fs_info: used only for generating a new devid, can be NULL if | |
6245 | * devid is provided (i.e. @devid != NULL). | |
6246 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6247 | * is generated. | |
6248 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6249 | * is generated. | |
6250 | * | |
6251 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6252 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6253 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6254 | */ |
6255 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6256 | const u64 *devid, | |
6257 | const u8 *uuid) | |
6258 | { | |
6259 | struct btrfs_device *dev; | |
6260 | u64 tmp; | |
6261 | ||
fae7f21c | 6262 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6263 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6264 | |
6265 | dev = __alloc_device(); | |
6266 | if (IS_ERR(dev)) | |
6267 | return dev; | |
6268 | ||
6269 | if (devid) | |
6270 | tmp = *devid; | |
6271 | else { | |
6272 | int ret; | |
6273 | ||
6274 | ret = find_next_devid(fs_info, &tmp); | |
6275 | if (ret) { | |
a425f9d4 | 6276 | btrfs_free_device(dev); |
12bd2fc0 ID |
6277 | return ERR_PTR(ret); |
6278 | } | |
6279 | } | |
6280 | dev->devid = tmp; | |
6281 | ||
6282 | if (uuid) | |
6283 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6284 | else | |
6285 | generate_random_uuid(dev->uuid); | |
6286 | ||
9e0af237 LB |
6287 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6288 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6289 | |
6290 | return dev; | |
6291 | } | |
6292 | ||
e06cd3dd | 6293 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6294 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6295 | struct extent_buffer *leaf, |
6296 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6297 | { |
0b86a832 | 6298 | u64 length; |
f04b772b | 6299 | u64 stripe_len; |
e06cd3dd LB |
6300 | u16 num_stripes; |
6301 | u16 sub_stripes; | |
6302 | u64 type; | |
315409b0 GJ |
6303 | u64 features; |
6304 | bool mixed = false; | |
0b86a832 | 6305 | |
e17cade2 | 6306 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6307 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6308 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6309 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6310 | type = btrfs_chunk_type(leaf, chunk); | |
6311 | ||
f04b772b | 6312 | if (!num_stripes) { |
0b246afa | 6313 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6314 | num_stripes); |
6315 | return -EIO; | |
6316 | } | |
0b246afa JM |
6317 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6318 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6319 | return -EIO; |
6320 | } | |
0b246afa JM |
6321 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6322 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6323 | btrfs_chunk_sector_size(leaf, chunk)); |
6324 | return -EIO; | |
6325 | } | |
0b246afa JM |
6326 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6327 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6328 | return -EIO; |
6329 | } | |
3d8da678 | 6330 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6331 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6332 | stripe_len); |
6333 | return -EIO; | |
6334 | } | |
6335 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6336 | type) { |
0b246afa | 6337 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6338 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6339 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6340 | btrfs_chunk_type(leaf, chunk)); | |
6341 | return -EIO; | |
6342 | } | |
315409b0 GJ |
6343 | |
6344 | if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) { | |
6345 | btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type); | |
6346 | return -EIO; | |
6347 | } | |
6348 | ||
6349 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
6350 | (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) { | |
6351 | btrfs_err(fs_info, | |
6352 | "system chunk with data or metadata type: 0x%llx", type); | |
6353 | return -EIO; | |
6354 | } | |
6355 | ||
6356 | features = btrfs_super_incompat_flags(fs_info->super_copy); | |
6357 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
6358 | mixed = true; | |
6359 | ||
6360 | if (!mixed) { | |
6361 | if ((type & BTRFS_BLOCK_GROUP_METADATA) && | |
6362 | (type & BTRFS_BLOCK_GROUP_DATA)) { | |
6363 | btrfs_err(fs_info, | |
6364 | "mixed chunk type in non-mixed mode: 0x%llx", type); | |
6365 | return -EIO; | |
6366 | } | |
6367 | } | |
6368 | ||
e06cd3dd LB |
6369 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6370 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6371 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6372 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6373 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6374 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6375 | num_stripes != 1)) { | |
0b246afa | 6376 | btrfs_err(fs_info, |
e06cd3dd LB |
6377 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6378 | num_stripes, sub_stripes, | |
6379 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6380 | return -EIO; | |
6381 | } | |
6382 | ||
6383 | return 0; | |
6384 | } | |
6385 | ||
5a2b8e60 | 6386 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6387 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6388 | { |
2b902dfc AJ |
6389 | if (error) |
6390 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6391 | devid, uuid); | |
6392 | else | |
6393 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6394 | devid, uuid); | |
5a2b8e60 AJ |
6395 | } |
6396 | ||
2ff7e61e | 6397 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6398 | struct extent_buffer *leaf, |
6399 | struct btrfs_chunk *chunk) | |
6400 | { | |
0b246afa | 6401 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6402 | struct map_lookup *map; |
