]>
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); | |
07e1ce09 | 1616 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
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 | ||
1da73967 AJ |
1857 | /* |
1858 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
1859 | * currently replaced. | |
1860 | */ | |
1861 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
1862 | { | |
1863 | u64 num_devices = fs_info->fs_devices->num_devices; | |
1864 | ||
1865 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); | |
1866 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
1867 | ASSERT(num_devices > 1); | |
1868 | num_devices--; | |
1869 | } | |
1870 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); | |
1871 | ||
1872 | return num_devices; | |
1873 | } | |
1874 | ||
da353f6b DS |
1875 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1876 | u64 devid) | |
f1fa7f26 AJ |
1877 | { |
1878 | struct btrfs_device *device; | |
1f78160c | 1879 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 1880 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 1881 | u64 num_devices; |
a061fc8d CM |
1882 | int ret = 0; |
1883 | ||
a061fc8d CM |
1884 | mutex_lock(&uuid_mutex); |
1885 | ||
1da73967 | 1886 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 1887 | |
0b246afa | 1888 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1889 | if (ret) |
a061fc8d | 1890 | goto out; |
a061fc8d | 1891 | |
a27a94c2 NB |
1892 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
1893 | ||
1894 | if (IS_ERR(device)) { | |
1895 | if (PTR_ERR(device) == -ENOENT && | |
1896 | strcmp(device_path, "missing") == 0) | |
1897 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
1898 | else | |
1899 | ret = PTR_ERR(device); | |
53b381b3 | 1900 | goto out; |
a27a94c2 | 1901 | } |
dfe25020 | 1902 | |
eede2bf3 OS |
1903 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
1904 | btrfs_warn_in_rcu(fs_info, | |
1905 | "cannot remove device %s (devid %llu) due to active swapfile", | |
1906 | rcu_str_deref(device->name), device->devid); | |
1907 | ret = -ETXTBSY; | |
1908 | goto out; | |
1909 | } | |
1910 | ||
401e29c1 | 1911 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1912 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1913 | goto out; |
63a212ab SB |
1914 | } |
1915 | ||
ebbede42 AJ |
1916 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1917 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1918 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1919 | goto out; |
2b82032c YZ |
1920 | } |
1921 | ||
ebbede42 | 1922 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1923 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1924 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1925 | device->fs_devices->rw_devices--; |
34441361 | 1926 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 1927 | } |
a061fc8d | 1928 | |
d7901554 | 1929 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1930 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1931 | mutex_lock(&uuid_mutex); |
a061fc8d | 1932 | if (ret) |
9b3517e9 | 1933 | goto error_undo; |
a061fc8d | 1934 | |
63a212ab SB |
1935 | /* |
1936 | * TODO: the superblock still includes this device in its num_devices | |
1937 | * counter although write_all_supers() is not locked out. This | |
1938 | * could give a filesystem state which requires a degraded mount. | |
1939 | */ | |
0b246afa | 1940 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1941 | if (ret) |
9b3517e9 | 1942 | goto error_undo; |
a061fc8d | 1943 | |
e12c9621 | 1944 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 1945 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1946 | |
1947 | /* | |
1948 | * the device list mutex makes sure that we don't change | |
1949 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1950 | * the device supers. Whoever is writing all supers, should |
1951 | * lock the device list mutex before getting the number of | |
1952 | * devices in the super block (super_copy). Conversely, | |
1953 | * whoever updates the number of devices in the super block | |
1954 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1955 | */ |
1f78160c | 1956 | |
41a52a0f AJ |
1957 | /* |
1958 | * In normal cases the cur_devices == fs_devices. But in case | |
1959 | * of deleting a seed device, the cur_devices should point to | |
1960 | * its own fs_devices listed under the fs_devices->seed. | |
1961 | */ | |
1f78160c | 1962 | cur_devices = device->fs_devices; |
b5185197 | 1963 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 1964 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1965 | |
41a52a0f AJ |
1966 | cur_devices->num_devices--; |
1967 | cur_devices->total_devices--; | |
b4993e64 AJ |
1968 | /* Update total_devices of the parent fs_devices if it's seed */ |
1969 | if (cur_devices != fs_devices) | |
1970 | fs_devices->total_devices--; | |
2b82032c | 1971 | |
e6e674bd | 1972 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 1973 | cur_devices->missing_devices--; |
cd02dca5 | 1974 | |
d6507cf1 | 1975 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 1976 | |
0bfaa9c5 | 1977 | if (device->bdev) { |
41a52a0f | 1978 | cur_devices->open_devices--; |
0bfaa9c5 | 1979 | /* remove sysfs entry */ |
b5185197 | 1980 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 1981 | } |
99994cde | 1982 | |
0b246afa JM |
1983 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
1984 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 1985 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 1986 | |
cea67ab9 JM |
1987 | /* |
1988 | * at this point, the device is zero sized and detached from | |
1989 | * the devices list. All that's left is to zero out the old | |
1990 | * supers and free the device. | |
1991 | */ | |
ebbede42 | 1992 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
1993 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
1994 | ||
1995 | btrfs_close_bdev(device); | |
f06c5965 | 1996 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 1997 | |
1f78160c | 1998 | if (cur_devices->open_devices == 0) { |
e4404d6e | 1999 | while (fs_devices) { |
8321cf25 RS |
2000 | if (fs_devices->seed == cur_devices) { |
2001 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2002 | break; |
8321cf25 | 2003 | } |
e4404d6e | 2004 | fs_devices = fs_devices->seed; |
2b82032c | 2005 | } |
1f78160c | 2006 | cur_devices->seed = NULL; |
0226e0eb | 2007 | close_fs_devices(cur_devices); |
1f78160c | 2008 | free_fs_devices(cur_devices); |
2b82032c YZ |
2009 | } |
2010 | ||
a061fc8d CM |
2011 | out: |
2012 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2013 | return ret; |
24fc572f | 2014 | |
9b3517e9 | 2015 | error_undo: |
ebbede42 | 2016 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2017 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2018 | list_add(&device->dev_alloc_list, |
b5185197 | 2019 | &fs_devices->alloc_list); |
c3929c36 | 2020 | device->fs_devices->rw_devices++; |
34441361 | 2021 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2022 | } |
24fc572f | 2023 | goto out; |
a061fc8d CM |
2024 | } |
2025 | ||
68a9db5f | 2026 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2027 | { |
d51908ce AJ |
2028 | struct btrfs_fs_devices *fs_devices; |
2029 | ||
68a9db5f | 2030 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2031 | |
25e8e911 AJ |
2032 | /* |
2033 | * in case of fs with no seed, srcdev->fs_devices will point | |
2034 | * to fs_devices of fs_info. However when the dev being replaced is | |
2035 | * a seed dev it will point to the seed's local fs_devices. In short | |
2036 | * srcdev will have its correct fs_devices in both the cases. | |
2037 | */ | |
2038 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2039 | |
e93c89c1 | 2040 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2041 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2042 | fs_devices->num_devices--; |
e6e674bd | 2043 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2044 | fs_devices->missing_devices--; |
e93c89c1 | 2045 | |
ebbede42 | 2046 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2047 | fs_devices->rw_devices--; |
1357272f | 2048 | |
82372bc8 | 2049 | if (srcdev->bdev) |
d51908ce | 2050 | fs_devices->open_devices--; |
084b6e7c QW |
2051 | } |
2052 | ||
2053 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2054 | struct btrfs_device *srcdev) | |
2055 | { | |
2056 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2057 | |
ebbede42 | 2058 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2059 | /* zero out the old super if it is writable */ |
2060 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2061 | } | |
14238819 AJ |
2062 | |
2063 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2064 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2065 | |
94d5f0c2 AJ |
2066 | /* if this is no devs we rather delete the fs_devices */ |
2067 | if (!fs_devices->num_devices) { | |
2068 | struct btrfs_fs_devices *tmp_fs_devices; | |
2069 | ||
6dd38f81 AJ |
2070 | /* |
2071 | * On a mounted FS, num_devices can't be zero unless it's a | |
2072 | * seed. In case of a seed device being replaced, the replace | |
2073 | * target added to the sprout FS, so there will be no more | |
2074 | * device left under the seed FS. | |
2075 | */ | |
2076 | ASSERT(fs_devices->seeding); | |
2077 | ||
94d5f0c2 AJ |
2078 | tmp_fs_devices = fs_info->fs_devices; |
2079 | while (tmp_fs_devices) { | |
2080 | if (tmp_fs_devices->seed == fs_devices) { | |
2081 | tmp_fs_devices->seed = fs_devices->seed; | |
2082 | break; | |
2083 | } | |
2084 | tmp_fs_devices = tmp_fs_devices->seed; | |
2085 | } | |
2086 | fs_devices->seed = NULL; | |
0226e0eb | 2087 | close_fs_devices(fs_devices); |
8bef8401 | 2088 | free_fs_devices(fs_devices); |
94d5f0c2 | 2089 | } |
e93c89c1 SB |
2090 | } |
2091 | ||
4f5ad7bd | 2092 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2093 | { |
4f5ad7bd | 2094 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2095 | |
e93c89c1 | 2096 | WARN_ON(!tgtdev); |
d9a071f0 | 2097 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2098 | |
d9a071f0 | 2099 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2100 | |
779bf3fe | 2101 | if (tgtdev->bdev) |
d9a071f0 | 2102 | fs_devices->open_devices--; |
779bf3fe | 2103 | |
d9a071f0 | 2104 | fs_devices->num_devices--; |
e93c89c1 | 2105 | |
d6507cf1 | 2106 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2107 | |
e93c89c1 | 2108 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2109 | |
d9a071f0 | 2110 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2111 | |
2112 | /* | |
2113 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2114 | * may lead to a call to btrfs_show_devname() which will try | |
2115 | * to hold device_list_mutex. And here this device | |
2116 | * is already out of device list, so we don't have to hold | |
2117 | * the device_list_mutex lock. | |
2118 | */ | |
2119 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2120 | |
2121 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2122 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2123 | } |
2124 | ||
b444ad46 NB |
2125 | static struct btrfs_device *btrfs_find_device_by_path( |
2126 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2127 | { |
2128 | int ret = 0; | |
2129 | struct btrfs_super_block *disk_super; | |
2130 | u64 devid; | |
2131 | u8 *dev_uuid; | |
2132 | struct block_device *bdev; | |
2133 | struct buffer_head *bh; | |
b444ad46 | 2134 | struct btrfs_device *device; |
7ba15b7d | 2135 | |
7ba15b7d | 2136 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2137 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2138 | if (ret) |
b444ad46 | 2139 | return ERR_PTR(ret); |
7ba15b7d SB |
2140 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2141 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2142 | dev_uuid = disk_super->dev_item.uuid; | |
b444ad46 | 2143 | device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d | 2144 | brelse(bh); |
b444ad46 NB |
2145 | if (!device) |
2146 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2147 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2148 | return device; |
7ba15b7d SB |
2149 | } |
2150 | ||
6c050407 NB |
2151 | static struct btrfs_device *btrfs_find_device_missing_or_by_path( |
2152 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d | 2153 | { |
6c050407 | 2154 | struct btrfs_device *device = NULL; |
7ba15b7d SB |
2155 | if (strcmp(device_path, "missing") == 0) { |
2156 | struct list_head *devices; | |
2157 | struct btrfs_device *tmp; | |
2158 | ||
0b246afa | 2159 | devices = &fs_info->fs_devices->devices; |
7ba15b7d | 2160 | list_for_each_entry(tmp, devices, dev_list) { |
e12c9621 AJ |
2161 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2162 | &tmp->dev_state) && !tmp->bdev) { | |
6c050407 | 2163 | device = tmp; |
7ba15b7d SB |
2164 | break; |
2165 | } | |
2166 | } | |
2167 | ||
6c050407 NB |
2168 | if (!device) |
2169 | return ERR_PTR(-ENOENT); | |
7ba15b7d | 2170 | } else { |
6c050407 | 2171 | device = btrfs_find_device_by_path(fs_info, device_path); |
7ba15b7d | 2172 | } |
b444ad46 | 2173 | |
6c050407 | 2174 | return device; |
7ba15b7d SB |
2175 | } |
2176 | ||
5c5c0df0 DS |
2177 | /* |
2178 | * Lookup a device given by device id, or the path if the id is 0. | |
2179 | */ | |
a27a94c2 NB |
2180 | struct btrfs_device *btrfs_find_device_by_devspec( |
2181 | struct btrfs_fs_info *fs_info, u64 devid, const char *devpath) | |
24e0474b | 2182 | { |
a27a94c2 | 2183 | struct btrfs_device *device; |
24e0474b | 2184 | |
5c5c0df0 | 2185 | if (devid) { |
a27a94c2 NB |
2186 | device = btrfs_find_device(fs_info, devid, NULL, NULL); |
2187 | if (!device) | |
2188 | return ERR_PTR(-ENOENT); | |
24e0474b | 2189 | } else { |
5c5c0df0 | 2190 | if (!devpath || !devpath[0]) |
a27a94c2 NB |
2191 | return ERR_PTR(-EINVAL); |
2192 | device = btrfs_find_device_missing_or_by_path(fs_info, devpath); | |
24e0474b | 2193 | } |
a27a94c2 | 2194 | return device; |
24e0474b AJ |
2195 | } |
2196 | ||
2b82032c YZ |
2197 | /* |
2198 | * does all the dirty work required for changing file system's UUID. | |
2199 | */ | |
2ff7e61e | 2200 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2201 | { |
0b246afa | 2202 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2203 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2204 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2205 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2206 | struct btrfs_device *device; |
2207 | u64 super_flags; | |
2208 | ||
a32bf9a3 | 2209 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2210 | if (!fs_devices->seeding) |
2b82032c YZ |
2211 | return -EINVAL; |
2212 | ||
2dfeca9b | 2213 | seed_devices = alloc_fs_devices(NULL); |
2208a378 ID |
2214 | if (IS_ERR(seed_devices)) |
2215 | return PTR_ERR(seed_devices); | |
2b82032c | 2216 | |
e4404d6e YZ |
2217 | old_devices = clone_fs_devices(fs_devices); |
2218 | if (IS_ERR(old_devices)) { | |
2219 | kfree(seed_devices); | |
2220 | return PTR_ERR(old_devices); | |
2b82032c | 2221 | } |
e4404d6e | 2222 | |
c4babc5e | 2223 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2224 | |
e4404d6e YZ |
2225 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2226 | seed_devices->opened = 1; | |
2227 | INIT_LIST_HEAD(&seed_devices->devices); | |
2228 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2229 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2230 | |
321a4bf7 | 2231 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2232 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2233 | synchronize_rcu); | |
2196d6e8 MX |
2234 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2235 | device->fs_devices = seed_devices; | |
c9513edb | 2236 | |
34441361 | 2237 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2238 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2239 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2240 | |
2b82032c YZ |
2241 | fs_devices->seeding = 0; |
2242 | fs_devices->num_devices = 0; | |
2243 | fs_devices->open_devices = 0; | |
69611ac8 | 2244 | fs_devices->missing_devices = 0; |
69611ac8 | 2245 | fs_devices->rotating = 0; |
e4404d6e | 2246 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2247 | |
2248 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2249 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2250 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2251 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2252 | |
2b82032c YZ |
2253 | super_flags = btrfs_super_flags(disk_super) & |
2254 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2255 | btrfs_set_super_flags(disk_super, super_flags); | |
2256 | ||
2257 | return 0; | |
2258 | } | |
2259 | ||
2260 | /* | |
01327610 | 2261 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2262 | */ |
2263 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2264 | struct btrfs_fs_info *fs_info) |
2b82032c | 2265 | { |
5b4aacef | 2266 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2267 | struct btrfs_path *path; |
2268 | struct extent_buffer *leaf; | |
2269 | struct btrfs_dev_item *dev_item; | |
2270 | struct btrfs_device *device; | |
2271 | struct btrfs_key key; | |
44880fdc | 2272 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2273 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2274 | u64 devid; | |
2275 | int ret; | |
2276 | ||
2277 | path = btrfs_alloc_path(); | |
2278 | if (!path) | |
2279 | return -ENOMEM; | |
2280 | ||
2b82032c YZ |
2281 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2282 | key.offset = 0; | |
2283 | key.type = BTRFS_DEV_ITEM_KEY; | |
2284 | ||
2285 | while (1) { | |
2286 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2287 | if (ret < 0) | |
2288 | goto error; | |
2289 | ||
2290 | leaf = path->nodes[0]; | |
2291 | next_slot: | |
2292 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2293 | ret = btrfs_next_leaf(root, path); | |
2294 | if (ret > 0) | |
2295 | break; | |
2296 | if (ret < 0) | |
2297 | goto error; | |
2298 | leaf = path->nodes[0]; | |
2299 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2300 | btrfs_release_path(path); |
2b82032c YZ |
2301 | continue; |
2302 | } | |
2303 | ||
2304 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2305 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2306 | key.type != BTRFS_DEV_ITEM_KEY) | |
2307 | break; | |
2308 | ||
2309 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2310 | struct btrfs_dev_item); | |
2311 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2312 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2313 | BTRFS_UUID_SIZE); |
1473b24e | 2314 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2315 | BTRFS_FSID_SIZE); |
0b246afa | 2316 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2317 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2318 | |
2319 | if (device->fs_devices->seeding) { | |
2320 | btrfs_set_device_generation(leaf, dev_item, | |
2321 | device->generation); | |
2322 | btrfs_mark_buffer_dirty(leaf); | |
2323 | } | |
2324 | ||
2325 | path->slots[0]++; | |
2326 | goto next_slot; | |
2327 | } | |
2328 | ret = 0; | |
2329 | error: | |
2330 | btrfs_free_path(path); | |
2331 | return ret; | |
2332 | } | |
2333 | ||
da353f6b | 2334 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2335 | { |
5112febb | 2336 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2337 | struct request_queue *q; |
788f20eb CM |
2338 | struct btrfs_trans_handle *trans; |
2339 | struct btrfs_device *device; | |
2340 | struct block_device *bdev; | |
0b246afa | 2341 | struct super_block *sb = fs_info->sb; |
606686ee | 2342 | struct rcu_string *name; |
5da54bc1 | 2343 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2344 | u64 orig_super_total_bytes; |
2345 | u64 orig_super_num_devices; | |
2b82032c | 2346 | int seeding_dev = 0; |
788f20eb | 2347 | int ret = 0; |
7132a262 | 2348 | bool unlocked = false; |
788f20eb | 2349 | |
5da54bc1 | 2350 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2351 | return -EROFS; |
788f20eb | 2352 | |
a5d16333 | 2353 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2354 | fs_info->bdev_holder); |
7f59203a JB |
2355 | if (IS_ERR(bdev)) |
2356 | return PTR_ERR(bdev); | |
a2135011 | 2357 | |
5da54bc1 | 2358 | if (fs_devices->seeding) { |
2b82032c YZ |
2359 | seeding_dev = 1; |
2360 | down_write(&sb->s_umount); | |
2361 | mutex_lock(&uuid_mutex); | |
2362 | } | |
2363 | ||
8c8bee1d | 2364 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2365 | |
5da54bc1 | 2366 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2367 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2368 | if (device->bdev == bdev) { |
2369 | ret = -EEXIST; | |
d25628bd | 2370 | mutex_unlock( |
5da54bc1 | 2371 | &fs_devices->device_list_mutex); |
2b82032c | 2372 | goto error; |
788f20eb CM |
2373 | } |
2374 | } | |
5da54bc1 | 2375 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2376 | |
0b246afa | 2377 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2378 | if (IS_ERR(device)) { |
788f20eb | 2379 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2380 | ret = PTR_ERR(device); |
2b82032c | 2381 | goto error; |
788f20eb CM |
2382 | } |
2383 | ||
78f2c9e6 | 2384 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2385 | if (!name) { |
2b82032c | 2386 | ret = -ENOMEM; |
5c4cf6c9 | 2387 | goto error_free_device; |
788f20eb | 2388 | } |
606686ee | 2389 | rcu_assign_pointer(device->name, name); |
2b82032c | 2390 | |
a22285a6 | 2391 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2392 | if (IS_ERR(trans)) { |
98d5dc13 | 2393 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2394 | goto error_free_device; |
98d5dc13 TI |
2395 | } |
2396 | ||
d5e2003c | 2397 | q = bdev_get_queue(bdev); |
ebbede42 | 2398 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2399 | device->generation = trans->transid; |
0b246afa JM |
2400 | device->io_width = fs_info->sectorsize; |
2401 | device->io_align = fs_info->sectorsize; | |
2402 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2403 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2404 | fs_info->sectorsize); | |
2cc3c559 | 2405 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2406 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2407 | device->fs_info = fs_info; |
788f20eb | 2408 | device->bdev = bdev; |
e12c9621 | 2409 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2410 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2411 | device->mode = FMODE_EXCL; |
27087f37 | 2412 | device->dev_stats_valid = 1; |
9f6d2510 | 2413 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2414 | |
2b82032c | 2415 | if (seeding_dev) { |
1751e8a6 | 2416 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2417 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2418 | if (ret) { |
2419 | btrfs_abort_transaction(trans, ret); | |
2420 | goto error_trans; | |
2421 | } | |
2b82032c | 2422 | } |
788f20eb | 2423 | |
5da54bc1 | 2424 | device->fs_devices = fs_devices; |
e5e9a520 | 2425 | |
5da54bc1 | 2426 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2427 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2428 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2429 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2430 | fs_devices->num_devices++; | |
2431 | fs_devices->open_devices++; | |
2432 | fs_devices->rw_devices++; | |
2433 | fs_devices->total_devices++; | |
2434 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2435 | |
a5ed45f8 | 2436 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2437 | |
e884f4f0 | 2438 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2439 | fs_devices->rotating = 1; |
c289811c | 2440 | |
39379faa | 2441 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2442 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2443 | round_down(orig_super_total_bytes + device->total_bytes, |
2444 | fs_info->sectorsize)); | |
788f20eb | 2445 | |
39379faa NA |
2446 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2447 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2448 | orig_super_num_devices + 1); | |
0d39376a AJ |
2449 | |
2450 | /* add sysfs device entry */ | |
5da54bc1 | 2451 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2452 | |
2196d6e8 MX |
2453 | /* |
2454 | * we've got more storage, clear any full flags on the space | |
2455 | * infos | |
2456 | */ | |
0b246afa | 2457 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2458 | |
34441361 | 2459 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2460 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2461 | |
2b82032c | 2462 | if (seeding_dev) { |
34441361 | 2463 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2464 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2465 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2466 | if (ret) { |
66642832 | 2467 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2468 | goto error_sysfs; |
005d6427 | 2469 | } |
2196d6e8 MX |
2470 | } |
2471 | ||
8e87e856 | 2472 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2473 | if (ret) { |
66642832 | 2474 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2475 | goto error_sysfs; |
2196d6e8 MX |
2476 | } |
2477 | ||
2478 | if (seeding_dev) { | |
2479 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2480 | ||
0b246afa | 2481 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2482 | if (ret) { |
66642832 | 2483 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2484 | goto error_sysfs; |
005d6427 | 2485 | } |
b2373f25 AJ |
2486 | |
2487 | /* Sprouting would change fsid of the mounted root, | |
2488 | * so rename the fsid on the sysfs | |
2489 | */ | |
2490 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa | 2491 | fs_info->fsid); |
5da54bc1 | 2492 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2493 | btrfs_warn(fs_info, |
2494 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2495 | } |
2496 | ||
3a45bb20 | 2497 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2498 | |
2b82032c YZ |
2499 | if (seeding_dev) { |
2500 | mutex_unlock(&uuid_mutex); | |
2501 | up_write(&sb->s_umount); | |
7132a262 | 2502 | unlocked = true; |
788f20eb | 2503 | |
79787eaa JM |
2504 | if (ret) /* transaction commit */ |
2505 | return ret; | |
2506 | ||
2ff7e61e | 2507 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2508 | if (ret < 0) |
0b246afa | 2509 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2510 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2511 | trans = btrfs_attach_transaction(root); |
2512 | if (IS_ERR(trans)) { | |
2513 | if (PTR_ERR(trans) == -ENOENT) | |
2514 | return 0; | |
7132a262 AJ |
2515 | ret = PTR_ERR(trans); |
2516 | trans = NULL; | |
2517 | goto error_sysfs; | |
671415b7 | 2518 | } |
3a45bb20 | 2519 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2520 | } |
c9e9f97b | 2521 | |
5a1972bd QW |
2522 | /* Update ctime/mtime for libblkid */ |
2523 | update_dev_time(device_path); | |
2b82032c | 2524 | return ret; |
79787eaa | 2525 | |
d31c32f6 | 2526 | error_sysfs: |
5da54bc1 | 2527 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2528 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2529 | mutex_lock(&fs_info->chunk_mutex); | |
2530 | list_del_rcu(&device->dev_list); | |
2531 | list_del(&device->dev_alloc_list); | |
2532 | fs_info->fs_devices->num_devices--; | |
2533 | fs_info->fs_devices->open_devices--; | |
