]>
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, | |
b50836ed | 40 | .nparity = 0, |
ed23467b | 41 | .raid_name = "raid10", |
41a6e891 | 42 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 43 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
44 | }, |
45 | [BTRFS_RAID_RAID1] = { | |
46 | .sub_stripes = 1, | |
47 | .dev_stripes = 1, | |
48 | .devs_max = 2, | |
49 | .devs_min = 2, | |
8789f4fe | 50 | .tolerated_failures = 1, |
af902047 ZL |
51 | .devs_increment = 2, |
52 | .ncopies = 2, | |
b50836ed | 53 | .nparity = 0, |
ed23467b | 54 | .raid_name = "raid1", |
41a6e891 | 55 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 56 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 ZL |
57 | }, |
58 | [BTRFS_RAID_DUP] = { | |
59 | .sub_stripes = 1, | |
60 | .dev_stripes = 2, | |
61 | .devs_max = 1, | |
62 | .devs_min = 1, | |
8789f4fe | 63 | .tolerated_failures = 0, |
af902047 ZL |
64 | .devs_increment = 1, |
65 | .ncopies = 2, | |
b50836ed | 66 | .nparity = 0, |
ed23467b | 67 | .raid_name = "dup", |
41a6e891 | 68 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 69 | .mindev_error = 0, |
af902047 ZL |
70 | }, |
71 | [BTRFS_RAID_RAID0] = { | |
72 | .sub_stripes = 1, | |
73 | .dev_stripes = 1, | |
74 | .devs_max = 0, | |
75 | .devs_min = 2, | |
8789f4fe | 76 | .tolerated_failures = 0, |
af902047 ZL |
77 | .devs_increment = 1, |
78 | .ncopies = 1, | |
b50836ed | 79 | .nparity = 0, |
ed23467b | 80 | .raid_name = "raid0", |
41a6e891 | 81 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 82 | .mindev_error = 0, |
af902047 ZL |
83 | }, |
84 | [BTRFS_RAID_SINGLE] = { | |
85 | .sub_stripes = 1, | |
86 | .dev_stripes = 1, | |
87 | .devs_max = 1, | |
88 | .devs_min = 1, | |
8789f4fe | 89 | .tolerated_failures = 0, |
af902047 ZL |
90 | .devs_increment = 1, |
91 | .ncopies = 1, | |
b50836ed | 92 | .nparity = 0, |
ed23467b | 93 | .raid_name = "single", |
41a6e891 | 94 | .bg_flag = 0, |
f9fbcaa2 | 95 | .mindev_error = 0, |
af902047 ZL |
96 | }, |
97 | [BTRFS_RAID_RAID5] = { | |
98 | .sub_stripes = 1, | |
99 | .dev_stripes = 1, | |
100 | .devs_max = 0, | |
101 | .devs_min = 2, | |
8789f4fe | 102 | .tolerated_failures = 1, |
af902047 | 103 | .devs_increment = 1, |
da612e31 | 104 | .ncopies = 1, |
b50836ed | 105 | .nparity = 1, |
ed23467b | 106 | .raid_name = "raid5", |
41a6e891 | 107 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 108 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
109 | }, |
110 | [BTRFS_RAID_RAID6] = { | |
111 | .sub_stripes = 1, | |
112 | .dev_stripes = 1, | |
113 | .devs_max = 0, | |
114 | .devs_min = 3, | |
8789f4fe | 115 | .tolerated_failures = 2, |
af902047 | 116 | .devs_increment = 1, |
da612e31 | 117 | .ncopies = 1, |
b50836ed | 118 | .nparity = 2, |
ed23467b | 119 | .raid_name = "raid6", |
41a6e891 | 120 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 121 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
122 | }, |
123 | }; | |
124 | ||
ed23467b AJ |
125 | const char *get_raid_name(enum btrfs_raid_types type) |
126 | { | |
127 | if (type >= BTRFS_NR_RAID_TYPES) | |
128 | return NULL; | |
129 | ||
130 | return btrfs_raid_array[type].raid_name; | |
131 | } | |
132 | ||
2b82032c | 133 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 134 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 135 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 136 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 137 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 138 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
139 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
140 | enum btrfs_map_op op, | |
141 | u64 logical, u64 *length, | |
142 | struct btrfs_bio **bbio_ret, | |
143 | int mirror_num, int need_raid_map); | |
2b82032c | 144 | |
9c6b1c4d DS |
145 | /* |
146 | * Device locking | |
147 | * ============== | |
148 | * | |
149 | * There are several mutexes that protect manipulation of devices and low-level | |
150 | * structures like chunks but not block groups, extents or files | |
151 | * | |
152 | * uuid_mutex (global lock) | |
153 | * ------------------------ | |
154 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
155 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
156 | * device) or requested by the device= mount option | |
157 | * | |
158 | * the mutex can be very coarse and can cover long-running operations | |
159 | * | |
160 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
161 | * seeding, structure cloning, openning/closing devices at mount/umount time | |
162 | * | |
163 | * global::fs_devs - add, remove, updates to the global list | |
164 | * | |
165 | * does not protect: manipulation of the fs_devices::devices list! | |
166 | * | |
167 | * btrfs_device::name - renames (write side), read is RCU | |
168 | * | |
169 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
170 | * ------------------------------------------------ | |
171 | * protects updates to fs_devices::devices, ie. adding and deleting | |
172 | * | |
173 | * simple list traversal with read-only actions can be done with RCU protection | |
174 | * | |
175 | * may be used to exclude some operations from running concurrently without any | |
176 | * modifications to the list (see write_all_supers) | |
177 | * | |
9c6b1c4d DS |
178 | * balance_mutex |
179 | * ------------- | |
180 | * protects balance structures (status, state) and context accessed from | |
181 | * several places (internally, ioctl) | |
182 | * | |
183 | * chunk_mutex | |
184 | * ----------- | |
185 | * protects chunks, adding or removing during allocation, trim or when a new | |
186 | * device is added/removed | |
187 | * | |
188 | * cleaner_mutex | |
189 | * ------------- | |
190 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
191 | * cleaning together with relocation or delayed iputs | |
192 | * | |
193 | * | |
194 | * Lock nesting | |
195 | * ============ | |
196 | * | |
197 | * uuid_mutex | |
198 | * volume_mutex | |
199 | * device_list_mutex | |
200 | * chunk_mutex | |
201 | * balance_mutex | |
89595e80 AJ |
202 | * |
203 | * | |
204 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
205 | * ====================================== | |
206 | * | |
207 | * Maintains the exclusivity of the following operations that apply to the | |
208 | * whole filesystem and cannot run in parallel. | |
209 | * | |
210 | * - Balance (*) | |
211 | * - Device add | |
212 | * - Device remove | |
213 | * - Device replace (*) | |
214 | * - Resize | |
215 | * | |
216 | * The device operations (as above) can be in one of the following states: | |
217 | * | |
218 | * - Running state | |
219 | * - Paused state | |
220 | * - Completed state | |
221 | * | |
222 | * Only device operations marked with (*) can go into the Paused state for the | |
223 | * following reasons: | |
224 | * | |
225 | * - ioctl (only Balance can be Paused through ioctl) | |
226 | * - filesystem remounted as read-only | |
227 | * - filesystem unmounted and mounted as read-only | |
228 | * - system power-cycle and filesystem mounted as read-only | |
229 | * - filesystem or device errors leading to forced read-only | |
230 | * | |
231 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
232 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
233 | * A device operation in Paused or Running state can be canceled or resumed | |
234 | * either by ioctl (Balance only) or when remounted as read-write. | |
235 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
236 | * completed. | |
9c6b1c4d DS |
237 | */ |
238 | ||
67a2c45e | 239 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 240 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
241 | struct list_head *btrfs_get_fs_uuids(void) |
242 | { | |
243 | return &fs_uuids; | |
244 | } | |
8a4b83cc | 245 | |
2dfeca9b DS |
246 | /* |
247 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
248 | * @fsid: if not NULL, copy the uuid to fs_devices::fsid | |
249 | * | |
250 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
251 | * The returned struct is not linked onto any lists and can be destroyed with | |
252 | * kfree() right away. | |
253 | */ | |
254 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
2208a378 ID |
255 | { |
256 | struct btrfs_fs_devices *fs_devs; | |
257 | ||
78f2c9e6 | 258 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
259 | if (!fs_devs) |
260 | return ERR_PTR(-ENOMEM); | |
261 | ||
262 | mutex_init(&fs_devs->device_list_mutex); | |
263 | ||
264 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 265 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 | 266 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
c4babc5e | 267 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
268 | if (fsid) |
269 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 ID |
270 | |
271 | return fs_devs; | |
272 | } | |
273 | ||
a425f9d4 | 274 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf DS |
275 | { |
276 | rcu_string_free(device->name); | |
277 | bio_put(device->flush_bio); | |
278 | kfree(device); | |
279 | } | |
280 | ||
e4404d6e YZ |
281 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
282 | { | |
283 | struct btrfs_device *device; | |
284 | WARN_ON(fs_devices->opened); | |
285 | while (!list_empty(&fs_devices->devices)) { | |
286 | device = list_entry(fs_devices->devices.next, | |
287 | struct btrfs_device, dev_list); | |
288 | list_del(&device->dev_list); | |
a425f9d4 | 289 | btrfs_free_device(device); |
e4404d6e YZ |
290 | } |
291 | kfree(fs_devices); | |
292 | } | |
293 | ||
b8b8ff59 LC |
294 | static void btrfs_kobject_uevent(struct block_device *bdev, |
295 | enum kobject_action action) | |
296 | { | |
297 | int ret; | |
298 | ||
299 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
300 | if (ret) | |
efe120a0 | 301 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
302 | action, |
303 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
304 | &disk_to_dev(bdev->bd_disk)->kobj); | |
305 | } | |
306 | ||
ffc5a379 | 307 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
308 | { |
309 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 310 | |
2b82032c YZ |
311 | while (!list_empty(&fs_uuids)) { |
312 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
313 | struct btrfs_fs_devices, fs_list); |
314 | list_del(&fs_devices->fs_list); | |
e4404d6e | 315 | free_fs_devices(fs_devices); |
8a4b83cc | 316 | } |
8a4b83cc CM |
317 | } |
318 | ||
48dae9cf DS |
319 | /* |
320 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
321 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 322 | * btrfs_free_device. |
48dae9cf | 323 | */ |
12bd2fc0 ID |
324 | static struct btrfs_device *__alloc_device(void) |
325 | { | |
326 | struct btrfs_device *dev; | |
327 | ||
78f2c9e6 | 328 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
329 | if (!dev) |
330 | return ERR_PTR(-ENOMEM); | |
331 | ||
e0ae9994 DS |
332 | /* |
333 | * Preallocate a bio that's always going to be used for flushing device | |
334 | * barriers and matches the device lifespan | |
335 | */ | |
336 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
337 | if (!dev->flush_bio) { | |
338 | kfree(dev); | |
339 | return ERR_PTR(-ENOMEM); | |
340 | } | |
e0ae9994 | 341 | |
12bd2fc0 ID |
342 | INIT_LIST_HEAD(&dev->dev_list); |
343 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 344 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
345 | |
346 | spin_lock_init(&dev->io_lock); | |
347 | ||
12bd2fc0 | 348 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 349 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 350 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 351 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 352 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
353 | |
354 | return dev; | |
355 | } | |
356 | ||
35c70103 DS |
357 | /* |
358 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
359 | * return NULL. | |
360 | * | |
361 | * If devid and uuid are both specified, the match must be exact, otherwise | |
362 | * only devid is used. | |
363 | */ | |
364 | static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices, | |
365 | u64 devid, const u8 *uuid) | |
8a4b83cc CM |
366 | { |
367 | struct btrfs_device *dev; | |
8a4b83cc | 368 | |
636d2c9d | 369 | list_for_each_entry(dev, &fs_devices->devices, dev_list) { |
a443755f | 370 | if (dev->devid == devid && |
8f18cf13 | 371 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 372 | return dev; |
a443755f | 373 | } |
8a4b83cc CM |
374 | } |
375 | return NULL; | |
376 | } | |
377 | ||
a1b32a59 | 378 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 379 | { |
8a4b83cc CM |
380 | struct btrfs_fs_devices *fs_devices; |
381 | ||
c4babc5e | 382 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
8a4b83cc CM |
383 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
384 | return fs_devices; | |
385 | } | |
386 | return NULL; | |
387 | } | |
388 | ||
beaf8ab3 SB |
389 | static int |
390 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
391 | int flush, struct block_device **bdev, | |
392 | struct buffer_head **bh) | |
393 | { | |
394 | int ret; | |
395 | ||
396 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
397 | ||
398 | if (IS_ERR(*bdev)) { | |
399 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
400 | goto error; |
401 | } | |
402 | ||
403 | if (flush) | |
404 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 405 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
406 | if (ret) { |
407 | blkdev_put(*bdev, flags); | |
408 | goto error; | |
409 | } | |
410 | invalidate_bdev(*bdev); | |
411 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
412 | if (IS_ERR(*bh)) { |
413 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
414 | blkdev_put(*bdev, flags); |
415 | goto error; | |
416 | } | |
417 | ||
418 | return 0; | |
419 | ||
420 | error: | |
421 | *bdev = NULL; | |
422 | *bh = NULL; | |
423 | return ret; | |
424 | } | |
425 | ||
ffbd517d CM |
426 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
427 | struct bio *head, struct bio *tail) | |
428 | { | |
429 | ||
430 | struct bio *old_head; | |
431 | ||
432 | old_head = pending_bios->head; | |
433 | pending_bios->head = head; | |
434 | if (pending_bios->tail) | |
435 | tail->bi_next = old_head; | |
436 | else | |
437 | pending_bios->tail = tail; | |
438 | } | |
439 | ||
8b712842 CM |
440 | /* |
441 | * we try to collect pending bios for a device so we don't get a large | |
442 | * number of procs sending bios down to the same device. This greatly | |
443 | * improves the schedulers ability to collect and merge the bios. | |
444 | * | |
445 | * But, it also turns into a long list of bios to process and that is sure | |
446 | * to eventually make the worker thread block. The solution here is to | |
447 | * make some progress and then put this work struct back at the end of | |
448 | * the list if the block device is congested. This way, multiple devices | |
449 | * can make progress from a single worker thread. | |
450 | */ | |
143bede5 | 451 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 452 | { |
0b246afa | 453 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
454 | struct bio *pending; |
455 | struct backing_dev_info *bdi; | |
ffbd517d | 456 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
457 | struct bio *tail; |
458 | struct bio *cur; | |
459 | int again = 0; | |
ffbd517d | 460 | unsigned long num_run; |
d644d8a1 | 461 | unsigned long batch_run = 0; |
b765ead5 | 462 | unsigned long last_waited = 0; |
d84275c9 | 463 | int force_reg = 0; |
0e588859 | 464 | int sync_pending = 0; |
211588ad CM |
465 | struct blk_plug plug; |
466 | ||
467 | /* | |
468 | * this function runs all the bios we've collected for | |
469 | * a particular device. We don't want to wander off to | |
470 | * another device without first sending all of these down. | |
471 | * So, setup a plug here and finish it off before we return | |
472 | */ | |
473 | blk_start_plug(&plug); | |
8b712842 | 474 | |
efa7c9f9 | 475 | bdi = device->bdev->bd_bdi; |
b64a2851 | 476 | |
8b712842 CM |
477 | loop: |
478 | spin_lock(&device->io_lock); | |
479 | ||
a6837051 | 480 | loop_lock: |
d84275c9 | 481 | num_run = 0; |
ffbd517d | 482 | |
8b712842 CM |
483 | /* take all the bios off the list at once and process them |
484 | * later on (without the lock held). But, remember the | |
485 | * tail and other pointers so the bios can be properly reinserted | |
486 | * into the list if we hit congestion | |
487 | */ | |
d84275c9 | 488 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 489 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
490 | force_reg = 1; |
491 | } else { | |
ffbd517d | 492 | pending_bios = &device->pending_bios; |
d84275c9 CM |
493 | force_reg = 0; |
494 | } | |
ffbd517d CM |
495 | |
496 | pending = pending_bios->head; | |
497 | tail = pending_bios->tail; | |
8b712842 | 498 | WARN_ON(pending && !tail); |
8b712842 CM |
499 | |
500 | /* | |
501 | * if pending was null this time around, no bios need processing | |
502 | * at all and we can stop. Otherwise it'll loop back up again | |
503 | * and do an additional check so no bios are missed. | |
504 | * | |
505 | * device->running_pending is used to synchronize with the | |
506 | * schedule_bio code. | |
507 | */ | |
ffbd517d CM |
508 | if (device->pending_sync_bios.head == NULL && |
509 | device->pending_bios.head == NULL) { | |
8b712842 CM |
510 | again = 0; |
511 | device->running_pending = 0; | |
ffbd517d CM |
512 | } else { |
513 | again = 1; | |
514 | device->running_pending = 1; | |
8b712842 | 515 | } |
ffbd517d CM |
516 | |
517 | pending_bios->head = NULL; | |
518 | pending_bios->tail = NULL; | |
519 | ||
8b712842 CM |
520 | spin_unlock(&device->io_lock); |
521 | ||
d397712b | 522 | while (pending) { |
ffbd517d CM |
523 | |
524 | rmb(); | |
d84275c9 CM |
525 | /* we want to work on both lists, but do more bios on the |
526 | * sync list than the regular list | |
527 | */ | |
528 | if ((num_run > 32 && | |
529 | pending_bios != &device->pending_sync_bios && | |
530 | device->pending_sync_bios.head) || | |
531 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
532 | device->pending_bios.head)) { | |
ffbd517d CM |
533 | spin_lock(&device->io_lock); |
534 | requeue_list(pending_bios, pending, tail); | |
535 | goto loop_lock; | |
536 | } | |
537 | ||
8b712842 CM |
538 | cur = pending; |
539 | pending = pending->bi_next; | |
540 | cur->bi_next = NULL; | |
b64a2851 | 541 | |
dac56212 | 542 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 543 | |
2ab1ba68 CM |
544 | /* |
545 | * if we're doing the sync list, record that our | |
546 | * plug has some sync requests on it | |
547 | * | |
548 | * If we're doing the regular list and there are | |
549 | * sync requests sitting around, unplug before | |
550 | * we add more | |
551 | */ | |
552 | if (pending_bios == &device->pending_sync_bios) { | |
553 | sync_pending = 1; | |
554 | } else if (sync_pending) { | |
555 | blk_finish_plug(&plug); | |
556 | blk_start_plug(&plug); | |
557 | sync_pending = 0; | |
558 | } | |
559 | ||
4e49ea4a | 560 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
561 | num_run++; |
562 | batch_run++; | |
853d8ec4 DS |
563 | |
564 | cond_resched(); | |
8b712842 CM |
565 | |
566 | /* | |
567 | * we made progress, there is more work to do and the bdi | |
568 | * is now congested. Back off and let other work structs | |
569 | * run instead | |
570 | */ | |
57fd5a5f | 571 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 572 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 573 | struct io_context *ioc; |
8b712842 | 574 | |
b765ead5 CM |
575 | ioc = current->io_context; |
576 | ||
577 | /* | |
578 | * the main goal here is that we don't want to | |
579 | * block if we're going to be able to submit | |
580 | * more requests without blocking. | |
581 | * | |
582 | * This code does two great things, it pokes into | |
583 | * the elevator code from a filesystem _and_ | |
584 | * it makes assumptions about how batching works. | |
585 | */ | |
586 | if (ioc && ioc->nr_batch_requests > 0 && | |
587 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
588 | (last_waited == 0 || | |
589 | ioc->last_waited == last_waited)) { | |
590 | /* | |
591 | * we want to go through our batch of | |
592 | * requests and stop. So, we copy out | |
593 | * the ioc->last_waited time and test | |
594 | * against it before looping | |
595 | */ | |
596 | last_waited = ioc->last_waited; | |
853d8ec4 | 597 | cond_resched(); |
b765ead5 CM |
598 | continue; |
599 | } | |
8b712842 | 600 | spin_lock(&device->io_lock); |
ffbd517d | 601 | requeue_list(pending_bios, pending, tail); |
a6837051 | 602 | device->running_pending = 1; |
8b712842 CM |
603 | |
604 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
605 | btrfs_queue_work(fs_info->submit_workers, |
606 | &device->work); | |
8b712842 CM |
607 | goto done; |
608 | } | |
609 | } | |
ffbd517d | 610 | |
51684082 CM |
611 | cond_resched(); |
612 | if (again) | |
613 | goto loop; | |
614 | ||
615 | spin_lock(&device->io_lock); | |
616 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
617 | goto loop_lock; | |
618 | spin_unlock(&device->io_lock); | |
619 | ||
8b712842 | 620 | done: |
211588ad | 621 | blk_finish_plug(&plug); |
8b712842 CM |
622 | } |
623 | ||
b2950863 | 624 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
625 | { |
626 | struct btrfs_device *device; | |
627 | ||
628 | device = container_of(work, struct btrfs_device, work); | |
629 | run_scheduled_bios(device); | |
630 | } | |
631 | ||
d8367db3 AJ |
632 | /* |
633 | * Search and remove all stale (devices which are not mounted) devices. | |
634 | * When both inputs are NULL, it will search and release all stale devices. | |
635 | * path: Optional. When provided will it release all unmounted devices | |
636 | * matching this path only. | |
637 | * skip_dev: Optional. Will skip this device when searching for the stale | |
638 | * devices. | |
639 | */ | |
640 | static void btrfs_free_stale_devices(const char *path, | |
fa6d2ae5 | 641 | struct btrfs_device *skip_device) |
4fde46f0 | 642 | { |
fa6d2ae5 AJ |
643 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
644 | struct btrfs_device *device, *tmp_device; | |
4fde46f0 | 645 | |
fa6d2ae5 | 646 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
7bcb8164 AJ |
647 | mutex_lock(&fs_devices->device_list_mutex); |
648 | if (fs_devices->opened) { | |
649 | mutex_unlock(&fs_devices->device_list_mutex); | |
4fde46f0 | 650 | continue; |
7bcb8164 | 651 | } |
4fde46f0 | 652 | |
fa6d2ae5 AJ |
653 | list_for_each_entry_safe(device, tmp_device, |
654 | &fs_devices->devices, dev_list) { | |
522f1b45 | 655 | int not_found = 0; |
4fde46f0 | 656 | |
fa6d2ae5 | 657 | if (skip_device && skip_device == device) |
d8367db3 | 658 | continue; |
fa6d2ae5 | 659 | if (path && !device->name) |
4fde46f0 AJ |
660 | continue; |
661 | ||
4fde46f0 | 662 | rcu_read_lock(); |
d8367db3 | 663 | if (path) |
fa6d2ae5 | 664 | not_found = strcmp(rcu_str_deref(device->name), |
d8367db3 | 665 | path); |
4fde46f0 | 666 | rcu_read_unlock(); |
38cf665d AJ |
667 | if (not_found) |
668 | continue; | |
4fde46f0 | 669 | |
4fde46f0 | 670 | /* delete the stale device */ |
7bcb8164 AJ |
671 | fs_devices->num_devices--; |
672 | list_del(&device->dev_list); | |
673 | btrfs_free_device(device); | |
674 | ||
675 | if (fs_devices->num_devices == 0) | |
fd649f10 | 676 | break; |
7bcb8164 AJ |
677 | } |
678 | mutex_unlock(&fs_devices->device_list_mutex); | |
679 | if (fs_devices->num_devices == 0) { | |
680 | btrfs_sysfs_remove_fsid(fs_devices); | |
681 | list_del(&fs_devices->fs_list); | |
682 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
683 | } |
684 | } | |
685 | } | |
686 | ||
0fb08bcc AJ |
687 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
688 | struct btrfs_device *device, fmode_t flags, | |
689 | void *holder) | |
690 | { | |
691 | struct request_queue *q; | |
692 | struct block_device *bdev; | |
693 | struct buffer_head *bh; | |
694 | struct btrfs_super_block *disk_super; | |
695 | u64 devid; | |
696 | int ret; | |
697 | ||
698 | if (device->bdev) | |
699 | return -EINVAL; | |
700 | if (!device->name) | |
701 | return -EINVAL; | |
702 | ||
703 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
704 | &bdev, &bh); | |
705 | if (ret) | |
706 | return ret; | |
707 | ||
708 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
709 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
710 | if (devid != device->devid) | |
711 | goto error_brelse; | |
712 | ||
713 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
714 | goto error_brelse; | |
715 | ||
716 | device->generation = btrfs_super_generation(disk_super); | |
717 | ||
718 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
ebbede42 | 719 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
720 | fs_devices->seeding = 1; |
721 | } else { | |
ebbede42 AJ |
722 | if (bdev_read_only(bdev)) |
723 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
724 | else | |
725 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
726 | } |
727 | ||
728 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
729 | if (!blk_queue_nonrot(q)) |
730 | fs_devices->rotating = 1; | |
731 | ||
732 | device->bdev = bdev; | |
e12c9621 | 733 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
734 | device->mode = flags; |
735 | ||
736 | fs_devices->open_devices++; | |
ebbede42 AJ |
737 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
738 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 739 | fs_devices->rw_devices++; |
b1b8e386 | 740 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
741 | } |
742 | brelse(bh); | |
743 | ||
744 | return 0; | |
745 | ||
746 | error_brelse: | |
747 | brelse(bh); | |
748 | blkdev_put(bdev, flags); | |
749 | ||
750 | return -EINVAL; | |
751 | } | |
752 | ||
60999ca4 DS |
753 | /* |
754 | * Add new device to list of registered devices | |
755 | * | |
756 | * Returns: | |
e124ece5 AJ |
757 | * device pointer which was just added or updated when successful |
758 | * error pointer when failed | |
60999ca4 | 759 | */ |
e124ece5 | 760 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
761 | struct btrfs_super_block *disk_super, |
762 | bool *new_device_added) | |
8a4b83cc CM |
763 | { |
764 | struct btrfs_device *device; | |
765 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 766 | struct rcu_string *name; |
8a4b83cc | 767 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 768 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
8a4b83cc CM |
769 | |
770 | fs_devices = find_fsid(disk_super->fsid); | |
771 | if (!fs_devices) { | |
2208a378 ID |
772 | fs_devices = alloc_fs_devices(disk_super->fsid); |
773 | if (IS_ERR(fs_devices)) | |
e124ece5 | 774 | return ERR_CAST(fs_devices); |
2208a378 | 775 | |
9c6d173e | 776 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 777 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 778 | |
8a4b83cc CM |
779 | device = NULL; |
780 | } else { | |
9c6d173e | 781 | mutex_lock(&fs_devices->device_list_mutex); |
35c70103 DS |
782 | device = find_device(fs_devices, devid, |
783 | disk_super->dev_item.uuid); | |
8a4b83cc | 784 | } |
443f24fe | 785 | |
8a4b83cc | 786 | if (!device) { |
9c6d173e AJ |
787 | if (fs_devices->opened) { |
788 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 789 | return ERR_PTR(-EBUSY); |
9c6d173e | 790 | } |
2b82032c | 791 | |
12bd2fc0 ID |
792 | device = btrfs_alloc_device(NULL, &devid, |
793 | disk_super->dev_item.uuid); | |
794 | if (IS_ERR(device)) { | |
9c6d173e | 795 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 796 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 797 | return device; |
8a4b83cc | 798 | } |
606686ee JB |
799 | |
800 | name = rcu_string_strdup(path, GFP_NOFS); | |
801 | if (!name) { | |
a425f9d4 | 802 | btrfs_free_device(device); |
9c6d173e | 803 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 804 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 805 | } |
606686ee | 806 | rcu_assign_pointer(device->name, name); |
90519d66 | 807 | |
1f78160c | 808 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 809 | fs_devices->num_devices++; |
e5e9a520 | 810 | |
2b82032c | 811 | device->fs_devices = fs_devices; |
4306a974 | 812 | *new_device_added = true; |
327f18cc AJ |
813 | |
814 | if (disk_super->label[0]) | |
815 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
816 | disk_super->label, devid, found_transid, path); | |
817 | else | |
818 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
819 | disk_super->fsid, devid, found_transid, path); | |
820 | ||
606686ee | 821 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
822 | /* |
823 | * When FS is already mounted. | |
824 | * 1. If you are here and if the device->name is NULL that | |
825 | * means this device was missing at time of FS mount. | |
826 | * 2. If you are here and if the device->name is different | |
827 | * from 'path' that means either | |
828 | * a. The same device disappeared and reappeared with | |
829 | * different name. or | |
830 | * b. The missing-disk-which-was-replaced, has | |
831 | * reappeared now. | |
832 | * | |
833 | * We must allow 1 and 2a above. But 2b would be a spurious | |
834 | * and unintentional. | |
835 | * | |
836 | * Further in case of 1 and 2a above, the disk at 'path' | |
837 | * would have missed some transaction when it was away and | |
838 | * in case of 2a the stale bdev has to be updated as well. | |
839 | * 2b must not be allowed at all time. | |
840 | */ | |
841 | ||
842 | /* | |
0f23ae74 CM |
843 | * For now, we do allow update to btrfs_fs_device through the |
844 | * btrfs dev scan cli after FS has been mounted. We're still | |
845 | * tracking a problem where systems fail mount by subvolume id | |
846 | * when we reject replacement on a mounted FS. | |
b96de000 | 847 | */ |
0f23ae74 | 848 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
849 | /* |
850 | * That is if the FS is _not_ mounted and if you | |
851 | * are here, that means there is more than one | |
852 | * disk with same uuid and devid.We keep the one | |
853 | * with larger generation number or the last-in if | |
854 | * generation are equal. | |
855 | */ | |
9c6d173e | 856 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 857 | return ERR_PTR(-EEXIST); |
77bdae4d | 858 | } |
b96de000 | 859 | |
a9261d41 AJ |
860 | /* |
861 | * We are going to replace the device path for a given devid, | |
862 | * make sure it's the same device if the device is mounted | |
863 | */ | |
864 | if (device->bdev) { | |
865 | struct block_device *path_bdev; | |
866 | ||
867 | path_bdev = lookup_bdev(path); | |
868 | if (IS_ERR(path_bdev)) { | |
869 | mutex_unlock(&fs_devices->device_list_mutex); | |
870 | return ERR_CAST(path_bdev); | |
871 | } | |
872 | ||
873 | if (device->bdev != path_bdev) { | |
874 | bdput(path_bdev); | |
875 | mutex_unlock(&fs_devices->device_list_mutex); | |