6403 | struct extent_map *em; | |
6404 | u64 logical; | |
6405 | u64 length; | |
e06cd3dd LB |
6406 | u64 devid; |
6407 | u8 uuid[BTRFS_UUID_SIZE]; | |
6408 | int num_stripes; | |
6409 | int ret; | |
6410 | int i; | |
6411 | ||
6412 | logical = key->offset; | |
6413 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6414 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6415 | ||
2ff7e61e | 6416 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6417 | if (ret) |
6418 | return ret; | |
a061fc8d | 6419 | |
890871be | 6420 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6421 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6422 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6423 | |
6424 | /* already mapped? */ | |
6425 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6426 | free_extent_map(em); | |
0b86a832 CM |
6427 | return 0; |
6428 | } else if (em) { | |
6429 | free_extent_map(em); | |
6430 | } | |
0b86a832 | 6431 | |
172ddd60 | 6432 | em = alloc_extent_map(); |
0b86a832 CM |
6433 | if (!em) |
6434 | return -ENOMEM; | |
593060d7 | 6435 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6436 | if (!map) { |
6437 | free_extent_map(em); | |
6438 | return -ENOMEM; | |
6439 | } | |
6440 | ||
298a8f9c | 6441 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6442 | em->map_lookup = map; |
0b86a832 CM |
6443 | em->start = logical; |
6444 | em->len = length; | |
70c8a91c | 6445 | em->orig_start = 0; |
0b86a832 | 6446 | em->block_start = 0; |
c8b97818 | 6447 | em->block_len = em->len; |
0b86a832 | 6448 | |
593060d7 CM |
6449 | map->num_stripes = num_stripes; |
6450 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6451 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6452 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6453 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6454 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6455 | map->verified_stripes = 0; |
593060d7 CM |
6456 | for (i = 0; i < num_stripes; i++) { |
6457 | map->stripes[i].physical = | |
6458 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6459 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6460 | read_extent_buffer(leaf, uuid, (unsigned long) |
6461 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6462 | BTRFS_UUID_SIZE); | |
0b246afa | 6463 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6464 | uuid, NULL); |
3cdde224 | 6465 | if (!map->stripes[i].dev && |
0b246afa | 6466 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6467 | free_extent_map(em); |
2b902dfc | 6468 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6469 | return -ENOENT; |
593060d7 | 6470 | } |
dfe25020 CM |
6471 | if (!map->stripes[i].dev) { |
6472 | map->stripes[i].dev = | |
2ff7e61e JM |
6473 | add_missing_dev(fs_info->fs_devices, devid, |
6474 | uuid); | |
adfb69af | 6475 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6476 | free_extent_map(em); |
adfb69af AJ |
6477 | btrfs_err(fs_info, |
6478 | "failed to init missing dev %llu: %ld", | |
6479 | devid, PTR_ERR(map->stripes[i].dev)); | |
6480 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6481 | } |
2b902dfc | 6482 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6483 | } |
e12c9621 AJ |
6484 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6485 | &(map->stripes[i].dev->dev_state)); | |
6486 | ||
0b86a832 CM |
6487 | } |
6488 | ||
890871be | 6489 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6490 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6491 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6492 | if (ret < 0) { |
6493 | btrfs_err(fs_info, | |
6494 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6495 | em->start, em->len, ret); | |
6496 | } | |
0b86a832 CM |
6497 | free_extent_map(em); |
6498 | ||
64f64f43 | 6499 | return ret; |
0b86a832 CM |
6500 | } |
6501 | ||
143bede5 | 6502 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6503 | struct btrfs_dev_item *dev_item, |
6504 | struct btrfs_device *device) | |
6505 | { | |
6506 | unsigned long ptr; | |
0b86a832 CM |
6507 | |
6508 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6509 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6510 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6511 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6512 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6513 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6514 | device->type = btrfs_device_type(leaf, dev_item); |
6515 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6516 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6517 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6518 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6519 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6520 | |
410ba3a2 | 6521 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6522 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6523 | } |
6524 | ||
2ff7e61e | 6525 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6526 | u8 *fsid) |
2b82032c YZ |
6527 | { |
6528 | struct btrfs_fs_devices *fs_devices; | |
6529 | int ret; | |
6530 | ||
a32bf9a3 | 6531 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6532 | ASSERT(fsid); |
2b82032c | 6533 | |
0b246afa | 6534 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6535 | while (fs_devices) { |
44880fdc | 6536 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6537 | return fs_devices; |
6538 | ||
2b82032c YZ |
6539 | fs_devices = fs_devices->seed; |
6540 | } | |
6541 | ||
6542 | fs_devices = find_fsid(fsid); | |
6543 | if (!fs_devices) { | |
0b246afa | 6544 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6545 | return ERR_PTR(-ENOENT); |
6546 | ||
6547 | fs_devices = alloc_fs_devices(fsid); | |
6548 | if (IS_ERR(fs_devices)) | |
6549 | return fs_devices; | |
6550 | ||
6551 | fs_devices->seeding = 1; | |
6552 | fs_devices->opened = 1; | |