2534 | fs_info->fs_devices->rw_devices--; | |
2535 | fs_info->fs_devices->total_devices--; | |
2536 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2537 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2538 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2539 | orig_super_total_bytes); | |
2540 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2541 | orig_super_num_devices); | |
2542 | mutex_unlock(&fs_info->chunk_mutex); | |
2543 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2544 | error_trans: |
0af2c4bf | 2545 | if (seeding_dev) |
1751e8a6 | 2546 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2547 | if (trans) |
2548 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2549 | error_free_device: |
a425f9d4 | 2550 | btrfs_free_device(device); |
2b82032c | 2551 | error: |
e525fd89 | 2552 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2553 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2554 | mutex_unlock(&uuid_mutex); |
2555 | up_write(&sb->s_umount); | |
2556 | } | |
c9e9f97b | 2557 | return ret; |
788f20eb CM |
2558 | } |
2559 | ||
d397712b CM |
2560 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2561 | struct btrfs_device *device) | |
0b86a832 CM |
2562 | { |
2563 | int ret; | |
2564 | struct btrfs_path *path; | |
0b246afa | 2565 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2566 | struct btrfs_dev_item *dev_item; |
2567 | struct extent_buffer *leaf; | |
2568 | struct btrfs_key key; | |
2569 | ||
0b86a832 CM |
2570 | path = btrfs_alloc_path(); |
2571 | if (!path) | |
2572 | return -ENOMEM; | |
2573 | ||
2574 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2575 | key.type = BTRFS_DEV_ITEM_KEY; | |
2576 | key.offset = device->devid; | |
2577 | ||
2578 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2579 | if (ret < 0) | |
2580 | goto out; | |
2581 | ||
2582 | if (ret > 0) { | |
2583 | ret = -ENOENT; | |
2584 | goto out; | |
2585 | } | |
2586 | ||
2587 | leaf = path->nodes[0]; | |
2588 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2589 | ||
2590 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2591 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2592 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2593 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2594 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2595 | btrfs_set_device_total_bytes(leaf, dev_item, |
2596 | btrfs_device_get_disk_total_bytes(device)); | |
2597 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2598 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2599 | btrfs_mark_buffer_dirty(leaf); |
2600 | ||
2601 | out: | |
2602 | btrfs_free_path(path); | |
2603 | return ret; | |
2604 | } | |
2605 | ||
2196d6e8 | 2606 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2607 | struct btrfs_device *device, u64 new_size) |
2608 | { | |
0b246afa JM |
2609 | struct btrfs_fs_info *fs_info = device->fs_info; |
2610 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2611 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2612 | u64 old_total; |
2613 | u64 diff; | |
8f18cf13 | 2614 | |
ebbede42 | 2615 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2616 | return -EACCES; |
2196d6e8 | 2617 | |
7dfb8be1 NB |
2618 | new_size = round_down(new_size, fs_info->sectorsize); |
2619 | ||
34441361 | 2620 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2621 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2622 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2623 | |
63a212ab | 2624 | if (new_size <= device->total_bytes || |
401e29c1 | 2625 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2626 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2627 | return -EINVAL; |
2196d6e8 | 2628 | } |
2b82032c | 2629 | |
0b246afa | 2630 | fs_devices = fs_info->fs_devices; |
2b82032c | 2631 | |
7dfb8be1 NB |
2632 | btrfs_set_super_total_bytes(super_copy, |
2633 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2634 | device->fs_devices->total_rw_bytes += diff; |
2635 | ||
7cc8e58d MX |
2636 | btrfs_device_set_total_bytes(device, new_size); |
2637 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2638 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2639 | if (list_empty(&device->resized_list)) |
2640 | list_add_tail(&device->resized_list, | |
2641 | &fs_devices->resized_devices); | |
34441361 | 2642 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2643 | |
8f18cf13 CM |
2644 | return btrfs_update_device(trans, device); |
2645 | } | |
2646 | ||
f4208794 | 2647 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2648 | { |
f4208794 | 2649 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2650 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2651 | int ret; |
2652 | struct btrfs_path *path; | |
2653 | struct btrfs_key key; | |
2654 | ||
8f18cf13 CM |
2655 | path = btrfs_alloc_path(); |
2656 | if (!path) | |
2657 | return -ENOMEM; | |
2658 | ||
408fbf19 | 2659 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2660 | key.offset = chunk_offset; |
2661 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2662 | ||
2663 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2664 | if (ret < 0) |
2665 | goto out; | |
2666 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2667 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2668 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2669 | ret = -ENOENT; |
2670 | goto out; | |
2671 | } | |
8f18cf13 CM |
2672 | |
2673 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2674 | if (ret < 0) |
0b246afa JM |
2675 | btrfs_handle_fs_error(fs_info, ret, |
2676 | "Failed to delete chunk item."); | |
79787eaa | 2677 | out: |
8f18cf13 | 2678 | btrfs_free_path(path); |
65a246c5 | 2679 | return ret; |
8f18cf13 CM |
2680 | } |
2681 | ||
408fbf19 | 2682 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2683 | { |
0b246afa | 2684 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2685 | struct btrfs_disk_key *disk_key; |
2686 | struct btrfs_chunk *chunk; | |
2687 | u8 *ptr; | |
2688 | int ret = 0; | |
2689 | u32 num_stripes; | |
2690 | u32 array_size; | |
2691 | u32 len = 0; | |
2692 | u32 cur; | |
2693 | struct btrfs_key key; | |
2694 | ||
34441361 | 2695 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2696 | array_size = btrfs_super_sys_array_size(super_copy); |
2697 | ||
2698 | ptr = super_copy->sys_chunk_array; | |
2699 | cur = 0; | |
2700 | ||
2701 | while (cur < array_size) { | |
2702 | disk_key = (struct btrfs_disk_key *)ptr; | |
2703 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2704 | ||
2705 | len = sizeof(*disk_key); | |
2706 | ||
2707 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2708 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2709 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2710 | len += btrfs_chunk_item_size(num_stripes); | |
2711 | } else { | |
2712 | ret = -EIO; | |
2713 | break; | |
2714 | } | |
408fbf19 | 2715 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2716 | key.offset == chunk_offset) { |
2717 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2718 | array_size -= len; | |
2719 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2720 | } else { | |
2721 | ptr += len; | |
2722 | cur += len; | |
2723 | } | |
2724 | } | |
34441361 | 2725 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2726 | return ret; |
2727 | } | |
2728 | ||
60ca842e OS |
2729 | /* |
2730 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2731 | * @logical: Logical block offset in bytes. | |
2732 | * @length: Length of extent in bytes. | |
2733 | * | |
2734 | * Return: Chunk mapping or ERR_PTR. | |
2735 | */ | |
2736 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2737 | u64 logical, u64 length) | |
592d92ee LB |
2738 | { |
2739 | struct extent_map_tree *em_tree; | |
2740 | struct extent_map *em; | |
2741 | ||
2742 | em_tree = &fs_info->mapping_tree.map_tree; | |
2743 | read_lock(&em_tree->lock); | |
2744 | em = lookup_extent_mapping(em_tree, logical, length); | |
2745 | read_unlock(&em_tree->lock); | |
2746 | ||
2747 | if (!em) { | |
2748 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2749 | logical, length); | |
2750 | return ERR_PTR(-EINVAL); | |
2751 | } | |
2752 | ||
2753 | if (em->start > logical || em->start + em->len < logical) { | |
2754 | btrfs_crit(fs_info, | |
2755 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2756 | logical, length, em->start, em->start + em->len); | |
2757 | free_extent_map(em); | |
2758 | return ERR_PTR(-EINVAL); | |
2759 | } | |
2760 | ||
2761 | /* callers are responsible for dropping em's ref. */ | |
2762 | return em; | |
2763 | } | |
2764 | ||
97aff912 | 2765 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2766 | { |
97aff912 | 2767 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2768 | struct extent_map *em; |
2769 | struct map_lookup *map; | |
2196d6e8 | 2770 | u64 dev_extent_len = 0; |
47ab2a6c | 2771 | int i, ret = 0; |
0b246afa | 2772 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2773 | |
60ca842e | 2774 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2775 | if (IS_ERR(em)) { |
47ab2a6c JB |
2776 | /* |
2777 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2778 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2779 | * do anything we still error out. |
2780 | */ | |
2781 | ASSERT(0); | |
592d92ee | 2782 | return PTR_ERR(em); |
47ab2a6c | 2783 | } |
95617d69 | 2784 | map = em->map_lookup; |
34441361 | 2785 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2786 | check_system_chunk(trans, map->type); |
34441361 | 2787 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2788 | |
57ba4cb8 FM |
2789 | /* |
2790 | * Take the device list mutex to prevent races with the final phase of | |
2791 | * a device replace operation that replaces the device object associated | |
2792 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2793 | */ | |
2794 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2795 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2796 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2797 | ret = btrfs_free_dev_extent(trans, device, |
2798 | map->stripes[i].physical, | |
2799 | &dev_extent_len); | |
47ab2a6c | 2800 | if (ret) { |
57ba4cb8 | 2801 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2802 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2803 | goto out; |
2804 | } | |
a061fc8d | 2805 | |
2196d6e8 | 2806 | if (device->bytes_used > 0) { |
34441361 | 2807 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2808 | btrfs_device_set_bytes_used(device, |
2809 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2810 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2811 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2812 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2813 | } |
a061fc8d | 2814 | |
dfe25020 CM |
2815 | if (map->stripes[i].dev) { |
2816 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2817 | if (ret) { |
57ba4cb8 | 2818 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2819 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2820 | goto out; |
2821 | } | |
dfe25020 | 2822 | } |
8f18cf13 | 2823 | } |
57ba4cb8 FM |
2824 | mutex_unlock(&fs_devices->device_list_mutex); |
2825 | ||
f4208794 | 2826 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 2827 | if (ret) { |
66642832 | 2828 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2829 | goto out; |
2830 | } | |
8f18cf13 | 2831 | |
6bccf3ab | 2832 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2833 | |
8f18cf13 | 2834 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2835 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2836 | if (ret) { |
66642832 | 2837 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2838 | goto out; |
2839 | } | |
8f18cf13 CM |
2840 | } |
2841 | ||
5a98ec01 | 2842 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 2843 | if (ret) { |
66642832 | 2844 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2845 | goto out; |
2846 | } | |
2b82032c | 2847 | |
47ab2a6c | 2848 | out: |
2b82032c YZ |
2849 | /* once for us */ |
2850 | free_extent_map(em); | |
47ab2a6c JB |
2851 | return ret; |
2852 | } | |
2b82032c | 2853 | |
5b4aacef | 2854 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2855 | { |
5b4aacef | 2856 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2857 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2858 | int ret; |
2b82032c | 2859 | |
67c5e7d4 FM |
2860 | /* |
2861 | * Prevent races with automatic removal of unused block groups. | |
2862 | * After we relocate and before we remove the chunk with offset | |
2863 | * chunk_offset, automatic removal of the block group can kick in, | |
2864 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2865 | * | |
2866 | * Make sure to acquire this mutex before doing a tree search (dev | |
2867 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2868 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2869 | * we release the path used to search the chunk/dev tree and before | |
2870 | * the current task acquires this mutex and calls us. | |
2871 | */ | |
a32bf9a3 | 2872 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 2873 | |
0b246afa | 2874 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2875 | if (ret) |
2876 | return -ENOSPC; | |
2877 | ||
2878 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 2879 | btrfs_scrub_pause(fs_info); |
0b246afa | 2880 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2881 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2882 | if (ret) |
2883 | return ret; | |
2884 | ||
75cb379d JM |
2885 | /* |
2886 | * We add the kobjects here (and after forcing data chunk creation) | |
2887 | * since relocation is the only place we'll create chunks of a new | |
2888 | * type at runtime. The only place where we'll remove the last | |
2889 | * chunk of a type is the call immediately below this one. Even | |
2890 | * so, we're protected against races with the cleaner thread since | |
2891 | * we're covered by the delete_unused_bgs_mutex. | |
2892 | */ | |
2893 | btrfs_add_raid_kobjects(fs_info); | |
2894 | ||
19c4d2f9 CM |
2895 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2896 | chunk_offset); | |
2897 | if (IS_ERR(trans)) { | |
2898 | ret = PTR_ERR(trans); | |
2899 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2900 | return ret; | |
2901 | } | |
2902 | ||
47ab2a6c | 2903 | /* |
19c4d2f9 CM |
2904 | * step two, delete the device extents and the |
2905 | * chunk tree entries | |
47ab2a6c | 2906 | */ |
97aff912 | 2907 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 2908 | btrfs_end_transaction(trans); |
19c4d2f9 | 2909 | return ret; |
2b82032c YZ |
2910 | } |
2911 | ||
2ff7e61e | 2912 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2913 | { |
0b246afa | 2914 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2915 | struct btrfs_path *path; |
2916 | struct extent_buffer *leaf; | |
2917 | struct btrfs_chunk *chunk; | |
2918 | struct btrfs_key key; | |
2919 | struct btrfs_key found_key; | |
2b82032c | 2920 | u64 chunk_type; |
ba1bf481 JB |
2921 | bool retried = false; |
2922 | int failed = 0; | |
2b82032c YZ |
2923 | int ret; |
2924 | ||
2925 | path = btrfs_alloc_path(); | |
2926 | if (!path) | |
2927 | return -ENOMEM; | |
2928 | ||
ba1bf481 | 2929 | again: |
2b82032c YZ |
2930 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2931 | key.offset = (u64)-1; | |
2932 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2933 | ||
2934 | while (1) { | |
0b246afa | 2935 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2936 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2937 | if (ret < 0) { |
0b246afa | 2938 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2939 | goto error; |
67c5e7d4 | 2940 | } |
79787eaa | 2941 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2942 | |
2943 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2944 | key.type); | |
67c5e7d4 | 2945 | if (ret) |
0b246afa | 2946 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
2947 | if (ret < 0) |
2948 | goto error; | |
2949 | if (ret > 0) | |
2950 | break; | |
1a40e23b | 2951 | |
2b82032c YZ |
2952 | leaf = path->nodes[0]; |
2953 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2954 | |
2b82032c YZ |
2955 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2956 | struct btrfs_chunk); | |
2957 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2958 | btrfs_release_path(path); |
8f18cf13 | 2959 | |
2b82032c | 2960 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 2961 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
2962 | if (ret == -ENOSPC) |
2963 | failed++; | |
14586651 HS |
2964 | else |
2965 | BUG_ON(ret); | |
2b82032c | 2966 | } |
0b246afa | 2967 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 2968 | |
2b82032c YZ |
2969 | if (found_key.offset == 0) |
2970 | break; | |
2971 | key.offset = found_key.offset - 1; | |
2972 | } | |
2973 | ret = 0; | |
ba1bf481 JB |
2974 | if (failed && !retried) { |
2975 | failed = 0; | |
2976 | retried = true; | |
2977 | goto again; | |
fae7f21c | 2978 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2979 | ret = -ENOSPC; |
2980 | } | |
2b82032c YZ |
2981 | error: |
2982 | btrfs_free_path(path); | |
2983 | return ret; | |
8f18cf13 CM |
2984 | } |
2985 | ||
a6f93c71 LB |
2986 | /* |
2987 | * return 1 : allocate a data chunk successfully, | |
2988 | * return <0: errors during allocating a data chunk, | |
2989 | * return 0 : no need to allocate a data chunk. | |
2990 | */ | |
2991 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
2992 | u64 chunk_offset) | |
2993 | { | |
2994 | struct btrfs_block_group_cache *cache; | |
2995 | u64 bytes_used; | |
2996 | u64 chunk_type; | |
2997 | ||
2998 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2999 | ASSERT(cache); | |
3000 | chunk_type = cache->flags; | |
3001 | btrfs_put_block_group(cache); | |
3002 | ||
3003 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3004 | spin_lock(&fs_info->data_sinfo->lock); | |
3005 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3006 | spin_unlock(&fs_info->data_sinfo->lock); | |
3007 | ||
3008 | if (!bytes_used) { | |
3009 | struct btrfs_trans_handle *trans; | |
3010 | int ret; | |
3011 | ||
3012 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3013 | if (IS_ERR(trans)) | |
3014 | return PTR_ERR(trans); | |
3015 | ||
43a7e99d | 3016 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3017 | BTRFS_BLOCK_GROUP_DATA); |
3018 | btrfs_end_transaction(trans); | |
3019 | if (ret < 0) | |
3020 | return ret; | |
3021 | ||
75cb379d JM |
3022 | btrfs_add_raid_kobjects(fs_info); |
3023 | ||
a6f93c71 LB |
3024 | return 1; |
3025 | } | |
3026 | } | |
3027 | return 0; | |
3028 | } | |
3029 | ||
6bccf3ab | 3030 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3031 | struct btrfs_balance_control *bctl) |
3032 | { | |
6bccf3ab | 3033 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3034 | struct btrfs_trans_handle *trans; |
3035 | struct btrfs_balance_item *item; | |
3036 | struct btrfs_disk_balance_args disk_bargs; | |
3037 | struct btrfs_path *path; | |
3038 | struct extent_buffer *leaf; | |
3039 | struct btrfs_key key; | |
3040 | int ret, err; | |
3041 | ||
3042 | path = btrfs_alloc_path(); | |
3043 | if (!path) | |
3044 | return -ENOMEM; | |
3045 | ||
3046 | trans = btrfs_start_transaction(root, 0); | |
3047 | if (IS_ERR(trans)) { | |
3048 | btrfs_free_path(path); | |
3049 | return PTR_ERR(trans); | |
3050 | } | |
3051 | ||
3052 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3053 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3054 | key.offset = 0; |
3055 | ||
3056 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3057 | sizeof(*item)); | |
3058 | if (ret) | |
3059 | goto out; | |
3060 | ||
3061 | leaf = path->nodes[0]; | |
3062 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3063 | ||
b159fa28 | 3064 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3065 | |
3066 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3067 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3068 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3069 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3070 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3071 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3072 | ||
3073 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3074 | ||
3075 | btrfs_mark_buffer_dirty(leaf); | |
3076 | out: | |
3077 | btrfs_free_path(path); | |
3a45bb20 | 3078 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3079 | if (err && !ret) |
3080 | ret = err; | |
3081 | return ret; | |
3082 | } | |
3083 | ||
6bccf3ab | 3084 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3085 | { |
6bccf3ab | 3086 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3087 | struct btrfs_trans_handle *trans; |
3088 | struct btrfs_path *path; | |
3089 | struct btrfs_key key; | |
3090 | int ret, err; | |
3091 | ||
3092 | path = btrfs_alloc_path(); | |
3093 | if (!path) | |
3094 | return -ENOMEM; | |
3095 | ||
3096 | trans = btrfs_start_transaction(root, 0); | |
3097 | if (IS_ERR(trans)) { | |
3098 | btrfs_free_path(path); | |
3099 | return PTR_ERR(trans); | |
3100 | } | |
3101 | ||
3102 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3103 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3104 | key.offset = 0; |
3105 | ||
3106 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3107 | if (ret < 0) | |
3108 | goto out; | |
3109 | if (ret > 0) { | |
3110 | ret = -ENOENT; | |
3111 | goto out; | |
3112 | } | |
3113 | ||
3114 | ret = btrfs_del_item(trans, root, path); | |
3115 | out: | |
3116 | btrfs_free_path(path); | |
3a45bb20 | 3117 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3118 | if (err && !ret) |
3119 | ret = err; | |
3120 | return ret; | |
3121 | } | |
3122 | ||
59641015 ID |
3123 | /* |
3124 | * This is a heuristic used to reduce the number of chunks balanced on | |
3125 | * resume after balance was interrupted. | |
3126 | */ | |
3127 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3128 | { | |
3129 | /* | |
3130 | * Turn on soft mode for chunk types that were being converted. | |
3131 | */ | |
3132 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3133 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3134 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3135 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3136 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3137 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3138 | ||
3139 | /* | |
3140 | * Turn on usage filter if is not already used. The idea is | |
3141 | * that chunks that we have already balanced should be | |
3142 | * reasonably full. Don't do it for chunks that are being | |
3143 | * converted - that will keep us from relocating unconverted | |
3144 | * (albeit full) chunks. | |
3145 | */ | |
3146 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3147 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3148 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3149 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3150 | bctl->data.usage = 90; | |
3151 | } | |
3152 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3153 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3154 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3155 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3156 | bctl->sys.usage = 90; | |
3157 | } | |
3158 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3159 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3160 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3161 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3162 | bctl->meta.usage = 90; | |
3163 | } | |
3164 | } | |
3165 | ||
149196a2 DS |
3166 | /* |
3167 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3168 | */ | |
3169 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3170 | { |
3171 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3172 | int ret; |
c9e9f97b ID |
3173 | |
3174 | BUG_ON(!fs_info->balance_ctl); | |
3175 | ||
3176 | spin_lock(&fs_info->balance_lock); | |
3177 | fs_info->balance_ctl = NULL; | |
3178 | spin_unlock(&fs_info->balance_lock); | |
3179 | ||
3180 | kfree(bctl); | |
149196a2 DS |
3181 | ret = del_balance_item(fs_info); |
3182 | if (ret) | |
3183 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3184 | } |
3185 | ||
ed25e9b2 ID |
3186 | /* |
3187 | * Balance filters. Return 1 if chunk should be filtered out | |
3188 | * (should not be balanced). | |
3189 | */ | |
899c81ea | 3190 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3191 | struct btrfs_balance_args *bargs) |
3192 | { | |
899c81ea ID |
3193 | chunk_type = chunk_to_extended(chunk_type) & |
3194 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3195 | |
899c81ea | 3196 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3197 | return 0; |
3198 | ||
3199 | return 1; | |
3200 | } | |
3201 | ||
dba72cb3 | 3202 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3203 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3204 | { |
3205 | struct btrfs_block_group_cache *cache; | |
3206 | u64 chunk_used; | |
3207 | u64 user_thresh_min; | |
3208 | u64 user_thresh_max; | |
3209 | int ret = 1; | |
3210 | ||
3211 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3212 | chunk_used = btrfs_block_group_used(&cache->item); | |
3213 | ||
3214 | if (bargs->usage_min == 0) | |
3215 | user_thresh_min = 0; | |
3216 | else | |
3217 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3218 | bargs->usage_min); | |
3219 | ||
3220 | if (bargs->usage_max == 0) | |
3221 | user_thresh_max = 1; | |
3222 | else if (bargs->usage_max > 100) | |
3223 | user_thresh_max = cache->key.offset; | |
3224 | else | |
3225 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3226 | bargs->usage_max); | |
3227 | ||
3228 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3229 | ret = 0; | |
3230 | ||
3231 | btrfs_put_block_group(cache); | |
3232 | return ret; | |
3233 | } | |
3234 | ||
dba72cb3 | 3235 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3236 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3237 | { |
3238 | struct btrfs_block_group_cache *cache; | |
3239 | u64 chunk_used, user_thresh; | |
3240 | int ret = 1; | |
3241 | ||
3242 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3243 | chunk_used = btrfs_block_group_used(&cache->item); | |
3244 | ||
bc309467 | 3245 | if (bargs->usage_min == 0) |
3e39cea6 | 3246 | user_thresh = 1; |
a105bb88 ID |
3247 | else if (bargs->usage > 100) |
3248 | user_thresh = cache->key.offset; | |
3249 | else | |
3250 | user_thresh = div_factor_fine(cache->key.offset, | |
3251 | bargs->usage); | |
3252 | ||
5ce5b3c0 ID |
3253 | if (chunk_used < user_thresh) |
3254 | ret = 0; | |
3255 | ||
3256 | btrfs_put_block_group(cache); | |
3257 | return ret; | |
3258 | } | |
3259 | ||
409d404b ID |
3260 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3261 | struct btrfs_chunk *chunk, | |
3262 | struct btrfs_balance_args *bargs) | |
3263 | { | |
3264 | struct btrfs_stripe *stripe; | |
3265 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3266 | int i; | |
3267 | ||
3268 | for (i = 0; i < num_stripes; i++) { | |
3269 | stripe = btrfs_stripe_nr(chunk, i); | |
3270 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3271 | return 0; | |
3272 | } | |
3273 | ||
3274 | return 1; | |
3275 | } | |
3276 | ||
94e60d5a ID |
3277 | /* [pstart, pend) */ |