876 | btrfs_warn_in_rcu(device->fs_info, | |
877 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
878 | disk_super->fsid, devid, | |
879 | rcu_str_deref(device->name), path); | |
880 | return ERR_PTR(-EEXIST); | |
881 | } | |
882 | bdput(path_bdev); | |
883 | btrfs_info_in_rcu(device->fs_info, | |
884 | "device fsid %pU devid %llu moved old:%s new:%s", | |
885 | disk_super->fsid, devid, | |
886 | rcu_str_deref(device->name), path); | |
887 | } | |
888 | ||
606686ee | 889 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
890 | if (!name) { |
891 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 892 | return ERR_PTR(-ENOMEM); |
9c6d173e | 893 | } |
606686ee JB |
894 | rcu_string_free(device->name); |
895 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 896 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 897 | fs_devices->missing_devices--; |
e6e674bd | 898 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 899 | } |
8a4b83cc CM |
900 | } |
901 | ||
77bdae4d AJ |
902 | /* |
903 | * Unmount does not free the btrfs_device struct but would zero | |
904 | * generation along with most of the other members. So just update | |
905 | * it back. We need it to pick the disk with largest generation | |
906 | * (as above). | |
907 | */ | |
908 | if (!fs_devices->opened) | |
909 | device->generation = found_transid; | |
910 | ||
f2788d2f AJ |
911 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
912 | ||
9c6d173e | 913 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 914 | return device; |
8a4b83cc CM |
915 | } |
916 | ||
e4404d6e YZ |
917 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
918 | { | |
919 | struct btrfs_fs_devices *fs_devices; | |
920 | struct btrfs_device *device; | |
921 | struct btrfs_device *orig_dev; | |
922 | ||
2208a378 ID |
923 | fs_devices = alloc_fs_devices(orig->fsid); |
924 | if (IS_ERR(fs_devices)) | |
925 | return fs_devices; | |
e4404d6e | 926 | |
adbbb863 | 927 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 928 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 929 | |
46224705 | 930 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 931 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
932 | struct rcu_string *name; |
933 | ||
12bd2fc0 ID |
934 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
935 | orig_dev->uuid); | |
936 | if (IS_ERR(device)) | |
e4404d6e YZ |
937 | goto error; |
938 | ||
606686ee JB |
939 | /* |
940 | * This is ok to do without rcu read locked because we hold the | |
941 | * uuid mutex so nothing we touch in here is going to disappear. | |
942 | */ | |
e755f780 | 943 | if (orig_dev->name) { |
78f2c9e6 DS |
944 | name = rcu_string_strdup(orig_dev->name->str, |
945 | GFP_KERNEL); | |
e755f780 | 946 | if (!name) { |
a425f9d4 | 947 | btrfs_free_device(device); |
e755f780 AJ |
948 | goto error; |
949 | } | |
950 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 951 | } |
e4404d6e | 952 | |
e4404d6e YZ |
953 | list_add(&device->dev_list, &fs_devices->devices); |
954 | device->fs_devices = fs_devices; | |
955 | fs_devices->num_devices++; | |
956 | } | |
adbbb863 | 957 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
958 | return fs_devices; |
959 | error: | |
adbbb863 | 960 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
961 | free_fs_devices(fs_devices); |
962 | return ERR_PTR(-ENOMEM); | |
963 | } | |
964 | ||
9b99b115 AJ |
965 | /* |
966 | * After we have read the system tree and know devids belonging to | |
967 | * this filesystem, remove the device which does not belong there. | |
968 | */ | |
969 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 970 | { |
c6e30871 | 971 | struct btrfs_device *device, *next; |
443f24fe | 972 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 973 | |
dfe25020 CM |
974 | mutex_lock(&uuid_mutex); |
975 | again: | |
46224705 | 976 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 977 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
978 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
979 | &device->dev_state)) { | |
401e29c1 AJ |
980 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
981 | &device->dev_state) && | |
982 | (!latest_dev || | |
983 | device->generation > latest_dev->generation)) { | |
443f24fe | 984 | latest_dev = device; |
a6b0d5c8 | 985 | } |
2b82032c | 986 | continue; |
a6b0d5c8 | 987 | } |
2b82032c | 988 | |
8dabb742 SB |
989 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
990 | /* | |
991 | * In the first step, keep the device which has | |
992 | * the correct fsid and the devid that is used | |
993 | * for the dev_replace procedure. | |
994 | * In the second step, the dev_replace state is | |
995 | * read from the device tree and it is known | |
996 | * whether the procedure is really active or | |
997 | * not, which means whether this device is | |
998 | * used or whether it should be removed. | |
999 | */ | |
401e29c1 AJ |
1000 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1001 | &device->dev_state)) { | |
8dabb742 SB |
1002 | continue; |
1003 | } | |
1004 | } | |
2b82032c | 1005 | if (device->bdev) { |
d4d77629 | 1006 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1007 | device->bdev = NULL; |
1008 | fs_devices->open_devices--; | |
1009 | } | |
ebbede42 | 1010 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1011 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1012 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1013 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1014 | &device->dev_state)) | |
8dabb742 | 1015 | fs_devices->rw_devices--; |
2b82032c | 1016 | } |
e4404d6e YZ |
1017 | list_del_init(&device->dev_list); |
1018 | fs_devices->num_devices--; | |
a425f9d4 | 1019 | btrfs_free_device(device); |
dfe25020 | 1020 | } |
2b82032c YZ |
1021 | |
1022 | if (fs_devices->seed) { | |
1023 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1024 | goto again; |
1025 | } | |
1026 | ||
443f24fe | 1027 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1028 | |
dfe25020 | 1029 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1030 | } |
a0af469b | 1031 | |
f06c5965 | 1032 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
1033 | { |
1034 | struct btrfs_device *device; | |
1035 | ||
9f5316c1 | 1036 | device = container_of(head, struct btrfs_device, rcu); |
a425f9d4 | 1037 | btrfs_free_device(device); |
1f78160c XG |
1038 | } |
1039 | ||
14238819 AJ |
1040 | static void btrfs_close_bdev(struct btrfs_device *device) |
1041 | { | |
08ffcae8 DS |
1042 | if (!device->bdev) |
1043 | return; | |
1044 | ||
ebbede42 | 1045 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1046 | sync_blockdev(device->bdev); |
1047 | invalidate_bdev(device->bdev); | |
1048 | } | |
1049 | ||
08ffcae8 | 1050 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1051 | } |
1052 | ||
959b1c04 | 1053 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1054 | { |
1055 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1056 | struct btrfs_device *new_device; | |
1057 | struct rcu_string *name; | |
1058 | ||
1059 | if (device->bdev) | |
1060 | fs_devices->open_devices--; | |
1061 | ||
ebbede42 | 1062 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1063 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1064 | list_del_init(&device->dev_alloc_list); | |
1065 | fs_devices->rw_devices--; | |
1066 | } | |
1067 | ||
e6e674bd | 1068 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1069 | fs_devices->missing_devices--; |
1070 | ||
959b1c04 NB |
1071 | btrfs_close_bdev(device); |
1072 | ||
f448341a AJ |
1073 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1074 | device->uuid); | |
1075 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1076 | ||
1077 | /* Safe because we are under uuid_mutex */ | |
1078 | if (device->name) { | |
1079 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1080 | BUG_ON(!name); /* -ENOMEM */ | |
1081 | rcu_assign_pointer(new_device->name, name); | |
1082 | } | |
1083 | ||
1084 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1085 | new_device->fs_devices = device->fs_devices; | |
959b1c04 NB |
1086 | |
1087 | call_rcu(&device->rcu, free_device_rcu); | |
f448341a AJ |
1088 | } |
1089 | ||
0226e0eb | 1090 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1091 | { |
2037a093 | 1092 | struct btrfs_device *device, *tmp; |
e4404d6e | 1093 | |
2b82032c YZ |
1094 | if (--fs_devices->opened > 0) |
1095 | return 0; | |
8a4b83cc | 1096 | |
c9513edb | 1097 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1098 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1099 | btrfs_close_one_device(device); |
8a4b83cc | 1100 | } |
c9513edb XG |
1101 | mutex_unlock(&fs_devices->device_list_mutex); |
1102 | ||
e4404d6e YZ |
1103 | WARN_ON(fs_devices->open_devices); |
1104 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1105 | fs_devices->opened = 0; |
1106 | fs_devices->seeding = 0; | |
2b82032c | 1107 | |
8a4b83cc CM |
1108 | return 0; |
1109 | } | |
1110 | ||
2b82032c YZ |
1111 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1112 | { | |
e4404d6e | 1113 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1114 | int ret; |
1115 | ||
1116 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1117 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1118 | if (!fs_devices->opened) { |
1119 | seed_devices = fs_devices->seed; | |
1120 | fs_devices->seed = NULL; | |
1121 | } | |
2b82032c | 1122 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1123 | |
1124 | while (seed_devices) { | |
1125 | fs_devices = seed_devices; | |
1126 | seed_devices = fs_devices->seed; | |
0226e0eb | 1127 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1128 | free_fs_devices(fs_devices); |
1129 | } | |
2b82032c YZ |
1130 | return ret; |
1131 | } | |
1132 | ||
897fb573 | 1133 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1134 | fmode_t flags, void *holder) |
8a4b83cc | 1135 | { |
8a4b83cc | 1136 | struct btrfs_device *device; |
443f24fe | 1137 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1138 | int ret = 0; |
8a4b83cc | 1139 | |
d4d77629 TH |
1140 | flags |= FMODE_EXCL; |
1141 | ||
f117e290 | 1142 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1143 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1144 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1145 | continue; |
a0af469b | 1146 | |
9f050db4 AJ |
1147 | if (!latest_dev || |
1148 | device->generation > latest_dev->generation) | |
1149 | latest_dev = device; | |
8a4b83cc | 1150 | } |
a0af469b | 1151 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1152 | ret = -EINVAL; |
a0af469b CM |
1153 | goto out; |
1154 | } | |
2b82032c | 1155 | fs_devices->opened = 1; |
443f24fe | 1156 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1157 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1158 | out: |
2b82032c YZ |
1159 | return ret; |
1160 | } | |
1161 | ||
f8e10cd3 AJ |
1162 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1163 | { | |
1164 | struct btrfs_device *dev1, *dev2; | |
1165 | ||
1166 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1167 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1168 | ||
1169 | if (dev1->devid < dev2->devid) | |
1170 | return -1; | |
1171 | else if (dev1->devid > dev2->devid) | |
1172 | return 1; | |
1173 | return 0; | |
1174 | } | |
1175 | ||
2b82032c | 1176 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1177 | fmode_t flags, void *holder) |
2b82032c YZ |
1178 | { |
1179 | int ret; | |
1180 | ||
f5194e34 DS |
1181 | lockdep_assert_held(&uuid_mutex); |
1182 | ||
542c5908 | 1183 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1184 | if (fs_devices->opened) { |
e4404d6e YZ |
1185 | fs_devices->opened++; |
1186 | ret = 0; | |
2b82032c | 1187 | } else { |
f8e10cd3 | 1188 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1189 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1190 | } |
542c5908 AJ |
1191 | mutex_unlock(&fs_devices->device_list_mutex); |
1192 | ||
8a4b83cc CM |
1193 | return ret; |
1194 | } | |
1195 | ||
c9162bdf | 1196 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1197 | { |
1198 | kunmap(page); | |
1199 | put_page(page); | |
1200 | } | |
1201 | ||
c9162bdf OS |
1202 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1203 | struct page **page, | |
1204 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1205 | { |
1206 | void *p; | |
1207 | pgoff_t index; | |
1208 | ||
1209 | /* make sure our super fits in the device */ | |
1210 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1211 | return 1; | |
1212 | ||
1213 | /* make sure our super fits in the page */ | |
1214 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1215 | return 1; | |
1216 | ||
1217 | /* make sure our super doesn't straddle pages on disk */ | |
1218 | index = bytenr >> PAGE_SHIFT; | |
1219 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1220 | return 1; | |
1221 | ||
1222 | /* pull in the page with our super */ | |
1223 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1224 | index, GFP_KERNEL); | |
1225 | ||
1226 | if (IS_ERR_OR_NULL(*page)) | |
1227 | return 1; | |
1228 | ||
1229 | p = kmap(*page); | |
1230 | ||
1231 | /* align our pointer to the offset of the super block */ | |
1232 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1233 | ||
1234 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1235 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1236 | btrfs_release_disk_super(*page); | |
1237 | return 1; | |
1238 | } | |
1239 | ||
1240 | if ((*disk_super)->label[0] && | |
1241 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1242 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1243 | ||
1244 | return 0; | |
1245 | } | |
1246 | ||
6f60cbd3 DS |
1247 | /* |
1248 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1249 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1250 | * is read via pagecache | |
1251 | */ | |
36350e95 GJ |
1252 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1253 | void *holder) | |
8a4b83cc CM |
1254 | { |
1255 | struct btrfs_super_block *disk_super; | |
4306a974 | 1256 | bool new_device_added = false; |
36350e95 | 1257 | struct btrfs_device *device = NULL; |
8a4b83cc | 1258 | struct block_device *bdev; |
6f60cbd3 | 1259 | struct page *page; |
6f60cbd3 | 1260 | u64 bytenr; |
8a4b83cc | 1261 | |
899f9307 DS |
1262 | lockdep_assert_held(&uuid_mutex); |
1263 | ||
6f60cbd3 DS |
1264 | /* |
1265 | * we would like to check all the supers, but that would make | |
1266 | * a btrfs mount succeed after a mkfs from a different FS. | |
1267 | * So, we need to add a special mount option to scan for | |
1268 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1269 | */ | |
1270 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1271 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1272 | |
1273 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1274 | if (IS_ERR(bdev)) |
36350e95 | 1275 | return ERR_CAST(bdev); |
6f60cbd3 | 1276 | |
05a5c55d | 1277 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1278 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1279 | goto error_bdev_put; |
05a5c55d | 1280 | } |
6f60cbd3 | 1281 | |
4306a974 | 1282 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1283 | if (!IS_ERR(device)) { |
4306a974 AJ |
1284 | if (new_device_added) |
1285 | btrfs_free_stale_devices(path, device); | |
1286 | } | |
6f60cbd3 | 1287 | |
6cf86a00 | 1288 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1289 | |
1290 | error_bdev_put: | |
d4d77629 | 1291 | blkdev_put(bdev, flags); |
b6ed73bc | 1292 | |
36350e95 | 1293 | return device; |
8a4b83cc | 1294 | } |
0b86a832 | 1295 | |
499f377f | 1296 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1297 | struct btrfs_device *device, |
1298 | u64 *start, u64 len) | |
1299 | { | |
fb456252 | 1300 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1301 | struct extent_map *em; |
499f377f | 1302 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1303 | int ret = 0; |
1b984508 | 1304 | u64 physical_start = *start; |
6df9a95e | 1305 | |
499f377f JM |
1306 | if (transaction) |
1307 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1308 | again: |
1309 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1310 | struct map_lookup *map; |
1311 | int i; | |
1312 | ||
95617d69 | 1313 | map = em->map_lookup; |
6df9a95e | 1314 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1315 | u64 end; |
1316 | ||
6df9a95e JB |
1317 | if (map->stripes[i].dev != device) |
1318 | continue; | |
1b984508 | 1319 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1320 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1321 | physical_start) |
6df9a95e | 1322 | continue; |
c152b63e FM |
1323 | /* |
1324 | * Make sure that while processing the pinned list we do | |
1325 | * not override our *start with a lower value, because | |
1326 | * we can have pinned chunks that fall within this | |
1327 | * device hole and that have lower physical addresses | |
1328 | * than the pending chunks we processed before. If we | |
1329 | * do not take this special care we can end up getting | |
1330 | * 2 pending chunks that start at the same physical | |
1331 | * device offsets because the end offset of a pinned | |
1332 | * chunk can be equal to the start offset of some | |
1333 | * pending chunk. | |
1334 | */ | |
1335 | end = map->stripes[i].physical + em->orig_block_len; | |
1336 | if (end > *start) { | |
1337 | *start = end; | |
1338 | ret = 1; | |
1339 | } | |
6df9a95e JB |
1340 | } |
1341 | } | |
499f377f JM |
1342 | if (search_list != &fs_info->pinned_chunks) { |
1343 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1344 | goto again; |
1345 | } | |
6df9a95e JB |
1346 | |
1347 | return ret; | |
1348 | } | |
1349 | ||
1350 | ||
0b86a832 | 1351 | /* |
499f377f JM |
1352 | * find_free_dev_extent_start - find free space in the specified device |
1353 | * @device: the device which we search the free space in | |
1354 | * @num_bytes: the size of the free space that we need | |
1355 | * @search_start: the position from which to begin the search | |
1356 | * @start: store the start of the free space. | |
1357 | * @len: the size of the free space. that we find, or the size | |
1358 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1359 | * |
0b86a832 CM |
1360 | * this uses a pretty simple search, the expectation is that it is |
1361 | * called very infrequently and that a given device has a small number | |
1362 | * of extents | |
7bfc837d MX |
1363 | * |
1364 | * @start is used to store the start of the free space if we find. But if we | |
1365 | * don't find suitable free space, it will be used to store the start position | |
1366 | * of the max free space. | |
1367 | * | |
1368 | * @len is used to store the size of the free space that we find. | |
1369 | * But if we don't find suitable free space, it is used to store the size of | |
1370 | * the max free space. | |
0b86a832 | 1371 | */ |
499f377f JM |
1372 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1373 | struct btrfs_device *device, u64 num_bytes, | |
1374 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1375 | { |
0b246afa JM |
1376 | struct btrfs_fs_info *fs_info = device->fs_info; |
1377 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1378 | struct btrfs_key key; |
7bfc837d | 1379 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1380 | struct btrfs_path *path; |
7bfc837d MX |
1381 | u64 hole_size; |
1382 | u64 max_hole_start; | |
1383 | u64 max_hole_size; | |
1384 | u64 extent_end; | |
0b86a832 CM |
1385 | u64 search_end = device->total_bytes; |
1386 | int ret; | |
7bfc837d | 1387 | int slot; |
0b86a832 | 1388 | struct extent_buffer *l; |
8cdc7c5b FM |
1389 | |
1390 | /* | |
1391 | * We don't want to overwrite the superblock on the drive nor any area | |
1392 | * used by the boot loader (grub for example), so we make sure to start | |
1393 | * at an offset of at least 1MB. | |
1394 | */ | |
0d0c71b3 | 1395 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1396 | |
6df9a95e JB |
1397 | path = btrfs_alloc_path(); |
1398 | if (!path) | |
1399 | return -ENOMEM; | |
f2ab7618 | 1400 | |
7bfc837d MX |
1401 | max_hole_start = search_start; |
1402 | max_hole_size = 0; | |
1403 | ||
f2ab7618 | 1404 | again: |
401e29c1 AJ |
1405 | if (search_start >= search_end || |
1406 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1407 | ret = -ENOSPC; |
6df9a95e | 1408 | goto out; |
7bfc837d MX |
1409 | } |
1410 | ||
e4058b54 | 1411 | path->reada = READA_FORWARD; |
6df9a95e JB |
1412 | path->search_commit_root = 1; |
1413 | path->skip_locking = 1; | |
7bfc837d | 1414 | |
0b86a832 CM |
1415 | key.objectid = device->devid; |
1416 | key.offset = search_start; | |
1417 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1418 | |
125ccb0a | 1419 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1420 | if (ret < 0) |
7bfc837d | 1421 | goto out; |
1fcbac58 YZ |
1422 | if (ret > 0) { |
1423 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1424 | if (ret < 0) | |
7bfc837d | 1425 | goto out; |
1fcbac58 | 1426 | } |
7bfc837d | 1427 | |
0b86a832 CM |
1428 | while (1) { |
1429 | l = path->nodes[0]; | |
1430 | slot = path->slots[0]; | |
1431 | if (slot >= btrfs_header_nritems(l)) { | |
1432 | ret = btrfs_next_leaf(root, path); | |
1433 | if (ret == 0) | |
1434 | continue; | |
1435 | if (ret < 0) | |
7bfc837d MX |
1436 | goto out; |
1437 | ||
1438 | break; | |
0b86a832 CM |
1439 | } |
1440 | btrfs_item_key_to_cpu(l, &key, slot); | |
1441 | ||
1442 | if (key.objectid < device->devid) | |
1443 | goto next; | |
1444 | ||
1445 | if (key.objectid > device->devid) | |
7bfc837d | 1446 | break; |
0b86a832 | 1447 | |
962a298f | 1448 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1449 | goto next; |
9779b72f | 1450 | |
7bfc837d MX |
1451 | if (key.offset > search_start) { |
1452 | hole_size = key.offset - search_start; | |
9779b72f | 1453 | |
6df9a95e JB |
1454 | /* |
1455 | * Have to check before we set max_hole_start, otherwise | |
1456 | * we could end up sending back this offset anyway. | |
1457 | */ | |
499f377f | 1458 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1459 | &search_start, |
1b984508 FL |
1460 | hole_size)) { |
1461 | if (key.offset >= search_start) { | |
1462 | hole_size = key.offset - search_start; | |
1463 | } else { | |
1464 | WARN_ON_ONCE(1); | |
1465 | hole_size = 0; | |
1466 | } | |
1467 | } | |
6df9a95e | 1468 | |
7bfc837d MX |
1469 | if (hole_size > max_hole_size) { |
1470 | max_hole_start = search_start; | |
1471 | max_hole_size = hole_size; | |
1472 | } | |
9779b72f | 1473 | |
7bfc837d MX |
1474 | /* |
1475 | * If this free space is greater than which we need, | |
1476 | * it must be the max free space that we have found | |
1477 | * until now, so max_hole_start must point to the start | |
1478 | * of this free space and the length of this free space | |
1479 | * is stored in max_hole_size. Thus, we return | |
1480 | * max_hole_start and max_hole_size and go back to the | |
1481 | * caller. | |
1482 | */ | |
1483 | if (hole_size >= num_bytes) { | |
1484 | ret = 0; | |
1485 | goto out; | |
0b86a832 CM |
1486 | } |
1487 | } | |
0b86a832 | 1488 | |
0b86a832 | 1489 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1490 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1491 | dev_extent); | |
1492 | if (extent_end > search_start) | |
1493 | search_start = extent_end; | |
0b86a832 CM |
1494 | next: |
1495 | path->slots[0]++; | |
1496 | cond_resched(); | |
1497 | } | |
0b86a832 | 1498 | |
38c01b96 | 1499 | /* |
1500 | * At this point, search_start should be the end of | |
1501 | * allocated dev extents, and when shrinking the device, | |
1502 | * search_end may be smaller than search_start. | |
1503 | */ | |
f2ab7618 | 1504 | if (search_end > search_start) { |
38c01b96 | 1505 | hole_size = search_end - search_start; |
1506 | ||
499f377f | 1507 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1508 | hole_size)) { |
1509 | btrfs_release_path(path); | |
1510 | goto again; | |
1511 | } | |
0b86a832 | 1512 | |
f2ab7618 ZL |
1513 | if (hole_size > max_hole_size) { |
1514 | max_hole_start = search_start; | |
1515 | max_hole_size = hole_size; | |
1516 | } | |
6df9a95e JB |
1517 | } |
1518 | ||
7bfc837d | 1519 | /* See above. */ |
f2ab7618 | 1520 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1521 | ret = -ENOSPC; |
1522 | else | |
1523 | ret = 0; | |
1524 | ||
1525 | out: | |
2b82032c | 1526 | btrfs_free_path(path); |
7bfc837d | 1527 | *start = max_hole_start; |
b2117a39 | 1528 | if (len) |
7bfc837d | 1529 | *len = max_hole_size; |
0b86a832 CM |
1530 | return ret; |
1531 | } | |
1532 | ||
499f377f JM |
1533 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1534 | struct btrfs_device *device, u64 num_bytes, | |
1535 | u64 *start, u64 *len) | |
1536 | { | |
499f377f | 1537 | /* FIXME use last free of some kind */ |
499f377f | 1538 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1539 | num_bytes, 0, start, len); |
499f377f JM |
1540 | } |
1541 | ||
b2950863 | 1542 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1543 | struct btrfs_device *device, |
2196d6e8 | 1544 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1545 | { |
0b246afa JM |
1546 | struct btrfs_fs_info *fs_info = device->fs_info; |
1547 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1548 | int ret; |
1549 | struct btrfs_path *path; | |
8f18cf13 | 1550 | struct btrfs_key key; |
a061fc8d CM |
1551 | struct btrfs_key found_key; |
1552 | struct extent_buffer *leaf = NULL; | |
1553 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1554 | |
1555 | path = btrfs_alloc_path(); | |
1556 | if (!path) | |
1557 | return -ENOMEM; | |
1558 | ||
1559 | key.objectid = device->devid; | |
1560 | key.offset = start; | |
1561 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1562 | again: |
8f18cf13 | 1563 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1564 | if (ret > 0) { |
1565 | ret = btrfs_previous_item(root, path, key.objectid, | |
1566 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1567 | if (ret) |
1568 | goto out; | |
a061fc8d CM |
1569 | leaf = path->nodes[0]; |
1570 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1571 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1572 | struct btrfs_dev_extent); | |
1573 | BUG_ON(found_key.offset > start || found_key.offset + | |
1574 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1575 | key = found_key; |
1576 | btrfs_release_path(path); | |
1577 | goto again; | |
a061fc8d CM |
1578 | } else if (ret == 0) { |
1579 | leaf = path->nodes[0]; | |
1580 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1581 | struct btrfs_dev_extent); | |
79787eaa | 1582 | } else { |
0b246afa | 1583 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1584 | goto out; |
a061fc8d | 1585 | } |
8f18cf13 | 1586 | |
2196d6e8 MX |
1587 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1588 | ||
8f18cf13 | 1589 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1590 | if (ret) { |
0b246afa JM |
1591 | btrfs_handle_fs_error(fs_info, ret, |
1592 | "Failed to remove dev extent item"); | |
13212b54 | 1593 | } else { |
3204d33c | 1594 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1595 | } |
b0b802d7 | 1596 | out: |
8f18cf13 CM |
1597 | btrfs_free_path(path); |
1598 | return ret; | |
1599 | } | |
1600 | ||
48a3b636 ES |
1601 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1602 | struct btrfs_device *device, | |
48a3b636 | 1603 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1604 | { |
1605 | int ret; | |
1606 | struct btrfs_path *path; | |
0b246afa JM |
1607 | struct btrfs_fs_info *fs_info = device->fs_info; |
1608 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1609 | struct btrfs_dev_extent *extent; |
1610 | struct extent_buffer *leaf; | |
1611 | struct btrfs_key key; | |
1612 | ||
e12c9621 | 1613 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1614 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1615 | path = btrfs_alloc_path(); |
1616 | if (!path) | |
1617 | return -ENOMEM; | |
1618 | ||
0b86a832 | 1619 | key.objectid = device->devid; |
2b82032c | 1620 | key.offset = start; |
0b86a832 CM |
1621 | key.type = BTRFS_DEV_EXTENT_KEY; |
1622 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1623 | sizeof(*extent)); | |
2cdcecbc MF |
1624 | if (ret) |
1625 | goto out; | |
0b86a832 CM |
1626 | |
1627 | leaf = path->nodes[0]; | |
1628 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1629 | struct btrfs_dev_extent); | |
b5d9071c NB |
1630 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1631 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1632 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1633 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1634 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1635 | ||
0b86a832 CM |
1636 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1637 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1638 | out: |
0b86a832 CM |
1639 | btrfs_free_path(path); |
1640 | return ret; | |
1641 | } | |
1642 | ||
6df9a95e | 1643 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1644 | { |
6df9a95e JB |
1645 | struct extent_map_tree *em_tree; |
1646 | struct extent_map *em; | |
1647 | struct rb_node *n; | |
1648 | u64 ret = 0; | |
0b86a832 | 1649 | |
6df9a95e JB |
1650 | em_tree = &fs_info->mapping_tree.map_tree; |
1651 | read_lock(&em_tree->lock); | |
07e1ce09 | 1652 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1653 | if (n) { |
1654 | em = rb_entry(n, struct extent_map, rb_node); | |
1655 | ret = em->start + em->len; | |
0b86a832 | 1656 | } |
6df9a95e JB |
1657 | read_unlock(&em_tree->lock); |
1658 | ||
0b86a832 CM |
1659 | return ret; |
1660 | } | |
1661 | ||
53f10659 ID |
1662 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1663 | u64 *devid_ret) | |
0b86a832 CM |
1664 | { |
1665 | int ret; | |
1666 | struct btrfs_key key; | |
1667 | struct btrfs_key found_key; | |
2b82032c YZ |
1668 | struct btrfs_path *path; |
1669 | ||
2b82032c YZ |
1670 | path = btrfs_alloc_path(); |
1671 | if (!path) | |
1672 | return -ENOMEM; | |
0b86a832 CM |
1673 | |
1674 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1675 | key.type = BTRFS_DEV_ITEM_KEY; | |
1676 | key.offset = (u64)-1; | |
1677 | ||
53f10659 | 1678 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1679 | if (ret < 0) |
1680 | goto error; | |
1681 | ||
79787eaa | 1682 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1683 | |
53f10659 ID |
1684 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1685 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1686 | BTRFS_DEV_ITEM_KEY); |
1687 | if (ret) { | |
53f10659 | 1688 | *devid_ret = 1; |
0b86a832 CM |
1689 | } else { |
1690 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1691 | path->slots[0]); | |
53f10659 | 1692 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1693 | } |
1694 | ret = 0; | |
1695 | error: | |
2b82032c | 1696 | btrfs_free_path(path); |
0b86a832 CM |