6553 | return fs_devices; | |
2b82032c | 6554 | } |
e4404d6e YZ |
6555 | |
6556 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6557 | if (IS_ERR(fs_devices)) |
6558 | return fs_devices; | |
2b82032c | 6559 | |
897fb573 | 6560 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6561 | if (ret) { |
6562 | free_fs_devices(fs_devices); | |
5f375835 | 6563 | fs_devices = ERR_PTR(ret); |
2b82032c | 6564 | goto out; |
48d28232 | 6565 | } |
2b82032c YZ |
6566 | |
6567 | if (!fs_devices->seeding) { | |
0226e0eb | 6568 | close_fs_devices(fs_devices); |
e4404d6e | 6569 | free_fs_devices(fs_devices); |
5f375835 | 6570 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6571 | goto out; |
6572 | } | |
6573 | ||
0b246afa JM |
6574 | fs_devices->seed = fs_info->fs_devices->seed; |
6575 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6576 | out: |
5f375835 | 6577 | return fs_devices; |
2b82032c YZ |
6578 | } |
6579 | ||
2ff7e61e | 6580 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6581 | struct extent_buffer *leaf, |
6582 | struct btrfs_dev_item *dev_item) | |
6583 | { | |
0b246afa | 6584 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6585 | struct btrfs_device *device; |
6586 | u64 devid; | |
6587 | int ret; | |
44880fdc | 6588 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6589 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6590 | ||
0b86a832 | 6591 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6592 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6593 | BTRFS_UUID_SIZE); |
1473b24e | 6594 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6595 | BTRFS_FSID_SIZE); |
2b82032c | 6596 | |
44880fdc | 6597 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6598 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6599 | if (IS_ERR(fs_devices)) |
6600 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6601 | } |
6602 | ||
0b246afa | 6603 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6604 | if (!device) { |
c5502451 | 6605 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6606 | btrfs_report_missing_device(fs_info, devid, |
6607 | dev_uuid, true); | |
45dbdbc9 | 6608 | return -ENOENT; |
c5502451 | 6609 | } |
2b82032c | 6610 | |
2ff7e61e | 6611 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6612 | if (IS_ERR(device)) { |
6613 | btrfs_err(fs_info, | |
6614 | "failed to add missing dev %llu: %ld", | |
6615 | devid, PTR_ERR(device)); | |
6616 | return PTR_ERR(device); | |
6617 | } | |
2b902dfc | 6618 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6619 | } else { |
c5502451 | 6620 | if (!device->bdev) { |
2b902dfc AJ |
6621 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6622 | btrfs_report_missing_device(fs_info, | |
6623 | devid, dev_uuid, true); | |
45dbdbc9 | 6624 | return -ENOENT; |
2b902dfc AJ |
6625 | } |
6626 | btrfs_report_missing_device(fs_info, devid, | |
6627 | dev_uuid, false); | |
c5502451 | 6628 | } |
5f375835 | 6629 | |
e6e674bd AJ |
6630 | if (!device->bdev && |
6631 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6632 | /* |
6633 | * this happens when a device that was properly setup | |
6634 | * in the device info lists suddenly goes bad. | |
6635 | * device->bdev is NULL, and so we have to set | |
6636 | * device->missing to one here | |
6637 | */ | |
5f375835 | 6638 | device->fs_devices->missing_devices++; |
e6e674bd | 6639 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6640 | } |
5f375835 MX |
6641 | |
6642 | /* Move the device to its own fs_devices */ | |
6643 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6644 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6645 | &device->dev_state)); | |
5f375835 MX |
6646 | |
6647 | list_move(&device->dev_list, &fs_devices->devices); | |
6648 | device->fs_devices->num_devices--; | |
6649 | fs_devices->num_devices++; | |
6650 | ||
6651 | device->fs_devices->missing_devices--; | |
6652 | fs_devices->missing_devices++; | |
6653 | ||
6654 | device->fs_devices = fs_devices; | |
6655 | } | |
2b82032c YZ |
6656 | } |
6657 | ||
0b246afa | 6658 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6659 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6660 | if (device->generation != |
6661 | btrfs_device_generation(leaf, dev_item)) | |
6662 | return -EINVAL; | |
6324fbf3 | 6663 | } |
0b86a832 CM |
6664 | |
6665 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6666 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6667 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6668 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6669 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6670 | atomic64_add(device->total_bytes - device->bytes_used, |
6671 | &fs_info->free_chunk_space); | |
2bf64758 | 6672 | } |
0b86a832 | 6673 | ret = 0; |
0b86a832 CM |
6674 | return ret; |
6675 | } | |
6676 | ||
6bccf3ab | 6677 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6678 | { |
6bccf3ab | 6679 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6680 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6681 | struct extent_buffer *sb; |
0b86a832 | 6682 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6683 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6684 | u8 *array_ptr; |
6685 | unsigned long sb_array_offset; | |
84eed90f | 6686 | int ret = 0; |
0b86a832 CM |
6687 | u32 num_stripes; |
6688 | u32 array_size; | |
6689 | u32 len = 0; | |
1ffb22cf | 6690 | u32 cur_offset; |
e06cd3dd | 6691 | u64 type; |
84eed90f | 6692 | struct btrfs_key key; |
0b86a832 | 6693 | |
0b246afa | 6694 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6695 | /* |
6696 | * This will create extent buffer of nodesize, superblock size is | |
6697 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6698 | * overallocate but we can keep it as-is, only the first page is used. | |
6699 | */ | |