3278 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3279 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3280 | struct btrfs_balance_args *bargs) |
3281 | { | |
3282 | struct btrfs_stripe *stripe; | |
3283 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3284 | u64 stripe_offset; | |
3285 | u64 stripe_length; | |
3286 | int factor; | |
3287 | int i; | |
3288 | ||
3289 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3290 | return 0; | |
3291 | ||
3292 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3293 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3294 | factor = num_stripes / 2; | |
3295 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3296 | factor = num_stripes - 1; | |
3297 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3298 | factor = num_stripes - 2; | |
3299 | } else { | |
3300 | factor = num_stripes; | |
3301 | } | |
94e60d5a ID |
3302 | |
3303 | for (i = 0; i < num_stripes; i++) { | |
3304 | stripe = btrfs_stripe_nr(chunk, i); | |
3305 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3306 | continue; | |
3307 | ||
3308 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3309 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3310 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3311 | |
3312 | if (stripe_offset < bargs->pend && | |
3313 | stripe_offset + stripe_length > bargs->pstart) | |
3314 | return 0; | |
3315 | } | |
3316 | ||
3317 | return 1; | |
3318 | } | |
3319 | ||
ea67176a ID |
3320 | /* [vstart, vend) */ |
3321 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3322 | struct btrfs_chunk *chunk, | |
3323 | u64 chunk_offset, | |
3324 | struct btrfs_balance_args *bargs) | |
3325 | { | |
3326 | if (chunk_offset < bargs->vend && | |
3327 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3328 | /* at least part of the chunk is inside this vrange */ | |
3329 | return 0; | |
3330 | ||
3331 | return 1; | |
3332 | } | |
3333 | ||
dee32d0a GAP |
3334 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3335 | struct btrfs_chunk *chunk, | |
3336 | struct btrfs_balance_args *bargs) | |
3337 | { | |
3338 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3339 | ||
3340 | if (bargs->stripes_min <= num_stripes | |
3341 | && num_stripes <= bargs->stripes_max) | |
3342 | return 0; | |
3343 | ||
3344 | return 1; | |
3345 | } | |
3346 | ||
899c81ea | 3347 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3348 | struct btrfs_balance_args *bargs) |
3349 | { | |
3350 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3351 | return 0; | |
3352 | ||
899c81ea ID |
3353 | chunk_type = chunk_to_extended(chunk_type) & |
3354 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3355 | |
899c81ea | 3356 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3357 | return 1; |
3358 | ||
3359 | return 0; | |
3360 | } | |
3361 | ||
2ff7e61e | 3362 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3363 | struct extent_buffer *leaf, |
3364 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3365 | { | |
0b246afa | 3366 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3367 | struct btrfs_balance_args *bargs = NULL; |
3368 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3369 | ||
3370 | /* type filter */ | |
3371 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3372 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3373 | return 0; | |
3374 | } | |
3375 | ||
3376 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3377 | bargs = &bctl->data; | |
3378 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3379 | bargs = &bctl->sys; | |
3380 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3381 | bargs = &bctl->meta; | |
3382 | ||
ed25e9b2 ID |
3383 | /* profiles filter */ |
3384 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3385 | chunk_profiles_filter(chunk_type, bargs)) { | |
3386 | return 0; | |
5ce5b3c0 ID |
3387 | } |
3388 | ||
3389 | /* usage filter */ | |
3390 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3391 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3392 | return 0; |
bc309467 | 3393 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3394 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3395 | return 0; |
409d404b ID |
3396 | } |
3397 | ||
3398 | /* devid filter */ | |
3399 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3400 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3401 | return 0; | |
94e60d5a ID |
3402 | } |
3403 | ||
3404 | /* drange filter, makes sense only with devid filter */ | |
3405 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3406 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3407 | return 0; |
ea67176a ID |
3408 | } |
3409 | ||
3410 | /* vrange filter */ | |
3411 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3412 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3413 | return 0; | |
ed25e9b2 ID |
3414 | } |
3415 | ||
dee32d0a GAP |
3416 | /* stripes filter */ |
3417 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3418 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3419 | return 0; | |
3420 | } | |
3421 | ||
cfa4c961 ID |
3422 | /* soft profile changing mode */ |
3423 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3424 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3425 | return 0; | |
3426 | } | |
3427 | ||
7d824b6f DS |
3428 | /* |
3429 | * limited by count, must be the last filter | |
3430 | */ | |
3431 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3432 | if (bargs->limit == 0) | |
3433 | return 0; | |
3434 | else | |
3435 | bargs->limit--; | |
12907fc7 DS |
3436 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3437 | /* | |
3438 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3439 | * determined here because we do not have the global information |
12907fc7 DS |
3440 | * about the count of all chunks that satisfy the filters. |
3441 | */ | |
3442 | if (bargs->limit_max == 0) | |
3443 | return 0; | |
3444 | else | |
3445 | bargs->limit_max--; | |
7d824b6f DS |
3446 | } |
3447 | ||
f43ffb60 ID |
3448 | return 1; |
3449 | } | |
3450 | ||
c9e9f97b | 3451 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3452 | { |
19a39dce | 3453 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3454 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3455 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3456 | struct list_head *devices; | |
ec44a35c CM |
3457 | struct btrfs_device *device; |
3458 | u64 old_size; | |
3459 | u64 size_to_free; | |
12907fc7 | 3460 | u64 chunk_type; |
f43ffb60 | 3461 | struct btrfs_chunk *chunk; |
5a488b9d | 3462 | struct btrfs_path *path = NULL; |
ec44a35c | 3463 | struct btrfs_key key; |
ec44a35c | 3464 | struct btrfs_key found_key; |
c9e9f97b | 3465 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3466 | struct extent_buffer *leaf; |
3467 | int slot; | |
c9e9f97b ID |
3468 | int ret; |
3469 | int enospc_errors = 0; | |
19a39dce | 3470 | bool counting = true; |
12907fc7 | 3471 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3472 | u64 limit_data = bctl->data.limit; |
3473 | u64 limit_meta = bctl->meta.limit; | |
3474 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3475 | u32 count_data = 0; |
3476 | u32 count_meta = 0; | |
3477 | u32 count_sys = 0; | |
2c9fe835 | 3478 | int chunk_reserved = 0; |
ec44a35c | 3479 | |
ec44a35c | 3480 | /* step one make some room on all the devices */ |
c9e9f97b | 3481 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3482 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3483 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3484 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3485 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3486 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3487 | btrfs_device_get_total_bytes(device) - |
3488 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3489 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3490 | continue; |
3491 | ||
3492 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3493 | if (ret == -ENOSPC) |
3494 | break; | |
5a488b9d LB |
3495 | if (ret) { |
3496 | /* btrfs_shrink_device never returns ret > 0 */ | |
3497 | WARN_ON(ret > 0); | |
3498 | goto error; | |
3499 | } | |
ec44a35c | 3500 | |
a22285a6 | 3501 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3502 | if (IS_ERR(trans)) { |
3503 | ret = PTR_ERR(trans); | |
3504 | btrfs_info_in_rcu(fs_info, | |
3505 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3506 | rcu_str_deref(device->name), ret, | |
3507 | old_size, old_size - size_to_free); | |
3508 | goto error; | |
3509 | } | |
ec44a35c CM |
3510 | |
3511 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3512 | if (ret) { |
3a45bb20 | 3513 | btrfs_end_transaction(trans); |
5a488b9d LB |
3514 | /* btrfs_grow_device never returns ret > 0 */ |
3515 | WARN_ON(ret > 0); | |
3516 | btrfs_info_in_rcu(fs_info, | |
3517 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3518 | rcu_str_deref(device->name), ret, | |
3519 | old_size, old_size - size_to_free); | |
3520 | goto error; | |
3521 | } | |
ec44a35c | 3522 | |
3a45bb20 | 3523 | btrfs_end_transaction(trans); |
ec44a35c CM |
3524 | } |
3525 | ||
3526 | /* step two, relocate all the chunks */ | |
3527 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3528 | if (!path) { |
3529 | ret = -ENOMEM; | |
3530 | goto error; | |
3531 | } | |
19a39dce ID |
3532 | |
3533 | /* zero out stat counters */ | |
3534 | spin_lock(&fs_info->balance_lock); | |
3535 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3536 | spin_unlock(&fs_info->balance_lock); | |
3537 | again: | |
7d824b6f | 3538 | if (!counting) { |
12907fc7 DS |
3539 | /* |
3540 | * The single value limit and min/max limits use the same bytes | |
3541 | * in the | |
3542 | */ | |
7d824b6f DS |
3543 | bctl->data.limit = limit_data; |
3544 | bctl->meta.limit = limit_meta; | |
3545 | bctl->sys.limit = limit_sys; | |
3546 | } | |
ec44a35c CM |
3547 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3548 | key.offset = (u64)-1; | |
3549 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3550 | ||
d397712b | 3551 | while (1) { |
19a39dce | 3552 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3553 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3554 | ret = -ECANCELED; |
3555 | goto error; | |
3556 | } | |
3557 | ||
67c5e7d4 | 3558 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3559 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3560 | if (ret < 0) { |
3561 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3562 | goto error; |
67c5e7d4 | 3563 | } |
ec44a35c CM |
3564 | |
3565 | /* | |
3566 | * this shouldn't happen, it means the last relocate | |
3567 | * failed | |
3568 | */ | |
3569 | if (ret == 0) | |
c9e9f97b | 3570 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3571 | |
3572 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3573 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3574 | if (ret) { |
67c5e7d4 | 3575 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3576 | ret = 0; |
ec44a35c | 3577 | break; |
c9e9f97b | 3578 | } |
7d9eb12c | 3579 | |
f43ffb60 ID |
3580 | leaf = path->nodes[0]; |
3581 | slot = path->slots[0]; | |
3582 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3583 | |
67c5e7d4 FM |
3584 | if (found_key.objectid != key.objectid) { |
3585 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3586 | break; |
67c5e7d4 | 3587 | } |
7d9eb12c | 3588 | |
f43ffb60 | 3589 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3590 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3591 | |
19a39dce ID |
3592 | if (!counting) { |
3593 | spin_lock(&fs_info->balance_lock); | |
3594 | bctl->stat.considered++; | |
3595 | spin_unlock(&fs_info->balance_lock); | |
3596 | } | |
3597 | ||
2ff7e61e | 3598 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3599 | found_key.offset); |
2c9fe835 | 3600 | |
b3b4aa74 | 3601 | btrfs_release_path(path); |
67c5e7d4 FM |
3602 | if (!ret) { |
3603 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3604 | goto loop; |
67c5e7d4 | 3605 | } |
f43ffb60 | 3606 | |
19a39dce | 3607 | if (counting) { |
67c5e7d4 | 3608 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3609 | spin_lock(&fs_info->balance_lock); |
3610 | bctl->stat.expected++; | |
3611 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3612 | |
3613 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3614 | count_data++; | |
3615 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3616 | count_sys++; | |
3617 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3618 | count_meta++; | |
3619 | ||
3620 | goto loop; | |
3621 | } | |
3622 | ||
3623 | /* | |
3624 | * Apply limit_min filter, no need to check if the LIMITS | |
3625 | * filter is used, limit_min is 0 by default | |
3626 | */ | |
3627 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3628 | count_data < bctl->data.limit_min) | |
3629 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3630 | count_meta < bctl->meta.limit_min) | |
3631 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3632 | count_sys < bctl->sys.limit_min)) { | |
3633 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3634 | goto loop; |
3635 | } | |
3636 | ||
a6f93c71 LB |
3637 | if (!chunk_reserved) { |
3638 | /* | |
3639 | * We may be relocating the only data chunk we have, | |
3640 | * which could potentially end up with losing data's | |
3641 | * raid profile, so lets allocate an empty one in | |
3642 | * advance. | |
3643 | */ | |
3644 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3645 | found_key.offset); | |
2c9fe835 ZL |
3646 | if (ret < 0) { |
3647 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3648 | goto error; | |
a6f93c71 LB |
3649 | } else if (ret == 1) { |
3650 | chunk_reserved = 1; | |
2c9fe835 | 3651 | } |
2c9fe835 ZL |
3652 | } |
3653 | ||
5b4aacef | 3654 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3655 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3656 | if (ret == -ENOSPC) { |
c9e9f97b | 3657 | enospc_errors++; |
eede2bf3 OS |
3658 | } else if (ret == -ETXTBSY) { |
3659 | btrfs_info(fs_info, | |
3660 | "skipping relocation of block group %llu due to active swapfile", | |
3661 | found_key.offset); | |
3662 | ret = 0; | |
3663 | } else if (ret) { | |
3664 | goto error; | |
19a39dce ID |
3665 | } else { |
3666 | spin_lock(&fs_info->balance_lock); | |
3667 | bctl->stat.completed++; | |
3668 | spin_unlock(&fs_info->balance_lock); | |
3669 | } | |
f43ffb60 | 3670 | loop: |
795a3321 ID |
3671 | if (found_key.offset == 0) |
3672 | break; | |
ba1bf481 | 3673 | key.offset = found_key.offset - 1; |
ec44a35c | 3674 | } |
c9e9f97b | 3675 | |
19a39dce ID |
3676 | if (counting) { |
3677 | btrfs_release_path(path); | |
3678 | counting = false; | |
3679 | goto again; | |
3680 | } | |
ec44a35c CM |
3681 | error: |
3682 | btrfs_free_path(path); | |
c9e9f97b | 3683 | if (enospc_errors) { |
efe120a0 | 3684 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3685 | enospc_errors); |
c9e9f97b ID |
3686 | if (!ret) |
3687 | ret = -ENOSPC; | |
3688 | } | |
3689 | ||
ec44a35c CM |
3690 | return ret; |
3691 | } | |
3692 | ||
0c460c0d ID |
3693 | /** |
3694 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3695 | * @flags: profile to validate | |
3696 | * @extended: if true @flags is treated as an extended profile | |
3697 | */ | |
3698 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3699 | { | |
3700 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3701 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3702 | ||
3703 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3704 | ||
3705 | /* 1) check that all other bits are zeroed */ | |
3706 | if (flags & ~mask) | |
3707 | return 0; | |
3708 | ||
3709 | /* 2) see if profile is reduced */ | |
3710 | if (flags == 0) | |
3711 | return !extended; /* "0" is valid for usual profiles */ | |
3712 | ||
3713 | /* true if exactly one bit set */ | |
818255fe | 3714 | return is_power_of_2(flags); |
0c460c0d ID |
3715 | } |
3716 | ||
837d5b6e ID |
3717 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3718 | { | |
a7e99c69 ID |
3719 | /* cancel requested || normal exit path */ |
3720 | return atomic_read(&fs_info->balance_cancel_req) || | |
3721 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3722 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3723 | } |
3724 | ||
bdcd3c97 AM |
3725 | /* Non-zero return value signifies invalidity */ |
3726 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3727 | u64 allowed) | |
3728 | { | |
3729 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3730 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3731 | (bctl_arg->target & ~allowed))); | |
3732 | } | |
3733 | ||
c9e9f97b | 3734 | /* |
dccdb07b | 3735 | * Should be called with balance mutexe held |
c9e9f97b | 3736 | */ |
6fcf6e2b DS |
3737 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
3738 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
3739 | struct btrfs_ioctl_balance_args *bargs) |
3740 | { | |
14506127 | 3741 | u64 meta_target, data_target; |
f43ffb60 | 3742 | u64 allowed; |
e4837f8f | 3743 | int mixed = 0; |
c9e9f97b | 3744 | int ret; |
8dabb742 | 3745 | u64 num_devices; |
de98ced9 | 3746 | unsigned seq; |
c9e9f97b | 3747 | |
837d5b6e | 3748 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3749 | atomic_read(&fs_info->balance_pause_req) || |
3750 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3751 | ret = -EINVAL; |
3752 | goto out; | |
3753 | } | |
3754 | ||
e4837f8f ID |
3755 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3756 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3757 | mixed = 1; | |
3758 | ||
f43ffb60 ID |
3759 | /* |
3760 | * In case of mixed groups both data and meta should be picked, | |
3761 | * and identical options should be given for both of them. | |
3762 | */ | |
e4837f8f ID |
3763 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3764 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3765 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3766 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3767 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 3768 | btrfs_err(fs_info, |
6dac13f8 | 3769 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
3770 | ret = -EINVAL; |
3771 | goto out; | |
3772 | } | |
3773 | } | |
3774 | ||
1da73967 AJ |
3775 | num_devices = btrfs_num_devices(fs_info); |
3776 | ||
88be159c AH |
3777 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3778 | if (num_devices > 1) | |
e4d8ec0f | 3779 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3780 | if (num_devices > 2) |
3781 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3782 | if (num_devices > 3) | |
3783 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3784 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3785 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
3786 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
3787 | ||
5d163e0e | 3788 | btrfs_err(fs_info, |
6dac13f8 AJ |
3789 | "balance: invalid convert data profile %s", |
3790 | get_raid_name(index)); | |
e4d8ec0f ID |
3791 | ret = -EINVAL; |
3792 | goto out; | |
3793 | } | |
bdcd3c97 | 3794 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
3795 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
3796 | ||
efe120a0 | 3797 | btrfs_err(fs_info, |
6dac13f8 AJ |
3798 | "balance: invalid convert metadata profile %s", |
3799 | get_raid_name(index)); | |
e4d8ec0f ID |
3800 | ret = -EINVAL; |
3801 | goto out; | |
3802 | } | |
bdcd3c97 | 3803 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
3804 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
3805 | ||
efe120a0 | 3806 | btrfs_err(fs_info, |
6dac13f8 AJ |
3807 | "balance: invalid convert system profile %s", |
3808 | get_raid_name(index)); | |
e4d8ec0f ID |
3809 | ret = -EINVAL; |
3810 | goto out; | |
3811 | } | |
3812 | ||
e4d8ec0f ID |
3813 | /* allow to reduce meta or sys integrity only if force set */ |
3814 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3815 | BTRFS_BLOCK_GROUP_RAID10 | |
3816 | BTRFS_BLOCK_GROUP_RAID5 | | |
3817 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3818 | do { |
3819 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3820 | ||
3821 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3822 | (fs_info->avail_system_alloc_bits & allowed) && | |
3823 | !(bctl->sys.target & allowed)) || | |
3824 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3825 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3826 | !(bctl->meta.target & allowed))) { | |
3827 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e | 3828 | btrfs_info(fs_info, |
6dac13f8 | 3829 | "balance: force reducing metadata integrity"); |
de98ced9 | 3830 | } else { |
5d163e0e | 3831 | btrfs_err(fs_info, |
6dac13f8 | 3832 | "balance: reduces metadata integrity, use --force if you want this"); |
de98ced9 MX |
3833 | ret = -EINVAL; |
3834 | goto out; | |
3835 | } | |
e4d8ec0f | 3836 | } |
de98ced9 | 3837 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3838 | |
14506127 AB |
3839 | /* if we're not converting, the target field is uninitialized */ |
3840 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3841 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3842 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3843 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3844 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3845 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
3846 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
3847 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
3848 | ||
ee592d07 | 3849 | btrfs_warn(fs_info, |
6dac13f8 AJ |
3850 | "balance: metadata profile %s has lower redundancy than data profile %s", |
3851 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
3852 | } |
3853 | ||
6bccf3ab | 3854 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3855 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3856 | goto out; |
3857 | ||
59641015 ID |
3858 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3859 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
3860 | BUG_ON(fs_info->balance_ctl); |
3861 | spin_lock(&fs_info->balance_lock); | |
3862 | fs_info->balance_ctl = bctl; | |
3863 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
3864 | } else { |
3865 | BUG_ON(ret != -EEXIST); | |
3866 | spin_lock(&fs_info->balance_lock); | |
3867 | update_balance_args(bctl); | |
3868 | spin_unlock(&fs_info->balance_lock); | |
3869 | } | |
c9e9f97b | 3870 | |
3009a62f DS |
3871 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
3872 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
c9e9f97b ID |
3873 | mutex_unlock(&fs_info->balance_mutex); |
3874 | ||
3875 | ret = __btrfs_balance(fs_info); | |
3876 | ||
3877 | mutex_lock(&fs_info->balance_mutex); | |
3009a62f | 3878 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
3879 | |
3880 | if (bargs) { | |
3881 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 3882 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
3883 | } |
3884 | ||
3a01aa7a ID |
3885 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3886 | balance_need_close(fs_info)) { | |
149196a2 | 3887 | reset_balance_state(fs_info); |
a17c95df | 3888 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
3889 | } |
3890 | ||
837d5b6e | 3891 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3892 | |
3893 | return ret; | |
3894 | out: | |
59641015 | 3895 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 3896 | reset_balance_state(fs_info); |
a17c95df | 3897 | else |
59641015 | 3898 | kfree(bctl); |
a17c95df DS |
3899 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3900 | ||
59641015 ID |
3901 | return ret; |
3902 | } | |
3903 | ||
3904 | static int balance_kthread(void *data) | |
3905 | { | |
2b6ba629 | 3906 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3907 | int ret = 0; |
59641015 | 3908 | |
59641015 | 3909 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 3910 | if (fs_info->balance_ctl) { |
6dac13f8 | 3911 | btrfs_info(fs_info, "balance: resuming"); |
6fcf6e2b | 3912 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
9555c6c1 | 3913 | } |
59641015 | 3914 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 3915 | |
59641015 ID |
3916 | return ret; |
3917 | } | |
3918 | ||
2b6ba629 ID |
3919 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3920 | { | |
3921 | struct task_struct *tsk; | |
3922 | ||
1354e1a1 | 3923 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 3924 | if (!fs_info->balance_ctl) { |
1354e1a1 | 3925 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
3926 | return 0; |
3927 | } | |
1354e1a1 | 3928 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 3929 | |
3cdde224 | 3930 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 3931 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
3932 | return 0; |
3933 | } | |
3934 | ||
02ee654d AJ |
3935 | /* |
3936 | * A ro->rw remount sequence should continue with the paused balance | |
3937 | * regardless of who pauses it, system or the user as of now, so set | |
3938 | * the resume flag. | |
3939 | */ | |
3940 | spin_lock(&fs_info->balance_lock); | |
3941 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
3942 | spin_unlock(&fs_info->balance_lock); | |
3943 | ||
2b6ba629 | 3944 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 3945 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3946 | } |
3947 | ||
68310a5e | 3948 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3949 | { |
59641015 ID |
3950 | struct btrfs_balance_control *bctl; |
3951 | struct btrfs_balance_item *item; | |
3952 | struct btrfs_disk_balance_args disk_bargs; | |
3953 | struct btrfs_path *path; | |
3954 | struct extent_buffer *leaf; | |
3955 | struct btrfs_key key; | |
3956 | int ret; | |
3957 | ||
3958 | path = btrfs_alloc_path(); | |
3959 | if (!path) | |
3960 | return -ENOMEM; | |
3961 | ||
59641015 | 3962 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3963 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
3964 | key.offset = 0; |
3965 | ||
68310a5e | 3966 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3967 | if (ret < 0) |
68310a5e | 3968 | goto out; |
59641015 ID |
3969 | if (ret > 0) { /* ret = -ENOENT; */ |
3970 | ret = 0; | |
68310a5e ID |
3971 | goto out; |
3972 | } | |
3973 | ||
3974 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3975 | if (!bctl) { | |
3976 | ret = -ENOMEM; | |
3977 | goto out; | |
59641015 ID |
3978 | } |
3979 | ||
3980 | leaf = path->nodes[0]; | |
3981 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3982 | ||
68310a5e ID |
3983 | bctl->flags = btrfs_balance_flags(leaf, item); |
3984 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3985 | |
3986 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3987 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3988 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3989 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3990 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3991 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3992 | ||
eee95e3f DS |
3993 | /* |
3994 | * This should never happen, as the paused balance state is recovered | |
3995 | * during mount without any chance of other exclusive ops to collide. | |
3996 | * | |
3997 | * This gives the exclusive op status to balance and keeps in paused | |
3998 | * state until user intervention (cancel or umount). If the ownership | |