1697 | return ret; |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * the device information is stored in the chunk root | |
1702 | * the btrfs_device struct should be fully filled in | |
1703 | */ | |
c74a0b02 | 1704 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1705 | struct btrfs_device *device) |
0b86a832 CM |
1706 | { |
1707 | int ret; | |
1708 | struct btrfs_path *path; | |
1709 | struct btrfs_dev_item *dev_item; | |
1710 | struct extent_buffer *leaf; | |
1711 | struct btrfs_key key; | |
1712 | unsigned long ptr; | |
0b86a832 | 1713 | |
0b86a832 CM |
1714 | path = btrfs_alloc_path(); |
1715 | if (!path) | |
1716 | return -ENOMEM; | |
1717 | ||
0b86a832 CM |
1718 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1719 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1720 | key.offset = device->devid; |
0b86a832 | 1721 | |
8e87e856 NB |
1722 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1723 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1724 | if (ret) |
1725 | goto out; | |
1726 | ||
1727 | leaf = path->nodes[0]; | |
1728 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1729 | ||
1730 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1731 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1732 | btrfs_set_device_type(leaf, dev_item, device->type); |
1733 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1734 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1735 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1736 | btrfs_set_device_total_bytes(leaf, dev_item, |
1737 | btrfs_device_get_disk_total_bytes(device)); | |
1738 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1739 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1740 | btrfs_set_device_group(leaf, dev_item, 0); |
1741 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1742 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1743 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1744 | |
410ba3a2 | 1745 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1746 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1747 | ptr = btrfs_device_fsid(dev_item); |
8e87e856 | 1748 | write_extent_buffer(leaf, trans->fs_info->fsid, ptr, BTRFS_FSID_SIZE); |
0b86a832 | 1749 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1750 | |
2b82032c | 1751 | ret = 0; |
0b86a832 CM |
1752 | out: |
1753 | btrfs_free_path(path); | |
1754 | return ret; | |
1755 | } | |
8f18cf13 | 1756 | |
5a1972bd QW |
1757 | /* |
1758 | * Function to update ctime/mtime for a given device path. | |
1759 | * Mainly used for ctime/mtime based probe like libblkid. | |
1760 | */ | |
da353f6b | 1761 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1762 | { |
1763 | struct file *filp; | |
1764 | ||
1765 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1766 | if (IS_ERR(filp)) |
5a1972bd QW |
1767 | return; |
1768 | file_update_time(filp); | |
1769 | filp_close(filp, NULL); | |
5a1972bd QW |
1770 | } |
1771 | ||
5b4aacef | 1772 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1773 | struct btrfs_device *device) |
1774 | { | |
5b4aacef | 1775 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1776 | int ret; |
1777 | struct btrfs_path *path; | |
a061fc8d | 1778 | struct btrfs_key key; |
a061fc8d CM |
1779 | struct btrfs_trans_handle *trans; |
1780 | ||
a061fc8d CM |
1781 | path = btrfs_alloc_path(); |
1782 | if (!path) | |
1783 | return -ENOMEM; | |
1784 | ||
a22285a6 | 1785 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1786 | if (IS_ERR(trans)) { |
1787 | btrfs_free_path(path); | |
1788 | return PTR_ERR(trans); | |
1789 | } | |
a061fc8d CM |
1790 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1791 | key.type = BTRFS_DEV_ITEM_KEY; | |
1792 | key.offset = device->devid; | |
1793 | ||
1794 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1795 | if (ret) { |
1796 | if (ret > 0) | |
1797 | ret = -ENOENT; | |
1798 | btrfs_abort_transaction(trans, ret); | |
1799 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1800 | goto out; |
1801 | } | |
1802 | ||
1803 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1804 | if (ret) { |
1805 | btrfs_abort_transaction(trans, ret); | |
1806 | btrfs_end_transaction(trans); | |
1807 | } | |
1808 | ||
a061fc8d CM |
1809 | out: |
1810 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1811 | if (!ret) |
1812 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1813 | return ret; |
1814 | } | |
1815 | ||
3cc31a0d DS |
1816 | /* |
1817 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1818 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1819 | * replace. | |
1820 | */ | |
1821 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1822 | u64 num_devices) | |
a061fc8d | 1823 | { |
a061fc8d | 1824 | u64 all_avail; |
de98ced9 | 1825 | unsigned seq; |
418775a2 | 1826 | int i; |
a061fc8d | 1827 | |
de98ced9 | 1828 | do { |
bd45ffbc | 1829 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1830 | |
bd45ffbc AJ |
1831 | all_avail = fs_info->avail_data_alloc_bits | |
1832 | fs_info->avail_system_alloc_bits | | |
1833 | fs_info->avail_metadata_alloc_bits; | |
1834 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1835 | |
418775a2 | 1836 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1837 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1838 | continue; |
a061fc8d | 1839 | |
418775a2 | 1840 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 1841 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 1842 | |
418775a2 DS |
1843 | if (ret) |
1844 | return ret; | |
1845 | } | |
53b381b3 DW |
1846 | } |
1847 | ||
bd45ffbc | 1848 | return 0; |
f1fa7f26 AJ |
1849 | } |
1850 | ||
c9162bdf OS |
1851 | static struct btrfs_device * btrfs_find_next_active_device( |
1852 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1853 | { |
2b82032c | 1854 | struct btrfs_device *next_device; |
88acff64 AJ |
1855 | |
1856 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1857 | if (next_device != device && | |
e6e674bd AJ |
1858 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1859 | && next_device->bdev) | |
88acff64 AJ |
1860 | return next_device; |
1861 | } | |
1862 | ||
1863 | return NULL; | |
1864 | } | |
1865 | ||
1866 | /* | |
1867 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1868 | * and replace it with the provided or the next active device, in the context | |
1869 | * where this function called, there should be always be another device (or | |
1870 | * this_dev) which is active. | |
1871 | */ | |
d6507cf1 NB |
1872 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
1873 | struct btrfs_device *this_dev) | |
88acff64 | 1874 | { |
d6507cf1 | 1875 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
1876 | struct btrfs_device *next_device; |
1877 | ||
1878 | if (this_dev) | |
1879 | next_device = this_dev; | |
1880 | else | |
1881 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1882 | device); | |
1883 | ASSERT(next_device); | |
1884 | ||
1885 | if (fs_info->sb->s_bdev && | |
1886 | (fs_info->sb->s_bdev == device->bdev)) | |
1887 | fs_info->sb->s_bdev = next_device->bdev; | |
1888 | ||
1889 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1890 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1891 | } | |
1892 | ||
1da73967 AJ |
1893 | /* |
1894 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
1895 | * currently replaced. | |
1896 | */ | |
1897 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
1898 | { | |
1899 | u64 num_devices = fs_info->fs_devices->num_devices; | |
1900 | ||
1901 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); | |
1902 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
1903 | ASSERT(num_devices > 1); | |
1904 | num_devices--; | |
1905 | } | |
1906 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); | |
1907 | ||
1908 | return num_devices; | |
1909 | } | |
1910 | ||
da353f6b DS |
1911 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1912 | u64 devid) | |
f1fa7f26 AJ |
1913 | { |
1914 | struct btrfs_device *device; | |
1f78160c | 1915 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 1916 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 1917 | u64 num_devices; |
a061fc8d CM |
1918 | int ret = 0; |
1919 | ||
a061fc8d CM |
1920 | mutex_lock(&uuid_mutex); |
1921 | ||
1da73967 | 1922 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 1923 | |
0b246afa | 1924 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1925 | if (ret) |
a061fc8d | 1926 | goto out; |
a061fc8d | 1927 | |
a27a94c2 NB |
1928 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
1929 | ||
1930 | if (IS_ERR(device)) { | |
1931 | if (PTR_ERR(device) == -ENOENT && | |
1932 | strcmp(device_path, "missing") == 0) | |
1933 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
1934 | else | |
1935 | ret = PTR_ERR(device); | |
53b381b3 | 1936 | goto out; |
a27a94c2 | 1937 | } |
dfe25020 | 1938 | |
eede2bf3 OS |
1939 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
1940 | btrfs_warn_in_rcu(fs_info, | |
1941 | "cannot remove device %s (devid %llu) due to active swapfile", | |
1942 | rcu_str_deref(device->name), device->devid); | |
1943 | ret = -ETXTBSY; | |
1944 | goto out; | |
1945 | } | |
1946 | ||
401e29c1 | 1947 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1948 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1949 | goto out; |
63a212ab SB |
1950 | } |
1951 | ||
ebbede42 AJ |
1952 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1953 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1954 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1955 | goto out; |
2b82032c YZ |
1956 | } |
1957 | ||
ebbede42 | 1958 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1959 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1960 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1961 | device->fs_devices->rw_devices--; |
34441361 | 1962 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 1963 | } |
a061fc8d | 1964 | |
d7901554 | 1965 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1966 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1967 | mutex_lock(&uuid_mutex); |
a061fc8d | 1968 | if (ret) |
9b3517e9 | 1969 | goto error_undo; |
a061fc8d | 1970 | |
63a212ab SB |
1971 | /* |
1972 | * TODO: the superblock still includes this device in its num_devices | |
1973 | * counter although write_all_supers() is not locked out. This | |
1974 | * could give a filesystem state which requires a degraded mount. | |
1975 | */ | |
0b246afa | 1976 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1977 | if (ret) |
9b3517e9 | 1978 | goto error_undo; |
a061fc8d | 1979 | |
e12c9621 | 1980 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 1981 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1982 | |
1983 | /* | |
1984 | * the device list mutex makes sure that we don't change | |
1985 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1986 | * the device supers. Whoever is writing all supers, should |
1987 | * lock the device list mutex before getting the number of | |
1988 | * devices in the super block (super_copy). Conversely, | |
1989 | * whoever updates the number of devices in the super block | |
1990 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1991 | */ |
1f78160c | 1992 | |
41a52a0f AJ |
1993 | /* |
1994 | * In normal cases the cur_devices == fs_devices. But in case | |
1995 | * of deleting a seed device, the cur_devices should point to | |
1996 | * its own fs_devices listed under the fs_devices->seed. | |
1997 | */ | |
1f78160c | 1998 | cur_devices = device->fs_devices; |
b5185197 | 1999 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2000 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2001 | |
41a52a0f AJ |
2002 | cur_devices->num_devices--; |
2003 | cur_devices->total_devices--; | |
b4993e64 AJ |
2004 | /* Update total_devices of the parent fs_devices if it's seed */ |
2005 | if (cur_devices != fs_devices) | |
2006 | fs_devices->total_devices--; | |
2b82032c | 2007 | |
e6e674bd | 2008 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2009 | cur_devices->missing_devices--; |
cd02dca5 | 2010 | |
d6507cf1 | 2011 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2012 | |
0bfaa9c5 | 2013 | if (device->bdev) { |
41a52a0f | 2014 | cur_devices->open_devices--; |
0bfaa9c5 | 2015 | /* remove sysfs entry */ |
b5185197 | 2016 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2017 | } |
99994cde | 2018 | |
0b246afa JM |
2019 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2020 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2021 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2022 | |
cea67ab9 JM |
2023 | /* |
2024 | * at this point, the device is zero sized and detached from | |
2025 | * the devices list. All that's left is to zero out the old | |
2026 | * supers and free the device. | |
2027 | */ | |
ebbede42 | 2028 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2029 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2030 | ||
2031 | btrfs_close_bdev(device); | |
f06c5965 | 2032 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2033 | |
1f78160c | 2034 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2035 | while (fs_devices) { |
8321cf25 RS |
2036 | if (fs_devices->seed == cur_devices) { |
2037 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2038 | break; |
8321cf25 | 2039 | } |
e4404d6e | 2040 | fs_devices = fs_devices->seed; |
2b82032c | 2041 | } |
1f78160c | 2042 | cur_devices->seed = NULL; |
0226e0eb | 2043 | close_fs_devices(cur_devices); |
1f78160c | 2044 | free_fs_devices(cur_devices); |
2b82032c YZ |
2045 | } |
2046 | ||
a061fc8d CM |
2047 | out: |
2048 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2049 | return ret; |
24fc572f | 2050 | |
9b3517e9 | 2051 | error_undo: |
ebbede42 | 2052 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2053 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2054 | list_add(&device->dev_alloc_list, |
b5185197 | 2055 | &fs_devices->alloc_list); |
c3929c36 | 2056 | device->fs_devices->rw_devices++; |
34441361 | 2057 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2058 | } |
24fc572f | 2059 | goto out; |
a061fc8d CM |
2060 | } |
2061 | ||
68a9db5f | 2062 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2063 | { |
d51908ce AJ |
2064 | struct btrfs_fs_devices *fs_devices; |
2065 | ||
68a9db5f | 2066 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2067 | |
25e8e911 AJ |
2068 | /* |
2069 | * in case of fs with no seed, srcdev->fs_devices will point | |
2070 | * to fs_devices of fs_info. However when the dev being replaced is | |
2071 | * a seed dev it will point to the seed's local fs_devices. In short | |
2072 | * srcdev will have its correct fs_devices in both the cases. | |
2073 | */ | |
2074 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2075 | |
e93c89c1 | 2076 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2077 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2078 | fs_devices->num_devices--; |
e6e674bd | 2079 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2080 | fs_devices->missing_devices--; |
e93c89c1 | 2081 | |
ebbede42 | 2082 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2083 | fs_devices->rw_devices--; |
1357272f | 2084 | |
82372bc8 | 2085 | if (srcdev->bdev) |
d51908ce | 2086 | fs_devices->open_devices--; |
084b6e7c QW |
2087 | } |
2088 | ||
2089 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2090 | struct btrfs_device *srcdev) | |
2091 | { | |
2092 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2093 | |
ebbede42 | 2094 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2095 | /* zero out the old super if it is writable */ |
2096 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2097 | } | |
14238819 AJ |
2098 | |
2099 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2100 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2101 | |
94d5f0c2 AJ |
2102 | /* if this is no devs we rather delete the fs_devices */ |
2103 | if (!fs_devices->num_devices) { | |
2104 | struct btrfs_fs_devices *tmp_fs_devices; | |
2105 | ||
6dd38f81 AJ |
2106 | /* |
2107 | * On a mounted FS, num_devices can't be zero unless it's a | |
2108 | * seed. In case of a seed device being replaced, the replace | |
2109 | * target added to the sprout FS, so there will be no more | |
2110 | * device left under the seed FS. | |
2111 | */ | |
2112 | ASSERT(fs_devices->seeding); | |
2113 | ||
94d5f0c2 AJ |
2114 | tmp_fs_devices = fs_info->fs_devices; |
2115 | while (tmp_fs_devices) { | |
2116 | if (tmp_fs_devices->seed == fs_devices) { | |
2117 | tmp_fs_devices->seed = fs_devices->seed; | |
2118 | break; | |
2119 | } | |
2120 | tmp_fs_devices = tmp_fs_devices->seed; | |
2121 | } | |
2122 | fs_devices->seed = NULL; | |
0226e0eb | 2123 | close_fs_devices(fs_devices); |
8bef8401 | 2124 | free_fs_devices(fs_devices); |
94d5f0c2 | 2125 | } |
e93c89c1 SB |
2126 | } |
2127 | ||
4f5ad7bd | 2128 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2129 | { |
4f5ad7bd | 2130 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2131 | |
e93c89c1 | 2132 | WARN_ON(!tgtdev); |
d9a071f0 | 2133 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2134 | |
d9a071f0 | 2135 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2136 | |
779bf3fe | 2137 | if (tgtdev->bdev) |
d9a071f0 | 2138 | fs_devices->open_devices--; |
779bf3fe | 2139 | |
d9a071f0 | 2140 | fs_devices->num_devices--; |
e93c89c1 | 2141 | |
d6507cf1 | 2142 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2143 | |
e93c89c1 | 2144 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2145 | |
d9a071f0 | 2146 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2147 | |
2148 | /* | |
2149 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2150 | * may lead to a call to btrfs_show_devname() which will try | |
2151 | * to hold device_list_mutex. And here this device | |
2152 | * is already out of device list, so we don't have to hold | |
2153 | * the device_list_mutex lock. | |
2154 | */ | |
2155 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2156 | |
2157 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2158 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2159 | } |
2160 | ||
b444ad46 NB |
2161 | static struct btrfs_device *btrfs_find_device_by_path( |
2162 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2163 | { |
2164 | int ret = 0; | |
2165 | struct btrfs_super_block *disk_super; | |
2166 | u64 devid; | |
2167 | u8 *dev_uuid; | |
2168 | struct block_device *bdev; | |
2169 | struct buffer_head *bh; | |
b444ad46 | 2170 | struct btrfs_device *device; |
7ba15b7d | 2171 | |
7ba15b7d | 2172 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2173 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2174 | if (ret) |
b444ad46 | 2175 | return ERR_PTR(ret); |
7ba15b7d SB |
2176 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2177 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2178 | dev_uuid = disk_super->dev_item.uuid; | |
b444ad46 | 2179 | device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d | 2180 | brelse(bh); |
b444ad46 NB |
2181 | if (!device) |
2182 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2183 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2184 | return device; |
7ba15b7d SB |
2185 | } |
2186 | ||
6c050407 NB |
2187 | static struct btrfs_device *btrfs_find_device_missing_or_by_path( |
2188 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d | 2189 | { |
6c050407 | 2190 | struct btrfs_device *device = NULL; |
7ba15b7d SB |
2191 | if (strcmp(device_path, "missing") == 0) { |
2192 | struct list_head *devices; | |
2193 | struct btrfs_device *tmp; | |
2194 | ||
0b246afa | 2195 | devices = &fs_info->fs_devices->devices; |
7ba15b7d | 2196 | list_for_each_entry(tmp, devices, dev_list) { |
e12c9621 AJ |
2197 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2198 | &tmp->dev_state) && !tmp->bdev) { | |
6c050407 | 2199 | device = tmp; |
7ba15b7d SB |
2200 | break; |
2201 | } | |
2202 | } | |
2203 | ||
6c050407 NB |
2204 | if (!device) |
2205 | return ERR_PTR(-ENOENT); | |
7ba15b7d | 2206 | } else { |
6c050407 | 2207 | device = btrfs_find_device_by_path(fs_info, device_path); |
7ba15b7d | 2208 | } |
b444ad46 | 2209 | |
6c050407 | 2210 | return device; |
7ba15b7d SB |
2211 | } |
2212 | ||
5c5c0df0 DS |
2213 | /* |
2214 | * Lookup a device given by device id, or the path if the id is 0. | |
2215 | */ | |
a27a94c2 NB |
2216 | struct btrfs_device *btrfs_find_device_by_devspec( |
2217 | struct btrfs_fs_info *fs_info, u64 devid, const char *devpath) | |
24e0474b | 2218 | { |
a27a94c2 | 2219 | struct btrfs_device *device; |
24e0474b | 2220 | |
5c5c0df0 | 2221 | if (devid) { |
a27a94c2 NB |
2222 | device = btrfs_find_device(fs_info, devid, NULL, NULL); |
2223 | if (!device) | |
2224 | return ERR_PTR(-ENOENT); | |
24e0474b | 2225 | } else { |
5c5c0df0 | 2226 | if (!devpath || !devpath[0]) |
a27a94c2 NB |
2227 | return ERR_PTR(-EINVAL); |
2228 | device = btrfs_find_device_missing_or_by_path(fs_info, devpath); | |
24e0474b | 2229 | } |
a27a94c2 | 2230 | return device; |
24e0474b AJ |
2231 | } |
2232 | ||
2b82032c YZ |
2233 | /* |
2234 | * does all the dirty work required for changing file system's UUID. | |
2235 | */ | |
2ff7e61e | 2236 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2237 | { |
0b246afa | 2238 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2239 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2240 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2241 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2242 | struct btrfs_device *device; |
2243 | u64 super_flags; | |
2244 | ||
a32bf9a3 | 2245 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2246 | if (!fs_devices->seeding) |
2b82032c YZ |
2247 | return -EINVAL; |
2248 | ||
2dfeca9b | 2249 | seed_devices = alloc_fs_devices(NULL); |
2208a378 ID |
2250 | if (IS_ERR(seed_devices)) |
2251 | return PTR_ERR(seed_devices); | |
2b82032c | 2252 | |
e4404d6e YZ |
2253 | old_devices = clone_fs_devices(fs_devices); |
2254 | if (IS_ERR(old_devices)) { | |
2255 | kfree(seed_devices); | |
2256 | return PTR_ERR(old_devices); | |
2b82032c | 2257 | } |
e4404d6e | 2258 | |
c4babc5e | 2259 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2260 | |
e4404d6e YZ |
2261 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2262 | seed_devices->opened = 1; | |
2263 | INIT_LIST_HEAD(&seed_devices->devices); | |
2264 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2265 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2266 | |
321a4bf7 | 2267 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2268 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2269 | synchronize_rcu); | |
2196d6e8 MX |
2270 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2271 | device->fs_devices = seed_devices; | |
c9513edb | 2272 | |
34441361 | 2273 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2274 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2275 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2276 | |
2b82032c YZ |
2277 | fs_devices->seeding = 0; |
2278 | fs_devices->num_devices = 0; | |
2279 | fs_devices->open_devices = 0; | |
69611ac8 | 2280 | fs_devices->missing_devices = 0; |
69611ac8 | 2281 | fs_devices->rotating = 0; |
e4404d6e | 2282 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2283 | |
2284 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2285 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2286 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2287 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2288 | |
2b82032c YZ |
2289 | super_flags = btrfs_super_flags(disk_super) & |
2290 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2291 | btrfs_set_super_flags(disk_super, super_flags); | |
2292 | ||
2293 | return 0; | |
2294 | } | |
2295 | ||
2296 | /* | |
01327610 | 2297 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2298 | */ |
2299 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2300 | struct btrfs_fs_info *fs_info) |
2b82032c | 2301 | { |
5b4aacef | 2302 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2303 | struct btrfs_path *path; |
2304 | struct extent_buffer *leaf; | |
2305 | struct btrfs_dev_item *dev_item; | |
2306 | struct btrfs_device *device; | |
2307 | struct btrfs_key key; | |
44880fdc | 2308 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2309 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2310 | u64 devid; | |
2311 | int ret; | |
2312 | ||
2313 | path = btrfs_alloc_path(); | |
2314 | if (!path) | |
2315 | return -ENOMEM; | |
2316 | ||
2b82032c YZ |
2317 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2318 | key.offset = 0; | |
2319 | key.type = BTRFS_DEV_ITEM_KEY; | |
2320 | ||
2321 | while (1) { | |
2322 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2323 | if (ret < 0) | |
2324 | goto error; | |
2325 | ||
2326 | leaf = path->nodes[0]; | |
2327 | next_slot: | |
2328 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2329 | ret = btrfs_next_leaf(root, path); | |
2330 | if (ret > 0) | |
2331 | break; | |
2332 | if (ret < 0) | |
2333 | goto error; | |
2334 | leaf = path->nodes[0]; | |
2335 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2336 | btrfs_release_path(path); |
2b82032c YZ |
2337 | continue; |
2338 | } | |
2339 | ||
2340 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2341 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2342 | key.type != BTRFS_DEV_ITEM_KEY) | |
2343 | break; | |
2344 | ||
2345 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2346 | struct btrfs_dev_item); | |
2347 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2348 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2349 | BTRFS_UUID_SIZE); |
1473b24e | 2350 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2351 | BTRFS_FSID_SIZE); |
0b246afa | 2352 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2353 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2354 | |
2355 | if (device->fs_devices->seeding) { | |
2356 | btrfs_set_device_generation(leaf, dev_item, | |
2357 | device->generation); | |
2358 | btrfs_mark_buffer_dirty(leaf); | |
2359 | } | |
2360 | ||
2361 | path->slots[0]++; | |
2362 | goto next_slot; | |
2363 | } | |
2364 | ret = 0; | |
2365 | error: | |
2366 | btrfs_free_path(path); | |
2367 | return ret; | |
2368 | } | |
2369 | ||
da353f6b | 2370 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2371 | { |
5112febb | 2372 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2373 | struct request_queue *q; |
788f20eb CM |
2374 | struct btrfs_trans_handle *trans; |
2375 | struct btrfs_device *device; | |
2376 | struct block_device *bdev; | |
0b246afa | 2377 | struct super_block *sb = fs_info->sb; |
606686ee | 2378 | struct rcu_string *name; |
5da54bc1 | 2379 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2380 | u64 orig_super_total_bytes; |
2381 | u64 orig_super_num_devices; | |
2b82032c | 2382 | int seeding_dev = 0; |
788f20eb | 2383 | int ret = 0; |
7132a262 | 2384 | bool unlocked = false; |
788f20eb | 2385 | |
5da54bc1 | 2386 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2387 | return -EROFS; |
788f20eb | 2388 | |
a5d16333 | 2389 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2390 | fs_info->bdev_holder); |
7f59203a JB |
2391 | if (IS_ERR(bdev)) |
2392 | return PTR_ERR(bdev); | |
a2135011 | 2393 | |
5da54bc1 | 2394 | if (fs_devices->seeding) { |
2b82032c YZ |
2395 | seeding_dev = 1; |
2396 | down_write(&sb->s_umount); | |
2397 | mutex_lock(&uuid_mutex); | |
2398 | } | |
2399 | ||
8c8bee1d | 2400 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2401 | |
5da54bc1 | 2402 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2403 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2404 | if (device->bdev == bdev) { |
2405 | ret = -EEXIST; | |
d25628bd | 2406 | mutex_unlock( |
5da54bc1 | 2407 | &fs_devices->device_list_mutex); |
2b82032c | 2408 | goto error; |
788f20eb CM |
2409 | } |
2410 | } | |
5da54bc1 | 2411 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2412 | |
0b246afa | 2413 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2414 | if (IS_ERR(device)) { |
788f20eb | 2415 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2416 | ret = PTR_ERR(device); |
2b82032c | 2417 | goto error; |
788f20eb CM |
2418 | } |
2419 | ||
78f2c9e6 | 2420 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2421 | if (!name) { |
2b82032c | 2422 | ret = -ENOMEM; |
5c4cf6c9 | 2423 | goto error_free_device; |
788f20eb | 2424 | } |
606686ee | 2425 | rcu_assign_pointer(device->name, name); |
2b82032c | 2426 | |
a22285a6 | 2427 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2428 | if (IS_ERR(trans)) { |
98d5dc13 | 2429 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2430 | goto error_free_device; |
98d5dc13 TI |
2431 | } |
2432 | ||
d5e2003c | 2433 | q = bdev_get_queue(bdev); |
ebbede42 | 2434 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2435 | device->generation = trans->transid; |
0b246afa JM |
2436 | device->io_width = fs_info->sectorsize; |
2437 | device->io_align = fs_info->sectorsize; | |
2438 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2439 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2440 | fs_info->sectorsize); | |
2cc3c559 | 2441 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2442 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2443 | device->fs_info = fs_info; |
788f20eb | 2444 | device->bdev = bdev; |
e12c9621 | 2445 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2446 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2447 | device->mode = FMODE_EXCL; |
27087f37 | 2448 | device->dev_stats_valid = 1; |
9f6d2510 | 2449 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2450 | |
2b82032c | 2451 | if (seeding_dev) { |
1751e8a6 | 2452 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2453 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2454 | if (ret) { |
2455 | btrfs_abort_transaction(trans, ret); | |
2456 | goto error_trans; | |
2457 | } | |
2b82032c | 2458 | } |
788f20eb | 2459 | |
5da54bc1 | 2460 | device->fs_devices = fs_devices; |
e5e9a520 | 2461 | |
5da54bc1 | 2462 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2463 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2464 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2465 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2466 | fs_devices->num_devices++; | |
2467 | fs_devices->open_devices++; | |
2468 | fs_devices->rw_devices++; | |
2469 | fs_devices->total_devices++; | |
2470 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2471 | |
a5ed45f8 | 2472 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2473 | |
e884f4f0 | 2474 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2475 | fs_devices->rotating = 1; |
c289811c | 2476 | |
39379faa | 2477 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2478 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2479 | round_down(orig_super_total_bytes + device->total_bytes, |
2480 | fs_info->sectorsize)); | |
788f20eb | 2481 | |