2ff7e61e | 6700 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6701 | if (IS_ERR(sb)) |
6702 | return PTR_ERR(sb); | |
4db8c528 | 6703 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6704 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6705 | /* |
01327610 | 6706 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6707 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6708 | * pages up-to-date when the page is larger: extent does not cover the |
6709 | * whole page and consequently check_page_uptodate does not find all | |
6710 | * the page's extents up-to-date (the hole beyond sb), | |
6711 | * write_extent_buffer then triggers a WARN_ON. | |
6712 | * | |
6713 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6714 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6715 | * to silence the warning eg. on PowerPC 64. | |
6716 | */ | |
09cbfeaf | 6717 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6718 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6719 | |
a061fc8d | 6720 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6721 | array_size = btrfs_super_sys_array_size(super_copy); |
6722 | ||
1ffb22cf DS |
6723 | array_ptr = super_copy->sys_chunk_array; |
6724 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6725 | cur_offset = 0; | |
0b86a832 | 6726 | |
1ffb22cf DS |
6727 | while (cur_offset < array_size) { |
6728 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6729 | len = sizeof(*disk_key); |
6730 | if (cur_offset + len > array_size) | |
6731 | goto out_short_read; | |
6732 | ||
0b86a832 CM |
6733 | btrfs_disk_key_to_cpu(&key, disk_key); |
6734 | ||
1ffb22cf DS |
6735 | array_ptr += len; |
6736 | sb_array_offset += len; | |
6737 | cur_offset += len; | |
0b86a832 | 6738 | |
0d81ba5d | 6739 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6740 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6741 | /* |
6742 | * At least one btrfs_chunk with one stripe must be | |
6743 | * present, exact stripe count check comes afterwards | |
6744 | */ | |
6745 | len = btrfs_chunk_item_size(1); | |
6746 | if (cur_offset + len > array_size) | |
6747 | goto out_short_read; | |
6748 | ||
6749 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6750 | if (!num_stripes) { |
ab8d0fc4 JM |
6751 | btrfs_err(fs_info, |
6752 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6753 | num_stripes, cur_offset); |
6754 | ret = -EIO; | |
6755 | break; | |
6756 | } | |
6757 | ||
e06cd3dd LB |
6758 | type = btrfs_chunk_type(sb, chunk); |
6759 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6760 | btrfs_err(fs_info, |
e06cd3dd LB |
6761 | "invalid chunk type %llu in sys_array at offset %u", |
6762 | type, cur_offset); | |
6763 | ret = -EIO; | |
6764 | break; | |
6765 | } | |
6766 | ||
e3540eab DS |
6767 | len = btrfs_chunk_item_size(num_stripes); |
6768 | if (cur_offset + len > array_size) | |
6769 | goto out_short_read; | |
6770 | ||
2ff7e61e | 6771 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6772 | if (ret) |
6773 | break; | |
0b86a832 | 6774 | } else { |
ab8d0fc4 JM |
6775 | btrfs_err(fs_info, |
6776 | "unexpected item type %u in sys_array at offset %u", | |
6777 | (u32)key.type, cur_offset); | |
84eed90f CM |
6778 | ret = -EIO; |
6779 | break; | |
0b86a832 | 6780 | } |
1ffb22cf DS |
6781 | array_ptr += len; |
6782 | sb_array_offset += len; | |
6783 | cur_offset += len; | |
0b86a832 | 6784 | } |
d865177a | 6785 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6786 | free_extent_buffer_stale(sb); |
84eed90f | 6787 | return ret; |
e3540eab DS |
6788 | |
6789 | out_short_read: | |
ab8d0fc4 | 6790 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6791 | len, cur_offset); |
d865177a | 6792 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6793 | free_extent_buffer_stale(sb); |
e3540eab | 6794 | return -EIO; |
0b86a832 CM |
6795 | } |
6796 | ||
21634a19 QW |
6797 | /* |
6798 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6799 | * | |
6528b99d AJ |
6800 | * If the @failing_dev is specified, it's accounted as missing. |
6801 | * | |
21634a19 QW |
6802 | * Return true if all chunks meet the minimal RW mount requirements. |
6803 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6804 | */ | |
6528b99d AJ |
6805 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6806 | struct btrfs_device *failing_dev) | |
21634a19 QW |
6807 | { |
6808 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
6809 | struct extent_map *em; | |
6810 | u64 next_start = 0; | |
6811 | bool ret = true; | |
6812 | ||
6813 | read_lock(&map_tree->map_tree.lock); | |
6814 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
6815 | read_unlock(&map_tree->map_tree.lock); | |
6816 | /* No chunk at all? Return false anyway */ | |
6817 | if (!em) { | |
6818 | ret = false; | |
6819 | goto out; | |
6820 | } | |
6821 | while (em) { | |
6822 | struct map_lookup *map; | |
6823 | int missing = 0; | |
6824 | int max_tolerated; | |
6825 | int i; | |
6826 | ||
6827 | map = em->map_lookup; | |
6828 | max_tolerated = | |
6829 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6830 | map->type); | |
6831 | for (i = 0; i < map->num_stripes; i++) { | |
6832 | struct btrfs_device *dev = map->stripes[i].dev; | |
6833 | ||
e6e674bd AJ |
6834 | if (!dev || !dev->bdev || |
6835 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6836 | dev->last_flush_error) |
6837 | missing++; | |
6528b99d AJ |
6838 | else if (failing_dev && failing_dev == dev) |
6839 | missing++; | |
21634a19 QW |
6840 | } |
6841 | if (missing > max_tolerated) { | |
6528b99d AJ |
6842 | if (!failing_dev) |
6843 | btrfs_warn(fs_info, | |
21634a19 QW |
6844 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
6845 | em->start, missing, max_tolerated); | |
6846 | free_extent_map(em); | |
6847 | ret = false; | |
6848 | goto out; | |
6849 | } | |
6850 | next_start = extent_map_end(em); | |
6851 | free_extent_map(em); | |