3999 | * cannot be assigned, show a message but do not fail. The balance | |
4000 | * is in a paused state and must have fs_info::balance_ctl properly | |
4001 | * set up. | |
4002 | */ | |
4003 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4004 | btrfs_warn(fs_info, | |
6dac13f8 | 4005 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4006 | |
68310a5e | 4007 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4008 | BUG_ON(fs_info->balance_ctl); |
4009 | spin_lock(&fs_info->balance_lock); | |
4010 | fs_info->balance_ctl = bctl; | |
4011 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4012 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4013 | out: |
4014 | btrfs_free_path(path); | |
ec44a35c CM |
4015 | return ret; |
4016 | } | |
4017 | ||
837d5b6e ID |
4018 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4019 | { | |
4020 | int ret = 0; | |
4021 | ||
4022 | mutex_lock(&fs_info->balance_mutex); | |
4023 | if (!fs_info->balance_ctl) { | |
4024 | mutex_unlock(&fs_info->balance_mutex); | |
4025 | return -ENOTCONN; | |
4026 | } | |
4027 | ||
3009a62f | 4028 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4029 | atomic_inc(&fs_info->balance_pause_req); |
4030 | mutex_unlock(&fs_info->balance_mutex); | |
4031 | ||
4032 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4033 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4034 | |
4035 | mutex_lock(&fs_info->balance_mutex); | |
4036 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4037 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4038 | atomic_dec(&fs_info->balance_pause_req); |
4039 | } else { | |
4040 | ret = -ENOTCONN; | |
4041 | } | |
4042 | ||
4043 | mutex_unlock(&fs_info->balance_mutex); | |
4044 | return ret; | |
4045 | } | |
4046 | ||
a7e99c69 ID |
4047 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4048 | { | |
4049 | mutex_lock(&fs_info->balance_mutex); | |
4050 | if (!fs_info->balance_ctl) { | |
4051 | mutex_unlock(&fs_info->balance_mutex); | |
4052 | return -ENOTCONN; | |
4053 | } | |
4054 | ||
cf7d20f4 DS |
4055 | /* |
4056 | * A paused balance with the item stored on disk can be resumed at | |
4057 | * mount time if the mount is read-write. Otherwise it's still paused | |
4058 | * and we must not allow cancelling as it deletes the item. | |
4059 | */ | |
4060 | if (sb_rdonly(fs_info->sb)) { | |
4061 | mutex_unlock(&fs_info->balance_mutex); | |
4062 | return -EROFS; | |
4063 | } | |
4064 | ||
a7e99c69 ID |
4065 | atomic_inc(&fs_info->balance_cancel_req); |
4066 | /* | |
4067 | * if we are running just wait and return, balance item is | |
4068 | * deleted in btrfs_balance in this case | |
4069 | */ | |
3009a62f | 4070 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4071 | mutex_unlock(&fs_info->balance_mutex); |
4072 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4073 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4074 | mutex_lock(&fs_info->balance_mutex); |
4075 | } else { | |
a7e99c69 | 4076 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4077 | /* |
4078 | * Lock released to allow other waiters to continue, we'll | |
4079 | * reexamine the status again. | |
4080 | */ | |
a7e99c69 ID |
4081 | mutex_lock(&fs_info->balance_mutex); |
4082 | ||
a17c95df | 4083 | if (fs_info->balance_ctl) { |
149196a2 | 4084 | reset_balance_state(fs_info); |
a17c95df | 4085 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4086 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4087 | } |
a7e99c69 ID |
4088 | } |
4089 | ||
3009a62f DS |
4090 | BUG_ON(fs_info->balance_ctl || |
4091 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4092 | atomic_dec(&fs_info->balance_cancel_req); |
4093 | mutex_unlock(&fs_info->balance_mutex); | |
4094 | return 0; | |
4095 | } | |
4096 | ||
803b2f54 SB |
4097 | static int btrfs_uuid_scan_kthread(void *data) |
4098 | { | |
4099 | struct btrfs_fs_info *fs_info = data; | |
4100 | struct btrfs_root *root = fs_info->tree_root; | |
4101 | struct btrfs_key key; | |
803b2f54 SB |
4102 | struct btrfs_path *path = NULL; |
4103 | int ret = 0; | |
4104 | struct extent_buffer *eb; | |
4105 | int slot; | |
4106 | struct btrfs_root_item root_item; | |
4107 | u32 item_size; | |
f45388f3 | 4108 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4109 | |
4110 | path = btrfs_alloc_path(); | |
4111 | if (!path) { | |
4112 | ret = -ENOMEM; | |
4113 | goto out; | |
4114 | } | |
4115 | ||
4116 | key.objectid = 0; | |
4117 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4118 | key.offset = 0; | |
4119 | ||
803b2f54 | 4120 | while (1) { |
7c829b72 AJ |
4121 | ret = btrfs_search_forward(root, &key, path, |
4122 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4123 | if (ret) { |
4124 | if (ret > 0) | |
4125 | ret = 0; | |
4126 | break; | |
4127 | } | |
4128 | ||
4129 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4130 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4131 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4132 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4133 | goto skip; | |
4134 | ||
4135 | eb = path->nodes[0]; | |
4136 | slot = path->slots[0]; | |
4137 | item_size = btrfs_item_size_nr(eb, slot); | |
4138 | if (item_size < sizeof(root_item)) | |
4139 | goto skip; | |
4140 | ||
803b2f54 SB |
4141 | read_extent_buffer(eb, &root_item, |
4142 | btrfs_item_ptr_offset(eb, slot), | |
4143 | (int)sizeof(root_item)); | |
4144 | if (btrfs_root_refs(&root_item) == 0) | |
4145 | goto skip; | |
f45388f3 FDBM |
4146 | |
4147 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4148 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4149 | if (trans) | |
4150 | goto update_tree; | |
4151 | ||
4152 | btrfs_release_path(path); | |
803b2f54 SB |
4153 | /* |
4154 | * 1 - subvol uuid item | |
4155 | * 1 - received_subvol uuid item | |
4156 | */ | |
4157 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4158 | if (IS_ERR(trans)) { | |
4159 | ret = PTR_ERR(trans); | |
4160 | break; | |
4161 | } | |
f45388f3 FDBM |
4162 | continue; |
4163 | } else { | |
4164 | goto skip; | |
4165 | } | |
4166 | update_tree: | |
4167 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4168 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4169 | BTRFS_UUID_KEY_SUBVOL, |
4170 | key.objectid); | |
4171 | if (ret < 0) { | |
efe120a0 | 4172 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4173 | ret); |
803b2f54 SB |
4174 | break; |
4175 | } | |
4176 | } | |
4177 | ||
4178 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4179 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4180 | root_item.received_uuid, |
4181 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4182 | key.objectid); | |
4183 | if (ret < 0) { | |
efe120a0 | 4184 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4185 | ret); |
803b2f54 SB |
4186 | break; |
4187 | } | |
4188 | } | |
4189 | ||
f45388f3 | 4190 | skip: |
803b2f54 | 4191 | if (trans) { |
3a45bb20 | 4192 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4193 | trans = NULL; |
803b2f54 SB |
4194 | if (ret) |
4195 | break; | |
4196 | } | |
4197 | ||
803b2f54 SB |
4198 | btrfs_release_path(path); |
4199 | if (key.offset < (u64)-1) { | |
4200 | key.offset++; | |
4201 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4202 | key.offset = 0; | |
4203 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4204 | } else if (key.objectid < (u64)-1) { | |
4205 | key.offset = 0; | |
4206 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4207 | key.objectid++; | |
4208 | } else { | |
4209 | break; | |
4210 | } | |
4211 | cond_resched(); | |
4212 | } | |
4213 | ||
4214 | out: | |
4215 | btrfs_free_path(path); | |
f45388f3 | 4216 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4217 | btrfs_end_transaction(trans); |
803b2f54 | 4218 | if (ret) |
efe120a0 | 4219 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4220 | else |
afcdd129 | 4221 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4222 | up(&fs_info->uuid_tree_rescan_sem); |
4223 | return 0; | |
4224 | } | |
4225 | ||
70f80175 SB |
4226 | /* |
4227 | * Callback for btrfs_uuid_tree_iterate(). | |
4228 | * returns: | |
4229 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4230 | * < 0 if an error occurred. |
70f80175 SB |
4231 | * > 0 if the check failed, which means the caller shall remove the entry. |
4232 | */ | |
4233 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4234 | u8 *uuid, u8 type, u64 subid) | |
4235 | { | |
4236 | struct btrfs_key key; | |
4237 | int ret = 0; | |
4238 | struct btrfs_root *subvol_root; | |
4239 | ||
4240 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4241 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4242 | goto out; | |
4243 | ||
4244 | key.objectid = subid; | |
4245 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4246 | key.offset = (u64)-1; | |
4247 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4248 | if (IS_ERR(subvol_root)) { | |
4249 | ret = PTR_ERR(subvol_root); | |
4250 | if (ret == -ENOENT) | |
4251 | ret = 1; | |
4252 | goto out; | |
4253 | } | |
4254 | ||
4255 | switch (type) { | |
4256 | case BTRFS_UUID_KEY_SUBVOL: | |
4257 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4258 | ret = 1; | |
4259 | break; | |
4260 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4261 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4262 | BTRFS_UUID_SIZE)) | |
4263 | ret = 1; | |
4264 | break; | |
4265 | } | |
4266 | ||
4267 | out: | |
4268 | return ret; | |
4269 | } | |
4270 | ||
4271 | static int btrfs_uuid_rescan_kthread(void *data) | |
4272 | { | |
4273 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4274 | int ret; | |
4275 | ||
4276 | /* | |
4277 | * 1st step is to iterate through the existing UUID tree and | |
4278 | * to delete all entries that contain outdated data. | |
4279 | * 2nd step is to add all missing entries to the UUID tree. | |
4280 | */ | |
4281 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4282 | if (ret < 0) { | |
efe120a0 | 4283 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4284 | up(&fs_info->uuid_tree_rescan_sem); |
4285 | return ret; | |
4286 | } | |
4287 | return btrfs_uuid_scan_kthread(data); | |
4288 | } | |
4289 | ||
f7a81ea4 SB |
4290 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4291 | { | |
4292 | struct btrfs_trans_handle *trans; | |
4293 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4294 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4295 | struct task_struct *task; |
4296 | int ret; | |
f7a81ea4 SB |
4297 | |
4298 | /* | |
4299 | * 1 - root node | |
4300 | * 1 - root item | |
4301 | */ | |
4302 | trans = btrfs_start_transaction(tree_root, 2); | |
4303 | if (IS_ERR(trans)) | |
4304 | return PTR_ERR(trans); | |
4305 | ||
4306 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4307 | BTRFS_UUID_TREE_OBJECTID); | |
4308 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4309 | ret = PTR_ERR(uuid_root); |
66642832 | 4310 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4311 | btrfs_end_transaction(trans); |
6d13f549 | 4312 | return ret; |
f7a81ea4 SB |
4313 | } |
4314 | ||
4315 | fs_info->uuid_root = uuid_root; | |
4316 | ||
3a45bb20 | 4317 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4318 | if (ret) |
4319 | return ret; | |
4320 | ||
4321 | down(&fs_info->uuid_tree_rescan_sem); | |
4322 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4323 | if (IS_ERR(task)) { | |
70f80175 | 4324 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4325 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4326 | up(&fs_info->uuid_tree_rescan_sem); |
4327 | return PTR_ERR(task); | |
4328 | } | |
4329 | ||
4330 | return 0; | |
f7a81ea4 | 4331 | } |
803b2f54 | 4332 | |
70f80175 SB |
4333 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4334 | { | |
4335 | struct task_struct *task; | |
4336 | ||
4337 | down(&fs_info->uuid_tree_rescan_sem); | |
4338 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4339 | if (IS_ERR(task)) { | |
4340 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4341 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4342 | up(&fs_info->uuid_tree_rescan_sem); |
4343 | return PTR_ERR(task); | |
4344 | } | |
4345 | ||
4346 | return 0; | |
4347 | } | |
4348 | ||
8f18cf13 CM |
4349 | /* |
4350 | * shrinking a device means finding all of the device extents past | |
4351 | * the new size, and then following the back refs to the chunks. | |
4352 | * The chunk relocation code actually frees the device extent | |
4353 | */ | |
4354 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4355 | { | |
0b246afa JM |
4356 | struct btrfs_fs_info *fs_info = device->fs_info; |
4357 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4358 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4359 | struct btrfs_dev_extent *dev_extent = NULL; |
4360 | struct btrfs_path *path; | |
4361 | u64 length; | |
8f18cf13 CM |
4362 | u64 chunk_offset; |
4363 | int ret; | |
4364 | int slot; | |
ba1bf481 JB |
4365 | int failed = 0; |
4366 | bool retried = false; | |
53e489bc | 4367 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4368 | struct extent_buffer *l; |
4369 | struct btrfs_key key; | |
0b246afa | 4370 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4371 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4372 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4373 | u64 diff; |
4374 | ||
4375 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4376 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4377 | |
401e29c1 | 4378 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4379 | return -EINVAL; |
4380 | ||
8f18cf13 CM |
4381 | path = btrfs_alloc_path(); |
4382 | if (!path) | |
4383 | return -ENOMEM; | |
4384 | ||
0338dff6 | 4385 | path->reada = READA_BACK; |
8f18cf13 | 4386 | |
34441361 | 4387 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4388 | |
7cc8e58d | 4389 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4390 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4391 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4392 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4393 | } |
34441361 | 4394 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4395 | |
ba1bf481 | 4396 | again: |
8f18cf13 CM |
4397 | key.objectid = device->devid; |
4398 | key.offset = (u64)-1; | |
4399 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4400 | ||
213e64da | 4401 | do { |
0b246afa | 4402 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4403 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4404 | if (ret < 0) { |
0b246afa | 4405 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4406 | goto done; |
67c5e7d4 | 4407 | } |
8f18cf13 CM |
4408 | |
4409 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4410 | if (ret) |
0b246afa | 4411 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4412 | if (ret < 0) |
4413 | goto done; | |
4414 | if (ret) { | |
4415 | ret = 0; | |
b3b4aa74 | 4416 | btrfs_release_path(path); |
bf1fb512 | 4417 | break; |
8f18cf13 CM |
4418 | } |
4419 | ||
4420 | l = path->nodes[0]; | |
4421 | slot = path->slots[0]; | |
4422 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4423 | ||
ba1bf481 | 4424 | if (key.objectid != device->devid) { |
0b246afa | 4425 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4426 | btrfs_release_path(path); |
bf1fb512 | 4427 | break; |
ba1bf481 | 4428 | } |
8f18cf13 CM |
4429 | |
4430 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4431 | length = btrfs_dev_extent_length(l, dev_extent); | |
4432 | ||
ba1bf481 | 4433 | if (key.offset + length <= new_size) { |
0b246afa | 4434 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4435 | btrfs_release_path(path); |
d6397bae | 4436 | break; |
ba1bf481 | 4437 | } |
8f18cf13 | 4438 | |
8f18cf13 | 4439 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4440 | btrfs_release_path(path); |
8f18cf13 | 4441 | |
a6f93c71 LB |
4442 | /* |
4443 | * We may be relocating the only data chunk we have, | |
4444 | * which could potentially end up with losing data's | |
4445 | * raid profile, so lets allocate an empty one in | |
4446 | * advance. | |
4447 | */ | |
4448 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4449 | if (ret < 0) { | |
4450 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4451 | goto done; | |
4452 | } | |
4453 | ||
0b246afa JM |
4454 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4455 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4456 | if (ret == -ENOSPC) { |
ba1bf481 | 4457 | failed++; |
eede2bf3 OS |
4458 | } else if (ret) { |
4459 | if (ret == -ETXTBSY) { | |
4460 | btrfs_warn(fs_info, | |
4461 | "could not shrink block group %llu due to active swapfile", | |
4462 | chunk_offset); | |
4463 | } | |
4464 | goto done; | |
4465 | } | |
213e64da | 4466 | } while (key.offset-- > 0); |
ba1bf481 JB |
4467 | |
4468 | if (failed && !retried) { | |
4469 | failed = 0; | |
4470 | retried = true; | |
4471 | goto again; | |
4472 | } else if (failed && retried) { | |
4473 | ret = -ENOSPC; | |
ba1bf481 | 4474 | goto done; |
8f18cf13 CM |
4475 | } |
4476 | ||
d6397bae | 4477 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4478 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4479 | if (IS_ERR(trans)) { |
4480 | ret = PTR_ERR(trans); | |
4481 | goto done; | |
4482 | } | |
4483 | ||
34441361 | 4484 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4485 | |
4486 | /* | |
4487 | * We checked in the above loop all device extents that were already in | |
4488 | * the device tree. However before we have updated the device's | |
4489 | * total_bytes to the new size, we might have had chunk allocations that | |
4490 | * have not complete yet (new block groups attached to transaction | |
4491 | * handles), and therefore their device extents were not yet in the | |
4492 | * device tree and we missed them in the loop above. So if we have any | |
4493 | * pending chunk using a device extent that overlaps the device range | |
4494 | * that we can not use anymore, commit the current transaction and | |
4495 | * repeat the search on the device tree - this way we guarantee we will | |
4496 | * not have chunks using device extents that end beyond 'new_size'. | |
4497 | */ | |
4498 | if (!checked_pending_chunks) { | |
4499 | u64 start = new_size; | |
4500 | u64 len = old_size - new_size; | |
4501 | ||
499f377f JM |
4502 | if (contains_pending_extent(trans->transaction, device, |
4503 | &start, len)) { | |
34441361 | 4504 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4505 | checked_pending_chunks = true; |
4506 | failed = 0; | |
4507 | retried = false; | |
3a45bb20 | 4508 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4509 | if (ret) |
4510 | goto done; | |
4511 | goto again; | |
4512 | } | |
4513 | } | |
4514 | ||
7cc8e58d | 4515 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4516 | if (list_empty(&device->resized_list)) |
4517 | list_add_tail(&device->resized_list, | |
0b246afa | 4518 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4519 | |
d6397bae | 4520 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4521 | btrfs_set_super_total_bytes(super_copy, |
4522 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4523 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4524 | |
4525 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4526 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4527 | if (ret < 0) { |
4528 | btrfs_abort_transaction(trans, ret); | |
4529 | btrfs_end_transaction(trans); | |
4530 | } else { | |
4531 | ret = btrfs_commit_transaction(trans); | |
4532 | } | |
8f18cf13 CM |
4533 | done: |
4534 | btrfs_free_path(path); | |
53e489bc | 4535 | if (ret) { |
34441361 | 4536 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4537 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4538 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4539 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4540 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4541 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4542 | } |
8f18cf13 CM |
4543 | return ret; |
4544 | } | |
4545 | ||
2ff7e61e | 4546 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4547 | struct btrfs_key *key, |
4548 | struct btrfs_chunk *chunk, int item_size) | |
4549 | { | |
0b246afa | 4550 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4551 | struct btrfs_disk_key disk_key; |
4552 | u32 array_size; | |
4553 | u8 *ptr; | |
4554 | ||
34441361 | 4555 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4556 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4557 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4558 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4559 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4560 | return -EFBIG; |
fe48a5c0 | 4561 | } |
0b86a832 CM |
4562 | |
4563 | ptr = super_copy->sys_chunk_array + array_size; | |
4564 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4565 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4566 | ptr += sizeof(disk_key); | |
4567 | memcpy(ptr, chunk, item_size); | |
4568 | item_size += sizeof(disk_key); | |
4569 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4570 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4571 | |
0b86a832 CM |
4572 | return 0; |
4573 | } | |
4574 | ||
73c5de00 AJ |
4575 | /* |
4576 | * sort the devices in descending order by max_avail, total_avail | |
4577 | */ | |
4578 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4579 | { |
73c5de00 AJ |
4580 | const struct btrfs_device_info *di_a = a; |
4581 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4582 | |
73c5de00 | 4583 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4584 | return -1; |
73c5de00 | 4585 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4586 | return 1; |
73c5de00 AJ |
4587 | if (di_a->total_avail > di_b->total_avail) |
4588 | return -1; | |
4589 | if (di_a->total_avail < di_b->total_avail) | |
4590 | return 1; | |
4591 | return 0; | |
b2117a39 | 4592 | } |
0b86a832 | 4593 | |
53b381b3 DW |
4594 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4595 | { | |
ffe2d203 | 4596 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4597 | return; |
4598 | ||
ceda0864 | 4599 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4600 | } |
4601 | ||
062d4d1f | 4602 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
4603 | - sizeof(struct btrfs_chunk)) \ |
4604 | / sizeof(struct btrfs_stripe) + 1) | |
4605 | ||
4606 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4607 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4608 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4609 | / sizeof(struct btrfs_stripe) + 1) | |
4610 | ||
73c5de00 | 4611 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4612 | u64 start, u64 type) |
b2117a39 | 4613 | { |
2ff7e61e | 4614 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4615 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4616 | struct btrfs_device *device; |
73c5de00 AJ |
4617 | struct map_lookup *map = NULL; |
4618 | struct extent_map_tree *em_tree; | |
4619 | struct extent_map *em; | |
4620 | struct btrfs_device_info *devices_info = NULL; | |
4621 | u64 total_avail; | |
4622 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4623 | int data_stripes; /* number of stripes that count for |
4624 | block group size */ | |
73c5de00 AJ |
4625 | int sub_stripes; /* sub_stripes info for map */ |
4626 | int dev_stripes; /* stripes per dev */ | |
4627 | int devs_max; /* max devs to use */ | |
4628 | int devs_min; /* min devs needed */ | |
4629 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4630 | int ncopies; /* how many copies to data has */ | |
4631 | int ret; | |
4632 | u64 max_stripe_size; | |
4633 | u64 max_chunk_size; | |
4634 | u64 stripe_size; | |
4635 | u64 num_bytes; | |
4636 | int ndevs; | |
4637 | int i; | |
4638 | int j; | |
31e50229 | 4639 | int index; |
593060d7 | 4640 | |
0c460c0d | 4641 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4642 | |
4117f207 QW |
4643 | if (list_empty(&fs_devices->alloc_list)) { |
4644 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4645 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4646 | return -ENOSPC; |
4117f207 | 4647 | } |
b2117a39 | 4648 | |
3e72ee88 | 4649 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4650 | |
31e50229 LB |
4651 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4652 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4653 | devs_max = btrfs_raid_array[index].devs_max; | |
4654 | devs_min = btrfs_raid_array[index].devs_min; | |
4655 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4656 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4657 | |
9b3f68b9 | 4658 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4659 | max_stripe_size = SZ_1G; |
fce466ea | 4660 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 4661 | if (!devs_max) |
062d4d1f | 4662 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 4663 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4664 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4665 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4666 | max_stripe_size = SZ_1G; | |
1100373f | 4667 | else |
ee22184b | 4668 | max_stripe_size = SZ_256M; |
73c5de00 | 4669 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 4670 | if (!devs_max) |
062d4d1f | 4671 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 4672 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4673 | max_stripe_size = SZ_32M; |
73c5de00 | 4674 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4675 | if (!devs_max) |
4676 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4677 | } else { |
351fd353 | 4678 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4679 | type); |
4680 | BUG_ON(1); | |
9b3f68b9 CM |
4681 | } |
4682 | ||
2b82032c YZ |
4683 | /* we don't want a chunk larger than 10% of writeable space */ |
4684 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4685 | max_chunk_size); | |
9b3f68b9 | 4686 | |
31e818fe | 4687 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4688 | GFP_NOFS); |
4689 | if (!devices_info) | |
4690 | return -ENOMEM; | |
0cad8a11 | 4691 | |
9f680ce0 | 4692 | /* |
73c5de00 AJ |
4693 | * in the first pass through the devices list, we gather information |
4694 | * about the available holes on each device. | |
9f680ce0 | 4695 | */ |
73c5de00 | 4696 | ndevs = 0; |
ebcc9301 | 4697 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4698 | u64 max_avail; |
4699 | u64 dev_offset; | |
b2117a39 | 4700 | |
ebbede42 | 4701 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4702 | WARN(1, KERN_ERR |
efe120a0 | 4703 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4704 | continue; |
4705 | } | |
b2117a39 | 4706 | |
e12c9621 AJ |
4707 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4708 | &device->dev_state) || | |
401e29c1 | 4709 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4710 | continue; |
b2117a39 | 4711 | |
73c5de00 AJ |
4712 | if (device->total_bytes > device->bytes_used) |
4713 | total_avail = device->total_bytes - device->bytes_used; | |
4714 | else | |
4715 | total_avail = 0; | |
38c01b96 | 4716 | |
4717 | /* If there is no space on this device, skip it. */ | |
4718 | if (total_avail == 0) | |
4719 | continue; | |
b2117a39 | 4720 | |
6df9a95e | 4721 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4722 | max_stripe_size * dev_stripes, |
4723 | &dev_offset, &max_avail); | |
4724 | if (ret && ret != -ENOSPC) | |
4725 | goto error; | |