39379faa NA |
2482 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2483 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2484 | orig_super_num_devices + 1); | |
0d39376a AJ |
2485 | |
2486 | /* add sysfs device entry */ | |
5da54bc1 | 2487 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2488 | |
2196d6e8 MX |
2489 | /* |
2490 | * we've got more storage, clear any full flags on the space | |
2491 | * infos | |
2492 | */ | |
0b246afa | 2493 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2494 | |
34441361 | 2495 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2496 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2497 | |
2b82032c | 2498 | if (seeding_dev) { |
34441361 | 2499 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2500 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2501 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2502 | if (ret) { |
66642832 | 2503 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2504 | goto error_sysfs; |
005d6427 | 2505 | } |
2196d6e8 MX |
2506 | } |
2507 | ||
8e87e856 | 2508 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2509 | if (ret) { |
66642832 | 2510 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2511 | goto error_sysfs; |
2196d6e8 MX |
2512 | } |
2513 | ||
2514 | if (seeding_dev) { | |
2515 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2516 | ||
0b246afa | 2517 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2518 | if (ret) { |
66642832 | 2519 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2520 | goto error_sysfs; |
005d6427 | 2521 | } |
b2373f25 AJ |
2522 | |
2523 | /* Sprouting would change fsid of the mounted root, | |
2524 | * so rename the fsid on the sysfs | |
2525 | */ | |
2526 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa | 2527 | fs_info->fsid); |
5da54bc1 | 2528 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2529 | btrfs_warn(fs_info, |
2530 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2531 | } |
2532 | ||
3a45bb20 | 2533 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2534 | |
2b82032c YZ |
2535 | if (seeding_dev) { |
2536 | mutex_unlock(&uuid_mutex); | |
2537 | up_write(&sb->s_umount); | |
7132a262 | 2538 | unlocked = true; |
788f20eb | 2539 | |
79787eaa JM |
2540 | if (ret) /* transaction commit */ |
2541 | return ret; | |
2542 | ||
2ff7e61e | 2543 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2544 | if (ret < 0) |
0b246afa | 2545 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2546 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2547 | trans = btrfs_attach_transaction(root); |
2548 | if (IS_ERR(trans)) { | |
2549 | if (PTR_ERR(trans) == -ENOENT) | |
2550 | return 0; | |
7132a262 AJ |
2551 | ret = PTR_ERR(trans); |
2552 | trans = NULL; | |
2553 | goto error_sysfs; | |
671415b7 | 2554 | } |
3a45bb20 | 2555 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2556 | } |
c9e9f97b | 2557 | |
5a1972bd QW |
2558 | /* Update ctime/mtime for libblkid */ |
2559 | update_dev_time(device_path); | |
2b82032c | 2560 | return ret; |
79787eaa | 2561 | |
d31c32f6 | 2562 | error_sysfs: |
5da54bc1 | 2563 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2564 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2565 | mutex_lock(&fs_info->chunk_mutex); | |
2566 | list_del_rcu(&device->dev_list); | |
2567 | list_del(&device->dev_alloc_list); | |
2568 | fs_info->fs_devices->num_devices--; | |
2569 | fs_info->fs_devices->open_devices--; | |
2570 | fs_info->fs_devices->rw_devices--; | |
2571 | fs_info->fs_devices->total_devices--; | |
2572 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2573 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2574 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2575 | orig_super_total_bytes); | |
2576 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2577 | orig_super_num_devices); | |
2578 | mutex_unlock(&fs_info->chunk_mutex); | |
2579 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2580 | error_trans: |
0af2c4bf | 2581 | if (seeding_dev) |
1751e8a6 | 2582 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2583 | if (trans) |
2584 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2585 | error_free_device: |
a425f9d4 | 2586 | btrfs_free_device(device); |
2b82032c | 2587 | error: |
e525fd89 | 2588 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2589 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2590 | mutex_unlock(&uuid_mutex); |
2591 | up_write(&sb->s_umount); | |
2592 | } | |
c9e9f97b | 2593 | return ret; |
788f20eb CM |
2594 | } |
2595 | ||
d397712b CM |
2596 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2597 | struct btrfs_device *device) | |
0b86a832 CM |
2598 | { |
2599 | int ret; | |
2600 | struct btrfs_path *path; | |
0b246afa | 2601 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2602 | struct btrfs_dev_item *dev_item; |
2603 | struct extent_buffer *leaf; | |
2604 | struct btrfs_key key; | |
2605 | ||
0b86a832 CM |
2606 | path = btrfs_alloc_path(); |
2607 | if (!path) | |
2608 | return -ENOMEM; | |
2609 | ||
2610 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2611 | key.type = BTRFS_DEV_ITEM_KEY; | |
2612 | key.offset = device->devid; | |
2613 | ||
2614 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2615 | if (ret < 0) | |
2616 | goto out; | |
2617 | ||
2618 | if (ret > 0) { | |
2619 | ret = -ENOENT; | |
2620 | goto out; | |
2621 | } | |
2622 | ||
2623 | leaf = path->nodes[0]; | |
2624 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2625 | ||
2626 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2627 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2628 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2629 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2630 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2631 | btrfs_set_device_total_bytes(leaf, dev_item, |
2632 | btrfs_device_get_disk_total_bytes(device)); | |
2633 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2634 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2635 | btrfs_mark_buffer_dirty(leaf); |
2636 | ||
2637 | out: | |
2638 | btrfs_free_path(path); | |
2639 | return ret; | |
2640 | } | |
2641 | ||
2196d6e8 | 2642 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2643 | struct btrfs_device *device, u64 new_size) |
2644 | { | |
0b246afa JM |
2645 | struct btrfs_fs_info *fs_info = device->fs_info; |
2646 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2647 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2648 | u64 old_total; |
2649 | u64 diff; | |
8f18cf13 | 2650 | |
ebbede42 | 2651 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2652 | return -EACCES; |
2196d6e8 | 2653 | |
7dfb8be1 NB |
2654 | new_size = round_down(new_size, fs_info->sectorsize); |
2655 | ||
34441361 | 2656 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2657 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2658 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2659 | |
63a212ab | 2660 | if (new_size <= device->total_bytes || |
401e29c1 | 2661 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2662 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2663 | return -EINVAL; |
2196d6e8 | 2664 | } |
2b82032c | 2665 | |
0b246afa | 2666 | fs_devices = fs_info->fs_devices; |
2b82032c | 2667 | |
7dfb8be1 NB |
2668 | btrfs_set_super_total_bytes(super_copy, |
2669 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2670 | device->fs_devices->total_rw_bytes += diff; |
2671 | ||
7cc8e58d MX |
2672 | btrfs_device_set_total_bytes(device, new_size); |
2673 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2674 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2675 | if (list_empty(&device->resized_list)) |
2676 | list_add_tail(&device->resized_list, | |
2677 | &fs_devices->resized_devices); | |
34441361 | 2678 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2679 | |
8f18cf13 CM |
2680 | return btrfs_update_device(trans, device); |
2681 | } | |
2682 | ||
f4208794 | 2683 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2684 | { |
f4208794 | 2685 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2686 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2687 | int ret; |
2688 | struct btrfs_path *path; | |
2689 | struct btrfs_key key; | |
2690 | ||
8f18cf13 CM |
2691 | path = btrfs_alloc_path(); |
2692 | if (!path) | |
2693 | return -ENOMEM; | |
2694 | ||
408fbf19 | 2695 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2696 | key.offset = chunk_offset; |
2697 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2698 | ||
2699 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2700 | if (ret < 0) |
2701 | goto out; | |
2702 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2703 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2704 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2705 | ret = -ENOENT; |
2706 | goto out; | |
2707 | } | |
8f18cf13 CM |
2708 | |
2709 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2710 | if (ret < 0) |
0b246afa JM |
2711 | btrfs_handle_fs_error(fs_info, ret, |
2712 | "Failed to delete chunk item."); | |
79787eaa | 2713 | out: |
8f18cf13 | 2714 | btrfs_free_path(path); |
65a246c5 | 2715 | return ret; |
8f18cf13 CM |
2716 | } |
2717 | ||
408fbf19 | 2718 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2719 | { |
0b246afa | 2720 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2721 | struct btrfs_disk_key *disk_key; |
2722 | struct btrfs_chunk *chunk; | |
2723 | u8 *ptr; | |
2724 | int ret = 0; | |
2725 | u32 num_stripes; | |
2726 | u32 array_size; | |
2727 | u32 len = 0; | |
2728 | u32 cur; | |
2729 | struct btrfs_key key; | |
2730 | ||
34441361 | 2731 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2732 | array_size = btrfs_super_sys_array_size(super_copy); |
2733 | ||
2734 | ptr = super_copy->sys_chunk_array; | |
2735 | cur = 0; | |
2736 | ||
2737 | while (cur < array_size) { | |
2738 | disk_key = (struct btrfs_disk_key *)ptr; | |
2739 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2740 | ||
2741 | len = sizeof(*disk_key); | |
2742 | ||
2743 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2744 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2745 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2746 | len += btrfs_chunk_item_size(num_stripes); | |
2747 | } else { | |
2748 | ret = -EIO; | |
2749 | break; | |
2750 | } | |
408fbf19 | 2751 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2752 | key.offset == chunk_offset) { |
2753 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2754 | array_size -= len; | |
2755 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2756 | } else { | |
2757 | ptr += len; | |
2758 | cur += len; | |
2759 | } | |
2760 | } | |
34441361 | 2761 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2762 | return ret; |
2763 | } | |
2764 | ||
60ca842e OS |
2765 | /* |
2766 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2767 | * @logical: Logical block offset in bytes. | |
2768 | * @length: Length of extent in bytes. | |
2769 | * | |
2770 | * Return: Chunk mapping or ERR_PTR. | |
2771 | */ | |
2772 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2773 | u64 logical, u64 length) | |
592d92ee LB |
2774 | { |
2775 | struct extent_map_tree *em_tree; | |
2776 | struct extent_map *em; | |
2777 | ||
2778 | em_tree = &fs_info->mapping_tree.map_tree; | |
2779 | read_lock(&em_tree->lock); | |
2780 | em = lookup_extent_mapping(em_tree, logical, length); | |
2781 | read_unlock(&em_tree->lock); | |
2782 | ||
2783 | if (!em) { | |
2784 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2785 | logical, length); | |
2786 | return ERR_PTR(-EINVAL); | |
2787 | } | |
2788 | ||
2789 | if (em->start > logical || em->start + em->len < logical) { | |
2790 | btrfs_crit(fs_info, | |
2791 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2792 | logical, length, em->start, em->start + em->len); | |
2793 | free_extent_map(em); | |
2794 | return ERR_PTR(-EINVAL); | |
2795 | } | |
2796 | ||
2797 | /* callers are responsible for dropping em's ref. */ | |
2798 | return em; | |
2799 | } | |
2800 | ||
97aff912 | 2801 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2802 | { |
97aff912 | 2803 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2804 | struct extent_map *em; |
2805 | struct map_lookup *map; | |
2196d6e8 | 2806 | u64 dev_extent_len = 0; |
47ab2a6c | 2807 | int i, ret = 0; |
0b246afa | 2808 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2809 | |
60ca842e | 2810 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2811 | if (IS_ERR(em)) { |
47ab2a6c JB |
2812 | /* |
2813 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2814 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2815 | * do anything we still error out. |
2816 | */ | |
2817 | ASSERT(0); | |
592d92ee | 2818 | return PTR_ERR(em); |
47ab2a6c | 2819 | } |
95617d69 | 2820 | map = em->map_lookup; |
34441361 | 2821 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2822 | check_system_chunk(trans, map->type); |
34441361 | 2823 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2824 | |
57ba4cb8 FM |
2825 | /* |
2826 | * Take the device list mutex to prevent races with the final phase of | |
2827 | * a device replace operation that replaces the device object associated | |
2828 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2829 | */ | |
2830 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2831 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2832 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2833 | ret = btrfs_free_dev_extent(trans, device, |
2834 | map->stripes[i].physical, | |
2835 | &dev_extent_len); | |
47ab2a6c | 2836 | if (ret) { |
57ba4cb8 | 2837 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2838 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2839 | goto out; |
2840 | } | |
a061fc8d | 2841 | |
2196d6e8 | 2842 | if (device->bytes_used > 0) { |
34441361 | 2843 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2844 | btrfs_device_set_bytes_used(device, |
2845 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2846 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2847 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2848 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2849 | } |
a061fc8d | 2850 | |
dfe25020 CM |
2851 | if (map->stripes[i].dev) { |
2852 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2853 | if (ret) { |
57ba4cb8 | 2854 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2855 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2856 | goto out; |
2857 | } | |
dfe25020 | 2858 | } |
8f18cf13 | 2859 | } |
57ba4cb8 FM |
2860 | mutex_unlock(&fs_devices->device_list_mutex); |
2861 | ||
f4208794 | 2862 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 2863 | if (ret) { |
66642832 | 2864 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2865 | goto out; |
2866 | } | |
8f18cf13 | 2867 | |
6bccf3ab | 2868 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2869 | |
8f18cf13 | 2870 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2871 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2872 | if (ret) { |
66642832 | 2873 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2874 | goto out; |
2875 | } | |
8f18cf13 CM |
2876 | } |
2877 | ||
5a98ec01 | 2878 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 2879 | if (ret) { |
66642832 | 2880 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2881 | goto out; |
2882 | } | |
2b82032c | 2883 | |
47ab2a6c | 2884 | out: |
2b82032c YZ |
2885 | /* once for us */ |
2886 | free_extent_map(em); | |
47ab2a6c JB |
2887 | return ret; |
2888 | } | |
2b82032c | 2889 | |
5b4aacef | 2890 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2891 | { |
5b4aacef | 2892 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2893 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2894 | int ret; |
2b82032c | 2895 | |
67c5e7d4 FM |
2896 | /* |
2897 | * Prevent races with automatic removal of unused block groups. | |
2898 | * After we relocate and before we remove the chunk with offset | |
2899 | * chunk_offset, automatic removal of the block group can kick in, | |
2900 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2901 | * | |
2902 | * Make sure to acquire this mutex before doing a tree search (dev | |
2903 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2904 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2905 | * we release the path used to search the chunk/dev tree and before | |
2906 | * the current task acquires this mutex and calls us. | |
2907 | */ | |
a32bf9a3 | 2908 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 2909 | |
0b246afa | 2910 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2911 | if (ret) |
2912 | return -ENOSPC; | |
2913 | ||
2914 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 2915 | btrfs_scrub_pause(fs_info); |
0b246afa | 2916 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2917 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2918 | if (ret) |
2919 | return ret; | |
2920 | ||
75cb379d JM |
2921 | /* |
2922 | * We add the kobjects here (and after forcing data chunk creation) | |
2923 | * since relocation is the only place we'll create chunks of a new | |
2924 | * type at runtime. The only place where we'll remove the last | |
2925 | * chunk of a type is the call immediately below this one. Even | |
2926 | * so, we're protected against races with the cleaner thread since | |
2927 | * we're covered by the delete_unused_bgs_mutex. | |
2928 | */ | |
2929 | btrfs_add_raid_kobjects(fs_info); | |
2930 | ||
19c4d2f9 CM |
2931 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2932 | chunk_offset); | |
2933 | if (IS_ERR(trans)) { | |
2934 | ret = PTR_ERR(trans); | |
2935 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2936 | return ret; | |
2937 | } | |
2938 | ||
47ab2a6c | 2939 | /* |
19c4d2f9 CM |
2940 | * step two, delete the device extents and the |
2941 | * chunk tree entries | |
47ab2a6c | 2942 | */ |
97aff912 | 2943 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 2944 | btrfs_end_transaction(trans); |
19c4d2f9 | 2945 | return ret; |
2b82032c YZ |
2946 | } |
2947 | ||
2ff7e61e | 2948 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2949 | { |
0b246afa | 2950 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2951 | struct btrfs_path *path; |
2952 | struct extent_buffer *leaf; | |
2953 | struct btrfs_chunk *chunk; | |
2954 | struct btrfs_key key; | |
2955 | struct btrfs_key found_key; | |
2b82032c | 2956 | u64 chunk_type; |
ba1bf481 JB |
2957 | bool retried = false; |
2958 | int failed = 0; | |
2b82032c YZ |
2959 | int ret; |
2960 | ||
2961 | path = btrfs_alloc_path(); | |
2962 | if (!path) | |
2963 | return -ENOMEM; | |
2964 | ||
ba1bf481 | 2965 | again: |
2b82032c YZ |
2966 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2967 | key.offset = (u64)-1; | |
2968 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2969 | ||
2970 | while (1) { | |
0b246afa | 2971 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2972 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2973 | if (ret < 0) { |
0b246afa | 2974 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2975 | goto error; |
67c5e7d4 | 2976 | } |
79787eaa | 2977 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2978 | |
2979 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2980 | key.type); | |
67c5e7d4 | 2981 | if (ret) |
0b246afa | 2982 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
2983 | if (ret < 0) |
2984 | goto error; | |
2985 | if (ret > 0) | |
2986 | break; | |
1a40e23b | 2987 | |
2b82032c YZ |
2988 | leaf = path->nodes[0]; |
2989 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2990 | |
2b82032c YZ |
2991 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2992 | struct btrfs_chunk); | |
2993 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2994 | btrfs_release_path(path); |
8f18cf13 | 2995 | |
2b82032c | 2996 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 2997 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
2998 | if (ret == -ENOSPC) |
2999 | failed++; | |
14586651 HS |
3000 | else |
3001 | BUG_ON(ret); | |
2b82032c | 3002 | } |
0b246afa | 3003 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3004 | |
2b82032c YZ |
3005 | if (found_key.offset == 0) |
3006 | break; | |
3007 | key.offset = found_key.offset - 1; | |
3008 | } | |
3009 | ret = 0; | |
ba1bf481 JB |
3010 | if (failed && !retried) { |
3011 | failed = 0; | |
3012 | retried = true; | |
3013 | goto again; | |
fae7f21c | 3014 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3015 | ret = -ENOSPC; |
3016 | } | |
2b82032c YZ |
3017 | error: |
3018 | btrfs_free_path(path); | |
3019 | return ret; | |
8f18cf13 CM |
3020 | } |
3021 | ||
a6f93c71 LB |
3022 | /* |
3023 | * return 1 : allocate a data chunk successfully, | |
3024 | * return <0: errors during allocating a data chunk, | |
3025 | * return 0 : no need to allocate a data chunk. | |
3026 | */ | |
3027 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3028 | u64 chunk_offset) | |
3029 | { | |
3030 | struct btrfs_block_group_cache *cache; | |
3031 | u64 bytes_used; | |
3032 | u64 chunk_type; | |
3033 | ||
3034 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3035 | ASSERT(cache); | |
3036 | chunk_type = cache->flags; | |
3037 | btrfs_put_block_group(cache); | |
3038 | ||
3039 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3040 | spin_lock(&fs_info->data_sinfo->lock); | |
3041 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3042 | spin_unlock(&fs_info->data_sinfo->lock); | |
3043 | ||
3044 | if (!bytes_used) { | |
3045 | struct btrfs_trans_handle *trans; | |
3046 | int ret; | |
3047 | ||
3048 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3049 | if (IS_ERR(trans)) | |
3050 | return PTR_ERR(trans); | |
3051 | ||
43a7e99d | 3052 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3053 | BTRFS_BLOCK_GROUP_DATA); |
3054 | btrfs_end_transaction(trans); | |
3055 | if (ret < 0) | |
3056 | return ret; | |
3057 | ||
75cb379d JM |
3058 | btrfs_add_raid_kobjects(fs_info); |
3059 | ||
a6f93c71 LB |
3060 | return 1; |
3061 | } | |
3062 | } | |
3063 | return 0; | |
3064 | } | |
3065 | ||
6bccf3ab | 3066 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3067 | struct btrfs_balance_control *bctl) |
3068 | { | |
6bccf3ab | 3069 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3070 | struct btrfs_trans_handle *trans; |
3071 | struct btrfs_balance_item *item; | |
3072 | struct btrfs_disk_balance_args disk_bargs; | |
3073 | struct btrfs_path *path; | |
3074 | struct extent_buffer *leaf; | |
3075 | struct btrfs_key key; | |
3076 | int ret, err; | |
3077 | ||
3078 | path = btrfs_alloc_path(); | |
3079 | if (!path) | |
3080 | return -ENOMEM; | |
3081 | ||
3082 | trans = btrfs_start_transaction(root, 0); | |
3083 | if (IS_ERR(trans)) { | |
3084 | btrfs_free_path(path); | |
3085 | return PTR_ERR(trans); | |
3086 | } | |
3087 | ||
3088 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3089 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3090 | key.offset = 0; |
3091 | ||
3092 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3093 | sizeof(*item)); | |
3094 | if (ret) | |
3095 | goto out; | |
3096 | ||
3097 | leaf = path->nodes[0]; | |
3098 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3099 | ||
b159fa28 | 3100 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3101 | |
3102 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3103 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3104 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3105 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3106 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3107 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3108 | ||
3109 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3110 | ||
3111 | btrfs_mark_buffer_dirty(leaf); | |
3112 | out: | |
3113 | btrfs_free_path(path); | |
3a45bb20 | 3114 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3115 | if (err && !ret) |
3116 | ret = err; | |
3117 | return ret; | |
3118 | } | |
3119 | ||
6bccf3ab | 3120 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3121 | { |
6bccf3ab | 3122 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3123 | struct btrfs_trans_handle *trans; |
3124 | struct btrfs_path *path; | |
3125 | struct btrfs_key key; | |
3126 | int ret, err; | |
3127 | ||
3128 | path = btrfs_alloc_path(); | |
3129 | if (!path) | |
3130 | return -ENOMEM; | |
3131 | ||
3132 | trans = btrfs_start_transaction(root, 0); | |
3133 | if (IS_ERR(trans)) { | |
3134 | btrfs_free_path(path); | |
3135 | return PTR_ERR(trans); | |
3136 | } | |
3137 | ||
3138 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3139 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3140 | key.offset = 0; |
3141 | ||
3142 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3143 | if (ret < 0) | |
3144 | goto out; | |
3145 | if (ret > 0) { | |
3146 | ret = -ENOENT; | |
3147 | goto out; | |
3148 | } | |
3149 | ||
3150 | ret = btrfs_del_item(trans, root, path); | |
3151 | out: | |
3152 | btrfs_free_path(path); | |
3a45bb20 | 3153 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3154 | if (err && !ret) |
3155 | ret = err; | |
3156 | return ret; | |
3157 | } | |
3158 | ||
59641015 ID |
3159 | /* |
3160 | * This is a heuristic used to reduce the number of chunks balanced on | |
3161 | * resume after balance was interrupted. | |
3162 | */ | |
3163 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3164 | { | |
3165 | /* | |
3166 | * Turn on soft mode for chunk types that were being converted. | |
3167 | */ | |
3168 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3169 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3170 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3171 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3172 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3173 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3174 | ||
3175 | /* | |
3176 | * Turn on usage filter if is not already used. The idea is | |
3177 | * that chunks that we have already balanced should be | |
3178 | * reasonably full. Don't do it for chunks that are being | |
3179 | * converted - that will keep us from relocating unconverted | |
3180 | * (albeit full) chunks. | |
3181 | */ | |
3182 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3183 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3184 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3185 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3186 | bctl->data.usage = 90; | |
3187 | } | |
3188 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3189 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3190 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3191 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3192 | bctl->sys.usage = 90; | |
3193 | } | |
3194 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3195 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3196 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3197 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3198 | bctl->meta.usage = 90; | |
3199 | } | |
3200 | } | |
3201 | ||
149196a2 DS |
3202 | /* |
3203 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3204 | */ | |
3205 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3206 | { |
3207 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3208 | int ret; |
c9e9f97b ID |
3209 | |
3210 | BUG_ON(!fs_info->balance_ctl); | |
3211 | ||
3212 | spin_lock(&fs_info->balance_lock); | |
3213 | fs_info->balance_ctl = NULL; | |
3214 | spin_unlock(&fs_info->balance_lock); | |
3215 | ||
3216 | kfree(bctl); | |
149196a2 DS |
3217 | ret = del_balance_item(fs_info); |
3218 | if (ret) | |
3219 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3220 | } |
3221 | ||
ed25e9b2 ID |
3222 | /* |
3223 | * Balance filters. Return 1 if chunk should be filtered out | |
3224 | * (should not be balanced). | |
3225 | */ | |
899c81ea | 3226 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3227 | struct btrfs_balance_args *bargs) |
3228 | { | |
899c81ea ID |
3229 | chunk_type = chunk_to_extended(chunk_type) & |
3230 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3231 | |
899c81ea | 3232 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3233 | return 0; |
3234 | ||
3235 | return 1; | |
3236 | } | |
3237 | ||
dba72cb3 | 3238 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3239 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3240 | { |
3241 | struct btrfs_block_group_cache *cache; | |
3242 | u64 chunk_used; | |
3243 | u64 user_thresh_min; | |
3244 | u64 user_thresh_max; | |
3245 | int ret = 1; | |
3246 | ||
3247 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3248 | chunk_used = btrfs_block_group_used(&cache->item); | |
3249 | ||
3250 | if (bargs->usage_min == 0) | |
3251 | user_thresh_min = 0; | |
3252 | else | |
3253 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3254 | bargs->usage_min); | |
3255 | ||
3256 | if (bargs->usage_max == 0) | |
3257 | user_thresh_max = 1; | |
3258 | else if (bargs->usage_max > 100) | |
3259 | user_thresh_max = cache->key.offset; | |
3260 | else | |
3261 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3262 | bargs->usage_max); | |
3263 | ||
3264 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3265 | ret = 0; | |
3266 | ||
3267 | btrfs_put_block_group(cache); | |
3268 | return ret; | |
3269 | } | |
3270 | ||
dba72cb3 | 3271 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3272 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3273 | { |
3274 | struct btrfs_block_group_cache *cache; | |
3275 | u64 chunk_used, user_thresh; | |
3276 | int ret = 1; | |
3277 | ||
3278 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3279 | chunk_used = btrfs_block_group_used(&cache->item); | |
3280 | ||
bc309467 | 3281 | if (bargs->usage_min == 0) |
3e39cea6 | 3282 | user_thresh = 1; |
a105bb88 ID |
3283 | else if (bargs->usage > 100) |
3284 | user_thresh = cache->key.offset; | |
3285 | else | |
3286 | user_thresh = div_factor_fine(cache->key.offset, | |
3287 | bargs->usage); | |
3288 | ||
5ce5b3c0 ID |
3289 | if (chunk_used < user_thresh) |
3290 | ret = 0; | |
3291 | ||
3292 | btrfs_put_block_group(cache); | |
3293 | return ret; | |
3294 | } | |
3295 | ||
409d404b ID |
3296 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3297 | struct btrfs_chunk *chunk, | |
3298 | struct btrfs_balance_args *bargs) | |
3299 | { | |
3300 | struct btrfs_stripe *stripe; | |
3301 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3302 | int i; | |
3303 | ||
3304 | for (i = 0; i < num_stripes; i++) { | |
3305 | stripe = btrfs_stripe_nr(chunk, i); | |
3306 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3307 | return 0; | |
3308 | } | |
3309 | ||
3310 | return 1; | |
3311 | } | |
3312 | ||
94e60d5a ID |
3313 | /* [pstart, pend) */ |
3314 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3315 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3316 | struct btrfs_balance_args *bargs) |