6852 | ||
6853 | read_lock(&map_tree->map_tree.lock); | |
6854 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
6855 | (u64)(-1) - next_start); | |
6856 | read_unlock(&map_tree->map_tree.lock); | |
6857 | } | |
6858 | out: | |
6859 | return ret; | |
6860 | } | |
6861 | ||
5b4aacef | 6862 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6863 | { |
5b4aacef | 6864 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6865 | struct btrfs_path *path; |
6866 | struct extent_buffer *leaf; | |
6867 | struct btrfs_key key; | |
6868 | struct btrfs_key found_key; | |
6869 | int ret; | |
6870 | int slot; | |
99e3ecfc | 6871 | u64 total_dev = 0; |
0b86a832 | 6872 | |
0b86a832 CM |
6873 | path = btrfs_alloc_path(); |
6874 | if (!path) | |
6875 | return -ENOMEM; | |
6876 | ||
3dd0f7a3 AJ |
6877 | /* |
6878 | * uuid_mutex is needed only if we are mounting a sprout FS | |
6879 | * otherwise we don't need it. | |
6880 | */ | |
b367e47f | 6881 | mutex_lock(&uuid_mutex); |
34441361 | 6882 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6883 | |
395927a9 FDBM |
6884 | /* |
6885 | * Read all device items, and then all the chunk items. All | |
6886 | * device items are found before any chunk item (their object id | |
6887 | * is smaller than the lowest possible object id for a chunk | |
6888 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6889 | */ |
6890 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6891 | key.offset = 0; | |
6892 | key.type = 0; | |
0b86a832 | 6893 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6894 | if (ret < 0) |
6895 | goto error; | |
d397712b | 6896 | while (1) { |
0b86a832 CM |
6897 | leaf = path->nodes[0]; |
6898 | slot = path->slots[0]; | |
6899 | if (slot >= btrfs_header_nritems(leaf)) { | |
6900 | ret = btrfs_next_leaf(root, path); | |
6901 | if (ret == 0) | |
6902 | continue; | |
6903 | if (ret < 0) | |
6904 | goto error; | |
6905 | break; | |
6906 | } | |
6907 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6908 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6909 | struct btrfs_dev_item *dev_item; | |
6910 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6911 | struct btrfs_dev_item); |
2ff7e61e | 6912 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
6913 | if (ret) |
6914 | goto error; | |
99e3ecfc | 6915 | total_dev++; |
0b86a832 CM |
6916 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6917 | struct btrfs_chunk *chunk; | |
6918 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 6919 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
6920 | if (ret) |
6921 | goto error; | |
0b86a832 CM |
6922 | } |
6923 | path->slots[0]++; | |
6924 | } | |
99e3ecfc LB |
6925 | |
6926 | /* | |
6927 | * After loading chunk tree, we've got all device information, | |
6928 | * do another round of validation checks. | |
6929 | */ | |
0b246afa JM |
6930 | if (total_dev != fs_info->fs_devices->total_devices) { |
6931 | btrfs_err(fs_info, | |
99e3ecfc | 6932 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 6933 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
6934 | total_dev); |
6935 | ret = -EINVAL; | |
6936 | goto error; | |
6937 | } | |
0b246afa JM |
6938 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
6939 | fs_info->fs_devices->total_rw_bytes) { | |
6940 | btrfs_err(fs_info, | |
99e3ecfc | 6941 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
6942 | btrfs_super_total_bytes(fs_info->super_copy), |
6943 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
6944 | ret = -EINVAL; |
6945 | goto error; | |
6946 | } | |
0b86a832 CM |
6947 | ret = 0; |
6948 | error: | |
34441361 | 6949 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
6950 | mutex_unlock(&uuid_mutex); |
6951 | ||
2b82032c | 6952 | btrfs_free_path(path); |
0b86a832 CM |
6953 | return ret; |
6954 | } | |
442a4f63 | 6955 | |
cb517eab MX |
6956 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6957 | { | |
6958 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6959 | struct btrfs_device *device; | |
6960 | ||
29cc83f6 LB |
6961 | while (fs_devices) { |
6962 | mutex_lock(&fs_devices->device_list_mutex); | |
6963 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 6964 | device->fs_info = fs_info; |
29cc83f6 LB |
6965 | mutex_unlock(&fs_devices->device_list_mutex); |
6966 | ||
6967 | fs_devices = fs_devices->seed; | |
6968 | } | |
cb517eab MX |
6969 | } |
6970 | ||
733f4fbb SB |
6971 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6972 | { | |
6973 | int i; | |
6974 | ||
6975 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6976 | btrfs_dev_stat_reset(dev, i); | |
6977 | } | |
6978 | ||
6979 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6980 | { | |
6981 | struct btrfs_key key; | |
6982 | struct btrfs_key found_key; | |
6983 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6984 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6985 | struct extent_buffer *eb; | |
6986 | int slot; | |
6987 | int ret = 0; | |
6988 | struct btrfs_device *device; | |
6989 | struct btrfs_path *path = NULL; | |
6990 | int i; | |
6991 | ||
6992 | path = btrfs_alloc_path(); | |
6993 | if (!path) { | |
6994 | ret = -ENOMEM; | |
6995 | goto out; | |
6996 | } | |
6997 | ||
6998 | mutex_lock(&fs_devices->device_list_mutex); | |
6999 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7000 | int item_size; | |
7001 | struct btrfs_dev_stats_item *ptr; | |
7002 | ||
242e2956 DS |
7003 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7004 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7005 | key.offset = device->devid; |
7006 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7007 | if (ret) { | |
733f4fbb SB |
7008 | __btrfs_reset_dev_stats(device); |
7009 | device->dev_stats_valid = 1; | |
7010 | btrfs_release_path(path); | |
7011 | continue; | |
7012 | } | |
7013 | slot = path->slots[0]; | |