b2117a39 | 4726 | |
73c5de00 AJ |
4727 | if (ret == 0) |
4728 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4729 | |
4117f207 QW |
4730 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4731 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4732 | btrfs_debug(info, | |
4733 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4734 | __func__, device->devid, max_avail, | |
4735 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4736 | continue; |
4117f207 | 4737 | } |
b2117a39 | 4738 | |
063d006f ES |
4739 | if (ndevs == fs_devices->rw_devices) { |
4740 | WARN(1, "%s: found more than %llu devices\n", | |
4741 | __func__, fs_devices->rw_devices); | |
4742 | break; | |
4743 | } | |
73c5de00 AJ |
4744 | devices_info[ndevs].dev_offset = dev_offset; |
4745 | devices_info[ndevs].max_avail = max_avail; | |
4746 | devices_info[ndevs].total_avail = total_avail; | |
4747 | devices_info[ndevs].dev = device; | |
4748 | ++ndevs; | |
4749 | } | |
b2117a39 | 4750 | |
73c5de00 AJ |
4751 | /* |
4752 | * now sort the devices by hole size / available space | |
4753 | */ | |
4754 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4755 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4756 | |
73c5de00 | 4757 | /* round down to number of usable stripes */ |
e5600fd6 | 4758 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4759 | |
ba89b802 | 4760 | if (ndevs < devs_min) { |
73c5de00 | 4761 | ret = -ENOSPC; |
4117f207 QW |
4762 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4763 | btrfs_debug(info, | |
4764 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 4765 | __func__, ndevs, devs_min); |
4117f207 | 4766 | } |
73c5de00 | 4767 | goto error; |
b2117a39 | 4768 | } |
9f680ce0 | 4769 | |
f148ef4d NB |
4770 | ndevs = min(ndevs, devs_max); |
4771 | ||
73c5de00 | 4772 | /* |
92e222df HK |
4773 | * The primary goal is to maximize the number of stripes, so use as |
4774 | * many devices as possible, even if the stripes are not maximum sized. | |
4775 | * | |
4776 | * The DUP profile stores more than one stripe per device, the | |
4777 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4778 | */ |
92e222df | 4779 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4780 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4781 | |
53b381b3 DW |
4782 | /* |
4783 | * this will have to be fixed for RAID1 and RAID10 over | |
4784 | * more drives | |
4785 | */ | |
4786 | data_stripes = num_stripes / ncopies; | |
4787 | ||
500ceed8 | 4788 | if (type & BTRFS_BLOCK_GROUP_RAID5) |
53b381b3 | 4789 | data_stripes = num_stripes - 1; |
500ceed8 NB |
4790 | |
4791 | if (type & BTRFS_BLOCK_GROUP_RAID6) | |
53b381b3 | 4792 | data_stripes = num_stripes - 2; |
86db2578 CM |
4793 | |
4794 | /* | |
4795 | * Use the number of data stripes to figure out how big this chunk | |
4796 | * is really going to be in terms of logical address space, | |
4797 | * and compare that answer with the max chunk size | |
4798 | */ | |
4799 | if (stripe_size * data_stripes > max_chunk_size) { | |
b8b93add | 4800 | stripe_size = div_u64(max_chunk_size, data_stripes); |
86db2578 CM |
4801 | |
4802 | /* bump the answer up to a 16MB boundary */ | |
793ff2c8 | 4803 | stripe_size = round_up(stripe_size, SZ_16M); |
86db2578 | 4804 | |
793ff2c8 QW |
4805 | /* |
4806 | * But don't go higher than the limits we found while searching | |
4807 | * for free extents | |
86db2578 | 4808 | */ |
793ff2c8 QW |
4809 | stripe_size = min(devices_info[ndevs - 1].max_avail, |
4810 | stripe_size); | |
86db2578 CM |
4811 | } |
4812 | ||
37db63a4 | 4813 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 4814 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4815 | |
4816 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4817 | if (!map) { | |
4818 | ret = -ENOMEM; | |
4819 | goto error; | |
4820 | } | |
4821 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4822 | |
73c5de00 AJ |
4823 | for (i = 0; i < ndevs; ++i) { |
4824 | for (j = 0; j < dev_stripes; ++j) { | |
4825 | int s = i * dev_stripes + j; | |
4826 | map->stripes[s].dev = devices_info[i].dev; | |
4827 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4828 | j * stripe_size; | |
6324fbf3 | 4829 | } |
6324fbf3 | 4830 | } |
500ceed8 NB |
4831 | map->stripe_len = BTRFS_STRIPE_LEN; |
4832 | map->io_align = BTRFS_STRIPE_LEN; | |
4833 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4834 | map->type = type; |
2b82032c | 4835 | map->sub_stripes = sub_stripes; |
0b86a832 | 4836 | |
53b381b3 | 4837 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4838 | |
6bccf3ab | 4839 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4840 | |
172ddd60 | 4841 | em = alloc_extent_map(); |
2b82032c | 4842 | if (!em) { |
298a8f9c | 4843 | kfree(map); |
b2117a39 MX |
4844 | ret = -ENOMEM; |
4845 | goto error; | |
593060d7 | 4846 | } |
298a8f9c | 4847 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4848 | em->map_lookup = map; |
2b82032c | 4849 | em->start = start; |
73c5de00 | 4850 | em->len = num_bytes; |
2b82032c YZ |
4851 | em->block_start = 0; |
4852 | em->block_len = em->len; | |
6df9a95e | 4853 | em->orig_block_len = stripe_size; |
593060d7 | 4854 | |
0b246afa | 4855 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4856 | write_lock(&em_tree->lock); |
09a2a8f9 | 4857 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 4858 | if (ret) { |
1efb72a3 | 4859 | write_unlock(&em_tree->lock); |
0f5d42b2 | 4860 | free_extent_map(em); |
1dd4602f | 4861 | goto error; |
0f5d42b2 | 4862 | } |
0b86a832 | 4863 | |
1efb72a3 NB |
4864 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
4865 | refcount_inc(&em->refs); | |
4866 | write_unlock(&em_tree->lock); | |
4867 | ||
e7e02096 | 4868 | ret = btrfs_make_block_group(trans, 0, type, start, num_bytes); |
6df9a95e JB |
4869 | if (ret) |
4870 | goto error_del_extent; | |
2b82032c | 4871 | |
7cc8e58d MX |
4872 | for (i = 0; i < map->num_stripes; i++) { |
4873 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4874 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4875 | } | |
43530c46 | 4876 | |
a5ed45f8 | 4877 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4878 | |
0f5d42b2 | 4879 | free_extent_map(em); |
0b246afa | 4880 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4881 | |
b2117a39 | 4882 | kfree(devices_info); |
2b82032c | 4883 | return 0; |
b2117a39 | 4884 | |
6df9a95e | 4885 | error_del_extent: |
0f5d42b2 JB |
4886 | write_lock(&em_tree->lock); |
4887 | remove_extent_mapping(em_tree, em); | |
4888 | write_unlock(&em_tree->lock); | |
4889 | ||
4890 | /* One for our allocation */ | |
4891 | free_extent_map(em); | |
4892 | /* One for the tree reference */ | |
4893 | free_extent_map(em); | |
495e64f4 FM |
4894 | /* One for the pending_chunks list reference */ |
4895 | free_extent_map(em); | |
b2117a39 | 4896 | error: |
b2117a39 MX |
4897 | kfree(devices_info); |
4898 | return ret; | |
2b82032c YZ |
4899 | } |
4900 | ||
6df9a95e | 4901 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 4902 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4903 | { |
97aff912 | 4904 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
4905 | struct btrfs_root *extent_root = fs_info->extent_root; |
4906 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4907 | struct btrfs_key key; |
2b82032c YZ |
4908 | struct btrfs_device *device; |
4909 | struct btrfs_chunk *chunk; | |
4910 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4911 | struct extent_map *em; |
4912 | struct map_lookup *map; | |
4913 | size_t item_size; | |
4914 | u64 dev_offset; | |
4915 | u64 stripe_size; | |
4916 | int i = 0; | |
140e639f | 4917 | int ret = 0; |
2b82032c | 4918 | |
60ca842e | 4919 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
4920 | if (IS_ERR(em)) |
4921 | return PTR_ERR(em); | |
6df9a95e | 4922 | |
95617d69 | 4923 | map = em->map_lookup; |
6df9a95e JB |
4924 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4925 | stripe_size = em->orig_block_len; | |
4926 | ||
2b82032c | 4927 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4928 | if (!chunk) { |
4929 | ret = -ENOMEM; | |
4930 | goto out; | |
4931 | } | |
4932 | ||
50460e37 FM |
4933 | /* |
4934 | * Take the device list mutex to prevent races with the final phase of | |
4935 | * a device replace operation that replaces the device object associated | |
4936 | * with the map's stripes, because the device object's id can change | |
4937 | * at any time during that final phase of the device replace operation | |
4938 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4939 | */ | |
0b246afa | 4940 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
4941 | for (i = 0; i < map->num_stripes; i++) { |
4942 | device = map->stripes[i].dev; | |
4943 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4944 | |
0b86a832 | 4945 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4946 | if (ret) |
50460e37 | 4947 | break; |
b5d9071c NB |
4948 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
4949 | dev_offset, stripe_size); | |
6df9a95e | 4950 | if (ret) |
50460e37 FM |
4951 | break; |
4952 | } | |
4953 | if (ret) { | |
0b246afa | 4954 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 4955 | goto out; |
2b82032c YZ |
4956 | } |
4957 | ||
2b82032c | 4958 | stripe = &chunk->stripe; |
6df9a95e JB |
4959 | for (i = 0; i < map->num_stripes; i++) { |
4960 | device = map->stripes[i].dev; | |
4961 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4962 | |
e17cade2 CM |
4963 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4964 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4965 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4966 | stripe++; |
0b86a832 | 4967 | } |
0b246afa | 4968 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4969 | |
2b82032c | 4970 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4971 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4972 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4973 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4974 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4975 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4976 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 4977 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 4978 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4979 | |
2b82032c YZ |
4980 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4981 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4982 | key.offset = chunk_offset; | |
0b86a832 | 4983 | |
2b82032c | 4984 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4985 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4986 | /* | |
4987 | * TODO: Cleanup of inserted chunk root in case of | |
4988 | * failure. | |
4989 | */ | |
2ff7e61e | 4990 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 4991 | } |
1abe9b8a | 4992 | |
6df9a95e | 4993 | out: |
0b86a832 | 4994 | kfree(chunk); |
6df9a95e | 4995 | free_extent_map(em); |
4ed1d16e | 4996 | return ret; |
2b82032c | 4997 | } |
0b86a832 | 4998 | |
2b82032c YZ |
4999 | /* |
5000 | * Chunk allocation falls into two parts. The first part does works | |
5001 | * that make the new allocated chunk useable, but not do any operation | |
5002 | * that modifies the chunk tree. The second part does the works that | |
5003 | * require modifying the chunk tree. This division is important for the | |
5004 | * bootstrap process of adding storage to a seed btrfs. | |
5005 | */ | |
c216b203 | 5006 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5007 | { |
5008 | u64 chunk_offset; | |
2b82032c | 5009 | |
c216b203 NB |
5010 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5011 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5012 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5013 | } |
5014 | ||
d397712b | 5015 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5016 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5017 | { |
5018 | u64 chunk_offset; | |
5019 | u64 sys_chunk_offset; | |
2b82032c | 5020 | u64 alloc_profile; |
2b82032c YZ |
5021 | int ret; |
5022 | ||
6df9a95e | 5023 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5024 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5025 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5026 | if (ret) |
5027 | return ret; | |
2b82032c | 5028 | |
0b246afa | 5029 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5030 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5031 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5032 | return ret; |
2b82032c YZ |
5033 | } |
5034 | ||
d20983b4 MX |
5035 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5036 | { | |
5037 | int max_errors; | |
5038 | ||
5039 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5040 | BTRFS_BLOCK_GROUP_RAID10 | | |
5041 | BTRFS_BLOCK_GROUP_RAID5 | | |
5042 | BTRFS_BLOCK_GROUP_DUP)) { | |
5043 | max_errors = 1; | |
5044 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5045 | max_errors = 2; | |
5046 | } else { | |
5047 | max_errors = 0; | |
005d6427 | 5048 | } |
2b82032c | 5049 | |
d20983b4 | 5050 | return max_errors; |
2b82032c YZ |
5051 | } |
5052 | ||
2ff7e61e | 5053 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5054 | { |
5055 | struct extent_map *em; | |
5056 | struct map_lookup *map; | |
2b82032c | 5057 | int readonly = 0; |
d20983b4 | 5058 | int miss_ndevs = 0; |
2b82032c YZ |
5059 | int i; |
5060 | ||
60ca842e | 5061 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5062 | if (IS_ERR(em)) |
2b82032c YZ |
5063 | return 1; |
5064 | ||
95617d69 | 5065 | map = em->map_lookup; |
2b82032c | 5066 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5067 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5068 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5069 | miss_ndevs++; |
5070 | continue; | |
5071 | } | |
ebbede42 AJ |
5072 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5073 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5074 | readonly = 1; |
d20983b4 | 5075 | goto end; |
2b82032c YZ |
5076 | } |
5077 | } | |
d20983b4 MX |
5078 | |
5079 | /* | |
5080 | * If the number of missing devices is larger than max errors, | |
5081 | * we can not write the data into that chunk successfully, so | |
5082 | * set it readonly. | |
5083 | */ | |
5084 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5085 | readonly = 1; | |
5086 | end: | |
0b86a832 | 5087 | free_extent_map(em); |
2b82032c | 5088 | return readonly; |
0b86a832 CM |
5089 | } |
5090 | ||
5091 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5092 | { | |
a8067e02 | 5093 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5094 | } |
5095 | ||
5096 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5097 | { | |
5098 | struct extent_map *em; | |
5099 | ||
d397712b | 5100 | while (1) { |
890871be | 5101 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5102 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5103 | if (em) | |
5104 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5105 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5106 | if (!em) |
5107 | break; | |
0b86a832 CM |
5108 | /* once for us */ |
5109 | free_extent_map(em); | |
5110 | /* once for the tree */ | |
5111 | free_extent_map(em); | |
5112 | } | |
5113 | } | |
5114 | ||
5d964051 | 5115 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5116 | { |
5117 | struct extent_map *em; | |
5118 | struct map_lookup *map; | |
f188591e CM |
5119 | int ret; |
5120 | ||
60ca842e | 5121 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5122 | if (IS_ERR(em)) |
5123 | /* | |
5124 | * We could return errors for these cases, but that could get | |
5125 | * ugly and we'd probably do the same thing which is just not do | |
5126 | * anything else and exit, so return 1 so the callers don't try | |
5127 | * to use other copies. | |
5128 | */ | |
fb7669b5 | 5129 | return 1; |
fb7669b5 | 5130 | |
95617d69 | 5131 | map = em->map_lookup; |
f188591e CM |
5132 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5133 | ret = map->num_stripes; | |
321aecc6 CM |
5134 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5135 | ret = map->sub_stripes; | |
53b381b3 DW |
5136 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5137 | ret = 2; | |
5138 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5139 | /* |
5140 | * There could be two corrupted data stripes, we need | |
5141 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5142 | * |
8810f751 LB |
5143 | * Fail a stripe at a time on every retry except the |
5144 | * stripe under reconstruction. | |
5145 | */ | |
5146 | ret = map->num_stripes; | |
f188591e CM |
5147 | else |
5148 | ret = 1; | |
5149 | free_extent_map(em); | |
ad6d620e | 5150 | |
7e79cb86 | 5151 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
6fad823f LB |
5152 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5153 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5154 | ret++; |
7e79cb86 | 5155 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
ad6d620e | 5156 | |
f188591e CM |
5157 | return ret; |
5158 | } | |
5159 | ||
2ff7e61e | 5160 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5161 | u64 logical) |
5162 | { | |
5163 | struct extent_map *em; | |
5164 | struct map_lookup *map; | |
0b246afa | 5165 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5166 | |
60ca842e | 5167 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5168 | |
69f03f13 NB |
5169 | if (!WARN_ON(IS_ERR(em))) { |
5170 | map = em->map_lookup; | |
5171 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5172 | len = map->stripe_len * nr_data_stripes(map); | |
5173 | free_extent_map(em); | |
5174 | } | |
53b381b3 DW |
5175 | return len; |
5176 | } | |
5177 | ||
e4ff5fb5 | 5178 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5179 | { |
5180 | struct extent_map *em; | |
5181 | struct map_lookup *map; | |
53b381b3 DW |
5182 | int ret = 0; |
5183 | ||
60ca842e | 5184 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5185 | |
69f03f13 NB |
5186 | if(!WARN_ON(IS_ERR(em))) { |
5187 | map = em->map_lookup; | |
5188 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5189 | ret = 1; | |
5190 | free_extent_map(em); | |
5191 | } | |
53b381b3 DW |
5192 | return ret; |
5193 | } | |
5194 | ||
30d9861f | 5195 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5196 | struct map_lookup *map, int first, |
8ba0ae78 | 5197 | int dev_replace_is_ongoing) |
dfe25020 CM |
5198 | { |
5199 | int i; | |
99f92a7c | 5200 | int num_stripes; |
8ba0ae78 | 5201 | int preferred_mirror; |
30d9861f SB |
5202 | int tolerance; |
5203 | struct btrfs_device *srcdev; | |
5204 | ||
99f92a7c AJ |
5205 | ASSERT((map->type & |
5206 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5207 | ||
5208 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5209 | num_stripes = map->sub_stripes; | |
5210 | else | |
5211 | num_stripes = map->num_stripes; | |
5212 | ||
8ba0ae78 AJ |
5213 | preferred_mirror = first + current->pid % num_stripes; |
5214 | ||
30d9861f SB |
5215 | if (dev_replace_is_ongoing && |
5216 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5217 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5218 | srcdev = fs_info->dev_replace.srcdev; | |
5219 | else | |
5220 | srcdev = NULL; | |
5221 | ||
5222 | /* | |
5223 | * try to avoid the drive that is the source drive for a | |
5224 | * dev-replace procedure, only choose it if no other non-missing | |
5225 | * mirror is available | |
5226 | */ | |
5227 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5228 | if (map->stripes[preferred_mirror].dev->bdev && |
5229 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5230 | return preferred_mirror; | |
99f92a7c | 5231 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5232 | if (map->stripes[i].dev->bdev && |
5233 | (tolerance || map->stripes[i].dev != srcdev)) | |
5234 | return i; | |
5235 | } | |
dfe25020 | 5236 | } |
30d9861f | 5237 | |
dfe25020 CM |
5238 | /* we couldn't find one that doesn't fail. Just return something |
5239 | * and the io error handling code will clean up eventually | |
5240 | */ | |
8ba0ae78 | 5241 | return preferred_mirror; |
dfe25020 CM |
5242 | } |
5243 | ||
53b381b3 DW |
5244 | static inline int parity_smaller(u64 a, u64 b) |
5245 | { | |
5246 | return a > b; | |
5247 | } | |
5248 | ||
5249 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5250 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5251 | { |
5252 | struct btrfs_bio_stripe s; | |
5253 | int i; | |
5254 | u64 l; | |
5255 | int again = 1; | |
5256 | ||
5257 | while (again) { | |
5258 | again = 0; | |
cc7539ed | 5259 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5260 | if (parity_smaller(bbio->raid_map[i], |
5261 | bbio->raid_map[i+1])) { | |
53b381b3 | 5262 | s = bbio->stripes[i]; |
8e5cfb55 | 5263 | l = bbio->raid_map[i]; |
53b381b3 | 5264 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5265 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5266 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5267 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5268 | |
53b381b3 DW |
5269 | again = 1; |
5270 | } | |
5271 | } | |
5272 | } | |
5273 | } | |
5274 | ||
6e9606d2 ZL |
5275 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5276 | { | |
5277 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5278 | /* the size of the btrfs_bio */ |
6e9606d2 | 5279 | sizeof(struct btrfs_bio) + |
e57cf21e | 5280 | /* plus the variable array for the stripes */ |
6e9606d2 | 5281 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5282 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5283 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5284 | /* |
5285 | * plus the raid_map, which includes both the tgt dev | |
5286 | * and the stripes | |
5287 | */ | |
5288 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5289 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5290 | |
5291 | atomic_set(&bbio->error, 0); | |
140475ae | 5292 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5293 | |
5294 | return bbio; | |
5295 | } | |
5296 | ||
5297 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5298 | { | |
140475ae ER |
5299 | WARN_ON(!refcount_read(&bbio->refs)); |
5300 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5301 | } |
5302 | ||
5303 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5304 | { | |
5305 | if (!bbio) | |
5306 | return; | |
140475ae | 5307 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5308 | kfree(bbio); |
5309 | } | |
5310 | ||
0b3d4cd3 LB |
5311 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5312 | /* | |
5313 | * Please note that, discard won't be sent to target device of device | |
5314 | * replace. | |
5315 | */ | |
5316 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5317 | u64 logical, u64 length, | |
5318 | struct btrfs_bio **bbio_ret) | |
5319 | { | |
5320 | struct extent_map *em; | |
5321 | struct map_lookup *map; | |
5322 | struct btrfs_bio *bbio; | |
5323 | u64 offset; | |
5324 | u64 stripe_nr; | |
5325 | u64 stripe_nr_end; | |
5326 | u64 stripe_end_offset; | |
5327 | u64 stripe_cnt; | |
5328 | u64 stripe_len; | |
5329 | u64 stripe_offset; | |
5330 | u64 num_stripes; | |
5331 | u32 stripe_index; | |
5332 | u32 factor = 0; | |
5333 | u32 sub_stripes = 0; | |
5334 | u64 stripes_per_dev = 0; | |
5335 | u32 remaining_stripes = 0; | |
5336 | u32 last_stripe = 0; | |
5337 | int ret = 0; | |
5338 | int i; | |
5339 | ||
5340 | /* discard always return a bbio */ | |
5341 | ASSERT(bbio_ret); | |
5342 | ||
60ca842e | 5343 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5344 | if (IS_ERR(em)) |
5345 | return PTR_ERR(em); | |
5346 | ||
5347 | map = em->map_lookup; | |
5348 | /* we don't discard raid56 yet */ | |
5349 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5350 | ret = -EOPNOTSUPP; | |
5351 | goto out; | |
5352 | } | |
5353 | ||
5354 | offset = logical - em->start; | |
5355 | length = min_t(u64, em->len - offset, length); | |
5356 | ||
5357 | stripe_len = map->stripe_len; | |
5358 | /* | |
5359 | * stripe_nr counts the total number of stripes we have to stride | |
5360 | * to get to this block | |
5361 | */ | |
5362 | stripe_nr = div64_u64(offset, stripe_len); | |
5363 | ||
5364 | /* stripe_offset is the offset of this block in its stripe */ | |
5365 | stripe_offset = offset - stripe_nr * stripe_len; | |
5366 | ||
5367 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5368 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5369 | stripe_cnt = stripe_nr_end - stripe_nr; |
5370 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5371 | (offset + length); | |
5372 | /* | |
5373 | * after this, stripe_nr is the number of stripes on this | |
5374 | * device we have to walk to find the data, and stripe_index is | |
5375 | * the number of our device in the stripe array | |
5376 | */ | |
5377 | num_stripes = 1; | |
5378 | stripe_index = 0; | |
5379 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5380 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5381 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5382 | sub_stripes = 1; | |
5383 | else | |
5384 | sub_stripes = map->sub_stripes; | |
5385 | ||
5386 | factor = map->num_stripes / sub_stripes; | |
5387 | num_stripes = min_t(u64, map->num_stripes, | |
5388 | sub_stripes * stripe_cnt); | |
5389 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5390 | stripe_index *= sub_stripes; | |
5391 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5392 | &remaining_stripes); | |
5393 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5394 | last_stripe *= sub_stripes; | |
5395 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5396 | BTRFS_BLOCK_GROUP_DUP)) { | |
5397 | num_stripes = map->num_stripes; | |
5398 | } else { | |
5399 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5400 | &stripe_index); | |
5401 | } | |
5402 | ||
5403 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5404 | if (!bbio) { | |
5405 | ret = -ENOMEM; | |
5406 | goto out; | |
5407 | } | |
5408 | ||
5409 | for (i = 0; i < num_stripes; i++) { | |
5410 | bbio->stripes[i].physical = | |
5411 | map->stripes[stripe_index].physical + | |
5412 | stripe_offset + stripe_nr * map->stripe_len; | |
5413 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5414 | ||
5415 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5416 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5417 | bbio->stripes[i].length = stripes_per_dev * | |
5418 | map->stripe_len; | |
5419 | ||
5420 | if (i / sub_stripes < remaining_stripes) | |
5421 | bbio->stripes[i].length += | |
5422 | map->stripe_len; | |
5423 | ||
5424 | /* | |
5425 | * Special for the first stripe and | |
5426 | * the last stripe: | |
5427 | * | |
5428 | * |-------|...|-------| | |
5429 | * |----------| | |
5430 | * off end_off | |
5431 | */ | |
5432 | if (i < sub_stripes) | |
5433 | bbio->stripes[i].length -= | |
5434 | stripe_offset; | |
5435 | ||
5436 | if (stripe_index >= last_stripe && | |
5437 | stripe_index <= (last_stripe + | |
5438 | sub_stripes - 1)) | |
5439 | bbio->stripes[i].length -= | |
5440 | stripe_end_offset; | |
5441 | ||
5442 | if (i == sub_stripes - 1) | |
5443 | stripe_offset = 0; | |
5444 | } else { | |
5445 | bbio->stripes[i].length = length; | |
5446 | } | |
5447 | ||
5448 | stripe_index++; | |
5449 | if (stripe_index == map->num_stripes) { | |
5450 | stripe_index = 0; | |
5451 | stripe_nr++; | |
5452 | } | |
5453 | } | |
5454 | ||
5455 | *bbio_ret = bbio; | |