3317 | { | |
3318 | struct btrfs_stripe *stripe; | |
3319 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3320 | u64 stripe_offset; | |
3321 | u64 stripe_length; | |
3322 | int factor; | |
3323 | int i; | |
3324 | ||
3325 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3326 | return 0; | |
3327 | ||
3328 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3329 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3330 | factor = num_stripes / 2; | |
3331 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3332 | factor = num_stripes - 1; | |
3333 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3334 | factor = num_stripes - 2; | |
3335 | } else { | |
3336 | factor = num_stripes; | |
3337 | } | |
94e60d5a ID |
3338 | |
3339 | for (i = 0; i < num_stripes; i++) { | |
3340 | stripe = btrfs_stripe_nr(chunk, i); | |
3341 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3342 | continue; | |
3343 | ||
3344 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3345 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3346 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3347 | |
3348 | if (stripe_offset < bargs->pend && | |
3349 | stripe_offset + stripe_length > bargs->pstart) | |
3350 | return 0; | |
3351 | } | |
3352 | ||
3353 | return 1; | |
3354 | } | |
3355 | ||
ea67176a ID |
3356 | /* [vstart, vend) */ |
3357 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3358 | struct btrfs_chunk *chunk, | |
3359 | u64 chunk_offset, | |
3360 | struct btrfs_balance_args *bargs) | |
3361 | { | |
3362 | if (chunk_offset < bargs->vend && | |
3363 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3364 | /* at least part of the chunk is inside this vrange */ | |
3365 | return 0; | |
3366 | ||
3367 | return 1; | |
3368 | } | |
3369 | ||
dee32d0a GAP |
3370 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3371 | struct btrfs_chunk *chunk, | |
3372 | struct btrfs_balance_args *bargs) | |
3373 | { | |
3374 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3375 | ||
3376 | if (bargs->stripes_min <= num_stripes | |
3377 | && num_stripes <= bargs->stripes_max) | |
3378 | return 0; | |
3379 | ||
3380 | return 1; | |
3381 | } | |
3382 | ||
899c81ea | 3383 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3384 | struct btrfs_balance_args *bargs) |
3385 | { | |
3386 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3387 | return 0; | |
3388 | ||
899c81ea ID |
3389 | chunk_type = chunk_to_extended(chunk_type) & |
3390 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3391 | |
899c81ea | 3392 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3393 | return 1; |
3394 | ||
3395 | return 0; | |
3396 | } | |
3397 | ||
2ff7e61e | 3398 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3399 | struct extent_buffer *leaf, |
3400 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3401 | { | |
0b246afa | 3402 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3403 | struct btrfs_balance_args *bargs = NULL; |
3404 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3405 | ||
3406 | /* type filter */ | |
3407 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3408 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3409 | return 0; | |
3410 | } | |
3411 | ||
3412 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3413 | bargs = &bctl->data; | |
3414 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3415 | bargs = &bctl->sys; | |
3416 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3417 | bargs = &bctl->meta; | |
3418 | ||
ed25e9b2 ID |
3419 | /* profiles filter */ |
3420 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3421 | chunk_profiles_filter(chunk_type, bargs)) { | |
3422 | return 0; | |
5ce5b3c0 ID |
3423 | } |
3424 | ||
3425 | /* usage filter */ | |
3426 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3427 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3428 | return 0; |
bc309467 | 3429 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3430 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3431 | return 0; |
409d404b ID |
3432 | } |
3433 | ||
3434 | /* devid filter */ | |
3435 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3436 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3437 | return 0; | |
94e60d5a ID |
3438 | } |
3439 | ||
3440 | /* drange filter, makes sense only with devid filter */ | |
3441 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3442 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3443 | return 0; |
ea67176a ID |
3444 | } |
3445 | ||
3446 | /* vrange filter */ | |
3447 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3448 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3449 | return 0; | |
ed25e9b2 ID |
3450 | } |
3451 | ||
dee32d0a GAP |
3452 | /* stripes filter */ |
3453 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3454 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3455 | return 0; | |
3456 | } | |
3457 | ||
cfa4c961 ID |
3458 | /* soft profile changing mode */ |
3459 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3460 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3461 | return 0; | |
3462 | } | |
3463 | ||
7d824b6f DS |
3464 | /* |
3465 | * limited by count, must be the last filter | |
3466 | */ | |
3467 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3468 | if (bargs->limit == 0) | |
3469 | return 0; | |
3470 | else | |
3471 | bargs->limit--; | |
12907fc7 DS |
3472 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3473 | /* | |
3474 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3475 | * determined here because we do not have the global information |
12907fc7 DS |
3476 | * about the count of all chunks that satisfy the filters. |
3477 | */ | |
3478 | if (bargs->limit_max == 0) | |
3479 | return 0; | |
3480 | else | |
3481 | bargs->limit_max--; | |
7d824b6f DS |
3482 | } |
3483 | ||
f43ffb60 ID |
3484 | return 1; |
3485 | } | |
3486 | ||
c9e9f97b | 3487 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3488 | { |
19a39dce | 3489 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3490 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3491 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3492 | struct list_head *devices; | |
ec44a35c CM |
3493 | struct btrfs_device *device; |
3494 | u64 old_size; | |
3495 | u64 size_to_free; | |
12907fc7 | 3496 | u64 chunk_type; |
f43ffb60 | 3497 | struct btrfs_chunk *chunk; |
5a488b9d | 3498 | struct btrfs_path *path = NULL; |
ec44a35c | 3499 | struct btrfs_key key; |
ec44a35c | 3500 | struct btrfs_key found_key; |
c9e9f97b | 3501 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3502 | struct extent_buffer *leaf; |
3503 | int slot; | |
c9e9f97b ID |
3504 | int ret; |
3505 | int enospc_errors = 0; | |
19a39dce | 3506 | bool counting = true; |
12907fc7 | 3507 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3508 | u64 limit_data = bctl->data.limit; |
3509 | u64 limit_meta = bctl->meta.limit; | |
3510 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3511 | u32 count_data = 0; |
3512 | u32 count_meta = 0; | |
3513 | u32 count_sys = 0; | |
2c9fe835 | 3514 | int chunk_reserved = 0; |
ec44a35c | 3515 | |
ec44a35c | 3516 | /* step one make some room on all the devices */ |
c9e9f97b | 3517 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3518 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3519 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3520 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3521 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3522 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3523 | btrfs_device_get_total_bytes(device) - |
3524 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3525 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3526 | continue; |
3527 | ||
3528 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3529 | if (ret == -ENOSPC) |
3530 | break; | |
5a488b9d LB |
3531 | if (ret) { |
3532 | /* btrfs_shrink_device never returns ret > 0 */ | |
3533 | WARN_ON(ret > 0); | |
3534 | goto error; | |
3535 | } | |
ec44a35c | 3536 | |
a22285a6 | 3537 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3538 | if (IS_ERR(trans)) { |
3539 | ret = PTR_ERR(trans); | |
3540 | btrfs_info_in_rcu(fs_info, | |
3541 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3542 | rcu_str_deref(device->name), ret, | |
3543 | old_size, old_size - size_to_free); | |
3544 | goto error; | |
3545 | } | |
ec44a35c CM |
3546 | |
3547 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3548 | if (ret) { |
3a45bb20 | 3549 | btrfs_end_transaction(trans); |
5a488b9d LB |
3550 | /* btrfs_grow_device never returns ret > 0 */ |
3551 | WARN_ON(ret > 0); | |
3552 | btrfs_info_in_rcu(fs_info, | |
3553 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3554 | rcu_str_deref(device->name), ret, | |
3555 | old_size, old_size - size_to_free); | |
3556 | goto error; | |
3557 | } | |
ec44a35c | 3558 | |
3a45bb20 | 3559 | btrfs_end_transaction(trans); |
ec44a35c CM |
3560 | } |
3561 | ||
3562 | /* step two, relocate all the chunks */ | |
3563 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3564 | if (!path) { |
3565 | ret = -ENOMEM; | |
3566 | goto error; | |
3567 | } | |
19a39dce ID |
3568 | |
3569 | /* zero out stat counters */ | |
3570 | spin_lock(&fs_info->balance_lock); | |
3571 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3572 | spin_unlock(&fs_info->balance_lock); | |
3573 | again: | |
7d824b6f | 3574 | if (!counting) { |
12907fc7 DS |
3575 | /* |
3576 | * The single value limit and min/max limits use the same bytes | |
3577 | * in the | |
3578 | */ | |
7d824b6f DS |
3579 | bctl->data.limit = limit_data; |
3580 | bctl->meta.limit = limit_meta; | |
3581 | bctl->sys.limit = limit_sys; | |
3582 | } | |
ec44a35c CM |
3583 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3584 | key.offset = (u64)-1; | |
3585 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3586 | ||
d397712b | 3587 | while (1) { |
19a39dce | 3588 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3589 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3590 | ret = -ECANCELED; |
3591 | goto error; | |
3592 | } | |
3593 | ||
67c5e7d4 | 3594 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3595 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3596 | if (ret < 0) { |
3597 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3598 | goto error; |
67c5e7d4 | 3599 | } |
ec44a35c CM |
3600 | |
3601 | /* | |
3602 | * this shouldn't happen, it means the last relocate | |
3603 | * failed | |
3604 | */ | |
3605 | if (ret == 0) | |
c9e9f97b | 3606 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3607 | |
3608 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3609 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3610 | if (ret) { |
67c5e7d4 | 3611 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3612 | ret = 0; |
ec44a35c | 3613 | break; |
c9e9f97b | 3614 | } |
7d9eb12c | 3615 | |
f43ffb60 ID |
3616 | leaf = path->nodes[0]; |
3617 | slot = path->slots[0]; | |
3618 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3619 | |
67c5e7d4 FM |
3620 | if (found_key.objectid != key.objectid) { |
3621 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3622 | break; |
67c5e7d4 | 3623 | } |
7d9eb12c | 3624 | |
f43ffb60 | 3625 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3626 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3627 | |
19a39dce ID |
3628 | if (!counting) { |
3629 | spin_lock(&fs_info->balance_lock); | |
3630 | bctl->stat.considered++; | |
3631 | spin_unlock(&fs_info->balance_lock); | |
3632 | } | |
3633 | ||
2ff7e61e | 3634 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3635 | found_key.offset); |
2c9fe835 | 3636 | |
b3b4aa74 | 3637 | btrfs_release_path(path); |
67c5e7d4 FM |
3638 | if (!ret) { |
3639 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3640 | goto loop; |
67c5e7d4 | 3641 | } |
f43ffb60 | 3642 | |
19a39dce | 3643 | if (counting) { |
67c5e7d4 | 3644 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3645 | spin_lock(&fs_info->balance_lock); |
3646 | bctl->stat.expected++; | |
3647 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3648 | |
3649 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3650 | count_data++; | |
3651 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3652 | count_sys++; | |
3653 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3654 | count_meta++; | |
3655 | ||
3656 | goto loop; | |
3657 | } | |
3658 | ||
3659 | /* | |
3660 | * Apply limit_min filter, no need to check if the LIMITS | |
3661 | * filter is used, limit_min is 0 by default | |
3662 | */ | |
3663 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3664 | count_data < bctl->data.limit_min) | |
3665 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3666 | count_meta < bctl->meta.limit_min) | |
3667 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3668 | count_sys < bctl->sys.limit_min)) { | |
3669 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3670 | goto loop; |
3671 | } | |
3672 | ||
a6f93c71 LB |
3673 | if (!chunk_reserved) { |
3674 | /* | |
3675 | * We may be relocating the only data chunk we have, | |
3676 | * which could potentially end up with losing data's | |
3677 | * raid profile, so lets allocate an empty one in | |
3678 | * advance. | |
3679 | */ | |
3680 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3681 | found_key.offset); | |
2c9fe835 ZL |
3682 | if (ret < 0) { |
3683 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3684 | goto error; | |
a6f93c71 LB |
3685 | } else if (ret == 1) { |
3686 | chunk_reserved = 1; | |
2c9fe835 | 3687 | } |
2c9fe835 ZL |
3688 | } |
3689 | ||
5b4aacef | 3690 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3691 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3692 | if (ret == -ENOSPC) { |
c9e9f97b | 3693 | enospc_errors++; |
eede2bf3 OS |
3694 | } else if (ret == -ETXTBSY) { |
3695 | btrfs_info(fs_info, | |
3696 | "skipping relocation of block group %llu due to active swapfile", | |
3697 | found_key.offset); | |
3698 | ret = 0; | |
3699 | } else if (ret) { | |
3700 | goto error; | |
19a39dce ID |
3701 | } else { |
3702 | spin_lock(&fs_info->balance_lock); | |
3703 | bctl->stat.completed++; | |
3704 | spin_unlock(&fs_info->balance_lock); | |
3705 | } | |
f43ffb60 | 3706 | loop: |
795a3321 ID |
3707 | if (found_key.offset == 0) |
3708 | break; | |
ba1bf481 | 3709 | key.offset = found_key.offset - 1; |
ec44a35c | 3710 | } |
c9e9f97b | 3711 | |
19a39dce ID |
3712 | if (counting) { |
3713 | btrfs_release_path(path); | |
3714 | counting = false; | |
3715 | goto again; | |
3716 | } | |
ec44a35c CM |
3717 | error: |
3718 | btrfs_free_path(path); | |
c9e9f97b | 3719 | if (enospc_errors) { |
efe120a0 | 3720 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3721 | enospc_errors); |
c9e9f97b ID |
3722 | if (!ret) |
3723 | ret = -ENOSPC; | |
3724 | } | |
3725 | ||
ec44a35c CM |
3726 | return ret; |
3727 | } | |
3728 | ||
0c460c0d ID |
3729 | /** |
3730 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3731 | * @flags: profile to validate | |
3732 | * @extended: if true @flags is treated as an extended profile | |
3733 | */ | |
3734 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3735 | { | |
3736 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3737 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3738 | ||
3739 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3740 | ||
3741 | /* 1) check that all other bits are zeroed */ | |
3742 | if (flags & ~mask) | |
3743 | return 0; | |
3744 | ||
3745 | /* 2) see if profile is reduced */ | |
3746 | if (flags == 0) | |
3747 | return !extended; /* "0" is valid for usual profiles */ | |
3748 | ||
3749 | /* true if exactly one bit set */ | |
818255fe | 3750 | return is_power_of_2(flags); |
0c460c0d ID |
3751 | } |
3752 | ||
837d5b6e ID |
3753 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3754 | { | |
a7e99c69 ID |
3755 | /* cancel requested || normal exit path */ |
3756 | return atomic_read(&fs_info->balance_cancel_req) || | |
3757 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3758 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3759 | } |
3760 | ||
bdcd3c97 AM |
3761 | /* Non-zero return value signifies invalidity */ |
3762 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3763 | u64 allowed) | |
3764 | { | |
3765 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3766 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3767 | (bctl_arg->target & ~allowed))); | |
3768 | } | |
3769 | ||
c9e9f97b | 3770 | /* |
dccdb07b | 3771 | * Should be called with balance mutexe held |
c9e9f97b | 3772 | */ |
6fcf6e2b DS |
3773 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
3774 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
3775 | struct btrfs_ioctl_balance_args *bargs) |
3776 | { | |
14506127 | 3777 | u64 meta_target, data_target; |
f43ffb60 | 3778 | u64 allowed; |
e4837f8f | 3779 | int mixed = 0; |
c9e9f97b | 3780 | int ret; |
8dabb742 | 3781 | u64 num_devices; |
de98ced9 | 3782 | unsigned seq; |
c9e9f97b | 3783 | |
837d5b6e | 3784 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3785 | atomic_read(&fs_info->balance_pause_req) || |
3786 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3787 | ret = -EINVAL; |
3788 | goto out; | |
3789 | } | |
3790 | ||
e4837f8f ID |
3791 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3792 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3793 | mixed = 1; | |
3794 | ||
f43ffb60 ID |
3795 | /* |
3796 | * In case of mixed groups both data and meta should be picked, | |
3797 | * and identical options should be given for both of them. | |
3798 | */ | |
e4837f8f ID |
3799 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3800 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3801 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3802 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3803 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 3804 | btrfs_err(fs_info, |
6dac13f8 | 3805 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
3806 | ret = -EINVAL; |
3807 | goto out; | |
3808 | } | |
3809 | } | |
3810 | ||
1da73967 AJ |
3811 | num_devices = btrfs_num_devices(fs_info); |
3812 | ||
88be159c AH |
3813 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3814 | if (num_devices > 1) | |
e4d8ec0f | 3815 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3816 | if (num_devices > 2) |
3817 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3818 | if (num_devices > 3) | |
3819 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3820 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3821 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
3822 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
3823 | ||
5d163e0e | 3824 | btrfs_err(fs_info, |
6dac13f8 AJ |
3825 | "balance: invalid convert data profile %s", |
3826 | get_raid_name(index)); | |
e4d8ec0f ID |
3827 | ret = -EINVAL; |
3828 | goto out; | |
3829 | } | |
bdcd3c97 | 3830 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
3831 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
3832 | ||
efe120a0 | 3833 | btrfs_err(fs_info, |
6dac13f8 AJ |
3834 | "balance: invalid convert metadata profile %s", |
3835 | get_raid_name(index)); | |
e4d8ec0f ID |
3836 | ret = -EINVAL; |
3837 | goto out; | |
3838 | } | |
bdcd3c97 | 3839 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
3840 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
3841 | ||
efe120a0 | 3842 | btrfs_err(fs_info, |
6dac13f8 AJ |
3843 | "balance: invalid convert system profile %s", |
3844 | get_raid_name(index)); | |
e4d8ec0f ID |
3845 | ret = -EINVAL; |
3846 | goto out; | |
3847 | } | |
3848 | ||
e4d8ec0f ID |
3849 | /* allow to reduce meta or sys integrity only if force set */ |
3850 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3851 | BTRFS_BLOCK_GROUP_RAID10 | |
3852 | BTRFS_BLOCK_GROUP_RAID5 | | |
3853 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3854 | do { |
3855 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3856 | ||
3857 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3858 | (fs_info->avail_system_alloc_bits & allowed) && | |
3859 | !(bctl->sys.target & allowed)) || | |
3860 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3861 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3862 | !(bctl->meta.target & allowed))) { | |
3863 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e | 3864 | btrfs_info(fs_info, |
6dac13f8 | 3865 | "balance: force reducing metadata integrity"); |
de98ced9 | 3866 | } else { |
5d163e0e | 3867 | btrfs_err(fs_info, |
6dac13f8 | 3868 | "balance: reduces metadata integrity, use --force if you want this"); |
de98ced9 MX |
3869 | ret = -EINVAL; |
3870 | goto out; | |
3871 | } | |
e4d8ec0f | 3872 | } |
de98ced9 | 3873 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3874 | |
14506127 AB |
3875 | /* if we're not converting, the target field is uninitialized */ |
3876 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3877 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3878 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3879 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
3880 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
3881 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
3882 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
3883 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
3884 | ||
ee592d07 | 3885 | btrfs_warn(fs_info, |
6dac13f8 AJ |
3886 | "balance: metadata profile %s has lower redundancy than data profile %s", |
3887 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
3888 | } |
3889 | ||
6bccf3ab | 3890 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3891 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3892 | goto out; |
3893 | ||
59641015 ID |
3894 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3895 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
3896 | BUG_ON(fs_info->balance_ctl); |
3897 | spin_lock(&fs_info->balance_lock); | |
3898 | fs_info->balance_ctl = bctl; | |
3899 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
3900 | } else { |
3901 | BUG_ON(ret != -EEXIST); | |
3902 | spin_lock(&fs_info->balance_lock); | |
3903 | update_balance_args(bctl); | |
3904 | spin_unlock(&fs_info->balance_lock); | |
3905 | } | |
c9e9f97b | 3906 | |
3009a62f DS |
3907 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
3908 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
c9e9f97b ID |
3909 | mutex_unlock(&fs_info->balance_mutex); |
3910 | ||
3911 | ret = __btrfs_balance(fs_info); | |
3912 | ||
3913 | mutex_lock(&fs_info->balance_mutex); | |
3009a62f | 3914 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
3915 | |
3916 | if (bargs) { | |
3917 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 3918 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
3919 | } |
3920 | ||
3a01aa7a ID |
3921 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3922 | balance_need_close(fs_info)) { | |
149196a2 | 3923 | reset_balance_state(fs_info); |
a17c95df | 3924 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
3925 | } |
3926 | ||
837d5b6e | 3927 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3928 | |
3929 | return ret; | |
3930 | out: | |
59641015 | 3931 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 3932 | reset_balance_state(fs_info); |
a17c95df | 3933 | else |
59641015 | 3934 | kfree(bctl); |
a17c95df DS |
3935 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3936 | ||
59641015 ID |
3937 | return ret; |
3938 | } | |
3939 | ||
3940 | static int balance_kthread(void *data) | |
3941 | { | |
2b6ba629 | 3942 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3943 | int ret = 0; |
59641015 | 3944 | |
59641015 | 3945 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 3946 | if (fs_info->balance_ctl) { |
6dac13f8 | 3947 | btrfs_info(fs_info, "balance: resuming"); |
6fcf6e2b | 3948 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
9555c6c1 | 3949 | } |
59641015 | 3950 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 3951 | |
59641015 ID |
3952 | return ret; |
3953 | } | |
3954 | ||
2b6ba629 ID |
3955 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3956 | { | |
3957 | struct task_struct *tsk; | |
3958 | ||
1354e1a1 | 3959 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 3960 | if (!fs_info->balance_ctl) { |
1354e1a1 | 3961 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
3962 | return 0; |
3963 | } | |
1354e1a1 | 3964 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 3965 | |
3cdde224 | 3966 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 3967 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
3968 | return 0; |
3969 | } | |
3970 | ||
02ee654d AJ |
3971 | /* |
3972 | * A ro->rw remount sequence should continue with the paused balance | |
3973 | * regardless of who pauses it, system or the user as of now, so set | |
3974 | * the resume flag. | |
3975 | */ | |
3976 | spin_lock(&fs_info->balance_lock); | |
3977 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
3978 | spin_unlock(&fs_info->balance_lock); | |
3979 | ||
2b6ba629 | 3980 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 3981 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3982 | } |
3983 | ||
68310a5e | 3984 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3985 | { |
59641015 ID |
3986 | struct btrfs_balance_control *bctl; |
3987 | struct btrfs_balance_item *item; | |
3988 | struct btrfs_disk_balance_args disk_bargs; | |
3989 | struct btrfs_path *path; | |
3990 | struct extent_buffer *leaf; | |
3991 | struct btrfs_key key; | |
3992 | int ret; | |
3993 | ||
3994 | path = btrfs_alloc_path(); | |
3995 | if (!path) | |
3996 | return -ENOMEM; | |
3997 | ||
59641015 | 3998 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3999 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4000 | key.offset = 0; |
4001 | ||
68310a5e | 4002 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4003 | if (ret < 0) |
68310a5e | 4004 | goto out; |
59641015 ID |
4005 | if (ret > 0) { /* ret = -ENOENT; */ |
4006 | ret = 0; | |
68310a5e ID |
4007 | goto out; |
4008 | } | |
4009 | ||
4010 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4011 | if (!bctl) { | |
4012 | ret = -ENOMEM; | |
4013 | goto out; | |
59641015 ID |
4014 | } |
4015 | ||
4016 | leaf = path->nodes[0]; | |
4017 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4018 | ||
68310a5e ID |
4019 | bctl->flags = btrfs_balance_flags(leaf, item); |
4020 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4021 | |
4022 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4023 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4024 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4025 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4026 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4027 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4028 | ||
eee95e3f DS |
4029 | /* |
4030 | * This should never happen, as the paused balance state is recovered | |
4031 | * during mount without any chance of other exclusive ops to collide. | |
4032 | * | |
4033 | * This gives the exclusive op status to balance and keeps in paused | |
4034 | * state until user intervention (cancel or umount). If the ownership | |
4035 | * cannot be assigned, show a message but do not fail. The balance | |
4036 | * is in a paused state and must have fs_info::balance_ctl properly | |
4037 | * set up. | |
4038 | */ | |
4039 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4040 | btrfs_warn(fs_info, | |
6dac13f8 | 4041 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4042 | |
68310a5e | 4043 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4044 | BUG_ON(fs_info->balance_ctl); |
4045 | spin_lock(&fs_info->balance_lock); | |
4046 | fs_info->balance_ctl = bctl; | |
4047 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4048 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4049 | out: |
4050 | btrfs_free_path(path); | |
ec44a35c CM |
4051 | return ret; |
4052 | } | |
4053 | ||
837d5b6e ID |
4054 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4055 | { | |
4056 | int ret = 0; | |
4057 | ||
4058 | mutex_lock(&fs_info->balance_mutex); | |
4059 | if (!fs_info->balance_ctl) { | |
4060 | mutex_unlock(&fs_info->balance_mutex); | |
4061 | return -ENOTCONN; | |
4062 | } | |
4063 | ||
3009a62f | 4064 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4065 | atomic_inc(&fs_info->balance_pause_req); |
4066 | mutex_unlock(&fs_info->balance_mutex); | |
4067 | ||
4068 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4069 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4070 | |
4071 | mutex_lock(&fs_info->balance_mutex); | |
4072 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4073 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4074 | atomic_dec(&fs_info->balance_pause_req); |
4075 | } else { | |
4076 | ret = -ENOTCONN; | |
4077 | } | |
4078 | ||
4079 | mutex_unlock(&fs_info->balance_mutex); | |
4080 | return ret; | |
4081 | } | |
4082 | ||
a7e99c69 ID |
4083 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4084 | { | |
4085 | mutex_lock(&fs_info->balance_mutex); | |
4086 | if (!fs_info->balance_ctl) { | |
4087 | mutex_unlock(&fs_info->balance_mutex); | |
4088 | return -ENOTCONN; | |
4089 | } | |
4090 | ||
cf7d20f4 DS |
4091 | /* |
4092 | * A paused balance with the item stored on disk can be resumed at | |
4093 | * mount time if the mount is read-write. Otherwise it's still paused | |
4094 | * and we must not allow cancelling as it deletes the item. | |
4095 | */ | |
4096 | if (sb_rdonly(fs_info->sb)) { | |
4097 | mutex_unlock(&fs_info->balance_mutex); | |
4098 | return -EROFS; | |
4099 | } | |
4100 | ||
a7e99c69 ID |
4101 | atomic_inc(&fs_info->balance_cancel_req); |
4102 | /* | |
4103 | * if we are running just wait and return, balance item is | |
4104 | * deleted in btrfs_balance in this case | |
4105 | */ | |
3009a62f | 4106 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4107 | mutex_unlock(&fs_info->balance_mutex); |
4108 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4109 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4110 | mutex_lock(&fs_info->balance_mutex); |
4111 | } else { | |
a7e99c69 | 4112 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4113 | /* |
4114 | * Lock released to allow other waiters to continue, we'll | |
4115 | * reexamine the status again. | |
4116 | */ | |
a7e99c69 ID |
4117 | mutex_lock(&fs_info->balance_mutex); |
4118 | ||
a17c95df | 4119 | if (fs_info->balance_ctl) { |
149196a2 | 4120 | reset_balance_state(fs_info); |
a17c95df | 4121 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4122 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4123 | } |
a7e99c69 ID |
4124 | } |
4125 | ||
3009a62f DS |
4126 | BUG_ON(fs_info->balance_ctl || |
4127 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4128 | atomic_dec(&fs_info->balance_cancel_req); |