7014 | eb = path->nodes[0]; | |
7015 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7016 | item_size = btrfs_item_size_nr(eb, slot); | |
7017 | ||
7018 | ptr = btrfs_item_ptr(eb, slot, | |
7019 | struct btrfs_dev_stats_item); | |
7020 | ||
7021 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7022 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7023 | btrfs_dev_stat_set(device, i, | |
7024 | btrfs_dev_stats_value(eb, ptr, i)); | |
7025 | else | |
7026 | btrfs_dev_stat_reset(device, i); | |
7027 | } | |
7028 | ||
7029 | device->dev_stats_valid = 1; | |
7030 | btrfs_dev_stat_print_on_load(device); | |
7031 | btrfs_release_path(path); | |
7032 | } | |
7033 | mutex_unlock(&fs_devices->device_list_mutex); | |
7034 | ||
7035 | out: | |
7036 | btrfs_free_path(path); | |
7037 | return ret < 0 ? ret : 0; | |
7038 | } | |
7039 | ||
7040 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7041 | struct btrfs_device *device) |
7042 | { | |
5495f195 | 7043 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7044 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7045 | struct btrfs_path *path; |
7046 | struct btrfs_key key; | |
7047 | struct extent_buffer *eb; | |
7048 | struct btrfs_dev_stats_item *ptr; | |
7049 | int ret; | |
7050 | int i; | |
7051 | ||
242e2956 DS |
7052 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7053 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7054 | key.offset = device->devid; |
7055 | ||
7056 | path = btrfs_alloc_path(); | |
fa252992 DS |
7057 | if (!path) |
7058 | return -ENOMEM; | |
733f4fbb SB |
7059 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7060 | if (ret < 0) { | |
0b246afa | 7061 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7062 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7063 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7064 | goto out; |
7065 | } | |
7066 | ||
7067 | if (ret == 0 && | |
7068 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7069 | /* need to delete old one and insert a new one */ | |
7070 | ret = btrfs_del_item(trans, dev_root, path); | |
7071 | if (ret != 0) { | |
0b246afa | 7072 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7073 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7074 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7075 | goto out; |
7076 | } | |
7077 | ret = 1; | |
7078 | } | |
7079 | ||
7080 | if (ret == 1) { | |
7081 | /* need to insert a new item */ | |
7082 | btrfs_release_path(path); | |
7083 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7084 | &key, sizeof(*ptr)); | |
7085 | if (ret < 0) { | |
0b246afa | 7086 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7087 | "insert dev_stats item for device %s failed %d", |
7088 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7089 | goto out; |
7090 | } | |
7091 | } | |
7092 | ||
7093 | eb = path->nodes[0]; | |
7094 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7095 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7096 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7097 | btrfs_dev_stat_read(device, i)); | |
7098 | btrfs_mark_buffer_dirty(eb); | |
7099 | ||
7100 | out: | |
7101 | btrfs_free_path(path); | |
7102 | return ret; | |
7103 | } | |
7104 | ||
7105 | /* | |
7106 | * called from commit_transaction. Writes all changed device stats to disk. | |
7107 | */ | |
7108 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7109 | struct btrfs_fs_info *fs_info) | |
7110 | { | |
733f4fbb SB |
7111 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7112 | struct btrfs_device *device; | |
addc3fa7 | 7113 | int stats_cnt; |
733f4fbb SB |
7114 | int ret = 0; |
7115 | ||
7116 | mutex_lock(&fs_devices->device_list_mutex); | |
7117 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7118 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7119 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7120 | continue; |
7121 | ||
9deae968 NB |
7122 | |
7123 | /* | |
7124 | * There is a LOAD-LOAD control dependency between the value of | |
7125 | * dev_stats_ccnt and updating the on-disk values which requires | |
7126 | * reading the in-memory counters. Such control dependencies | |
7127 | * require explicit read memory barriers. | |
7128 | * | |
7129 | * This memory barriers pairs with smp_mb__before_atomic in | |
7130 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7131 | * barrier implied by atomic_xchg in | |
7132 | * btrfs_dev_stats_read_and_reset | |
7133 | */ | |
7134 | smp_rmb(); | |
7135 | ||
5495f195 | 7136 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7137 | if (!ret) |
addc3fa7 | 7138 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7139 | } |
7140 | mutex_unlock(&fs_devices->device_list_mutex); | |
7141 | ||
7142 | return ret; | |
7143 | } | |
7144 | ||
442a4f63 SB |
7145 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7146 | { | |
7147 | btrfs_dev_stat_inc(dev, index); | |
7148 | btrfs_dev_stat_print_on_error(dev); | |
7149 | } | |
7150 | ||
48a3b636 | 7151 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7152 | { |
733f4fbb SB |
7153 | if (!dev->dev_stats_valid) |
7154 | return; | |
fb456252 | 7155 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7156 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7157 | rcu_str_deref(dev->name), |
442a4f63 SB |
7158 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7159 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7160 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7161 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7162 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7163 | } |
c11d2c23 | 7164 | |
733f4fbb SB |
7165 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7166 | { | |
a98cdb85 SB |
7167 | int i; |
7168 | ||
7169 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7170 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7171 | break; | |