5456 | bbio->map_type = map->type; | |
5457 | bbio->num_stripes = num_stripes; | |
5458 | out: | |
5459 | free_extent_map(em); | |
5460 | return ret; | |
5461 | } | |
5462 | ||
5ab56090 LB |
5463 | /* |
5464 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5465 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5466 | * left cursor can also be read from the target drive. | |
5467 | * | |
5468 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5469 | * array of stripes. | |
5470 | * For READ, it also needs to be supported using the same mirror number. | |
5471 | * | |
5472 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5473 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5474 | * case. | |
5475 | */ | |
5476 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5477 | u64 logical, u64 length, | |
5478 | u64 srcdev_devid, int *mirror_num, | |
5479 | u64 *physical) | |
5480 | { | |
5481 | struct btrfs_bio *bbio = NULL; | |
5482 | int num_stripes; | |
5483 | int index_srcdev = 0; | |
5484 | int found = 0; | |
5485 | u64 physical_of_found = 0; | |
5486 | int i; | |
5487 | int ret = 0; | |
5488 | ||
5489 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5490 | logical, &length, &bbio, 0, 0); | |
5491 | if (ret) { | |
5492 | ASSERT(bbio == NULL); | |
5493 | return ret; | |
5494 | } | |
5495 | ||
5496 | num_stripes = bbio->num_stripes; | |
5497 | if (*mirror_num > num_stripes) { | |
5498 | /* | |
5499 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5500 | * that means that the requested area is not left of the left | |
5501 | * cursor | |
5502 | */ | |
5503 | btrfs_put_bbio(bbio); | |
5504 | return -EIO; | |
5505 | } | |
5506 | ||
5507 | /* | |
5508 | * process the rest of the function using the mirror_num of the source | |
5509 | * drive. Therefore look it up first. At the end, patch the device | |
5510 | * pointer to the one of the target drive. | |
5511 | */ | |
5512 | for (i = 0; i < num_stripes; i++) { | |
5513 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5514 | continue; | |
5515 | ||
5516 | /* | |
5517 | * In case of DUP, in order to keep it simple, only add the | |
5518 | * mirror with the lowest physical address | |
5519 | */ | |
5520 | if (found && | |
5521 | physical_of_found <= bbio->stripes[i].physical) | |
5522 | continue; | |
5523 | ||
5524 | index_srcdev = i; | |
5525 | found = 1; | |
5526 | physical_of_found = bbio->stripes[i].physical; | |
5527 | } | |
5528 | ||
5529 | btrfs_put_bbio(bbio); | |
5530 | ||
5531 | ASSERT(found); | |
5532 | if (!found) | |
5533 | return -EIO; | |
5534 | ||
5535 | *mirror_num = index_srcdev + 1; | |
5536 | *physical = physical_of_found; | |
5537 | return ret; | |
5538 | } | |
5539 | ||
73c0f228 LB |
5540 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5541 | struct btrfs_bio **bbio_ret, | |
5542 | struct btrfs_dev_replace *dev_replace, | |
5543 | int *num_stripes_ret, int *max_errors_ret) | |
5544 | { | |
5545 | struct btrfs_bio *bbio = *bbio_ret; | |
5546 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5547 | int tgtdev_indexes = 0; | |
5548 | int num_stripes = *num_stripes_ret; | |
5549 | int max_errors = *max_errors_ret; | |
5550 | int i; | |
5551 | ||
5552 | if (op == BTRFS_MAP_WRITE) { | |
5553 | int index_where_to_add; | |
5554 | ||
5555 | /* | |
5556 | * duplicate the write operations while the dev replace | |
5557 | * procedure is running. Since the copying of the old disk to | |
5558 | * the new disk takes place at run time while the filesystem is | |
5559 | * mounted writable, the regular write operations to the old | |
5560 | * disk have to be duplicated to go to the new disk as well. | |
5561 | * | |
5562 | * Note that device->missing is handled by the caller, and that | |
5563 | * the write to the old disk is already set up in the stripes | |
5564 | * array. | |
5565 | */ | |
5566 | index_where_to_add = num_stripes; | |
5567 | for (i = 0; i < num_stripes; i++) { | |
5568 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5569 | /* write to new disk, too */ | |
5570 | struct btrfs_bio_stripe *new = | |
5571 | bbio->stripes + index_where_to_add; | |
5572 | struct btrfs_bio_stripe *old = | |
5573 | bbio->stripes + i; | |
5574 | ||
5575 | new->physical = old->physical; | |
5576 | new->length = old->length; | |
5577 | new->dev = dev_replace->tgtdev; | |
5578 | bbio->tgtdev_map[i] = index_where_to_add; | |
5579 | index_where_to_add++; | |
5580 | max_errors++; | |
5581 | tgtdev_indexes++; | |
5582 | } | |
5583 | } | |
5584 | num_stripes = index_where_to_add; | |
5585 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5586 | int index_srcdev = 0; | |
5587 | int found = 0; | |
5588 | u64 physical_of_found = 0; | |
5589 | ||
5590 | /* | |
5591 | * During the dev-replace procedure, the target drive can also | |
5592 | * be used to read data in case it is needed to repair a corrupt | |
5593 | * block elsewhere. This is possible if the requested area is | |
5594 | * left of the left cursor. In this area, the target drive is a | |
5595 | * full copy of the source drive. | |
5596 | */ | |
5597 | for (i = 0; i < num_stripes; i++) { | |
5598 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5599 | /* | |
5600 | * In case of DUP, in order to keep it simple, | |
5601 | * only add the mirror with the lowest physical | |
5602 | * address | |
5603 | */ | |
5604 | if (found && | |
5605 | physical_of_found <= | |
5606 | bbio->stripes[i].physical) | |
5607 | continue; | |
5608 | index_srcdev = i; | |
5609 | found = 1; | |
5610 | physical_of_found = bbio->stripes[i].physical; | |
5611 | } | |
5612 | } | |
5613 | if (found) { | |
5614 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5615 | bbio->stripes + num_stripes; | |
5616 | ||
5617 | tgtdev_stripe->physical = physical_of_found; | |
5618 | tgtdev_stripe->length = | |
5619 | bbio->stripes[index_srcdev].length; | |
5620 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5621 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5622 | ||
5623 | tgtdev_indexes++; | |
5624 | num_stripes++; | |
5625 | } | |
5626 | } | |
5627 | ||
5628 | *num_stripes_ret = num_stripes; | |
5629 | *max_errors_ret = max_errors; | |
5630 | bbio->num_tgtdevs = tgtdev_indexes; | |
5631 | *bbio_ret = bbio; | |
5632 | } | |
5633 | ||
2b19a1fe LB |
5634 | static bool need_full_stripe(enum btrfs_map_op op) |
5635 | { | |
5636 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5637 | } | |
5638 | ||
cf8cddd3 CH |
5639 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5640 | enum btrfs_map_op op, | |
f2d8d74d | 5641 | u64 logical, u64 *length, |
a1d3c478 | 5642 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5643 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5644 | { |
5645 | struct extent_map *em; | |
5646 | struct map_lookup *map; | |
0b86a832 | 5647 | u64 offset; |
593060d7 CM |
5648 | u64 stripe_offset; |
5649 | u64 stripe_nr; | |
53b381b3 | 5650 | u64 stripe_len; |
9d644a62 | 5651 | u32 stripe_index; |
cea9e445 | 5652 | int i; |
de11cc12 | 5653 | int ret = 0; |
f2d8d74d | 5654 | int num_stripes; |
a236aed1 | 5655 | int max_errors = 0; |
2c8cdd6e | 5656 | int tgtdev_indexes = 0; |
a1d3c478 | 5657 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5658 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5659 | int dev_replace_is_ongoing = 0; | |
5660 | int num_alloc_stripes; | |
ad6d620e SB |
5661 | int patch_the_first_stripe_for_dev_replace = 0; |
5662 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5663 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5664 | |
0b3d4cd3 LB |
5665 | if (op == BTRFS_MAP_DISCARD) |
5666 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5667 | *length, bbio_ret); | |
5668 | ||
60ca842e | 5669 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
592d92ee LB |
5670 | if (IS_ERR(em)) |
5671 | return PTR_ERR(em); | |
0b86a832 | 5672 | |
95617d69 | 5673 | map = em->map_lookup; |
0b86a832 | 5674 | offset = logical - em->start; |
593060d7 | 5675 | |
53b381b3 | 5676 | stripe_len = map->stripe_len; |
593060d7 CM |
5677 | stripe_nr = offset; |
5678 | /* | |
5679 | * stripe_nr counts the total number of stripes we have to stride | |
5680 | * to get to this block | |
5681 | */ | |
47c5713f | 5682 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5683 | |
53b381b3 | 5684 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5685 | if (offset < stripe_offset) { |
5d163e0e JM |
5686 | btrfs_crit(fs_info, |
5687 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5688 | stripe_offset, offset, em->start, logical, |
5689 | stripe_len); | |
5690 | free_extent_map(em); | |
5691 | return -EINVAL; | |
5692 | } | |
593060d7 CM |
5693 | |
5694 | /* stripe_offset is the offset of this block in its stripe*/ | |
5695 | stripe_offset = offset - stripe_offset; | |
5696 | ||
53b381b3 | 5697 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5698 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5699 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5700 | raid56_full_stripe_start = offset; | |
5701 | ||
5702 | /* allow a write of a full stripe, but make sure we don't | |
5703 | * allow straddling of stripes | |
5704 | */ | |
47c5713f DS |
5705 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5706 | full_stripe_len); | |
53b381b3 DW |
5707 | raid56_full_stripe_start *= full_stripe_len; |
5708 | } | |
5709 | ||
0b3d4cd3 | 5710 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5711 | u64 max_len; |
5712 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5713 | For other RAID types and for RAID[56] reads, just allow a single | |
5714 | stripe (on a single disk). */ | |
ffe2d203 | 5715 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5716 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5717 | max_len = stripe_len * nr_data_stripes(map) - |
5718 | (offset - raid56_full_stripe_start); | |
5719 | } else { | |
5720 | /* we limit the length of each bio to what fits in a stripe */ | |
5721 | max_len = stripe_len - stripe_offset; | |
5722 | } | |
5723 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5724 | } else { |
5725 | *length = em->len - offset; | |
5726 | } | |
f2d8d74d | 5727 | |
53b381b3 DW |
5728 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5729 | it cares about is the length */ | |
a1d3c478 | 5730 | if (!bbio_ret) |
cea9e445 CM |
5731 | goto out; |
5732 | ||
7e79cb86 | 5733 | btrfs_dev_replace_read_lock(dev_replace); |
472262f3 SB |
5734 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5735 | if (!dev_replace_is_ongoing) | |
7e79cb86 | 5736 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 LB |
5737 | else |
5738 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5739 | |
ad6d620e | 5740 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5741 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5742 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5743 | dev_replace->srcdev->devid, | |
5744 | &mirror_num, | |
5745 | &physical_to_patch_in_first_stripe); | |
5746 | if (ret) | |
ad6d620e | 5747 | goto out; |
5ab56090 LB |
5748 | else |
5749 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5750 | } else if (mirror_num > map->num_stripes) { |
5751 | mirror_num = 0; | |
5752 | } | |
5753 | ||
f2d8d74d | 5754 | num_stripes = 1; |
cea9e445 | 5755 | stripe_index = 0; |
fce3bb9a | 5756 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5757 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5758 | &stripe_index); | |
de483734 | 5759 | if (!need_full_stripe(op)) |
28e1cc7d | 5760 | mirror_num = 1; |
fce3bb9a | 5761 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 5762 | if (need_full_stripe(op)) |
f2d8d74d | 5763 | num_stripes = map->num_stripes; |
2fff734f | 5764 | else if (mirror_num) |
f188591e | 5765 | stripe_index = mirror_num - 1; |
dfe25020 | 5766 | else { |
30d9861f | 5767 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 5768 | dev_replace_is_ongoing); |
a1d3c478 | 5769 | mirror_num = stripe_index + 1; |
dfe25020 | 5770 | } |
2fff734f | 5771 | |
611f0e00 | 5772 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5773 | if (need_full_stripe(op)) { |
f2d8d74d | 5774 | num_stripes = map->num_stripes; |
a1d3c478 | 5775 | } else if (mirror_num) { |
f188591e | 5776 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5777 | } else { |
5778 | mirror_num = 1; | |
5779 | } | |
2fff734f | 5780 | |
321aecc6 | 5781 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5782 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5783 | |
47c5713f | 5784 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5785 | stripe_index *= map->sub_stripes; |
5786 | ||
de483734 | 5787 | if (need_full_stripe(op)) |
f2d8d74d | 5788 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5789 | else if (mirror_num) |
5790 | stripe_index += mirror_num - 1; | |
dfe25020 | 5791 | else { |
3e74317a | 5792 | int old_stripe_index = stripe_index; |
30d9861f SB |
5793 | stripe_index = find_live_mirror(fs_info, map, |
5794 | stripe_index, | |
30d9861f | 5795 | dev_replace_is_ongoing); |
3e74317a | 5796 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5797 | } |
53b381b3 | 5798 | |
ffe2d203 | 5799 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5800 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5801 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5802 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5803 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5804 | |
5805 | /* RAID[56] write or recovery. Return all stripes */ | |
5806 | num_stripes = map->num_stripes; | |
5807 | max_errors = nr_parity_stripes(map); | |
5808 | ||
53b381b3 DW |
5809 | *length = map->stripe_len; |
5810 | stripe_index = 0; | |
5811 | stripe_offset = 0; | |
5812 | } else { | |
5813 | /* | |
5814 | * Mirror #0 or #1 means the original data block. | |
5815 | * Mirror #2 is RAID5 parity block. | |
5816 | * Mirror #3 is RAID6 Q block. | |
5817 | */ | |
47c5713f DS |
5818 | stripe_nr = div_u64_rem(stripe_nr, |
5819 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5820 | if (mirror_num > 1) |
5821 | stripe_index = nr_data_stripes(map) + | |
5822 | mirror_num - 2; | |
5823 | ||
5824 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5825 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5826 | &stripe_index); | |
de483734 | 5827 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 5828 | mirror_num = 1; |
53b381b3 | 5829 | } |
8790d502 CM |
5830 | } else { |
5831 | /* | |
47c5713f DS |
5832 | * after this, stripe_nr is the number of stripes on this |
5833 | * device we have to walk to find the data, and stripe_index is | |
5834 | * the number of our device in the stripe array | |
8790d502 | 5835 | */ |
47c5713f DS |
5836 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5837 | &stripe_index); | |
a1d3c478 | 5838 | mirror_num = stripe_index + 1; |
8790d502 | 5839 | } |
e042d1ec | 5840 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5841 | btrfs_crit(fs_info, |
5842 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5843 | stripe_index, map->num_stripes); |
5844 | ret = -EINVAL; | |
5845 | goto out; | |
5846 | } | |
cea9e445 | 5847 | |
472262f3 | 5848 | num_alloc_stripes = num_stripes; |
6fad823f | 5849 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5850 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5851 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5852 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5853 | num_alloc_stripes++; |
2c8cdd6e | 5854 | tgtdev_indexes = num_stripes; |
ad6d620e | 5855 | } |
2c8cdd6e | 5856 | |
6e9606d2 | 5857 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5858 | if (!bbio) { |
5859 | ret = -ENOMEM; | |
5860 | goto out; | |
5861 | } | |
6fad823f | 5862 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5863 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5864 | |
8e5cfb55 | 5865 | /* build raid_map */ |
2b19a1fe LB |
5866 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5867 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5868 | u64 tmp; |
9d644a62 | 5869 | unsigned rot; |
8e5cfb55 ZL |
5870 | |
5871 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5872 | sizeof(struct btrfs_bio_stripe) * | |
5873 | num_alloc_stripes + | |
5874 | sizeof(int) * tgtdev_indexes); | |
5875 | ||
5876 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5877 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5878 | |
5879 | /* Fill in the logical address of each stripe */ | |
5880 | tmp = stripe_nr * nr_data_stripes(map); | |
5881 | for (i = 0; i < nr_data_stripes(map); i++) | |
5882 | bbio->raid_map[(i+rot) % num_stripes] = | |
5883 | em->start + (tmp + i) * map->stripe_len; | |
5884 | ||
5885 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5886 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5887 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5888 | RAID6_Q_STRIPE; | |
5889 | } | |
5890 | ||
b89203f7 | 5891 | |
0b3d4cd3 LB |
5892 | for (i = 0; i < num_stripes; i++) { |
5893 | bbio->stripes[i].physical = | |
5894 | map->stripes[stripe_index].physical + | |
5895 | stripe_offset + | |
5896 | stripe_nr * map->stripe_len; | |
5897 | bbio->stripes[i].dev = | |
5898 | map->stripes[stripe_index].dev; | |
5899 | stripe_index++; | |
593060d7 | 5900 | } |
de11cc12 | 5901 | |
2b19a1fe | 5902 | if (need_full_stripe(op)) |
d20983b4 | 5903 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5904 | |
8e5cfb55 ZL |
5905 | if (bbio->raid_map) |
5906 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5907 | |
73c0f228 | 5908 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5909 | need_full_stripe(op)) { |
73c0f228 LB |
5910 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5911 | &max_errors); | |
472262f3 SB |
5912 | } |
5913 | ||
de11cc12 | 5914 | *bbio_ret = bbio; |
10f11900 | 5915 | bbio->map_type = map->type; |
de11cc12 LZ |
5916 | bbio->num_stripes = num_stripes; |
5917 | bbio->max_errors = max_errors; | |
5918 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5919 | |
5920 | /* | |
5921 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5922 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5923 | * available as a mirror | |
5924 | */ | |
5925 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5926 | WARN_ON(num_stripes > 1); | |
5927 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5928 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5929 | bbio->mirror_num = map->num_stripes + 1; | |
5930 | } | |
cea9e445 | 5931 | out: |
73beece9 | 5932 | if (dev_replace_is_ongoing) { |
7fb2eced DS |
5933 | ASSERT(atomic_read(&dev_replace->blocking_readers) > 0); |
5934 | btrfs_dev_replace_read_lock(dev_replace); | |
5935 | /* Barrier implied by atomic_dec_and_test */ | |
5936 | if (atomic_dec_and_test(&dev_replace->blocking_readers)) | |
5937 | cond_wake_up_nomb(&dev_replace->read_lock_wq); | |
7e79cb86 | 5938 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 | 5939 | } |
0b86a832 | 5940 | free_extent_map(em); |
de11cc12 | 5941 | return ret; |
0b86a832 CM |
5942 | } |
5943 | ||
cf8cddd3 | 5944 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 5945 | u64 logical, u64 *length, |
a1d3c478 | 5946 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5947 | { |
b3d3fa51 | 5948 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5949 | mirror_num, 0); |
f2d8d74d CM |
5950 | } |
5951 | ||
af8e2d1d | 5952 | /* For Scrub/replace */ |
cf8cddd3 | 5953 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 5954 | u64 logical, u64 *length, |
825ad4c9 | 5955 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 5956 | { |
825ad4c9 | 5957 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
5958 | } |
5959 | ||
63a9c7b9 NB |
5960 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
5961 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 5962 | { |
a512bbf8 YZ |
5963 | struct extent_map *em; |
5964 | struct map_lookup *map; | |
5965 | u64 *buf; | |
5966 | u64 bytenr; | |
5967 | u64 length; | |
5968 | u64 stripe_nr; | |
53b381b3 | 5969 | u64 rmap_len; |
a512bbf8 YZ |
5970 | int i, j, nr = 0; |
5971 | ||
60ca842e | 5972 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 5973 | if (IS_ERR(em)) |
835d974f | 5974 | return -EIO; |
835d974f | 5975 | |
95617d69 | 5976 | map = em->map_lookup; |
a512bbf8 | 5977 | length = em->len; |
53b381b3 DW |
5978 | rmap_len = map->stripe_len; |
5979 | ||
a512bbf8 | 5980 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5981 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5982 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5983 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5984 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5985 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5986 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5987 | } | |
a512bbf8 | 5988 | |
31e818fe | 5989 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5990 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5991 | |
5992 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
5993 | if (map->stripes[i].physical > physical || |
5994 | map->stripes[i].physical + length <= physical) | |
5995 | continue; | |
5996 | ||
5997 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 5998 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5999 | |
6000 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6001 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6002 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6003 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6004 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6005 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6006 | * Alternatively, just use rmap_len below instead of | |
6007 | * map->stripe_len */ | |
6008 | ||
6009 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6010 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6011 | for (j = 0; j < nr; j++) { |
6012 | if (buf[j] == bytenr) | |
6013 | break; | |
6014 | } | |
934d375b CM |
6015 | if (j == nr) { |
6016 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6017 | buf[nr++] = bytenr; |
934d375b | 6018 | } |
a512bbf8 YZ |
6019 | } |
6020 | ||
a512bbf8 YZ |
6021 | *logical = buf; |
6022 | *naddrs = nr; | |
53b381b3 | 6023 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6024 | |
6025 | free_extent_map(em); | |
6026 | return 0; | |
f2d8d74d CM |
6027 | } |
6028 | ||
4246a0b6 | 6029 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6030 | { |
326e1dbb MS |
6031 | bio->bi_private = bbio->private; |
6032 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6033 | bio_endio(bio); |
326e1dbb | 6034 | |
6e9606d2 | 6035 | btrfs_put_bbio(bbio); |
8408c716 MX |
6036 | } |
6037 | ||
4246a0b6 | 6038 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6039 | { |
9be3395b | 6040 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6041 | int is_orig_bio = 0; |
8790d502 | 6042 | |
4e4cbee9 | 6043 | if (bio->bi_status) { |
a1d3c478 | 6044 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6045 | if (bio->bi_status == BLK_STS_IOERR || |
6046 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6047 | unsigned int stripe_index = |
9be3395b | 6048 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6049 | struct btrfs_device *dev; |
442a4f63 SB |
6050 | |
6051 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6052 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6053 | if (dev->bdev) { |
37226b21 | 6054 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6055 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6056 | BTRFS_DEV_STAT_WRITE_ERRS); |
6057 | else | |
1cb34c8e | 6058 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6059 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6060 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6061 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6062 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6063 | } |
442a4f63 SB |
6064 | } |
6065 | } | |
8790d502 | 6066 | |
a1d3c478 | 6067 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6068 | is_orig_bio = 1; |
6069 | ||
c404e0dc MX |
6070 | btrfs_bio_counter_dec(bbio->fs_info); |
6071 | ||
a1d3c478 | 6072 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6073 | if (!is_orig_bio) { |
6074 | bio_put(bio); | |
a1d3c478 | 6075 | bio = bbio->orig_bio; |
7d2b4daa | 6076 | } |
c7b22bb1 | 6077 | |
9be3395b | 6078 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6079 | /* only send an error to the higher layers if it is |
53b381b3 | 6080 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6081 | */ |
a1d3c478 | 6082 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6083 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6084 | } else { |
1259ab75 CM |
6085 | /* |
6086 | * this bio is actually up to date, we didn't | |
6087 | * go over the max number of errors | |
6088 | */ | |
2dbe0c77 | 6089 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6090 | } |
c55f1396 | 6091 | |
4246a0b6 | 6092 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6093 | } else if (!is_orig_bio) { |
8790d502 CM |
6094 | bio_put(bio); |
6095 | } | |
8790d502 CM |
6096 | } |
6097 | ||
8b712842 CM |
6098 | /* |
6099 | * see run_scheduled_bios for a description of why bios are collected for | |
6100 | * async submit. | |
6101 | * | |
6102 | * This will add one bio to the pending list for a device and make sure | |
6103 | * the work struct is scheduled. | |
6104 | */ | |
2ff7e61e | 6105 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6106 | struct bio *bio) |
8b712842 | 6107 | { |
0b246afa | 6108 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6109 | int should_queue = 1; |
ffbd517d | 6110 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6111 | |
e6e674bd AJ |
6112 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) || |
6113 | !device->bdev) { | |
4246a0b6 | 6114 | bio_io_error(bio); |
53b381b3 DW |
6115 | return; |
6116 | } | |
6117 | ||
8b712842 | 6118 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6119 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6120 | btrfsic_submit_bio(bio); |
143bede5 | 6121 | return; |
8b712842 CM |
6122 | } |
6123 | ||
492bb6de | 6124 | WARN_ON(bio->bi_next); |
8b712842 | 6125 | bio->bi_next = NULL; |
8b712842 CM |
6126 | |
6127 | spin_lock(&device->io_lock); | |
67f055c7 | 6128 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6129 | pending_bios = &device->pending_sync_bios; |
6130 | else | |
6131 | pending_bios = &device->pending_bios; | |
8b712842 | 6132 | |
ffbd517d CM |
6133 | if (pending_bios->tail) |
6134 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6135 | |
ffbd517d CM |
6136 | pending_bios->tail = bio; |
6137 | if (!pending_bios->head) | |
6138 | pending_bios->head = bio; | |
8b712842 CM |
6139 | if (device->running_pending) |
6140 | should_queue = 0; | |
6141 | ||
6142 | spin_unlock(&device->io_lock); | |
6143 | ||
6144 | if (should_queue) | |
0b246afa | 6145 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6146 | } |
6147 | ||
2ff7e61e JM |
6148 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6149 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6150 | { |
6151 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6152 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6153 | |
6154 | bio->bi_private = bbio; | |
9be3395b | 6155 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6156 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6157 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6158 | btrfs_debug_in_rcu(fs_info, |