4129 | mutex_unlock(&fs_info->balance_mutex); | |
4130 | return 0; | |
4131 | } | |
4132 | ||
803b2f54 SB |
4133 | static int btrfs_uuid_scan_kthread(void *data) |
4134 | { | |
4135 | struct btrfs_fs_info *fs_info = data; | |
4136 | struct btrfs_root *root = fs_info->tree_root; | |
4137 | struct btrfs_key key; | |
803b2f54 SB |
4138 | struct btrfs_path *path = NULL; |
4139 | int ret = 0; | |
4140 | struct extent_buffer *eb; | |
4141 | int slot; | |
4142 | struct btrfs_root_item root_item; | |
4143 | u32 item_size; | |
f45388f3 | 4144 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4145 | |
4146 | path = btrfs_alloc_path(); | |
4147 | if (!path) { | |
4148 | ret = -ENOMEM; | |
4149 | goto out; | |
4150 | } | |
4151 | ||
4152 | key.objectid = 0; | |
4153 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4154 | key.offset = 0; | |
4155 | ||
803b2f54 | 4156 | while (1) { |
7c829b72 AJ |
4157 | ret = btrfs_search_forward(root, &key, path, |
4158 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4159 | if (ret) { |
4160 | if (ret > 0) | |
4161 | ret = 0; | |
4162 | break; | |
4163 | } | |
4164 | ||
4165 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4166 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4167 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4168 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4169 | goto skip; | |
4170 | ||
4171 | eb = path->nodes[0]; | |
4172 | slot = path->slots[0]; | |
4173 | item_size = btrfs_item_size_nr(eb, slot); | |
4174 | if (item_size < sizeof(root_item)) | |
4175 | goto skip; | |
4176 | ||
803b2f54 SB |
4177 | read_extent_buffer(eb, &root_item, |
4178 | btrfs_item_ptr_offset(eb, slot), | |
4179 | (int)sizeof(root_item)); | |
4180 | if (btrfs_root_refs(&root_item) == 0) | |
4181 | goto skip; | |
f45388f3 FDBM |
4182 | |
4183 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4184 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4185 | if (trans) | |
4186 | goto update_tree; | |
4187 | ||
4188 | btrfs_release_path(path); | |
803b2f54 SB |
4189 | /* |
4190 | * 1 - subvol uuid item | |
4191 | * 1 - received_subvol uuid item | |
4192 | */ | |
4193 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4194 | if (IS_ERR(trans)) { | |
4195 | ret = PTR_ERR(trans); | |
4196 | break; | |
4197 | } | |
f45388f3 FDBM |
4198 | continue; |
4199 | } else { | |
4200 | goto skip; | |
4201 | } | |
4202 | update_tree: | |
4203 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4204 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4205 | BTRFS_UUID_KEY_SUBVOL, |
4206 | key.objectid); | |
4207 | if (ret < 0) { | |
efe120a0 | 4208 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4209 | ret); |
803b2f54 SB |
4210 | break; |
4211 | } | |
4212 | } | |
4213 | ||
4214 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4215 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4216 | root_item.received_uuid, |
4217 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4218 | key.objectid); | |
4219 | if (ret < 0) { | |
efe120a0 | 4220 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4221 | ret); |
803b2f54 SB |
4222 | break; |
4223 | } | |
4224 | } | |
4225 | ||
f45388f3 | 4226 | skip: |
803b2f54 | 4227 | if (trans) { |
3a45bb20 | 4228 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4229 | trans = NULL; |
803b2f54 SB |
4230 | if (ret) |
4231 | break; | |
4232 | } | |
4233 | ||
803b2f54 SB |
4234 | btrfs_release_path(path); |
4235 | if (key.offset < (u64)-1) { | |
4236 | key.offset++; | |
4237 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4238 | key.offset = 0; | |
4239 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4240 | } else if (key.objectid < (u64)-1) { | |
4241 | key.offset = 0; | |
4242 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4243 | key.objectid++; | |
4244 | } else { | |
4245 | break; | |
4246 | } | |
4247 | cond_resched(); | |
4248 | } | |
4249 | ||
4250 | out: | |
4251 | btrfs_free_path(path); | |
f45388f3 | 4252 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4253 | btrfs_end_transaction(trans); |
803b2f54 | 4254 | if (ret) |
efe120a0 | 4255 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4256 | else |
afcdd129 | 4257 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4258 | up(&fs_info->uuid_tree_rescan_sem); |
4259 | return 0; | |
4260 | } | |
4261 | ||
70f80175 SB |
4262 | /* |
4263 | * Callback for btrfs_uuid_tree_iterate(). | |
4264 | * returns: | |
4265 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4266 | * < 0 if an error occurred. |
70f80175 SB |
4267 | * > 0 if the check failed, which means the caller shall remove the entry. |
4268 | */ | |
4269 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4270 | u8 *uuid, u8 type, u64 subid) | |
4271 | { | |
4272 | struct btrfs_key key; | |
4273 | int ret = 0; | |
4274 | struct btrfs_root *subvol_root; | |
4275 | ||
4276 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4277 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4278 | goto out; | |
4279 | ||
4280 | key.objectid = subid; | |
4281 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4282 | key.offset = (u64)-1; | |
4283 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4284 | if (IS_ERR(subvol_root)) { | |
4285 | ret = PTR_ERR(subvol_root); | |
4286 | if (ret == -ENOENT) | |
4287 | ret = 1; | |
4288 | goto out; | |
4289 | } | |
4290 | ||
4291 | switch (type) { | |
4292 | case BTRFS_UUID_KEY_SUBVOL: | |
4293 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4294 | ret = 1; | |
4295 | break; | |
4296 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4297 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4298 | BTRFS_UUID_SIZE)) | |
4299 | ret = 1; | |
4300 | break; | |
4301 | } | |
4302 | ||
4303 | out: | |
4304 | return ret; | |
4305 | } | |
4306 | ||
4307 | static int btrfs_uuid_rescan_kthread(void *data) | |
4308 | { | |
4309 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4310 | int ret; | |
4311 | ||
4312 | /* | |
4313 | * 1st step is to iterate through the existing UUID tree and | |
4314 | * to delete all entries that contain outdated data. | |
4315 | * 2nd step is to add all missing entries to the UUID tree. | |
4316 | */ | |
4317 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4318 | if (ret < 0) { | |
efe120a0 | 4319 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4320 | up(&fs_info->uuid_tree_rescan_sem); |
4321 | return ret; | |
4322 | } | |
4323 | return btrfs_uuid_scan_kthread(data); | |
4324 | } | |
4325 | ||
f7a81ea4 SB |
4326 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4327 | { | |
4328 | struct btrfs_trans_handle *trans; | |
4329 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4330 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4331 | struct task_struct *task; |
4332 | int ret; | |
f7a81ea4 SB |
4333 | |
4334 | /* | |
4335 | * 1 - root node | |
4336 | * 1 - root item | |
4337 | */ | |
4338 | trans = btrfs_start_transaction(tree_root, 2); | |
4339 | if (IS_ERR(trans)) | |
4340 | return PTR_ERR(trans); | |
4341 | ||
4342 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4343 | BTRFS_UUID_TREE_OBJECTID); | |
4344 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4345 | ret = PTR_ERR(uuid_root); |
66642832 | 4346 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4347 | btrfs_end_transaction(trans); |
6d13f549 | 4348 | return ret; |
f7a81ea4 SB |
4349 | } |
4350 | ||
4351 | fs_info->uuid_root = uuid_root; | |
4352 | ||
3a45bb20 | 4353 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4354 | if (ret) |
4355 | return ret; | |
4356 | ||
4357 | down(&fs_info->uuid_tree_rescan_sem); | |
4358 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4359 | if (IS_ERR(task)) { | |
70f80175 | 4360 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4361 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4362 | up(&fs_info->uuid_tree_rescan_sem); |
4363 | return PTR_ERR(task); | |
4364 | } | |
4365 | ||
4366 | return 0; | |
f7a81ea4 | 4367 | } |
803b2f54 | 4368 | |
70f80175 SB |
4369 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4370 | { | |
4371 | struct task_struct *task; | |
4372 | ||
4373 | down(&fs_info->uuid_tree_rescan_sem); | |
4374 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4375 | if (IS_ERR(task)) { | |
4376 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4377 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4378 | up(&fs_info->uuid_tree_rescan_sem); |
4379 | return PTR_ERR(task); | |
4380 | } | |
4381 | ||
4382 | return 0; | |
4383 | } | |
4384 | ||
8f18cf13 CM |
4385 | /* |
4386 | * shrinking a device means finding all of the device extents past | |
4387 | * the new size, and then following the back refs to the chunks. | |
4388 | * The chunk relocation code actually frees the device extent | |
4389 | */ | |
4390 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4391 | { | |
0b246afa JM |
4392 | struct btrfs_fs_info *fs_info = device->fs_info; |
4393 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4394 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4395 | struct btrfs_dev_extent *dev_extent = NULL; |
4396 | struct btrfs_path *path; | |
4397 | u64 length; | |
8f18cf13 CM |
4398 | u64 chunk_offset; |
4399 | int ret; | |
4400 | int slot; | |
ba1bf481 JB |
4401 | int failed = 0; |
4402 | bool retried = false; | |
53e489bc | 4403 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4404 | struct extent_buffer *l; |
4405 | struct btrfs_key key; | |
0b246afa | 4406 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4407 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4408 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4409 | u64 diff; |
4410 | ||
4411 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4412 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4413 | |
401e29c1 | 4414 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4415 | return -EINVAL; |
4416 | ||
8f18cf13 CM |
4417 | path = btrfs_alloc_path(); |
4418 | if (!path) | |
4419 | return -ENOMEM; | |
4420 | ||
0338dff6 | 4421 | path->reada = READA_BACK; |
8f18cf13 | 4422 | |
34441361 | 4423 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4424 | |
7cc8e58d | 4425 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4426 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4427 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4428 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4429 | } |
34441361 | 4430 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4431 | |
ba1bf481 | 4432 | again: |
8f18cf13 CM |
4433 | key.objectid = device->devid; |
4434 | key.offset = (u64)-1; | |
4435 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4436 | ||
213e64da | 4437 | do { |
0b246afa | 4438 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4439 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4440 | if (ret < 0) { |
0b246afa | 4441 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4442 | goto done; |
67c5e7d4 | 4443 | } |
8f18cf13 CM |
4444 | |
4445 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4446 | if (ret) |
0b246afa | 4447 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4448 | if (ret < 0) |
4449 | goto done; | |
4450 | if (ret) { | |
4451 | ret = 0; | |
b3b4aa74 | 4452 | btrfs_release_path(path); |
bf1fb512 | 4453 | break; |
8f18cf13 CM |
4454 | } |
4455 | ||
4456 | l = path->nodes[0]; | |
4457 | slot = path->slots[0]; | |
4458 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4459 | ||
ba1bf481 | 4460 | if (key.objectid != device->devid) { |
0b246afa | 4461 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4462 | btrfs_release_path(path); |
bf1fb512 | 4463 | break; |
ba1bf481 | 4464 | } |
8f18cf13 CM |
4465 | |
4466 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4467 | length = btrfs_dev_extent_length(l, dev_extent); | |
4468 | ||
ba1bf481 | 4469 | if (key.offset + length <= new_size) { |
0b246afa | 4470 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4471 | btrfs_release_path(path); |
d6397bae | 4472 | break; |
ba1bf481 | 4473 | } |
8f18cf13 | 4474 | |
8f18cf13 | 4475 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4476 | btrfs_release_path(path); |
8f18cf13 | 4477 | |
a6f93c71 LB |
4478 | /* |
4479 | * We may be relocating the only data chunk we have, | |
4480 | * which could potentially end up with losing data's | |
4481 | * raid profile, so lets allocate an empty one in | |
4482 | * advance. | |
4483 | */ | |
4484 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4485 | if (ret < 0) { | |
4486 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4487 | goto done; | |
4488 | } | |
4489 | ||
0b246afa JM |
4490 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4491 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4492 | if (ret == -ENOSPC) { |
ba1bf481 | 4493 | failed++; |
eede2bf3 OS |
4494 | } else if (ret) { |
4495 | if (ret == -ETXTBSY) { | |
4496 | btrfs_warn(fs_info, | |
4497 | "could not shrink block group %llu due to active swapfile", | |
4498 | chunk_offset); | |
4499 | } | |
4500 | goto done; | |
4501 | } | |
213e64da | 4502 | } while (key.offset-- > 0); |
ba1bf481 JB |
4503 | |
4504 | if (failed && !retried) { | |
4505 | failed = 0; | |
4506 | retried = true; | |
4507 | goto again; | |
4508 | } else if (failed && retried) { | |
4509 | ret = -ENOSPC; | |
ba1bf481 | 4510 | goto done; |
8f18cf13 CM |
4511 | } |
4512 | ||
d6397bae | 4513 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4514 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4515 | if (IS_ERR(trans)) { |
4516 | ret = PTR_ERR(trans); | |
4517 | goto done; | |
4518 | } | |
4519 | ||
34441361 | 4520 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4521 | |
4522 | /* | |
4523 | * We checked in the above loop all device extents that were already in | |
4524 | * the device tree. However before we have updated the device's | |
4525 | * total_bytes to the new size, we might have had chunk allocations that | |
4526 | * have not complete yet (new block groups attached to transaction | |
4527 | * handles), and therefore their device extents were not yet in the | |
4528 | * device tree and we missed them in the loop above. So if we have any | |
4529 | * pending chunk using a device extent that overlaps the device range | |
4530 | * that we can not use anymore, commit the current transaction and | |
4531 | * repeat the search on the device tree - this way we guarantee we will | |
4532 | * not have chunks using device extents that end beyond 'new_size'. | |
4533 | */ | |
4534 | if (!checked_pending_chunks) { | |
4535 | u64 start = new_size; | |
4536 | u64 len = old_size - new_size; | |
4537 | ||
499f377f JM |
4538 | if (contains_pending_extent(trans->transaction, device, |
4539 | &start, len)) { | |
34441361 | 4540 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4541 | checked_pending_chunks = true; |
4542 | failed = 0; | |
4543 | retried = false; | |
3a45bb20 | 4544 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4545 | if (ret) |
4546 | goto done; | |
4547 | goto again; | |
4548 | } | |
4549 | } | |
4550 | ||
7cc8e58d | 4551 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4552 | if (list_empty(&device->resized_list)) |
4553 | list_add_tail(&device->resized_list, | |
0b246afa | 4554 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4555 | |
d6397bae | 4556 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4557 | btrfs_set_super_total_bytes(super_copy, |
4558 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4559 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4560 | |
4561 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4562 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4563 | if (ret < 0) { |
4564 | btrfs_abort_transaction(trans, ret); | |
4565 | btrfs_end_transaction(trans); | |
4566 | } else { | |
4567 | ret = btrfs_commit_transaction(trans); | |
4568 | } | |
8f18cf13 CM |
4569 | done: |
4570 | btrfs_free_path(path); | |
53e489bc | 4571 | if (ret) { |
34441361 | 4572 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4573 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4574 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4575 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4576 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4577 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4578 | } |
8f18cf13 CM |
4579 | return ret; |
4580 | } | |
4581 | ||
2ff7e61e | 4582 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4583 | struct btrfs_key *key, |
4584 | struct btrfs_chunk *chunk, int item_size) | |
4585 | { | |
0b246afa | 4586 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4587 | struct btrfs_disk_key disk_key; |
4588 | u32 array_size; | |
4589 | u8 *ptr; | |
4590 | ||
34441361 | 4591 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4592 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4593 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4594 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4595 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4596 | return -EFBIG; |
fe48a5c0 | 4597 | } |
0b86a832 CM |
4598 | |
4599 | ptr = super_copy->sys_chunk_array + array_size; | |
4600 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4601 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4602 | ptr += sizeof(disk_key); | |
4603 | memcpy(ptr, chunk, item_size); | |
4604 | item_size += sizeof(disk_key); | |
4605 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4606 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4607 | |
0b86a832 CM |
4608 | return 0; |
4609 | } | |
4610 | ||
73c5de00 AJ |
4611 | /* |
4612 | * sort the devices in descending order by max_avail, total_avail | |
4613 | */ | |
4614 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4615 | { |
73c5de00 AJ |
4616 | const struct btrfs_device_info *di_a = a; |
4617 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4618 | |
73c5de00 | 4619 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4620 | return -1; |
73c5de00 | 4621 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4622 | return 1; |
73c5de00 AJ |
4623 | if (di_a->total_avail > di_b->total_avail) |
4624 | return -1; | |
4625 | if (di_a->total_avail < di_b->total_avail) | |
4626 | return 1; | |
4627 | return 0; | |
b2117a39 | 4628 | } |
0b86a832 | 4629 | |
53b381b3 DW |
4630 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4631 | { | |
ffe2d203 | 4632 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4633 | return; |
4634 | ||
ceda0864 | 4635 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4636 | } |
4637 | ||
062d4d1f | 4638 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
4639 | - sizeof(struct btrfs_chunk)) \ |
4640 | / sizeof(struct btrfs_stripe) + 1) | |
4641 | ||
4642 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4643 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4644 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4645 | / sizeof(struct btrfs_stripe) + 1) | |
4646 | ||
73c5de00 | 4647 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4648 | u64 start, u64 type) |
b2117a39 | 4649 | { |
2ff7e61e | 4650 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4651 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4652 | struct btrfs_device *device; |
73c5de00 AJ |
4653 | struct map_lookup *map = NULL; |
4654 | struct extent_map_tree *em_tree; | |
4655 | struct extent_map *em; | |
4656 | struct btrfs_device_info *devices_info = NULL; | |
4657 | u64 total_avail; | |
4658 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4659 | int data_stripes; /* number of stripes that count for |
4660 | block group size */ | |
73c5de00 AJ |
4661 | int sub_stripes; /* sub_stripes info for map */ |
4662 | int dev_stripes; /* stripes per dev */ | |
4663 | int devs_max; /* max devs to use */ | |
4664 | int devs_min; /* min devs needed */ | |
4665 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4666 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4667 | int nparity; /* number of stripes worth of bytes to |
4668 | store parity information */ | |
73c5de00 AJ |
4669 | int ret; |
4670 | u64 max_stripe_size; | |
4671 | u64 max_chunk_size; | |
4672 | u64 stripe_size; | |
23f0ff1e | 4673 | u64 chunk_size; |
73c5de00 AJ |
4674 | int ndevs; |
4675 | int i; | |
4676 | int j; | |
31e50229 | 4677 | int index; |
593060d7 | 4678 | |
0c460c0d | 4679 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4680 | |
4117f207 QW |
4681 | if (list_empty(&fs_devices->alloc_list)) { |
4682 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4683 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4684 | return -ENOSPC; |
4117f207 | 4685 | } |
b2117a39 | 4686 | |
3e72ee88 | 4687 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4688 | |
31e50229 LB |
4689 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4690 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4691 | devs_max = btrfs_raid_array[index].devs_max; | |
4692 | devs_min = btrfs_raid_array[index].devs_min; | |
4693 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4694 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 4695 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 4696 | |
9b3f68b9 | 4697 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4698 | max_stripe_size = SZ_1G; |
fce466ea | 4699 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 4700 | if (!devs_max) |
062d4d1f | 4701 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 4702 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4703 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4704 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4705 | max_stripe_size = SZ_1G; | |
1100373f | 4706 | else |
ee22184b | 4707 | max_stripe_size = SZ_256M; |
73c5de00 | 4708 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 4709 | if (!devs_max) |
062d4d1f | 4710 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 4711 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4712 | max_stripe_size = SZ_32M; |
73c5de00 | 4713 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4714 | if (!devs_max) |
4715 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4716 | } else { |
351fd353 | 4717 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4718 | type); |
4719 | BUG_ON(1); | |
9b3f68b9 CM |
4720 | } |
4721 | ||
2b82032c YZ |
4722 | /* we don't want a chunk larger than 10% of writeable space */ |
4723 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4724 | max_chunk_size); | |
9b3f68b9 | 4725 | |
31e818fe | 4726 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4727 | GFP_NOFS); |
4728 | if (!devices_info) | |
4729 | return -ENOMEM; | |
0cad8a11 | 4730 | |
9f680ce0 | 4731 | /* |
73c5de00 AJ |
4732 | * in the first pass through the devices list, we gather information |
4733 | * about the available holes on each device. | |
9f680ce0 | 4734 | */ |
73c5de00 | 4735 | ndevs = 0; |
ebcc9301 | 4736 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4737 | u64 max_avail; |
4738 | u64 dev_offset; | |
b2117a39 | 4739 | |
ebbede42 | 4740 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4741 | WARN(1, KERN_ERR |
efe120a0 | 4742 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4743 | continue; |
4744 | } | |
b2117a39 | 4745 | |
e12c9621 AJ |
4746 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4747 | &device->dev_state) || | |
401e29c1 | 4748 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4749 | continue; |
b2117a39 | 4750 | |
73c5de00 AJ |
4751 | if (device->total_bytes > device->bytes_used) |
4752 | total_avail = device->total_bytes - device->bytes_used; | |
4753 | else | |
4754 | total_avail = 0; | |
38c01b96 | 4755 | |
4756 | /* If there is no space on this device, skip it. */ | |
4757 | if (total_avail == 0) | |
4758 | continue; | |
b2117a39 | 4759 | |
6df9a95e | 4760 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4761 | max_stripe_size * dev_stripes, |
4762 | &dev_offset, &max_avail); | |
4763 | if (ret && ret != -ENOSPC) | |
4764 | goto error; | |
b2117a39 | 4765 | |
73c5de00 AJ |
4766 | if (ret == 0) |
4767 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4768 | |
4117f207 QW |
4769 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4770 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4771 | btrfs_debug(info, | |
4772 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4773 | __func__, device->devid, max_avail, | |
4774 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4775 | continue; |
4117f207 | 4776 | } |
b2117a39 | 4777 | |
063d006f ES |
4778 | if (ndevs == fs_devices->rw_devices) { |
4779 | WARN(1, "%s: found more than %llu devices\n", | |
4780 | __func__, fs_devices->rw_devices); | |
4781 | break; | |
4782 | } | |
73c5de00 AJ |
4783 | devices_info[ndevs].dev_offset = dev_offset; |
4784 | devices_info[ndevs].max_avail = max_avail; | |
4785 | devices_info[ndevs].total_avail = total_avail; | |
4786 | devices_info[ndevs].dev = device; | |
4787 | ++ndevs; | |
4788 | } | |
b2117a39 | 4789 | |
73c5de00 AJ |
4790 | /* |
4791 | * now sort the devices by hole size / available space | |
4792 | */ | |
4793 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4794 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4795 | |
73c5de00 | 4796 | /* round down to number of usable stripes */ |
e5600fd6 | 4797 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4798 | |
ba89b802 | 4799 | if (ndevs < devs_min) { |
73c5de00 | 4800 | ret = -ENOSPC; |
4117f207 QW |
4801 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4802 | btrfs_debug(info, | |
4803 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 4804 | __func__, ndevs, devs_min); |
4117f207 | 4805 | } |
73c5de00 | 4806 | goto error; |
b2117a39 | 4807 | } |
9f680ce0 | 4808 | |
f148ef4d NB |
4809 | ndevs = min(ndevs, devs_max); |
4810 | ||
73c5de00 | 4811 | /* |
92e222df HK |
4812 | * The primary goal is to maximize the number of stripes, so use as |
4813 | * many devices as possible, even if the stripes are not maximum sized. | |
4814 | * | |
4815 | * The DUP profile stores more than one stripe per device, the | |
4816 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4817 | */ |
92e222df | 4818 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4819 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4820 | |
53b381b3 DW |
4821 | /* |
4822 | * this will have to be fixed for RAID1 and RAID10 over | |
4823 | * more drives | |
4824 | */ | |
b50836ed | 4825 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
4826 | |
4827 | /* | |
4828 | * Use the number of data stripes to figure out how big this chunk | |
4829 | * is really going to be in terms of logical address space, | |
baf92114 HK |
4830 | * and compare that answer with the max chunk size. If it's higher, |
4831 | * we try to reduce stripe_size. | |
86db2578 CM |
4832 | */ |
4833 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 4834 | /* |
baf92114 HK |
4835 | * Reduce stripe_size, round it up to a 16MB boundary again and |
4836 | * then use it, unless it ends up being even bigger than the | |
4837 | * previous value we had already. | |
86db2578 | 4838 | */ |
baf92114 HK |
4839 | stripe_size = min(round_up(div_u64(max_chunk_size, |
4840 | data_stripes), SZ_16M), | |
793ff2c8 | 4841 | stripe_size); |
86db2578 CM |
4842 | } |
4843 | ||
37db63a4 | 4844 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 4845 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4846 | |
4847 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4848 | if (!map) { | |
4849 | ret = -ENOMEM; | |
4850 | goto error; | |
4851 | } | |
4852 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4853 | |
73c5de00 AJ |
4854 | for (i = 0; i < ndevs; ++i) { |
4855 | for (j = 0; j < dev_stripes; ++j) { | |
4856 | int s = i * dev_stripes + j; | |
4857 | map->stripes[s].dev = devices_info[i].dev; | |
4858 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4859 | j * stripe_size; | |
6324fbf3 | 4860 | } |
6324fbf3 | 4861 | } |
500ceed8 NB |
4862 | map->stripe_len = BTRFS_STRIPE_LEN; |
4863 | map->io_align = BTRFS_STRIPE_LEN; | |
4864 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4865 | map->type = type; |
2b82032c | 4866 | map->sub_stripes = sub_stripes; |
0b86a832 | 4867 | |
23f0ff1e | 4868 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 4869 | |
23f0ff1e | 4870 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 4871 | |
172ddd60 | 4872 | em = alloc_extent_map(); |
2b82032c | 4873 | if (!em) { |
298a8f9c | 4874 | kfree(map); |
b2117a39 MX |
4875 | ret = -ENOMEM; |
4876 | goto error; | |
593060d7 | 4877 | } |
298a8f9c | 4878 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4879 | em->map_lookup = map; |
2b82032c | 4880 | em->start = start; |
23f0ff1e | 4881 | em->len = chunk_size; |
2b82032c YZ |
4882 | em->block_start = 0; |
4883 | em->block_len = em->len; | |
6df9a95e | 4884 | em->orig_block_len = stripe_size; |
593060d7 | 4885 | |
0b246afa | 4886 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4887 | write_lock(&em_tree->lock); |
09a2a8f9 | 4888 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 4889 | if (ret) { |
1efb72a3 | 4890 | write_unlock(&em_tree->lock); |
0f5d42b2 | 4891 | free_extent_map(em); |
1dd4602f | 4892 | goto error; |
0f5d42b2 | 4893 | } |
0b86a832 | 4894 | |
1efb72a3 NB |
4895 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
4896 | refcount_inc(&em->refs); | |
4897 | write_unlock(&em_tree->lock); | |
4898 | ||
23f0ff1e | 4899 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
4900 | if (ret) |
4901 | goto error_del_extent; | |
2b82032c | 4902 | |
2f29df4f HK |
4903 | for (i = 0; i < map->num_stripes; i++) |
4904 | btrfs_device_set_bytes_used(map->stripes[i].dev, | |
4905 | map->stripes[i].dev->bytes_used + stripe_size); | |
43530c46 | 4906 | |
a5ed45f8 | 4907 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4908 | |
0f5d42b2 | 4909 | free_extent_map(em); |
0b246afa | 4910 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4911 | |
b2117a39 | 4912 | kfree(devices_info); |
2b82032c | 4913 | return 0; |
b2117a39 | 4914 | |
6df9a95e | 4915 | error_del_extent: |
0f5d42b2 JB |
4916 | write_lock(&em_tree->lock); |
4917 | remove_extent_mapping(em_tree, em); | |
4918 | write_unlock(&em_tree->lock); | |
4919 | ||
4920 | /* One for our allocation */ | |
4921 | free_extent_map(em); | |
4922 | /* One for the tree reference */ | |
4923 | free_extent_map(em); | |
495e64f4 FM |
4924 | /* One for the pending_chunks list reference */ |
4925 | free_extent_map(em); | |
b2117a39 | 4926 | error: |
b2117a39 MX |
4927 | kfree(devices_info); |
4928 | return ret; | |
2b82032c YZ |
4929 | } |
4930 | ||
6df9a95e | 4931 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 4932 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4933 | { |
97aff912 | 4934 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
4935 | struct btrfs_root *extent_root = fs_info->extent_root; |
4936 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4937 | struct btrfs_key key; |