7172 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7173 | return; /* all values == 0, suppress message */ | |
7174 | ||
fb456252 | 7175 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7176 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7177 | rcu_str_deref(dev->name), |
733f4fbb SB |
7178 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7179 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7180 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7181 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7182 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7183 | } | |
7184 | ||
2ff7e61e | 7185 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7186 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7187 | { |
7188 | struct btrfs_device *dev; | |
0b246afa | 7189 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7190 | int i; |
7191 | ||
7192 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7193 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7194 | mutex_unlock(&fs_devices->device_list_mutex); |
7195 | ||
7196 | if (!dev) { | |
0b246afa | 7197 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7198 | return -ENODEV; |
733f4fbb | 7199 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7200 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7201 | return -ENODEV; |
b27f7c0c | 7202 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7203 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7204 | if (stats->nr_items > i) | |
7205 | stats->values[i] = | |
7206 | btrfs_dev_stat_read_and_reset(dev, i); | |
7207 | else | |
7208 | btrfs_dev_stat_reset(dev, i); | |
7209 | } | |
7210 | } else { | |
7211 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7212 | if (stats->nr_items > i) | |
7213 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7214 | } | |
7215 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7216 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7217 | return 0; | |
7218 | } | |
a8a6dab7 | 7219 | |
da353f6b | 7220 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7221 | { |
7222 | struct buffer_head *bh; | |
7223 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7224 | int copy_num; |
a8a6dab7 | 7225 | |
12b1c263 AJ |
7226 | if (!bdev) |
7227 | return; | |
a8a6dab7 | 7228 | |
12b1c263 AJ |
7229 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7230 | copy_num++) { | |
a8a6dab7 | 7231 | |
12b1c263 AJ |
7232 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7233 | continue; | |
7234 | ||
7235 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7236 | ||
7237 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7238 | set_buffer_dirty(bh); | |
7239 | sync_dirty_buffer(bh); | |
7240 | brelse(bh); | |
7241 | } | |
7242 | ||
7243 | /* Notify udev that device has changed */ | |
7244 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7245 | ||
7246 | /* Update ctime/mtime for device path for libblkid */ | |
7247 | update_dev_time(device_path); | |
a8a6dab7 | 7248 | } |
935e5cc9 MX |
7249 | |
7250 | /* | |
7251 | * Update the size of all devices, which is used for writing out the | |
7252 | * super blocks. | |
7253 | */ | |
7254 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7255 | { | |
7256 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7257 | struct btrfs_device *curr, *next; | |
7258 | ||
7259 | if (list_empty(&fs_devices->resized_devices)) | |
7260 | return; | |
7261 | ||
7262 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7263 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7264 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7265 | resized_list) { | |
7266 | list_del_init(&curr->resized_list); | |
7267 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7268 | } | |
34441361 | 7269 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7270 | mutex_unlock(&fs_devices->device_list_mutex); |
7271 | } | |
ce7213c7 MX |
7272 | |
7273 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7274 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7275 | { |
e9b919b1 | 7276 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7277 | struct extent_map *em; |
7278 | struct map_lookup *map; | |
7279 | struct btrfs_device *dev; | |
7280 | int i; | |
7281 | ||
e9b919b1 | 7282 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7283 | return; |
7284 | ||
7285 | /* In order to kick the device replace finish process */ | |
34441361 | 7286 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7287 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7288 | map = em->map_lookup; |
ce7213c7 MX |
7289 | |
7290 | for (i = 0; i < map->num_stripes; i++) { | |
7291 | dev = map->stripes[i].dev; | |
7292 | dev->commit_bytes_used = dev->bytes_used; | |
7293 | } | |
7294 | } | |
34441361 | 7295 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7296 | } |
5a13f430 AJ |
7297 | |
7298 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7299 | { | |
7300 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7301 | while (fs_devices) { | |
7302 | fs_devices->fs_info = fs_info; | |
7303 | fs_devices = fs_devices->seed; | |
7304 | } | |
7305 | } | |
7306 | ||
7307 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7308 | { | |
7309 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7310 | while (fs_devices) { | |
7311 | fs_devices->fs_info = NULL; | |
7312 | fs_devices = fs_devices->seed; | |
7313 | } | |
7314 | } | |
46df06b8 DS |
7315 | |
7316 | /* | |
7317 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7318 | */ | |
7319 | int btrfs_bg_type_to_factor(u64 flags) | |
7320 | { | |
7321 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7322 | BTRFS_BLOCK_GROUP_RAID10)) | |
7323 | return 2; | |
7324 | return 1; | |
7325 | } | |
cf90d884 QW |
7326 | |
7327 | ||
7328 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) | |
7329 | { | |