6159 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6160 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6161 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6162 | bio->bi_iter.bi_size); | |
74d46992 | 6163 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6164 | |
2ff7e61e | 6165 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6166 | |
de1ee92a | 6167 | if (async) |
2ff7e61e | 6168 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6169 | else |
4e49ea4a | 6170 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6171 | } |
6172 | ||
de1ee92a JB |
6173 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6174 | { | |
6175 | atomic_inc(&bbio->error); | |
6176 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6177 | /* Should be the original bio. */ |
8408c716 MX |
6178 | WARN_ON(bio != bbio->orig_bio); |
6179 | ||
9be3395b | 6180 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6181 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6182 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6183 | bio->bi_status = BLK_STS_IOERR; | |
6184 | else | |
6185 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6186 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6187 | } |
6188 | } | |
6189 | ||
58efbc9f OS |
6190 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6191 | int mirror_num, int async_submit) | |
0b86a832 | 6192 | { |
0b86a832 | 6193 | struct btrfs_device *dev; |
8790d502 | 6194 | struct bio *first_bio = bio; |
4f024f37 | 6195 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6196 | u64 length = 0; |
6197 | u64 map_length; | |
0b86a832 | 6198 | int ret; |
08da757d ZL |
6199 | int dev_nr; |
6200 | int total_devs; | |
a1d3c478 | 6201 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6202 | |
4f024f37 | 6203 | length = bio->bi_iter.bi_size; |
0b86a832 | 6204 | map_length = length; |
cea9e445 | 6205 | |
0b246afa | 6206 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6207 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6208 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6209 | if (ret) { |
0b246afa | 6210 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6211 | return errno_to_blk_status(ret); |
c404e0dc | 6212 | } |
cea9e445 | 6213 | |
a1d3c478 | 6214 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6215 | bbio->orig_bio = first_bio; |
6216 | bbio->private = first_bio->bi_private; | |
6217 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6218 | bbio->fs_info = fs_info; |
53b381b3 DW |
6219 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6220 | ||
ad1ba2a0 | 6221 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6222 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6223 | /* In this case, map_length has been set to the length of |
6224 | a single stripe; not the whole write */ | |
37226b21 | 6225 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6226 | ret = raid56_parity_write(fs_info, bio, bbio, |
6227 | map_length); | |
53b381b3 | 6228 | } else { |
2ff7e61e JM |
6229 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6230 | map_length, mirror_num, 1); | |
53b381b3 | 6231 | } |
4245215d | 6232 | |
0b246afa | 6233 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6234 | return errno_to_blk_status(ret); |
53b381b3 DW |
6235 | } |
6236 | ||
cea9e445 | 6237 | if (map_length < length) { |
0b246afa | 6238 | btrfs_crit(fs_info, |
5d163e0e JM |
6239 | "mapping failed logical %llu bio len %llu len %llu", |
6240 | logical, length, map_length); | |
cea9e445 CM |
6241 | BUG(); |
6242 | } | |
a1d3c478 | 6243 | |
08da757d | 6244 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6245 | dev = bbio->stripes[dev_nr].dev; |
37226b21 | 6246 | if (!dev || !dev->bdev || |
ebbede42 AJ |
6247 | (bio_op(first_bio) == REQ_OP_WRITE && |
6248 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6249 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6250 | continue; |
6251 | } | |
6252 | ||
3aa8e074 | 6253 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6254 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6255 | else |
a1d3c478 | 6256 | bio = first_bio; |
de1ee92a | 6257 | |
2ff7e61e JM |
6258 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6259 | dev_nr, async_submit); | |
8790d502 | 6260 | } |
0b246afa | 6261 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6262 | return BLK_STS_OK; |
0b86a832 CM |
6263 | } |
6264 | ||
aa1b8cd4 | 6265 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6266 | u8 *uuid, u8 *fsid) |
0b86a832 | 6267 | { |
2b82032c YZ |
6268 | struct btrfs_device *device; |
6269 | struct btrfs_fs_devices *cur_devices; | |
6270 | ||
aa1b8cd4 | 6271 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6272 | while (cur_devices) { |
6273 | if (!fsid || | |
44880fdc | 6274 | !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6275 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6276 | if (device) |
6277 | return device; | |
6278 | } | |
6279 | cur_devices = cur_devices->seed; | |
6280 | } | |
6281 | return NULL; | |
0b86a832 CM |
6282 | } |
6283 | ||
2ff7e61e | 6284 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6285 | u64 devid, u8 *dev_uuid) |
6286 | { | |
6287 | struct btrfs_device *device; | |
dfe25020 | 6288 | |
12bd2fc0 ID |
6289 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6290 | if (IS_ERR(device)) | |
adfb69af | 6291 | return device; |
12bd2fc0 ID |
6292 | |
6293 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6294 | device->fs_devices = fs_devices; |
dfe25020 | 6295 | fs_devices->num_devices++; |
12bd2fc0 | 6296 | |
e6e674bd | 6297 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6298 | fs_devices->missing_devices++; |
12bd2fc0 | 6299 | |
dfe25020 CM |
6300 | return device; |
6301 | } | |
6302 | ||
12bd2fc0 ID |
6303 | /** |
6304 | * btrfs_alloc_device - allocate struct btrfs_device | |
6305 | * @fs_info: used only for generating a new devid, can be NULL if | |
6306 | * devid is provided (i.e. @devid != NULL). | |
6307 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6308 | * is generated. | |
6309 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6310 | * is generated. | |
6311 | * | |
6312 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6313 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6314 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6315 | */ |
6316 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6317 | const u64 *devid, | |
6318 | const u8 *uuid) | |
6319 | { | |
6320 | struct btrfs_device *dev; | |
6321 | u64 tmp; | |
6322 | ||
fae7f21c | 6323 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6324 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6325 | |
6326 | dev = __alloc_device(); | |
6327 | if (IS_ERR(dev)) | |
6328 | return dev; | |
6329 | ||
6330 | if (devid) | |
6331 | tmp = *devid; | |
6332 | else { | |
6333 | int ret; | |
6334 | ||
6335 | ret = find_next_devid(fs_info, &tmp); | |
6336 | if (ret) { | |
a425f9d4 | 6337 | btrfs_free_device(dev); |
12bd2fc0 ID |
6338 | return ERR_PTR(ret); |
6339 | } | |
6340 | } | |
6341 | dev->devid = tmp; | |
6342 | ||
6343 | if (uuid) | |
6344 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6345 | else | |
6346 | generate_random_uuid(dev->uuid); | |
6347 | ||
9e0af237 LB |
6348 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6349 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6350 | |
6351 | return dev; | |
6352 | } | |
6353 | ||
e06cd3dd | 6354 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6355 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6356 | struct extent_buffer *leaf, |
6357 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6358 | { |
0b86a832 | 6359 | u64 length; |
f04b772b | 6360 | u64 stripe_len; |
e06cd3dd LB |
6361 | u16 num_stripes; |
6362 | u16 sub_stripes; | |
6363 | u64 type; | |
315409b0 GJ |
6364 | u64 features; |
6365 | bool mixed = false; | |
0b86a832 | 6366 | |
e17cade2 | 6367 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6368 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6369 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6370 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6371 | type = btrfs_chunk_type(leaf, chunk); | |
6372 | ||
f04b772b | 6373 | if (!num_stripes) { |
0b246afa | 6374 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6375 | num_stripes); |
6376 | return -EIO; | |
6377 | } | |
0b246afa JM |
6378 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6379 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6380 | return -EIO; |
6381 | } | |
0b246afa JM |
6382 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6383 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6384 | btrfs_chunk_sector_size(leaf, chunk)); |
6385 | return -EIO; | |
6386 | } | |
0b246afa JM |
6387 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6388 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6389 | return -EIO; |
6390 | } | |
3d8da678 | 6391 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6392 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6393 | stripe_len); |
6394 | return -EIO; | |
6395 | } | |
6396 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6397 | type) { |
0b246afa | 6398 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6399 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6400 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6401 | btrfs_chunk_type(leaf, chunk)); | |
6402 | return -EIO; | |
6403 | } | |
315409b0 GJ |
6404 | |
6405 | if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) { | |
6406 | btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type); | |
6407 | return -EIO; | |
6408 | } | |
6409 | ||
6410 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
6411 | (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) { | |
6412 | btrfs_err(fs_info, | |
6413 | "system chunk with data or metadata type: 0x%llx", type); | |
6414 | return -EIO; | |
6415 | } | |
6416 | ||
6417 | features = btrfs_super_incompat_flags(fs_info->super_copy); | |
6418 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
6419 | mixed = true; | |
6420 | ||
6421 | if (!mixed) { | |
6422 | if ((type & BTRFS_BLOCK_GROUP_METADATA) && | |
6423 | (type & BTRFS_BLOCK_GROUP_DATA)) { | |
6424 | btrfs_err(fs_info, | |
6425 | "mixed chunk type in non-mixed mode: 0x%llx", type); | |
6426 | return -EIO; | |
6427 | } | |
6428 | } | |
6429 | ||
e06cd3dd LB |
6430 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6431 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6432 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6433 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6434 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6435 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6436 | num_stripes != 1)) { | |
0b246afa | 6437 | btrfs_err(fs_info, |
e06cd3dd LB |
6438 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6439 | num_stripes, sub_stripes, | |
6440 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6441 | return -EIO; | |
6442 | } | |
6443 | ||
6444 | return 0; | |
6445 | } | |
6446 | ||
5a2b8e60 | 6447 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6448 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6449 | { |
2b902dfc AJ |
6450 | if (error) |
6451 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6452 | devid, uuid); | |
6453 | else | |
6454 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6455 | devid, uuid); | |
5a2b8e60 AJ |
6456 | } |
6457 | ||
2ff7e61e | 6458 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6459 | struct extent_buffer *leaf, |
6460 | struct btrfs_chunk *chunk) | |
6461 | { | |
0b246afa | 6462 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6463 | struct map_lookup *map; |
6464 | struct extent_map *em; | |
6465 | u64 logical; | |
6466 | u64 length; | |
e06cd3dd LB |
6467 | u64 devid; |
6468 | u8 uuid[BTRFS_UUID_SIZE]; | |
6469 | int num_stripes; | |
6470 | int ret; | |
6471 | int i; | |
6472 | ||
6473 | logical = key->offset; | |
6474 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6475 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6476 | ||
2ff7e61e | 6477 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6478 | if (ret) |
6479 | return ret; | |
a061fc8d | 6480 | |
890871be | 6481 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6482 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6483 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6484 | |
6485 | /* already mapped? */ | |
6486 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6487 | free_extent_map(em); | |
0b86a832 CM |
6488 | return 0; |
6489 | } else if (em) { | |
6490 | free_extent_map(em); | |
6491 | } | |
0b86a832 | 6492 | |
172ddd60 | 6493 | em = alloc_extent_map(); |
0b86a832 CM |
6494 | if (!em) |
6495 | return -ENOMEM; | |
593060d7 | 6496 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6497 | if (!map) { |
6498 | free_extent_map(em); | |
6499 | return -ENOMEM; | |
6500 | } | |
6501 | ||
298a8f9c | 6502 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6503 | em->map_lookup = map; |
0b86a832 CM |
6504 | em->start = logical; |
6505 | em->len = length; | |
70c8a91c | 6506 | em->orig_start = 0; |
0b86a832 | 6507 | em->block_start = 0; |
c8b97818 | 6508 | em->block_len = em->len; |
0b86a832 | 6509 | |
593060d7 CM |
6510 | map->num_stripes = num_stripes; |
6511 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6512 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6513 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6514 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6515 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6516 | map->verified_stripes = 0; |
593060d7 CM |
6517 | for (i = 0; i < num_stripes; i++) { |
6518 | map->stripes[i].physical = | |
6519 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6520 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6521 | read_extent_buffer(leaf, uuid, (unsigned long) |
6522 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6523 | BTRFS_UUID_SIZE); | |
0b246afa | 6524 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6525 | uuid, NULL); |
3cdde224 | 6526 | if (!map->stripes[i].dev && |
0b246afa | 6527 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6528 | free_extent_map(em); |
2b902dfc | 6529 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6530 | return -ENOENT; |
593060d7 | 6531 | } |
dfe25020 CM |
6532 | if (!map->stripes[i].dev) { |
6533 | map->stripes[i].dev = | |
2ff7e61e JM |
6534 | add_missing_dev(fs_info->fs_devices, devid, |
6535 | uuid); | |
adfb69af | 6536 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6537 | free_extent_map(em); |
adfb69af AJ |
6538 | btrfs_err(fs_info, |
6539 | "failed to init missing dev %llu: %ld", | |
6540 | devid, PTR_ERR(map->stripes[i].dev)); | |
6541 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6542 | } |
2b902dfc | 6543 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6544 | } |
e12c9621 AJ |
6545 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6546 | &(map->stripes[i].dev->dev_state)); | |
6547 | ||
0b86a832 CM |
6548 | } |
6549 | ||
890871be | 6550 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6551 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6552 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6553 | if (ret < 0) { |
6554 | btrfs_err(fs_info, | |
6555 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6556 | em->start, em->len, ret); | |
6557 | } | |
0b86a832 CM |
6558 | free_extent_map(em); |
6559 | ||
64f64f43 | 6560 | return ret; |
0b86a832 CM |
6561 | } |
6562 | ||
143bede5 | 6563 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6564 | struct btrfs_dev_item *dev_item, |
6565 | struct btrfs_device *device) | |
6566 | { | |
6567 | unsigned long ptr; | |
0b86a832 CM |
6568 | |
6569 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6570 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6571 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6572 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6573 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6574 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6575 | device->type = btrfs_device_type(leaf, dev_item); |
6576 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6577 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6578 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6579 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6580 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6581 | |
410ba3a2 | 6582 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6583 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6584 | } |
6585 | ||
2ff7e61e | 6586 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6587 | u8 *fsid) |
2b82032c YZ |
6588 | { |
6589 | struct btrfs_fs_devices *fs_devices; | |
6590 | int ret; | |
6591 | ||
a32bf9a3 | 6592 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6593 | ASSERT(fsid); |
2b82032c | 6594 | |
0b246afa | 6595 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6596 | while (fs_devices) { |
44880fdc | 6597 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6598 | return fs_devices; |
6599 | ||
2b82032c YZ |
6600 | fs_devices = fs_devices->seed; |
6601 | } | |
6602 | ||
6603 | fs_devices = find_fsid(fsid); | |
6604 | if (!fs_devices) { | |
0b246afa | 6605 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6606 | return ERR_PTR(-ENOENT); |
6607 | ||
6608 | fs_devices = alloc_fs_devices(fsid); | |
6609 | if (IS_ERR(fs_devices)) | |
6610 | return fs_devices; | |
6611 | ||
6612 | fs_devices->seeding = 1; | |
6613 | fs_devices->opened = 1; | |
6614 | return fs_devices; | |
2b82032c | 6615 | } |
e4404d6e YZ |
6616 | |
6617 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6618 | if (IS_ERR(fs_devices)) |
6619 | return fs_devices; | |
2b82032c | 6620 | |
897fb573 | 6621 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6622 | if (ret) { |
6623 | free_fs_devices(fs_devices); | |
5f375835 | 6624 | fs_devices = ERR_PTR(ret); |
2b82032c | 6625 | goto out; |
48d28232 | 6626 | } |
2b82032c YZ |
6627 | |
6628 | if (!fs_devices->seeding) { | |
0226e0eb | 6629 | close_fs_devices(fs_devices); |
e4404d6e | 6630 | free_fs_devices(fs_devices); |
5f375835 | 6631 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6632 | goto out; |
6633 | } | |
6634 | ||
0b246afa JM |
6635 | fs_devices->seed = fs_info->fs_devices->seed; |
6636 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6637 | out: |
5f375835 | 6638 | return fs_devices; |
2b82032c YZ |
6639 | } |
6640 | ||
2ff7e61e | 6641 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6642 | struct extent_buffer *leaf, |
6643 | struct btrfs_dev_item *dev_item) | |
6644 | { | |
0b246afa | 6645 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6646 | struct btrfs_device *device; |
6647 | u64 devid; | |
6648 | int ret; | |
44880fdc | 6649 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6650 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6651 | ||
0b86a832 | 6652 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6653 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6654 | BTRFS_UUID_SIZE); |
1473b24e | 6655 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6656 | BTRFS_FSID_SIZE); |
2b82032c | 6657 | |
44880fdc | 6658 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6659 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6660 | if (IS_ERR(fs_devices)) |
6661 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6662 | } |
6663 | ||
0b246afa | 6664 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6665 | if (!device) { |
c5502451 | 6666 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6667 | btrfs_report_missing_device(fs_info, devid, |
6668 | dev_uuid, true); | |
45dbdbc9 | 6669 | return -ENOENT; |
c5502451 | 6670 | } |
2b82032c | 6671 | |
2ff7e61e | 6672 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6673 | if (IS_ERR(device)) { |
6674 | btrfs_err(fs_info, | |
6675 | "failed to add missing dev %llu: %ld", | |
6676 | devid, PTR_ERR(device)); | |
6677 | return PTR_ERR(device); | |
6678 | } | |
2b902dfc | 6679 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6680 | } else { |
c5502451 | 6681 | if (!device->bdev) { |
2b902dfc AJ |
6682 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6683 | btrfs_report_missing_device(fs_info, | |
6684 | devid, dev_uuid, true); | |
45dbdbc9 | 6685 | return -ENOENT; |
2b902dfc AJ |
6686 | } |
6687 | btrfs_report_missing_device(fs_info, devid, | |
6688 | dev_uuid, false); | |
c5502451 | 6689 | } |
5f375835 | 6690 | |
e6e674bd AJ |
6691 | if (!device->bdev && |
6692 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6693 | /* |
6694 | * this happens when a device that was properly setup | |
6695 | * in the device info lists suddenly goes bad. | |
6696 | * device->bdev is NULL, and so we have to set | |
6697 | * device->missing to one here | |
6698 | */ | |
5f375835 | 6699 | device->fs_devices->missing_devices++; |
e6e674bd | 6700 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6701 | } |
5f375835 MX |
6702 | |
6703 | /* Move the device to its own fs_devices */ | |
6704 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6705 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6706 | &device->dev_state)); | |
5f375835 MX |
6707 | |
6708 | list_move(&device->dev_list, &fs_devices->devices); | |
6709 | device->fs_devices->num_devices--; | |
6710 | fs_devices->num_devices++; | |
6711 | ||
6712 | device->fs_devices->missing_devices--; | |
6713 | fs_devices->missing_devices++; | |
6714 | ||
6715 | device->fs_devices = fs_devices; | |
6716 | } | |
2b82032c YZ |
6717 | } |
6718 | ||
0b246afa | 6719 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6720 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6721 | if (device->generation != |
6722 | btrfs_device_generation(leaf, dev_item)) | |
6723 | return -EINVAL; | |
6324fbf3 | 6724 | } |
0b86a832 CM |
6725 | |
6726 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6727 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6728 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6729 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6730 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6731 | atomic64_add(device->total_bytes - device->bytes_used, |
6732 | &fs_info->free_chunk_space); | |
2bf64758 | 6733 | } |
0b86a832 | 6734 | ret = 0; |
0b86a832 CM |
6735 | return ret; |
6736 | } | |
6737 | ||
6bccf3ab | 6738 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6739 | { |
6bccf3ab | 6740 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6741 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6742 | struct extent_buffer *sb; |
0b86a832 | 6743 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6744 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6745 | u8 *array_ptr; |
6746 | unsigned long sb_array_offset; | |
84eed90f | 6747 | int ret = 0; |
0b86a832 CM |
6748 | u32 num_stripes; |
6749 | u32 array_size; | |
6750 | u32 len = 0; | |
1ffb22cf | 6751 | u32 cur_offset; |
e06cd3dd | 6752 | u64 type; |
84eed90f | 6753 | struct btrfs_key key; |
0b86a832 | 6754 | |
0b246afa | 6755 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6756 | /* |
6757 | * This will create extent buffer of nodesize, superblock size is | |
6758 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6759 | * overallocate but we can keep it as-is, only the first page is used. | |
6760 | */ | |
2ff7e61e | 6761 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6762 | if (IS_ERR(sb)) |
6763 | return PTR_ERR(sb); | |
4db8c528 | 6764 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6765 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6766 | /* |
01327610 | 6767 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6768 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6769 | * pages up-to-date when the page is larger: extent does not cover the |
6770 | * whole page and consequently check_page_uptodate does not find all | |
6771 | * the page's extents up-to-date (the hole beyond sb), | |
6772 | * write_extent_buffer then triggers a WARN_ON. | |
6773 | * | |
6774 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6775 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6776 | * to silence the warning eg. on PowerPC 64. | |
6777 | */ | |
09cbfeaf | 6778 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6779 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6780 | |
a061fc8d | 6781 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6782 | array_size = btrfs_super_sys_array_size(super_copy); |
6783 | ||
1ffb22cf DS |
6784 | array_ptr = super_copy->sys_chunk_array; |
6785 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6786 | cur_offset = 0; | |
0b86a832 | 6787 | |
1ffb22cf DS |
6788 | while (cur_offset < array_size) { |
6789 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6790 | len = sizeof(*disk_key); |
6791 | if (cur_offset + len > array_size) | |
6792 | goto out_short_read; | |
6793 | ||
0b86a832 CM |
6794 | btrfs_disk_key_to_cpu(&key, disk_key); |
6795 | ||
1ffb22cf DS |
6796 | array_ptr += len; |
6797 | sb_array_offset += len; | |
6798 | cur_offset += len; | |
0b86a832 | 6799 | |
0d81ba5d | 6800 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6801 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6802 | /* |
6803 | * At least one btrfs_chunk with one stripe must be | |
6804 | * present, exact stripe count check comes afterwards | |
6805 | */ | |
6806 | len = btrfs_chunk_item_size(1); | |
6807 | if (cur_offset + len > array_size) | |
6808 | goto out_short_read; | |
6809 | ||
6810 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6811 | if (!num_stripes) { |
ab8d0fc4 JM |
6812 | btrfs_err(fs_info, |
6813 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6814 | num_stripes, cur_offset); |
6815 | ret = -EIO; | |
6816 | break; | |
6817 | } | |
6818 | ||
e06cd3dd LB |
6819 | type = btrfs_chunk_type(sb, chunk); |
6820 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6821 | btrfs_err(fs_info, |
e06cd3dd LB |
6822 | "invalid chunk type %llu in sys_array at offset %u", |
6823 | type, cur_offset); | |
6824 | ret = -EIO; | |
6825 | break; | |
6826 | } | |
6827 | ||
e3540eab DS |
6828 | len = btrfs_chunk_item_size(num_stripes); |
6829 | if (cur_offset + len > array_size) | |
6830 | goto out_short_read; | |
6831 | ||
2ff7e61e | 6832 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6833 | if (ret) |
6834 | break; | |
0b86a832 | 6835 | } else { |
ab8d0fc4 JM |
6836 | btrfs_err(fs_info, |
6837 | "unexpected item type %u in sys_array at offset %u", | |
6838 | (u32)key.type, cur_offset); | |
84eed90f CM |
6839 | ret = -EIO; |
6840 | break; | |
0b86a832 | 6841 | } |
1ffb22cf DS |
6842 | array_ptr += len; |
6843 | sb_array_offset += len; | |
6844 | cur_offset += len; | |
0b86a832 | 6845 | } |
d865177a | 6846 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6847 | free_extent_buffer_stale(sb); |
84eed90f | 6848 | return ret; |
e3540eab DS |
6849 | |
6850 | out_short_read: | |
ab8d0fc4 | 6851 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6852 | len, cur_offset); |
d865177a | 6853 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6854 | free_extent_buffer_stale(sb); |
e3540eab | 6855 | return -EIO; |
0b86a832 CM |
6856 | } |
6857 | ||
21634a19 QW |
6858 | /* |
6859 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6860 | * | |