2b82032c YZ |
4938 | struct btrfs_device *device; |
4939 | struct btrfs_chunk *chunk; | |
4940 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4941 | struct extent_map *em; |
4942 | struct map_lookup *map; | |
4943 | size_t item_size; | |
4944 | u64 dev_offset; | |
4945 | u64 stripe_size; | |
4946 | int i = 0; | |
140e639f | 4947 | int ret = 0; |
2b82032c | 4948 | |
60ca842e | 4949 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
4950 | if (IS_ERR(em)) |
4951 | return PTR_ERR(em); | |
6df9a95e | 4952 | |
95617d69 | 4953 | map = em->map_lookup; |
6df9a95e JB |
4954 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4955 | stripe_size = em->orig_block_len; | |
4956 | ||
2b82032c | 4957 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4958 | if (!chunk) { |
4959 | ret = -ENOMEM; | |
4960 | goto out; | |
4961 | } | |
4962 | ||
50460e37 FM |
4963 | /* |
4964 | * Take the device list mutex to prevent races with the final phase of | |
4965 | * a device replace operation that replaces the device object associated | |
4966 | * with the map's stripes, because the device object's id can change | |
4967 | * at any time during that final phase of the device replace operation | |
4968 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4969 | */ | |
0b246afa | 4970 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
4971 | for (i = 0; i < map->num_stripes; i++) { |
4972 | device = map->stripes[i].dev; | |
4973 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4974 | |
0b86a832 | 4975 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4976 | if (ret) |
50460e37 | 4977 | break; |
b5d9071c NB |
4978 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
4979 | dev_offset, stripe_size); | |
6df9a95e | 4980 | if (ret) |
50460e37 FM |
4981 | break; |
4982 | } | |
4983 | if (ret) { | |
0b246afa | 4984 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 4985 | goto out; |
2b82032c YZ |
4986 | } |
4987 | ||
2b82032c | 4988 | stripe = &chunk->stripe; |
6df9a95e JB |
4989 | for (i = 0; i < map->num_stripes; i++) { |
4990 | device = map->stripes[i].dev; | |
4991 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4992 | |
e17cade2 CM |
4993 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4994 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4995 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4996 | stripe++; |
0b86a832 | 4997 | } |
0b246afa | 4998 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4999 | |
2b82032c | 5000 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5001 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5002 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5003 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5004 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5005 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5006 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5007 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5008 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5009 | |
2b82032c YZ |
5010 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5011 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5012 | key.offset = chunk_offset; | |
0b86a832 | 5013 | |
2b82032c | 5014 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5015 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5016 | /* | |
5017 | * TODO: Cleanup of inserted chunk root in case of | |
5018 | * failure. | |
5019 | */ | |
2ff7e61e | 5020 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5021 | } |
1abe9b8a | 5022 | |
6df9a95e | 5023 | out: |
0b86a832 | 5024 | kfree(chunk); |
6df9a95e | 5025 | free_extent_map(em); |
4ed1d16e | 5026 | return ret; |
2b82032c | 5027 | } |
0b86a832 | 5028 | |
2b82032c YZ |
5029 | /* |
5030 | * Chunk allocation falls into two parts. The first part does works | |
5031 | * that make the new allocated chunk useable, but not do any operation | |
5032 | * that modifies the chunk tree. The second part does the works that | |
5033 | * require modifying the chunk tree. This division is important for the | |
5034 | * bootstrap process of adding storage to a seed btrfs. | |
5035 | */ | |
c216b203 | 5036 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5037 | { |
5038 | u64 chunk_offset; | |
2b82032c | 5039 | |
c216b203 NB |
5040 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5041 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5042 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5043 | } |
5044 | ||
d397712b | 5045 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5046 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5047 | { |
5048 | u64 chunk_offset; | |
5049 | u64 sys_chunk_offset; | |
2b82032c | 5050 | u64 alloc_profile; |
2b82032c YZ |
5051 | int ret; |
5052 | ||
6df9a95e | 5053 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5054 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5055 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5056 | if (ret) |
5057 | return ret; | |
2b82032c | 5058 | |
0b246afa | 5059 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5060 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5061 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5062 | return ret; |
2b82032c YZ |
5063 | } |
5064 | ||
d20983b4 MX |
5065 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5066 | { | |
5067 | int max_errors; | |
5068 | ||
5069 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5070 | BTRFS_BLOCK_GROUP_RAID10 | | |
5071 | BTRFS_BLOCK_GROUP_RAID5 | | |
5072 | BTRFS_BLOCK_GROUP_DUP)) { | |
5073 | max_errors = 1; | |
5074 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5075 | max_errors = 2; | |
5076 | } else { | |
5077 | max_errors = 0; | |
005d6427 | 5078 | } |
2b82032c | 5079 | |
d20983b4 | 5080 | return max_errors; |
2b82032c YZ |
5081 | } |
5082 | ||
2ff7e61e | 5083 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5084 | { |
5085 | struct extent_map *em; | |
5086 | struct map_lookup *map; | |
2b82032c | 5087 | int readonly = 0; |
d20983b4 | 5088 | int miss_ndevs = 0; |
2b82032c YZ |
5089 | int i; |
5090 | ||
60ca842e | 5091 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5092 | if (IS_ERR(em)) |
2b82032c YZ |
5093 | return 1; |
5094 | ||
95617d69 | 5095 | map = em->map_lookup; |
2b82032c | 5096 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5097 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5098 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5099 | miss_ndevs++; |
5100 | continue; | |
5101 | } | |
ebbede42 AJ |
5102 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5103 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5104 | readonly = 1; |
d20983b4 | 5105 | goto end; |
2b82032c YZ |
5106 | } |
5107 | } | |
d20983b4 MX |
5108 | |
5109 | /* | |
5110 | * If the number of missing devices is larger than max errors, | |
5111 | * we can not write the data into that chunk successfully, so | |
5112 | * set it readonly. | |
5113 | */ | |
5114 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5115 | readonly = 1; | |
5116 | end: | |
0b86a832 | 5117 | free_extent_map(em); |
2b82032c | 5118 | return readonly; |
0b86a832 CM |
5119 | } |
5120 | ||
5121 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5122 | { | |
a8067e02 | 5123 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5124 | } |
5125 | ||
5126 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5127 | { | |
5128 | struct extent_map *em; | |
5129 | ||
d397712b | 5130 | while (1) { |
890871be | 5131 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5132 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5133 | if (em) | |
5134 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5135 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5136 | if (!em) |
5137 | break; | |
0b86a832 CM |
5138 | /* once for us */ |
5139 | free_extent_map(em); | |
5140 | /* once for the tree */ | |
5141 | free_extent_map(em); | |
5142 | } | |
5143 | } | |
5144 | ||
5d964051 | 5145 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5146 | { |
5147 | struct extent_map *em; | |
5148 | struct map_lookup *map; | |
f188591e CM |
5149 | int ret; |
5150 | ||
60ca842e | 5151 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5152 | if (IS_ERR(em)) |
5153 | /* | |
5154 | * We could return errors for these cases, but that could get | |
5155 | * ugly and we'd probably do the same thing which is just not do | |
5156 | * anything else and exit, so return 1 so the callers don't try | |
5157 | * to use other copies. | |
5158 | */ | |
fb7669b5 | 5159 | return 1; |
fb7669b5 | 5160 | |
95617d69 | 5161 | map = em->map_lookup; |
f188591e CM |
5162 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5163 | ret = map->num_stripes; | |
321aecc6 CM |
5164 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5165 | ret = map->sub_stripes; | |
53b381b3 DW |
5166 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5167 | ret = 2; | |
5168 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5169 | /* |
5170 | * There could be two corrupted data stripes, we need | |
5171 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5172 | * |
8810f751 LB |
5173 | * Fail a stripe at a time on every retry except the |
5174 | * stripe under reconstruction. | |
5175 | */ | |
5176 | ret = map->num_stripes; | |
f188591e CM |
5177 | else |
5178 | ret = 1; | |
5179 | free_extent_map(em); | |
ad6d620e | 5180 | |
7e79cb86 | 5181 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
6fad823f LB |
5182 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5183 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5184 | ret++; |
7e79cb86 | 5185 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
ad6d620e | 5186 | |
f188591e CM |
5187 | return ret; |
5188 | } | |
5189 | ||
2ff7e61e | 5190 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5191 | u64 logical) |
5192 | { | |
5193 | struct extent_map *em; | |
5194 | struct map_lookup *map; | |
0b246afa | 5195 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5196 | |
60ca842e | 5197 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5198 | |
69f03f13 NB |
5199 | if (!WARN_ON(IS_ERR(em))) { |
5200 | map = em->map_lookup; | |
5201 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5202 | len = map->stripe_len * nr_data_stripes(map); | |
5203 | free_extent_map(em); | |
5204 | } | |
53b381b3 DW |
5205 | return len; |
5206 | } | |
5207 | ||
e4ff5fb5 | 5208 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5209 | { |
5210 | struct extent_map *em; | |
5211 | struct map_lookup *map; | |
53b381b3 DW |
5212 | int ret = 0; |
5213 | ||
60ca842e | 5214 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5215 | |
69f03f13 NB |
5216 | if(!WARN_ON(IS_ERR(em))) { |
5217 | map = em->map_lookup; | |
5218 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5219 | ret = 1; | |
5220 | free_extent_map(em); | |
5221 | } | |
53b381b3 DW |
5222 | return ret; |
5223 | } | |
5224 | ||
30d9861f | 5225 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5226 | struct map_lookup *map, int first, |
8ba0ae78 | 5227 | int dev_replace_is_ongoing) |
dfe25020 CM |
5228 | { |
5229 | int i; | |
99f92a7c | 5230 | int num_stripes; |
8ba0ae78 | 5231 | int preferred_mirror; |
30d9861f SB |
5232 | int tolerance; |
5233 | struct btrfs_device *srcdev; | |
5234 | ||
99f92a7c AJ |
5235 | ASSERT((map->type & |
5236 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5237 | ||
5238 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5239 | num_stripes = map->sub_stripes; | |
5240 | else | |
5241 | num_stripes = map->num_stripes; | |
5242 | ||
8ba0ae78 AJ |
5243 | preferred_mirror = first + current->pid % num_stripes; |
5244 | ||
30d9861f SB |
5245 | if (dev_replace_is_ongoing && |
5246 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5247 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5248 | srcdev = fs_info->dev_replace.srcdev; | |
5249 | else | |
5250 | srcdev = NULL; | |
5251 | ||
5252 | /* | |
5253 | * try to avoid the drive that is the source drive for a | |
5254 | * dev-replace procedure, only choose it if no other non-missing | |
5255 | * mirror is available | |
5256 | */ | |
5257 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5258 | if (map->stripes[preferred_mirror].dev->bdev && |
5259 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5260 | return preferred_mirror; | |
99f92a7c | 5261 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5262 | if (map->stripes[i].dev->bdev && |
5263 | (tolerance || map->stripes[i].dev != srcdev)) | |
5264 | return i; | |
5265 | } | |
dfe25020 | 5266 | } |
30d9861f | 5267 | |
dfe25020 CM |
5268 | /* we couldn't find one that doesn't fail. Just return something |
5269 | * and the io error handling code will clean up eventually | |
5270 | */ | |
8ba0ae78 | 5271 | return preferred_mirror; |
dfe25020 CM |
5272 | } |
5273 | ||
53b381b3 DW |
5274 | static inline int parity_smaller(u64 a, u64 b) |
5275 | { | |
5276 | return a > b; | |
5277 | } | |
5278 | ||
5279 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5280 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5281 | { |
5282 | struct btrfs_bio_stripe s; | |
5283 | int i; | |
5284 | u64 l; | |
5285 | int again = 1; | |
5286 | ||
5287 | while (again) { | |
5288 | again = 0; | |
cc7539ed | 5289 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5290 | if (parity_smaller(bbio->raid_map[i], |
5291 | bbio->raid_map[i+1])) { | |
53b381b3 | 5292 | s = bbio->stripes[i]; |
8e5cfb55 | 5293 | l = bbio->raid_map[i]; |
53b381b3 | 5294 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5295 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5296 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5297 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5298 | |
53b381b3 DW |
5299 | again = 1; |
5300 | } | |
5301 | } | |
5302 | } | |
5303 | } | |
5304 | ||
6e9606d2 ZL |
5305 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5306 | { | |
5307 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5308 | /* the size of the btrfs_bio */ |
6e9606d2 | 5309 | sizeof(struct btrfs_bio) + |
e57cf21e | 5310 | /* plus the variable array for the stripes */ |
6e9606d2 | 5311 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5312 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5313 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5314 | /* |
5315 | * plus the raid_map, which includes both the tgt dev | |
5316 | * and the stripes | |
5317 | */ | |
5318 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5319 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5320 | |
5321 | atomic_set(&bbio->error, 0); | |
140475ae | 5322 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5323 | |
5324 | return bbio; | |
5325 | } | |
5326 | ||
5327 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5328 | { | |
140475ae ER |
5329 | WARN_ON(!refcount_read(&bbio->refs)); |
5330 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5331 | } |
5332 | ||
5333 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5334 | { | |
5335 | if (!bbio) | |
5336 | return; | |
140475ae | 5337 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5338 | kfree(bbio); |
5339 | } | |
5340 | ||
0b3d4cd3 LB |
5341 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5342 | /* | |
5343 | * Please note that, discard won't be sent to target device of device | |
5344 | * replace. | |
5345 | */ | |
5346 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5347 | u64 logical, u64 length, | |
5348 | struct btrfs_bio **bbio_ret) | |
5349 | { | |
5350 | struct extent_map *em; | |
5351 | struct map_lookup *map; | |
5352 | struct btrfs_bio *bbio; | |
5353 | u64 offset; | |
5354 | u64 stripe_nr; | |
5355 | u64 stripe_nr_end; | |
5356 | u64 stripe_end_offset; | |
5357 | u64 stripe_cnt; | |
5358 | u64 stripe_len; | |
5359 | u64 stripe_offset; | |
5360 | u64 num_stripes; | |
5361 | u32 stripe_index; | |
5362 | u32 factor = 0; | |
5363 | u32 sub_stripes = 0; | |
5364 | u64 stripes_per_dev = 0; | |
5365 | u32 remaining_stripes = 0; | |
5366 | u32 last_stripe = 0; | |
5367 | int ret = 0; | |
5368 | int i; | |
5369 | ||
5370 | /* discard always return a bbio */ | |
5371 | ASSERT(bbio_ret); | |
5372 | ||
60ca842e | 5373 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5374 | if (IS_ERR(em)) |
5375 | return PTR_ERR(em); | |
5376 | ||
5377 | map = em->map_lookup; | |
5378 | /* we don't discard raid56 yet */ | |
5379 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5380 | ret = -EOPNOTSUPP; | |
5381 | goto out; | |
5382 | } | |
5383 | ||
5384 | offset = logical - em->start; | |
5385 | length = min_t(u64, em->len - offset, length); | |
5386 | ||
5387 | stripe_len = map->stripe_len; | |
5388 | /* | |
5389 | * stripe_nr counts the total number of stripes we have to stride | |
5390 | * to get to this block | |
5391 | */ | |
5392 | stripe_nr = div64_u64(offset, stripe_len); | |
5393 | ||
5394 | /* stripe_offset is the offset of this block in its stripe */ | |
5395 | stripe_offset = offset - stripe_nr * stripe_len; | |
5396 | ||
5397 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5398 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5399 | stripe_cnt = stripe_nr_end - stripe_nr; |
5400 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5401 | (offset + length); | |
5402 | /* | |
5403 | * after this, stripe_nr is the number of stripes on this | |
5404 | * device we have to walk to find the data, and stripe_index is | |
5405 | * the number of our device in the stripe array | |
5406 | */ | |
5407 | num_stripes = 1; | |
5408 | stripe_index = 0; | |
5409 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5410 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5411 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5412 | sub_stripes = 1; | |
5413 | else | |
5414 | sub_stripes = map->sub_stripes; | |
5415 | ||
5416 | factor = map->num_stripes / sub_stripes; | |
5417 | num_stripes = min_t(u64, map->num_stripes, | |
5418 | sub_stripes * stripe_cnt); | |
5419 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5420 | stripe_index *= sub_stripes; | |
5421 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5422 | &remaining_stripes); | |
5423 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5424 | last_stripe *= sub_stripes; | |
5425 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5426 | BTRFS_BLOCK_GROUP_DUP)) { | |
5427 | num_stripes = map->num_stripes; | |
5428 | } else { | |
5429 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5430 | &stripe_index); | |
5431 | } | |
5432 | ||
5433 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5434 | if (!bbio) { | |
5435 | ret = -ENOMEM; | |
5436 | goto out; | |
5437 | } | |
5438 | ||
5439 | for (i = 0; i < num_stripes; i++) { | |
5440 | bbio->stripes[i].physical = | |
5441 | map->stripes[stripe_index].physical + | |
5442 | stripe_offset + stripe_nr * map->stripe_len; | |
5443 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5444 | ||
5445 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5446 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5447 | bbio->stripes[i].length = stripes_per_dev * | |
5448 | map->stripe_len; | |
5449 | ||
5450 | if (i / sub_stripes < remaining_stripes) | |
5451 | bbio->stripes[i].length += | |
5452 | map->stripe_len; | |
5453 | ||
5454 | /* | |
5455 | * Special for the first stripe and | |
5456 | * the last stripe: | |
5457 | * | |
5458 | * |-------|...|-------| | |
5459 | * |----------| | |
5460 | * off end_off | |
5461 | */ | |
5462 | if (i < sub_stripes) | |
5463 | bbio->stripes[i].length -= | |
5464 | stripe_offset; | |
5465 | ||
5466 | if (stripe_index >= last_stripe && | |
5467 | stripe_index <= (last_stripe + | |
5468 | sub_stripes - 1)) | |
5469 | bbio->stripes[i].length -= | |
5470 | stripe_end_offset; | |
5471 | ||
5472 | if (i == sub_stripes - 1) | |
5473 | stripe_offset = 0; | |
5474 | } else { | |
5475 | bbio->stripes[i].length = length; | |
5476 | } | |
5477 | ||
5478 | stripe_index++; | |
5479 | if (stripe_index == map->num_stripes) { | |
5480 | stripe_index = 0; | |
5481 | stripe_nr++; | |
5482 | } | |
5483 | } | |
5484 | ||
5485 | *bbio_ret = bbio; | |
5486 | bbio->map_type = map->type; | |
5487 | bbio->num_stripes = num_stripes; | |
5488 | out: | |
5489 | free_extent_map(em); | |
5490 | return ret; | |
5491 | } | |
5492 | ||
5ab56090 LB |
5493 | /* |
5494 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5495 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5496 | * left cursor can also be read from the target drive. | |
5497 | * | |
5498 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5499 | * array of stripes. | |
5500 | * For READ, it also needs to be supported using the same mirror number. | |
5501 | * | |
5502 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5503 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5504 | * case. | |
5505 | */ | |
5506 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5507 | u64 logical, u64 length, | |
5508 | u64 srcdev_devid, int *mirror_num, | |
5509 | u64 *physical) | |
5510 | { | |
5511 | struct btrfs_bio *bbio = NULL; | |
5512 | int num_stripes; | |
5513 | int index_srcdev = 0; | |
5514 | int found = 0; | |
5515 | u64 physical_of_found = 0; | |
5516 | int i; | |
5517 | int ret = 0; | |
5518 | ||
5519 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5520 | logical, &length, &bbio, 0, 0); | |
5521 | if (ret) { | |
5522 | ASSERT(bbio == NULL); | |
5523 | return ret; | |
5524 | } | |
5525 | ||
5526 | num_stripes = bbio->num_stripes; | |
5527 | if (*mirror_num > num_stripes) { | |
5528 | /* | |
5529 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5530 | * that means that the requested area is not left of the left | |
5531 | * cursor | |
5532 | */ | |
5533 | btrfs_put_bbio(bbio); | |
5534 | return -EIO; | |
5535 | } | |
5536 | ||
5537 | /* | |
5538 | * process the rest of the function using the mirror_num of the source | |
5539 | * drive. Therefore look it up first. At the end, patch the device | |
5540 | * pointer to the one of the target drive. | |
5541 | */ | |
5542 | for (i = 0; i < num_stripes; i++) { | |
5543 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5544 | continue; | |
5545 | ||
5546 | /* | |
5547 | * In case of DUP, in order to keep it simple, only add the | |
5548 | * mirror with the lowest physical address | |
5549 | */ | |
5550 | if (found && | |
5551 | physical_of_found <= bbio->stripes[i].physical) | |
5552 | continue; | |
5553 | ||
5554 | index_srcdev = i; | |
5555 | found = 1; | |
5556 | physical_of_found = bbio->stripes[i].physical; | |
5557 | } | |
5558 | ||
5559 | btrfs_put_bbio(bbio); | |
5560 | ||
5561 | ASSERT(found); | |
5562 | if (!found) | |
5563 | return -EIO; | |
5564 | ||
5565 | *mirror_num = index_srcdev + 1; | |
5566 | *physical = physical_of_found; | |
5567 | return ret; | |
5568 | } | |
5569 | ||
73c0f228 LB |
5570 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5571 | struct btrfs_bio **bbio_ret, | |
5572 | struct btrfs_dev_replace *dev_replace, | |
5573 | int *num_stripes_ret, int *max_errors_ret) | |
5574 | { | |
5575 | struct btrfs_bio *bbio = *bbio_ret; | |
5576 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5577 | int tgtdev_indexes = 0; | |
5578 | int num_stripes = *num_stripes_ret; | |
5579 | int max_errors = *max_errors_ret; | |
5580 | int i; | |
5581 | ||
5582 | if (op == BTRFS_MAP_WRITE) { | |
5583 | int index_where_to_add; | |
5584 | ||
5585 | /* | |
5586 | * duplicate the write operations while the dev replace | |
5587 | * procedure is running. Since the copying of the old disk to | |
5588 | * the new disk takes place at run time while the filesystem is | |
5589 | * mounted writable, the regular write operations to the old | |
5590 | * disk have to be duplicated to go to the new disk as well. | |
5591 | * | |
5592 | * Note that device->missing is handled by the caller, and that | |
5593 | * the write to the old disk is already set up in the stripes | |
5594 | * array. | |
5595 | */ | |
5596 | index_where_to_add = num_stripes; | |
5597 | for (i = 0; i < num_stripes; i++) { | |
5598 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5599 | /* write to new disk, too */ | |
5600 | struct btrfs_bio_stripe *new = | |
5601 | bbio->stripes + index_where_to_add; | |
5602 | struct btrfs_bio_stripe *old = | |
5603 | bbio->stripes + i; | |
5604 | ||
5605 | new->physical = old->physical; | |
5606 | new->length = old->length; | |
5607 | new->dev = dev_replace->tgtdev; | |
5608 | bbio->tgtdev_map[i] = index_where_to_add; | |
5609 | index_where_to_add++; | |
5610 | max_errors++; | |
5611 | tgtdev_indexes++; | |
5612 | } | |
5613 | } | |
5614 | num_stripes = index_where_to_add; | |
5615 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5616 | int index_srcdev = 0; | |
5617 | int found = 0; | |
5618 | u64 physical_of_found = 0; | |
5619 | ||
5620 | /* | |
5621 | * During the dev-replace procedure, the target drive can also | |
5622 | * be used to read data in case it is needed to repair a corrupt | |
5623 | * block elsewhere. This is possible if the requested area is | |
5624 | * left of the left cursor. In this area, the target drive is a | |
5625 | * full copy of the source drive. | |
5626 | */ | |
5627 | for (i = 0; i < num_stripes; i++) { | |
5628 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5629 | /* | |
5630 | * In case of DUP, in order to keep it simple, | |
5631 | * only add the mirror with the lowest physical | |
5632 | * address | |
5633 | */ | |
5634 | if (found && | |
5635 | physical_of_found <= | |
5636 | bbio->stripes[i].physical) | |
5637 | continue; | |
5638 | index_srcdev = i; | |
5639 | found = 1; | |
5640 | physical_of_found = bbio->stripes[i].physical; | |
5641 | } | |
5642 | } | |
5643 | if (found) { | |
5644 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5645 | bbio->stripes + num_stripes; | |
5646 | ||
5647 | tgtdev_stripe->physical = physical_of_found; | |
5648 | tgtdev_stripe->length = | |
5649 | bbio->stripes[index_srcdev].length; | |
5650 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5651 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5652 | ||
5653 | tgtdev_indexes++; | |
5654 | num_stripes++; | |
5655 | } | |
5656 | } | |
5657 | ||
5658 | *num_stripes_ret = num_stripes; | |
5659 | *max_errors_ret = max_errors; | |
5660 | bbio->num_tgtdevs = tgtdev_indexes; | |
5661 | *bbio_ret = bbio; | |
5662 | } | |
5663 | ||
2b19a1fe LB |
5664 | static bool need_full_stripe(enum btrfs_map_op op) |
5665 | { | |
5666 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5667 | } | |
5668 | ||
cf8cddd3 CH |
5669 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5670 | enum btrfs_map_op op, | |
f2d8d74d | 5671 | u64 logical, u64 *length, |
a1d3c478 | 5672 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5673 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5674 | { |
5675 | struct extent_map *em; | |
5676 | struct map_lookup *map; | |
0b86a832 | 5677 | u64 offset; |
593060d7 CM |
5678 | u64 stripe_offset; |
5679 | u64 stripe_nr; | |
53b381b3 | 5680 | u64 stripe_len; |
9d644a62 | 5681 | u32 stripe_index; |
cea9e445 | 5682 | int i; |
de11cc12 | 5683 | int ret = 0; |
f2d8d74d | 5684 | int num_stripes; |
a236aed1 | 5685 | int max_errors = 0; |
2c8cdd6e | 5686 | int tgtdev_indexes = 0; |
a1d3c478 | 5687 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5688 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5689 | int dev_replace_is_ongoing = 0; | |
5690 | int num_alloc_stripes; | |
ad6d620e SB |
5691 | int patch_the_first_stripe_for_dev_replace = 0; |
5692 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5693 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5694 | |
0b3d4cd3 LB |
5695 | if (op == BTRFS_MAP_DISCARD) |
5696 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5697 | *length, bbio_ret); | |
5698 | ||
60ca842e | 5699 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
592d92ee LB |
5700 | if (IS_ERR(em)) |
5701 | return PTR_ERR(em); | |
0b86a832 | 5702 | |
95617d69 | 5703 | map = em->map_lookup; |
0b86a832 | 5704 | offset = logical - em->start; |
593060d7 | 5705 | |
53b381b3 | 5706 | stripe_len = map->stripe_len; |
593060d7 CM |
5707 | stripe_nr = offset; |
5708 | /* | |
5709 | * stripe_nr counts the total number of stripes we have to stride | |
5710 | * to get to this block | |
5711 | */ | |
47c5713f | 5712 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5713 | |
53b381b3 | 5714 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5715 | if (offset < stripe_offset) { |
5d163e0e JM |
5716 | btrfs_crit(fs_info, |
5717 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5718 | stripe_offset, offset, em->start, logical, |
5719 | stripe_len); | |
5720 | free_extent_map(em); | |
5721 | return -EINVAL; | |
5722 | } | |
593060d7 CM |
5723 | |
5724 | /* stripe_offset is the offset of this block in its stripe*/ | |
5725 | stripe_offset = offset - stripe_offset; | |
5726 | ||
53b381b3 | 5727 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5728 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5729 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5730 | raid56_full_stripe_start = offset; | |
5731 | ||
5732 | /* allow a write of a full stripe, but make sure we don't | |
5733 | * allow straddling of stripes | |
5734 | */ | |
47c5713f DS |
5735 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5736 | full_stripe_len); | |
53b381b3 DW |
5737 | raid56_full_stripe_start *= full_stripe_len; |
5738 | } | |
5739 | ||
0b3d4cd3 | 5740 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5741 | u64 max_len; |
5742 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5743 | For other RAID types and for RAID[56] reads, just allow a single | |
5744 | stripe (on a single disk). */ | |
ffe2d203 | 5745 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5746 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5747 | max_len = stripe_len * nr_data_stripes(map) - |
5748 | (offset - raid56_full_stripe_start); | |
5749 | } else { | |
5750 | /* we limit the length of each bio to what fits in a stripe */ | |
5751 | max_len = stripe_len - stripe_offset; | |
5752 | } | |
5753 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5754 | } else { |
5755 | *length = em->len - offset; | |
5756 | } | |
f2d8d74d | 5757 | |
53b381b3 DW |
5758 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5759 | it cares about is the length */ | |
a1d3c478 | 5760 | if (!bbio_ret) |
cea9e445 CM |
5761 | goto out; |
5762 | ||
7e79cb86 | 5763 | btrfs_dev_replace_read_lock(dev_replace); |
472262f3 SB |
5764 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5765 | if (!dev_replace_is_ongoing) | |
7e79cb86 | 5766 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 LB |
5767 | else |
5768 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5769 | |
ad6d620e | 5770 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5771 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5772 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5773 | dev_replace->srcdev->devid, | |