7330 | int index = btrfs_bg_flags_to_raid_index(type); | |
7331 | int ncopies = btrfs_raid_array[index].ncopies; | |
7332 | int data_stripes; | |
7333 | ||
7334 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
7335 | case BTRFS_BLOCK_GROUP_RAID5: | |
7336 | data_stripes = num_stripes - 1; | |
7337 | break; | |
7338 | case BTRFS_BLOCK_GROUP_RAID6: | |
7339 | data_stripes = num_stripes - 2; | |
7340 | break; | |
7341 | default: | |
7342 | data_stripes = num_stripes / ncopies; | |
7343 | break; | |
7344 | } | |
7345 | return div_u64(chunk_len, data_stripes); | |
7346 | } | |
7347 | ||
7348 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7349 | u64 chunk_offset, u64 devid, | |
7350 | u64 physical_offset, u64 physical_len) | |
7351 | { | |
7352 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7353 | struct extent_map *em; | |
7354 | struct map_lookup *map; | |
7355 | u64 stripe_len; | |
7356 | bool found = false; | |
7357 | int ret = 0; | |
7358 | int i; | |
7359 | ||
7360 | read_lock(&em_tree->lock); | |
7361 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7362 | read_unlock(&em_tree->lock); | |
7363 | ||
7364 | if (!em) { | |
7365 | btrfs_err(fs_info, | |
7366 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7367 | physical_offset, devid); | |
7368 | ret = -EUCLEAN; | |
7369 | goto out; | |
7370 | } | |
7371 | ||
7372 | map = em->map_lookup; | |
7373 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7374 | if (physical_len != stripe_len) { | |
7375 | btrfs_err(fs_info, | |
7376 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7377 | physical_offset, devid, em->start, physical_len, | |
7378 | stripe_len); | |
7379 | ret = -EUCLEAN; | |
7380 | goto out; | |
7381 | } | |
7382 | ||
7383 | for (i = 0; i < map->num_stripes; i++) { | |
7384 | if (map->stripes[i].dev->devid == devid && | |
7385 | map->stripes[i].physical == physical_offset) { | |
7386 | found = true; | |
7387 | if (map->verified_stripes >= map->num_stripes) { | |
7388 | btrfs_err(fs_info, | |
7389 | "too many dev extents for chunk %llu found", | |
7390 | em->start); | |
7391 | ret = -EUCLEAN; | |
7392 | goto out; | |
7393 | } | |
7394 | map->verified_stripes++; | |
7395 | break; | |
7396 | } | |
7397 | } | |
7398 | if (!found) { | |
7399 | btrfs_err(fs_info, | |
7400 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7401 | physical_offset, devid); | |
7402 | ret = -EUCLEAN; | |
7403 | } | |
7404 | out: | |
7405 | free_extent_map(em); | |
7406 | return ret; | |
7407 | } | |
7408 | ||
7409 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7410 | { | |
7411 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7412 | struct extent_map *em; | |
7413 | struct rb_node *node; | |
7414 | int ret = 0; | |
7415 | ||
7416 | read_lock(&em_tree->lock); | |
7417 | for (node = rb_first(&em_tree->map); node; node = rb_next(node)) { | |
7418 | em = rb_entry(node, struct extent_map, rb_node); | |
7419 | if (em->map_lookup->num_stripes != | |
7420 | em->map_lookup->verified_stripes) { | |
7421 | btrfs_err(fs_info, | |
7422 | "chunk %llu has missing dev extent, have %d expect %d", | |
7423 | em->start, em->map_lookup->verified_stripes, | |
7424 | em->map_lookup->num_stripes); | |
7425 | ret = -EUCLEAN; | |
7426 | goto out; | |
7427 | } | |
7428 | } | |
7429 | out: | |
7430 | read_unlock(&em_tree->lock); | |
7431 | return ret; | |
7432 | } | |
7433 | ||
7434 | /* | |
7435 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7436 | * later chunk allocation/free would cause unexpected behavior. | |
7437 | * | |
7438 | * NOTE: This will iterate through the whole device tree, which should be of | |
7439 | * the same size level as the chunk tree. This slightly increases mount time. | |
7440 | */ | |
7441 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7442 | { | |
7443 | struct btrfs_path *path; | |
7444 | struct btrfs_root *root = fs_info->dev_root; | |
7445 | struct btrfs_key key; | |
7446 | int ret = 0; | |
7447 | ||
7448 | key.objectid = 1; | |
7449 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7450 | key.offset = 0; | |
7451 | ||
7452 | path = btrfs_alloc_path(); | |
7453 | if (!path) | |
7454 | return -ENOMEM; | |
7455 | ||
7456 | path->reada = READA_FORWARD; | |
7457 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7458 | if (ret < 0) | |
7459 | goto out; | |
7460 | ||
7461 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7462 | ret = btrfs_next_item(root, path); | |
7463 | if (ret < 0) | |
7464 | goto out; | |
7465 | /* No dev extents at all? Not good */ | |
7466 | if (ret > 0) { | |
7467 | ret = -EUCLEAN; | |
7468 | goto out; | |
7469 | } | |
7470 | } | |
7471 | while (1) { | |
7472 | struct extent_buffer *leaf = path->nodes[0]; | |
7473 | struct btrfs_dev_extent *dext; | |
7474 | int slot = path->slots[0]; | |
7475 | u64 chunk_offset; | |
7476 | u64 physical_offset; | |
7477 | u64 physical_len; | |
7478 | u64 devid; | |
7479 | ||
7480 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7481 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7482 | break; | |
7483 | devid = key.objectid; | |
7484 | physical_offset = key.offset; | |
7485 | ||
7486 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7487 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7488 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7489 | ||
7490 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, | |
7491 | physical_offset, physical_len); | |
7492 | if (ret < 0) | |
7493 | goto out; | |
7494 | ret = btrfs_next_item(root, path); | |
7495 | if (ret < 0) | |
7496 | goto out; | |
7497 | if (ret > 0) { | |
7498 | ret = 0; | |
7499 | break; | |
7500 | } | |
7501 | } | |
7502 | ||
7503 | /* Ensure all chunks have corresponding dev extents */ | |
7504 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7505 | out: | |
7506 | btrfs_free_path(path); | |
7507 | return ret; | |
7508 | } |