6528b99d AJ |
6861 | * If the @failing_dev is specified, it's accounted as missing. |
6862 | * | |
21634a19 QW |
6863 | * Return true if all chunks meet the minimal RW mount requirements. |
6864 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6865 | */ | |
6528b99d AJ |
6866 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6867 | struct btrfs_device *failing_dev) | |
21634a19 QW |
6868 | { |
6869 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
6870 | struct extent_map *em; | |
6871 | u64 next_start = 0; | |
6872 | bool ret = true; | |
6873 | ||
6874 | read_lock(&map_tree->map_tree.lock); | |
6875 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
6876 | read_unlock(&map_tree->map_tree.lock); | |
6877 | /* No chunk at all? Return false anyway */ | |
6878 | if (!em) { | |
6879 | ret = false; | |
6880 | goto out; | |
6881 | } | |
6882 | while (em) { | |
6883 | struct map_lookup *map; | |
6884 | int missing = 0; | |
6885 | int max_tolerated; | |
6886 | int i; | |
6887 | ||
6888 | map = em->map_lookup; | |
6889 | max_tolerated = | |
6890 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6891 | map->type); | |
6892 | for (i = 0; i < map->num_stripes; i++) { | |
6893 | struct btrfs_device *dev = map->stripes[i].dev; | |
6894 | ||
e6e674bd AJ |
6895 | if (!dev || !dev->bdev || |
6896 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6897 | dev->last_flush_error) |
6898 | missing++; | |
6528b99d AJ |
6899 | else if (failing_dev && failing_dev == dev) |
6900 | missing++; | |
21634a19 QW |
6901 | } |
6902 | if (missing > max_tolerated) { | |
6528b99d AJ |
6903 | if (!failing_dev) |
6904 | btrfs_warn(fs_info, | |
21634a19 QW |
6905 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
6906 | em->start, missing, max_tolerated); | |
6907 | free_extent_map(em); | |
6908 | ret = false; | |
6909 | goto out; | |
6910 | } | |
6911 | next_start = extent_map_end(em); | |
6912 | free_extent_map(em); | |
6913 | ||
6914 | read_lock(&map_tree->map_tree.lock); | |
6915 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
6916 | (u64)(-1) - next_start); | |
6917 | read_unlock(&map_tree->map_tree.lock); | |
6918 | } | |
6919 | out: | |
6920 | return ret; | |
6921 | } | |
6922 | ||
5b4aacef | 6923 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6924 | { |
5b4aacef | 6925 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6926 | struct btrfs_path *path; |
6927 | struct extent_buffer *leaf; | |
6928 | struct btrfs_key key; | |
6929 | struct btrfs_key found_key; | |
6930 | int ret; | |
6931 | int slot; | |
99e3ecfc | 6932 | u64 total_dev = 0; |
0b86a832 | 6933 | |
0b86a832 CM |
6934 | path = btrfs_alloc_path(); |
6935 | if (!path) | |
6936 | return -ENOMEM; | |
6937 | ||
3dd0f7a3 AJ |
6938 | /* |
6939 | * uuid_mutex is needed only if we are mounting a sprout FS | |
6940 | * otherwise we don't need it. | |
6941 | */ | |
b367e47f | 6942 | mutex_lock(&uuid_mutex); |
34441361 | 6943 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6944 | |
395927a9 FDBM |
6945 | /* |
6946 | * Read all device items, and then all the chunk items. All | |
6947 | * device items are found before any chunk item (their object id | |
6948 | * is smaller than the lowest possible object id for a chunk | |
6949 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6950 | */ |
6951 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6952 | key.offset = 0; | |
6953 | key.type = 0; | |
0b86a832 | 6954 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6955 | if (ret < 0) |
6956 | goto error; | |
d397712b | 6957 | while (1) { |
0b86a832 CM |
6958 | leaf = path->nodes[0]; |
6959 | slot = path->slots[0]; | |
6960 | if (slot >= btrfs_header_nritems(leaf)) { | |
6961 | ret = btrfs_next_leaf(root, path); | |
6962 | if (ret == 0) | |
6963 | continue; | |
6964 | if (ret < 0) | |
6965 | goto error; | |
6966 | break; | |
6967 | } | |
6968 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6969 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6970 | struct btrfs_dev_item *dev_item; | |
6971 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6972 | struct btrfs_dev_item); |
2ff7e61e | 6973 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
6974 | if (ret) |
6975 | goto error; | |
99e3ecfc | 6976 | total_dev++; |
0b86a832 CM |
6977 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6978 | struct btrfs_chunk *chunk; | |
6979 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 6980 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
6981 | if (ret) |
6982 | goto error; | |
0b86a832 CM |
6983 | } |
6984 | path->slots[0]++; | |
6985 | } | |
99e3ecfc LB |
6986 | |
6987 | /* | |
6988 | * After loading chunk tree, we've got all device information, | |
6989 | * do another round of validation checks. | |
6990 | */ | |
0b246afa JM |
6991 | if (total_dev != fs_info->fs_devices->total_devices) { |
6992 | btrfs_err(fs_info, | |
99e3ecfc | 6993 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 6994 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
6995 | total_dev); |
6996 | ret = -EINVAL; | |
6997 | goto error; | |
6998 | } | |
0b246afa JM |
6999 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7000 | fs_info->fs_devices->total_rw_bytes) { | |
7001 | btrfs_err(fs_info, | |
99e3ecfc | 7002 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7003 | btrfs_super_total_bytes(fs_info->super_copy), |
7004 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7005 | ret = -EINVAL; |
7006 | goto error; | |
7007 | } | |
0b86a832 CM |
7008 | ret = 0; |
7009 | error: | |
34441361 | 7010 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7011 | mutex_unlock(&uuid_mutex); |
7012 | ||
2b82032c | 7013 | btrfs_free_path(path); |
0b86a832 CM |
7014 | return ret; |
7015 | } | |
442a4f63 | 7016 | |
cb517eab MX |
7017 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7018 | { | |
7019 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7020 | struct btrfs_device *device; | |
7021 | ||
29cc83f6 LB |
7022 | while (fs_devices) { |
7023 | mutex_lock(&fs_devices->device_list_mutex); | |
7024 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7025 | device->fs_info = fs_info; |
29cc83f6 LB |
7026 | mutex_unlock(&fs_devices->device_list_mutex); |
7027 | ||
7028 | fs_devices = fs_devices->seed; | |
7029 | } | |
cb517eab MX |
7030 | } |
7031 | ||
733f4fbb SB |
7032 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7033 | { | |
7034 | int i; | |
7035 | ||
7036 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7037 | btrfs_dev_stat_reset(dev, i); | |
7038 | } | |
7039 | ||
7040 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7041 | { | |
7042 | struct btrfs_key key; | |
7043 | struct btrfs_key found_key; | |
7044 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7045 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7046 | struct extent_buffer *eb; | |
7047 | int slot; | |
7048 | int ret = 0; | |
7049 | struct btrfs_device *device; | |
7050 | struct btrfs_path *path = NULL; | |
7051 | int i; | |
7052 | ||
7053 | path = btrfs_alloc_path(); | |
7054 | if (!path) { | |
7055 | ret = -ENOMEM; | |
7056 | goto out; | |
7057 | } | |
7058 | ||
7059 | mutex_lock(&fs_devices->device_list_mutex); | |
7060 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7061 | int item_size; | |
7062 | struct btrfs_dev_stats_item *ptr; | |
7063 | ||
242e2956 DS |
7064 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7065 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7066 | key.offset = device->devid; |
7067 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7068 | if (ret) { | |
733f4fbb SB |
7069 | __btrfs_reset_dev_stats(device); |
7070 | device->dev_stats_valid = 1; | |
7071 | btrfs_release_path(path); | |
7072 | continue; | |
7073 | } | |
7074 | slot = path->slots[0]; | |
7075 | eb = path->nodes[0]; | |
7076 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7077 | item_size = btrfs_item_size_nr(eb, slot); | |
7078 | ||
7079 | ptr = btrfs_item_ptr(eb, slot, | |
7080 | struct btrfs_dev_stats_item); | |
7081 | ||
7082 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7083 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7084 | btrfs_dev_stat_set(device, i, | |
7085 | btrfs_dev_stats_value(eb, ptr, i)); | |
7086 | else | |
7087 | btrfs_dev_stat_reset(device, i); | |
7088 | } | |
7089 | ||
7090 | device->dev_stats_valid = 1; | |
7091 | btrfs_dev_stat_print_on_load(device); | |
7092 | btrfs_release_path(path); | |
7093 | } | |
7094 | mutex_unlock(&fs_devices->device_list_mutex); | |
7095 | ||
7096 | out: | |
7097 | btrfs_free_path(path); | |
7098 | return ret < 0 ? ret : 0; | |
7099 | } | |
7100 | ||
7101 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7102 | struct btrfs_device *device) |
7103 | { | |
5495f195 | 7104 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7105 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7106 | struct btrfs_path *path; |
7107 | struct btrfs_key key; | |
7108 | struct extent_buffer *eb; | |
7109 | struct btrfs_dev_stats_item *ptr; | |
7110 | int ret; | |
7111 | int i; | |
7112 | ||
242e2956 DS |
7113 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7114 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7115 | key.offset = device->devid; |
7116 | ||
7117 | path = btrfs_alloc_path(); | |
fa252992 DS |
7118 | if (!path) |
7119 | return -ENOMEM; | |
733f4fbb SB |
7120 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7121 | if (ret < 0) { | |
0b246afa | 7122 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7123 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7124 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7125 | goto out; |
7126 | } | |
7127 | ||
7128 | if (ret == 0 && | |
7129 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7130 | /* need to delete old one and insert a new one */ | |
7131 | ret = btrfs_del_item(trans, dev_root, path); | |
7132 | if (ret != 0) { | |
0b246afa | 7133 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7134 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7135 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7136 | goto out; |
7137 | } | |
7138 | ret = 1; | |
7139 | } | |
7140 | ||
7141 | if (ret == 1) { | |
7142 | /* need to insert a new item */ | |
7143 | btrfs_release_path(path); | |
7144 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7145 | &key, sizeof(*ptr)); | |
7146 | if (ret < 0) { | |
0b246afa | 7147 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7148 | "insert dev_stats item for device %s failed %d", |
7149 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7150 | goto out; |
7151 | } | |
7152 | } | |
7153 | ||
7154 | eb = path->nodes[0]; | |
7155 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7156 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7157 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7158 | btrfs_dev_stat_read(device, i)); | |
7159 | btrfs_mark_buffer_dirty(eb); | |
7160 | ||
7161 | out: | |
7162 | btrfs_free_path(path); | |
7163 | return ret; | |
7164 | } | |
7165 | ||
7166 | /* | |
7167 | * called from commit_transaction. Writes all changed device stats to disk. | |
7168 | */ | |
7169 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7170 | struct btrfs_fs_info *fs_info) | |
7171 | { | |
733f4fbb SB |
7172 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7173 | struct btrfs_device *device; | |
addc3fa7 | 7174 | int stats_cnt; |
733f4fbb SB |
7175 | int ret = 0; |
7176 | ||
7177 | mutex_lock(&fs_devices->device_list_mutex); | |
7178 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7179 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7180 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7181 | continue; |
7182 | ||
9deae968 NB |
7183 | |
7184 | /* | |
7185 | * There is a LOAD-LOAD control dependency between the value of | |
7186 | * dev_stats_ccnt and updating the on-disk values which requires | |
7187 | * reading the in-memory counters. Such control dependencies | |
7188 | * require explicit read memory barriers. | |
7189 | * | |
7190 | * This memory barriers pairs with smp_mb__before_atomic in | |
7191 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7192 | * barrier implied by atomic_xchg in | |
7193 | * btrfs_dev_stats_read_and_reset | |
7194 | */ | |
7195 | smp_rmb(); | |
7196 | ||
5495f195 | 7197 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7198 | if (!ret) |
addc3fa7 | 7199 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7200 | } |
7201 | mutex_unlock(&fs_devices->device_list_mutex); | |
7202 | ||
7203 | return ret; | |
7204 | } | |
7205 | ||
442a4f63 SB |
7206 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7207 | { | |
7208 | btrfs_dev_stat_inc(dev, index); | |
7209 | btrfs_dev_stat_print_on_error(dev); | |
7210 | } | |
7211 | ||
48a3b636 | 7212 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7213 | { |
733f4fbb SB |
7214 | if (!dev->dev_stats_valid) |
7215 | return; | |
fb456252 | 7216 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7217 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7218 | rcu_str_deref(dev->name), |
442a4f63 SB |
7219 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7220 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7221 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7222 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7223 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7224 | } |
c11d2c23 | 7225 | |
733f4fbb SB |
7226 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7227 | { | |
a98cdb85 SB |
7228 | int i; |
7229 | ||
7230 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7231 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7232 | break; | |
7233 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7234 | return; /* all values == 0, suppress message */ | |
7235 | ||
fb456252 | 7236 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7237 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7238 | rcu_str_deref(dev->name), |
733f4fbb SB |
7239 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7240 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7241 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7242 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7243 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7244 | } | |
7245 | ||
2ff7e61e | 7246 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7247 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7248 | { |
7249 | struct btrfs_device *dev; | |
0b246afa | 7250 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7251 | int i; |
7252 | ||
7253 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7254 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7255 | mutex_unlock(&fs_devices->device_list_mutex); |
7256 | ||
7257 | if (!dev) { | |
0b246afa | 7258 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7259 | return -ENODEV; |
733f4fbb | 7260 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7261 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7262 | return -ENODEV; |
b27f7c0c | 7263 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7264 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7265 | if (stats->nr_items > i) | |
7266 | stats->values[i] = | |
7267 | btrfs_dev_stat_read_and_reset(dev, i); | |
7268 | else | |
7269 | btrfs_dev_stat_reset(dev, i); | |
7270 | } | |
7271 | } else { | |
7272 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7273 | if (stats->nr_items > i) | |
7274 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7275 | } | |
7276 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7277 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7278 | return 0; | |
7279 | } | |
a8a6dab7 | 7280 | |
da353f6b | 7281 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7282 | { |
7283 | struct buffer_head *bh; | |
7284 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7285 | int copy_num; |
a8a6dab7 | 7286 | |
12b1c263 AJ |
7287 | if (!bdev) |
7288 | return; | |
a8a6dab7 | 7289 | |
12b1c263 AJ |
7290 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7291 | copy_num++) { | |
a8a6dab7 | 7292 | |
12b1c263 AJ |
7293 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7294 | continue; | |
7295 | ||
7296 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7297 | ||
7298 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7299 | set_buffer_dirty(bh); | |
7300 | sync_dirty_buffer(bh); | |
7301 | brelse(bh); | |
7302 | } | |
7303 | ||
7304 | /* Notify udev that device has changed */ | |
7305 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7306 | ||
7307 | /* Update ctime/mtime for device path for libblkid */ | |
7308 | update_dev_time(device_path); | |
a8a6dab7 | 7309 | } |
935e5cc9 MX |
7310 | |
7311 | /* | |
7312 | * Update the size of all devices, which is used for writing out the | |
7313 | * super blocks. | |
7314 | */ | |
7315 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7316 | { | |
7317 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7318 | struct btrfs_device *curr, *next; | |
7319 | ||
7320 | if (list_empty(&fs_devices->resized_devices)) | |
7321 | return; | |
7322 | ||
7323 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7324 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7325 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7326 | resized_list) { | |
7327 | list_del_init(&curr->resized_list); | |
7328 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7329 | } | |
34441361 | 7330 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7331 | mutex_unlock(&fs_devices->device_list_mutex); |
7332 | } | |
ce7213c7 MX |
7333 | |
7334 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7335 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7336 | { |
e9b919b1 | 7337 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7338 | struct extent_map *em; |
7339 | struct map_lookup *map; | |
7340 | struct btrfs_device *dev; | |
7341 | int i; | |
7342 | ||
e9b919b1 | 7343 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7344 | return; |
7345 | ||
7346 | /* In order to kick the device replace finish process */ | |
34441361 | 7347 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7348 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7349 | map = em->map_lookup; |
ce7213c7 MX |
7350 | |
7351 | for (i = 0; i < map->num_stripes; i++) { | |
7352 | dev = map->stripes[i].dev; | |
7353 | dev->commit_bytes_used = dev->bytes_used; | |
7354 | } | |
7355 | } | |
34441361 | 7356 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7357 | } |
5a13f430 AJ |
7358 | |
7359 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7360 | { | |
7361 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7362 | while (fs_devices) { | |
7363 | fs_devices->fs_info = fs_info; | |
7364 | fs_devices = fs_devices->seed; | |
7365 | } | |
7366 | } | |
7367 | ||
7368 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7369 | { | |
7370 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7371 | while (fs_devices) { | |
7372 | fs_devices->fs_info = NULL; | |
7373 | fs_devices = fs_devices->seed; | |
7374 | } | |
7375 | } | |
46df06b8 DS |
7376 | |
7377 | /* | |
7378 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7379 | */ | |
7380 | int btrfs_bg_type_to_factor(u64 flags) | |
7381 | { | |
7382 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7383 | BTRFS_BLOCK_GROUP_RAID10)) | |
7384 | return 2; | |
7385 | return 1; | |
7386 | } | |
cf90d884 QW |
7387 | |
7388 | ||
7389 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) | |
7390 | { | |
7391 | int index = btrfs_bg_flags_to_raid_index(type); | |
7392 | int ncopies = btrfs_raid_array[index].ncopies; | |
7393 | int data_stripes; | |
7394 | ||
7395 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
7396 | case BTRFS_BLOCK_GROUP_RAID5: | |
7397 | data_stripes = num_stripes - 1; | |
7398 | break; | |
7399 | case BTRFS_BLOCK_GROUP_RAID6: | |
7400 | data_stripes = num_stripes - 2; | |
7401 | break; | |
7402 | default: | |
7403 | data_stripes = num_stripes / ncopies; | |
7404 | break; | |
7405 | } | |
7406 | return div_u64(chunk_len, data_stripes); | |
7407 | } | |
7408 | ||
7409 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7410 | u64 chunk_offset, u64 devid, | |
7411 | u64 physical_offset, u64 physical_len) | |
7412 | { | |
7413 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7414 | struct extent_map *em; | |
7415 | struct map_lookup *map; | |
05a37c48 | 7416 | struct btrfs_device *dev; |
cf90d884 QW |
7417 | u64 stripe_len; |
7418 | bool found = false; | |
7419 | int ret = 0; | |
7420 | int i; | |
7421 | ||
7422 | read_lock(&em_tree->lock); | |
7423 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7424 | read_unlock(&em_tree->lock); | |
7425 | ||
7426 | if (!em) { | |
7427 | btrfs_err(fs_info, | |
7428 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7429 | physical_offset, devid); | |
7430 | ret = -EUCLEAN; | |
7431 | goto out; | |
7432 | } | |
7433 | ||
7434 | map = em->map_lookup; | |
7435 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7436 | if (physical_len != stripe_len) { | |
7437 | btrfs_err(fs_info, | |
7438 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7439 | physical_offset, devid, em->start, physical_len, | |
7440 | stripe_len); | |
7441 | ret = -EUCLEAN; | |
7442 | goto out; | |
7443 | } | |
7444 | ||
7445 | for (i = 0; i < map->num_stripes; i++) { | |
7446 | if (map->stripes[i].dev->devid == devid && | |
7447 | map->stripes[i].physical == physical_offset) { | |
7448 | found = true; | |
7449 | if (map->verified_stripes >= map->num_stripes) { | |
7450 | btrfs_err(fs_info, | |
7451 | "too many dev extents for chunk %llu found", | |
7452 | em->start); | |
7453 | ret = -EUCLEAN; | |
7454 | goto out; | |
7455 | } | |
7456 | map->verified_stripes++; | |
7457 | break; | |
7458 | } | |
7459 | } | |
7460 | if (!found) { | |
7461 | btrfs_err(fs_info, | |
7462 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7463 | physical_offset, devid); | |
7464 | ret = -EUCLEAN; | |
7465 | } | |
05a37c48 QW |
7466 | |
7467 | /* Make sure no dev extent is beyond device bondary */ | |
7468 | dev = btrfs_find_device(fs_info, devid, NULL, NULL); | |
7469 | if (!dev) { | |
7470 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7471 | ret = -EUCLEAN; | |
7472 | goto out; | |
7473 | } | |
7474 | if (physical_offset + physical_len > dev->disk_total_bytes) { | |
7475 | btrfs_err(fs_info, | |
7476 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7477 | devid, physical_offset, physical_len, | |
7478 | dev->disk_total_bytes); | |
7479 | ret = -EUCLEAN; | |
7480 | goto out; | |
7481 | } | |
cf90d884 QW |
7482 | out: |
7483 | free_extent_map(em); | |
7484 | return ret; | |
7485 | } | |
7486 | ||
7487 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7488 | { | |
7489 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7490 | struct extent_map *em; | |
7491 | struct rb_node *node; | |
7492 | int ret = 0; | |
7493 | ||
7494 | read_lock(&em_tree->lock); | |
07e1ce09 | 7495 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7496 | em = rb_entry(node, struct extent_map, rb_node); |
7497 | if (em->map_lookup->num_stripes != | |
7498 | em->map_lookup->verified_stripes) { | |
7499 | btrfs_err(fs_info, | |
7500 | "chunk %llu has missing dev extent, have %d expect %d", | |
7501 | em->start, em->map_lookup->verified_stripes, | |
7502 | em->map_lookup->num_stripes); | |
7503 | ret = -EUCLEAN; | |
7504 | goto out; | |
7505 | } | |
7506 | } | |
7507 | out: | |
7508 | read_unlock(&em_tree->lock); | |
7509 | return ret; | |
7510 | } | |
7511 | ||
7512 | /* | |
7513 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7514 | * later chunk allocation/free would cause unexpected behavior. | |
7515 | * | |
7516 | * NOTE: This will iterate through the whole device tree, which should be of | |
7517 | * the same size level as the chunk tree. This slightly increases mount time. | |
7518 | */ | |
7519 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7520 | { | |
7521 | struct btrfs_path *path; | |
7522 | struct btrfs_root *root = fs_info->dev_root; | |
7523 | struct btrfs_key key; | |
5eb19381 QW |
7524 | u64 prev_devid = 0; |
7525 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7526 | int ret = 0; |
7527 | ||
7528 | key.objectid = 1; | |
7529 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7530 | key.offset = 0; | |
7531 | ||
7532 | path = btrfs_alloc_path(); | |
7533 | if (!path) | |
7534 | return -ENOMEM; | |
7535 | ||
7536 | path->reada = READA_FORWARD; | |
7537 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7538 | if (ret < 0) | |
7539 | goto out; | |
7540 | ||
7541 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7542 | ret = btrfs_next_item(root, path); | |
7543 | if (ret < 0) | |
7544 | goto out; | |
7545 | /* No dev extents at all? Not good */ | |
7546 | if (ret > 0) { | |
7547 | ret = -EUCLEAN; | |
7548 | goto out; | |
7549 | } | |
7550 | } | |
7551 | while (1) { | |
7552 | struct extent_buffer *leaf = path->nodes[0]; | |
7553 | struct btrfs_dev_extent *dext; | |
7554 | int slot = path->slots[0]; | |
7555 | u64 chunk_offset; | |
7556 | u64 physical_offset; | |
7557 | u64 physical_len; | |
7558 | u64 devid; | |
7559 | ||
7560 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7561 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7562 | break; | |
7563 | devid = key.objectid; | |
7564 | physical_offset = key.offset; | |
7565 | ||
7566 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7567 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7568 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7569 | ||
5eb19381 QW |
7570 | /* Check if this dev extent overlaps with the previous one */ |
7571 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7572 | btrfs_err(fs_info, | |
7573 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7574 | devid, physical_offset, prev_dev_ext_end); | |
7575 | ret = -EUCLEAN; | |
7576 | goto out; | |
7577 | } | |
7578 | ||
cf90d884 QW |
7579 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7580 | physical_offset, physical_len); | |
7581 | if (ret < 0) | |
7582 | goto out; | |
5eb19381 QW |
7583 | prev_devid = devid; |
7584 | prev_dev_ext_end = physical_offset + physical_len; | |
7585 | ||
cf90d884 QW |
7586 | ret = btrfs_next_item(root, path); |
7587 | if (ret < 0) | |
7588 | goto out; | |
7589 | if (ret > 0) { | |
7590 | ret = 0; | |
7591 | break; | |
7592 | } | |
7593 | } | |
7594 | ||
7595 | /* Ensure all chunks have corresponding dev extents */ | |
7596 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7597 | out: | |
7598 | btrfs_free_path(path); | |
7599 | return ret; | |
7600 | } | |
eede2bf3 OS |
7601 | |
7602 | /* | |
7603 | * Check whether the given block group or device is pinned by any inode being | |
7604 | * used as a swapfile. | |
7605 | */ | |
7606 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7607 | { | |
7608 | struct btrfs_swapfile_pin *sp; | |
7609 | struct rb_node *node; | |
7610 | ||
7611 | spin_lock(&fs_info->swapfile_pins_lock); | |
7612 | node = fs_info->swapfile_pins.rb_node; | |
7613 | while (node) { | |
7614 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7615 | if (ptr < sp->ptr) | |
7616 | node = node->rb_left; | |
7617 | else if (ptr > sp->ptr) | |
7618 | node = node->rb_right; | |
7619 | else | |
7620 | break; | |
7621 | } | |
7622 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7623 | return node != NULL; | |
7624 | } |