5774 | &mirror_num, | |
5775 | &physical_to_patch_in_first_stripe); | |
5776 | if (ret) | |
ad6d620e | 5777 | goto out; |
5ab56090 LB |
5778 | else |
5779 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5780 | } else if (mirror_num > map->num_stripes) { |
5781 | mirror_num = 0; | |
5782 | } | |
5783 | ||
f2d8d74d | 5784 | num_stripes = 1; |
cea9e445 | 5785 | stripe_index = 0; |
fce3bb9a | 5786 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5787 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5788 | &stripe_index); | |
de483734 | 5789 | if (!need_full_stripe(op)) |
28e1cc7d | 5790 | mirror_num = 1; |
fce3bb9a | 5791 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 5792 | if (need_full_stripe(op)) |
f2d8d74d | 5793 | num_stripes = map->num_stripes; |
2fff734f | 5794 | else if (mirror_num) |
f188591e | 5795 | stripe_index = mirror_num - 1; |
dfe25020 | 5796 | else { |
30d9861f | 5797 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 5798 | dev_replace_is_ongoing); |
a1d3c478 | 5799 | mirror_num = stripe_index + 1; |
dfe25020 | 5800 | } |
2fff734f | 5801 | |
611f0e00 | 5802 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5803 | if (need_full_stripe(op)) { |
f2d8d74d | 5804 | num_stripes = map->num_stripes; |
a1d3c478 | 5805 | } else if (mirror_num) { |
f188591e | 5806 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5807 | } else { |
5808 | mirror_num = 1; | |
5809 | } | |
2fff734f | 5810 | |
321aecc6 | 5811 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5812 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5813 | |
47c5713f | 5814 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5815 | stripe_index *= map->sub_stripes; |
5816 | ||
de483734 | 5817 | if (need_full_stripe(op)) |
f2d8d74d | 5818 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5819 | else if (mirror_num) |
5820 | stripe_index += mirror_num - 1; | |
dfe25020 | 5821 | else { |
3e74317a | 5822 | int old_stripe_index = stripe_index; |
30d9861f SB |
5823 | stripe_index = find_live_mirror(fs_info, map, |
5824 | stripe_index, | |
30d9861f | 5825 | dev_replace_is_ongoing); |
3e74317a | 5826 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5827 | } |
53b381b3 | 5828 | |
ffe2d203 | 5829 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5830 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5831 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5832 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5833 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5834 | |
5835 | /* RAID[56] write or recovery. Return all stripes */ | |
5836 | num_stripes = map->num_stripes; | |
5837 | max_errors = nr_parity_stripes(map); | |
5838 | ||
53b381b3 DW |
5839 | *length = map->stripe_len; |
5840 | stripe_index = 0; | |
5841 | stripe_offset = 0; | |
5842 | } else { | |
5843 | /* | |
5844 | * Mirror #0 or #1 means the original data block. | |
5845 | * Mirror #2 is RAID5 parity block. | |
5846 | * Mirror #3 is RAID6 Q block. | |
5847 | */ | |
47c5713f DS |
5848 | stripe_nr = div_u64_rem(stripe_nr, |
5849 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5850 | if (mirror_num > 1) |
5851 | stripe_index = nr_data_stripes(map) + | |
5852 | mirror_num - 2; | |
5853 | ||
5854 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5855 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5856 | &stripe_index); | |
de483734 | 5857 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 5858 | mirror_num = 1; |
53b381b3 | 5859 | } |
8790d502 CM |
5860 | } else { |
5861 | /* | |
47c5713f DS |
5862 | * after this, stripe_nr is the number of stripes on this |
5863 | * device we have to walk to find the data, and stripe_index is | |
5864 | * the number of our device in the stripe array | |
8790d502 | 5865 | */ |
47c5713f DS |
5866 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5867 | &stripe_index); | |
a1d3c478 | 5868 | mirror_num = stripe_index + 1; |
8790d502 | 5869 | } |
e042d1ec | 5870 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5871 | btrfs_crit(fs_info, |
5872 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5873 | stripe_index, map->num_stripes); |
5874 | ret = -EINVAL; | |
5875 | goto out; | |
5876 | } | |
cea9e445 | 5877 | |
472262f3 | 5878 | num_alloc_stripes = num_stripes; |
6fad823f | 5879 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 5880 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 5881 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5882 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5883 | num_alloc_stripes++; |
2c8cdd6e | 5884 | tgtdev_indexes = num_stripes; |
ad6d620e | 5885 | } |
2c8cdd6e | 5886 | |
6e9606d2 | 5887 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5888 | if (!bbio) { |
5889 | ret = -ENOMEM; | |
5890 | goto out; | |
5891 | } | |
6fad823f | 5892 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 5893 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 5894 | |
8e5cfb55 | 5895 | /* build raid_map */ |
2b19a1fe LB |
5896 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
5897 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 5898 | u64 tmp; |
9d644a62 | 5899 | unsigned rot; |
8e5cfb55 ZL |
5900 | |
5901 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5902 | sizeof(struct btrfs_bio_stripe) * | |
5903 | num_alloc_stripes + | |
5904 | sizeof(int) * tgtdev_indexes); | |
5905 | ||
5906 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5907 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5908 | |
5909 | /* Fill in the logical address of each stripe */ | |
5910 | tmp = stripe_nr * nr_data_stripes(map); | |
5911 | for (i = 0; i < nr_data_stripes(map); i++) | |
5912 | bbio->raid_map[(i+rot) % num_stripes] = | |
5913 | em->start + (tmp + i) * map->stripe_len; | |
5914 | ||
5915 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5916 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5917 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5918 | RAID6_Q_STRIPE; | |
5919 | } | |
5920 | ||
b89203f7 | 5921 | |
0b3d4cd3 LB |
5922 | for (i = 0; i < num_stripes; i++) { |
5923 | bbio->stripes[i].physical = | |
5924 | map->stripes[stripe_index].physical + | |
5925 | stripe_offset + | |
5926 | stripe_nr * map->stripe_len; | |
5927 | bbio->stripes[i].dev = | |
5928 | map->stripes[stripe_index].dev; | |
5929 | stripe_index++; | |
593060d7 | 5930 | } |
de11cc12 | 5931 | |
2b19a1fe | 5932 | if (need_full_stripe(op)) |
d20983b4 | 5933 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5934 | |
8e5cfb55 ZL |
5935 | if (bbio->raid_map) |
5936 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5937 | |
73c0f228 | 5938 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 5939 | need_full_stripe(op)) { |
73c0f228 LB |
5940 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
5941 | &max_errors); | |
472262f3 SB |
5942 | } |
5943 | ||
de11cc12 | 5944 | *bbio_ret = bbio; |
10f11900 | 5945 | bbio->map_type = map->type; |
de11cc12 LZ |
5946 | bbio->num_stripes = num_stripes; |
5947 | bbio->max_errors = max_errors; | |
5948 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
5949 | |
5950 | /* | |
5951 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5952 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5953 | * available as a mirror | |
5954 | */ | |
5955 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5956 | WARN_ON(num_stripes > 1); | |
5957 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5958 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5959 | bbio->mirror_num = map->num_stripes + 1; | |
5960 | } | |
cea9e445 | 5961 | out: |
73beece9 | 5962 | if (dev_replace_is_ongoing) { |
7fb2eced DS |
5963 | ASSERT(atomic_read(&dev_replace->blocking_readers) > 0); |
5964 | btrfs_dev_replace_read_lock(dev_replace); | |
5965 | /* Barrier implied by atomic_dec_and_test */ | |
5966 | if (atomic_dec_and_test(&dev_replace->blocking_readers)) | |
5967 | cond_wake_up_nomb(&dev_replace->read_lock_wq); | |
7e79cb86 | 5968 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 | 5969 | } |
0b86a832 | 5970 | free_extent_map(em); |
de11cc12 | 5971 | return ret; |
0b86a832 CM |
5972 | } |
5973 | ||
cf8cddd3 | 5974 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 5975 | u64 logical, u64 *length, |
a1d3c478 | 5976 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5977 | { |
b3d3fa51 | 5978 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5979 | mirror_num, 0); |
f2d8d74d CM |
5980 | } |
5981 | ||
af8e2d1d | 5982 | /* For Scrub/replace */ |
cf8cddd3 | 5983 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 5984 | u64 logical, u64 *length, |
825ad4c9 | 5985 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 5986 | { |
825ad4c9 | 5987 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
5988 | } |
5989 | ||
63a9c7b9 NB |
5990 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
5991 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 5992 | { |
a512bbf8 YZ |
5993 | struct extent_map *em; |
5994 | struct map_lookup *map; | |
5995 | u64 *buf; | |
5996 | u64 bytenr; | |
5997 | u64 length; | |
5998 | u64 stripe_nr; | |
53b381b3 | 5999 | u64 rmap_len; |
a512bbf8 YZ |
6000 | int i, j, nr = 0; |
6001 | ||
60ca842e | 6002 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6003 | if (IS_ERR(em)) |
835d974f | 6004 | return -EIO; |
835d974f | 6005 | |
95617d69 | 6006 | map = em->map_lookup; |
a512bbf8 | 6007 | length = em->len; |
53b381b3 DW |
6008 | rmap_len = map->stripe_len; |
6009 | ||
a512bbf8 | 6010 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6011 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6012 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6013 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6014 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6015 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6016 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6017 | } | |
a512bbf8 | 6018 | |
31e818fe | 6019 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6020 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6021 | |
6022 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6023 | if (map->stripes[i].physical > physical || |
6024 | map->stripes[i].physical + length <= physical) | |
6025 | continue; | |
6026 | ||
6027 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6028 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6029 | |
6030 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6031 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6032 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6033 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6034 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6035 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6036 | * Alternatively, just use rmap_len below instead of | |
6037 | * map->stripe_len */ | |
6038 | ||
6039 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6040 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6041 | for (j = 0; j < nr; j++) { |
6042 | if (buf[j] == bytenr) | |
6043 | break; | |
6044 | } | |
934d375b CM |
6045 | if (j == nr) { |
6046 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6047 | buf[nr++] = bytenr; |
934d375b | 6048 | } |
a512bbf8 YZ |
6049 | } |
6050 | ||
a512bbf8 YZ |
6051 | *logical = buf; |
6052 | *naddrs = nr; | |
53b381b3 | 6053 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6054 | |
6055 | free_extent_map(em); | |
6056 | return 0; | |
f2d8d74d CM |
6057 | } |
6058 | ||
4246a0b6 | 6059 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6060 | { |
326e1dbb MS |
6061 | bio->bi_private = bbio->private; |
6062 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6063 | bio_endio(bio); |
326e1dbb | 6064 | |
6e9606d2 | 6065 | btrfs_put_bbio(bbio); |
8408c716 MX |
6066 | } |
6067 | ||
4246a0b6 | 6068 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6069 | { |
9be3395b | 6070 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6071 | int is_orig_bio = 0; |
8790d502 | 6072 | |
4e4cbee9 | 6073 | if (bio->bi_status) { |
a1d3c478 | 6074 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6075 | if (bio->bi_status == BLK_STS_IOERR || |
6076 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6077 | unsigned int stripe_index = |
9be3395b | 6078 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6079 | struct btrfs_device *dev; |
442a4f63 SB |
6080 | |
6081 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6082 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6083 | if (dev->bdev) { |
37226b21 | 6084 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6085 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6086 | BTRFS_DEV_STAT_WRITE_ERRS); |
6087 | else | |
1cb34c8e | 6088 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6089 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6090 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6091 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6092 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6093 | } |
442a4f63 SB |
6094 | } |
6095 | } | |
8790d502 | 6096 | |
a1d3c478 | 6097 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6098 | is_orig_bio = 1; |
6099 | ||
c404e0dc MX |
6100 | btrfs_bio_counter_dec(bbio->fs_info); |
6101 | ||
a1d3c478 | 6102 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6103 | if (!is_orig_bio) { |
6104 | bio_put(bio); | |
a1d3c478 | 6105 | bio = bbio->orig_bio; |
7d2b4daa | 6106 | } |
c7b22bb1 | 6107 | |
9be3395b | 6108 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6109 | /* only send an error to the higher layers if it is |
53b381b3 | 6110 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6111 | */ |
a1d3c478 | 6112 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6113 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6114 | } else { |
1259ab75 CM |
6115 | /* |
6116 | * this bio is actually up to date, we didn't | |
6117 | * go over the max number of errors | |
6118 | */ | |
2dbe0c77 | 6119 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6120 | } |
c55f1396 | 6121 | |
4246a0b6 | 6122 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6123 | } else if (!is_orig_bio) { |
8790d502 CM |
6124 | bio_put(bio); |
6125 | } | |
8790d502 CM |
6126 | } |
6127 | ||
8b712842 CM |
6128 | /* |
6129 | * see run_scheduled_bios for a description of why bios are collected for | |
6130 | * async submit. | |
6131 | * | |
6132 | * This will add one bio to the pending list for a device and make sure | |
6133 | * the work struct is scheduled. | |
6134 | */ | |
2ff7e61e | 6135 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6136 | struct bio *bio) |
8b712842 | 6137 | { |
0b246afa | 6138 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6139 | int should_queue = 1; |
ffbd517d | 6140 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6141 | |
6142 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6143 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6144 | btrfsic_submit_bio(bio); |
143bede5 | 6145 | return; |
8b712842 CM |
6146 | } |
6147 | ||
492bb6de | 6148 | WARN_ON(bio->bi_next); |
8b712842 | 6149 | bio->bi_next = NULL; |
8b712842 CM |
6150 | |
6151 | spin_lock(&device->io_lock); | |
67f055c7 | 6152 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6153 | pending_bios = &device->pending_sync_bios; |
6154 | else | |
6155 | pending_bios = &device->pending_bios; | |
8b712842 | 6156 | |
ffbd517d CM |
6157 | if (pending_bios->tail) |
6158 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6159 | |
ffbd517d CM |
6160 | pending_bios->tail = bio; |
6161 | if (!pending_bios->head) | |
6162 | pending_bios->head = bio; | |
8b712842 CM |
6163 | if (device->running_pending) |
6164 | should_queue = 0; | |
6165 | ||
6166 | spin_unlock(&device->io_lock); | |
6167 | ||
6168 | if (should_queue) | |
0b246afa | 6169 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6170 | } |
6171 | ||
2ff7e61e JM |
6172 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6173 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6174 | { |
6175 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6176 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6177 | |
6178 | bio->bi_private = bbio; | |
9be3395b | 6179 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6180 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6181 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6182 | btrfs_debug_in_rcu(fs_info, |
6183 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6184 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6185 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6186 | bio->bi_iter.bi_size); | |
74d46992 | 6187 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6188 | |
2ff7e61e | 6189 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6190 | |
de1ee92a | 6191 | if (async) |
2ff7e61e | 6192 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6193 | else |
4e49ea4a | 6194 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6195 | } |
6196 | ||
de1ee92a JB |
6197 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6198 | { | |
6199 | atomic_inc(&bbio->error); | |
6200 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6201 | /* Should be the original bio. */ |
8408c716 MX |
6202 | WARN_ON(bio != bbio->orig_bio); |
6203 | ||
9be3395b | 6204 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6205 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6206 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6207 | bio->bi_status = BLK_STS_IOERR; | |
6208 | else | |
6209 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6210 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6211 | } |
6212 | } | |
6213 | ||
58efbc9f OS |
6214 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6215 | int mirror_num, int async_submit) | |
0b86a832 | 6216 | { |
0b86a832 | 6217 | struct btrfs_device *dev; |
8790d502 | 6218 | struct bio *first_bio = bio; |
4f024f37 | 6219 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6220 | u64 length = 0; |
6221 | u64 map_length; | |
0b86a832 | 6222 | int ret; |
08da757d ZL |
6223 | int dev_nr; |
6224 | int total_devs; | |
a1d3c478 | 6225 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6226 | |
4f024f37 | 6227 | length = bio->bi_iter.bi_size; |
0b86a832 | 6228 | map_length = length; |
cea9e445 | 6229 | |
0b246afa | 6230 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6231 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6232 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6233 | if (ret) { |
0b246afa | 6234 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6235 | return errno_to_blk_status(ret); |
c404e0dc | 6236 | } |
cea9e445 | 6237 | |
a1d3c478 | 6238 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6239 | bbio->orig_bio = first_bio; |
6240 | bbio->private = first_bio->bi_private; | |
6241 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6242 | bbio->fs_info = fs_info; |
53b381b3 DW |
6243 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6244 | ||
ad1ba2a0 | 6245 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6246 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6247 | /* In this case, map_length has been set to the length of |
6248 | a single stripe; not the whole write */ | |
37226b21 | 6249 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6250 | ret = raid56_parity_write(fs_info, bio, bbio, |
6251 | map_length); | |
53b381b3 | 6252 | } else { |
2ff7e61e JM |
6253 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6254 | map_length, mirror_num, 1); | |
53b381b3 | 6255 | } |
4245215d | 6256 | |
0b246afa | 6257 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6258 | return errno_to_blk_status(ret); |
53b381b3 DW |
6259 | } |
6260 | ||
cea9e445 | 6261 | if (map_length < length) { |
0b246afa | 6262 | btrfs_crit(fs_info, |
5d163e0e JM |
6263 | "mapping failed logical %llu bio len %llu len %llu", |
6264 | logical, length, map_length); | |
cea9e445 CM |
6265 | BUG(); |
6266 | } | |
a1d3c478 | 6267 | |
08da757d | 6268 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6269 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6270 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6271 | &dev->dev_state) || | |
ebbede42 AJ |
6272 | (bio_op(first_bio) == REQ_OP_WRITE && |
6273 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6274 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6275 | continue; |
6276 | } | |
6277 | ||
3aa8e074 | 6278 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6279 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6280 | else |
a1d3c478 | 6281 | bio = first_bio; |
de1ee92a | 6282 | |
2ff7e61e JM |
6283 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6284 | dev_nr, async_submit); | |
8790d502 | 6285 | } |
0b246afa | 6286 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6287 | return BLK_STS_OK; |
0b86a832 CM |
6288 | } |
6289 | ||
aa1b8cd4 | 6290 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6291 | u8 *uuid, u8 *fsid) |
0b86a832 | 6292 | { |
2b82032c YZ |
6293 | struct btrfs_device *device; |
6294 | struct btrfs_fs_devices *cur_devices; | |
6295 | ||
aa1b8cd4 | 6296 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6297 | while (cur_devices) { |
6298 | if (!fsid || | |
44880fdc | 6299 | !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6300 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6301 | if (device) |
6302 | return device; | |
6303 | } | |
6304 | cur_devices = cur_devices->seed; | |
6305 | } | |
6306 | return NULL; | |
0b86a832 CM |
6307 | } |
6308 | ||
2ff7e61e | 6309 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6310 | u64 devid, u8 *dev_uuid) |
6311 | { | |
6312 | struct btrfs_device *device; | |
dfe25020 | 6313 | |
12bd2fc0 ID |
6314 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6315 | if (IS_ERR(device)) | |
adfb69af | 6316 | return device; |
12bd2fc0 ID |
6317 | |
6318 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6319 | device->fs_devices = fs_devices; |
dfe25020 | 6320 | fs_devices->num_devices++; |
12bd2fc0 | 6321 | |
e6e674bd | 6322 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6323 | fs_devices->missing_devices++; |
12bd2fc0 | 6324 | |
dfe25020 CM |
6325 | return device; |
6326 | } | |
6327 | ||
12bd2fc0 ID |
6328 | /** |
6329 | * btrfs_alloc_device - allocate struct btrfs_device | |
6330 | * @fs_info: used only for generating a new devid, can be NULL if | |
6331 | * devid is provided (i.e. @devid != NULL). | |
6332 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6333 | * is generated. | |
6334 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6335 | * is generated. | |
6336 | * | |
6337 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6338 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6339 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6340 | */ |
6341 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6342 | const u64 *devid, | |
6343 | const u8 *uuid) | |
6344 | { | |
6345 | struct btrfs_device *dev; | |
6346 | u64 tmp; | |
6347 | ||
fae7f21c | 6348 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6349 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6350 | |
6351 | dev = __alloc_device(); | |
6352 | if (IS_ERR(dev)) | |
6353 | return dev; | |
6354 | ||
6355 | if (devid) | |
6356 | tmp = *devid; | |
6357 | else { | |
6358 | int ret; | |
6359 | ||
6360 | ret = find_next_devid(fs_info, &tmp); | |
6361 | if (ret) { | |
a425f9d4 | 6362 | btrfs_free_device(dev); |
12bd2fc0 ID |
6363 | return ERR_PTR(ret); |
6364 | } | |
6365 | } | |
6366 | dev->devid = tmp; | |
6367 | ||
6368 | if (uuid) | |
6369 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6370 | else | |
6371 | generate_random_uuid(dev->uuid); | |
6372 | ||
9e0af237 LB |
6373 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6374 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6375 | |
6376 | return dev; | |
6377 | } | |
6378 | ||
e06cd3dd | 6379 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6380 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6381 | struct extent_buffer *leaf, |
6382 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6383 | { |
0b86a832 | 6384 | u64 length; |
f04b772b | 6385 | u64 stripe_len; |
e06cd3dd LB |
6386 | u16 num_stripes; |
6387 | u16 sub_stripes; | |
6388 | u64 type; | |
315409b0 GJ |
6389 | u64 features; |
6390 | bool mixed = false; | |
0b86a832 | 6391 | |
e17cade2 | 6392 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6393 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6394 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6395 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6396 | type = btrfs_chunk_type(leaf, chunk); | |
6397 | ||
f04b772b | 6398 | if (!num_stripes) { |
0b246afa | 6399 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6400 | num_stripes); |
6401 | return -EIO; | |
6402 | } | |
0b246afa JM |
6403 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6404 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6405 | return -EIO; |
6406 | } | |
0b246afa JM |
6407 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6408 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6409 | btrfs_chunk_sector_size(leaf, chunk)); |
6410 | return -EIO; | |
6411 | } | |
0b246afa JM |
6412 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6413 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6414 | return -EIO; |
6415 | } | |
3d8da678 | 6416 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6417 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6418 | stripe_len); |
6419 | return -EIO; | |
6420 | } | |
6421 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6422 | type) { |
0b246afa | 6423 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6424 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6425 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6426 | btrfs_chunk_type(leaf, chunk)); | |
6427 | return -EIO; | |
6428 | } | |
315409b0 GJ |
6429 | |
6430 | if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) { | |
6431 | btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type); | |
6432 | return -EIO; | |
6433 | } | |
6434 | ||
6435 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
6436 | (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) { | |
6437 | btrfs_err(fs_info, | |
6438 | "system chunk with data or metadata type: 0x%llx", type); | |
6439 | return -EIO; | |
6440 | } | |
6441 | ||
6442 | features = btrfs_super_incompat_flags(fs_info->super_copy); | |
6443 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
6444 | mixed = true; | |
6445 | ||
6446 | if (!mixed) { | |
6447 | if ((type & BTRFS_BLOCK_GROUP_METADATA) && | |
6448 | (type & BTRFS_BLOCK_GROUP_DATA)) { | |
6449 | btrfs_err(fs_info, | |
6450 | "mixed chunk type in non-mixed mode: 0x%llx", type); | |
6451 | return -EIO; | |
6452 | } | |
6453 | } | |
6454 | ||
e06cd3dd LB |
6455 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6456 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6457 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6458 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6459 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6460 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6461 | num_stripes != 1)) { | |
0b246afa | 6462 | btrfs_err(fs_info, |
e06cd3dd LB |
6463 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6464 | num_stripes, sub_stripes, | |
6465 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6466 | return -EIO; | |
6467 | } | |
6468 | ||
6469 | return 0; | |
6470 | } | |
6471 | ||
5a2b8e60 | 6472 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6473 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6474 | { |
2b902dfc AJ |
6475 | if (error) |
6476 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6477 | devid, uuid); | |
6478 | else | |
6479 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6480 | devid, uuid); | |
5a2b8e60 AJ |
6481 | } |
6482 | ||
2ff7e61e | 6483 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6484 | struct extent_buffer *leaf, |
6485 | struct btrfs_chunk *chunk) | |
6486 | { | |
0b246afa | 6487 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6488 | struct map_lookup *map; |
6489 | struct extent_map *em; | |
6490 | u64 logical; | |
6491 | u64 length; | |
e06cd3dd LB |
6492 | u64 devid; |
6493 | u8 uuid[BTRFS_UUID_SIZE]; | |
6494 | int num_stripes; | |
6495 | int ret; | |
6496 | int i; | |
6497 | ||
6498 | logical = key->offset; | |
6499 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6500 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6501 | ||
2ff7e61e | 6502 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6503 | if (ret) |
6504 | return ret; | |
a061fc8d | 6505 | |
890871be | 6506 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6507 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6508 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6509 | |
6510 | /* already mapped? */ | |
6511 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6512 | free_extent_map(em); | |
0b86a832 CM |
6513 | return 0; |
6514 | } else if (em) { | |
6515 | free_extent_map(em); | |
6516 | } | |
0b86a832 | 6517 | |
172ddd60 | 6518 | em = alloc_extent_map(); |
0b86a832 CM |
6519 | if (!em) |
6520 | return -ENOMEM; | |
593060d7 | 6521 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6522 | if (!map) { |
6523 | free_extent_map(em); | |
6524 | return -ENOMEM; | |
6525 | } | |
6526 | ||
298a8f9c | 6527 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6528 | em->map_lookup = map; |
0b86a832 CM |
6529 | em->start = logical; |
6530 | em->len = length; | |
70c8a91c | 6531 | em->orig_start = 0; |
0b86a832 | 6532 | em->block_start = 0; |
c8b97818 | 6533 | em->block_len = em->len; |
0b86a832 | 6534 | |
593060d7 CM |
6535 | map->num_stripes = num_stripes; |
6536 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6537 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6538 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6539 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6540 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6541 | map->verified_stripes = 0; |
593060d7 CM |
6542 | for (i = 0; i < num_stripes; i++) { |
6543 | map->stripes[i].physical = | |
6544 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6545 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6546 | read_extent_buffer(leaf, uuid, (unsigned long) |
6547 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6548 | BTRFS_UUID_SIZE); | |
0b246afa | 6549 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6550 | uuid, NULL); |
3cdde224 | 6551 | if (!map->stripes[i].dev && |
0b246afa | 6552 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6553 | free_extent_map(em); |
2b902dfc | 6554 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6555 | return -ENOENT; |
593060d7 | 6556 | } |
dfe25020 CM |
6557 | if (!map->stripes[i].dev) { |
6558 | map->stripes[i].dev = | |
2ff7e61e JM |
6559 | add_missing_dev(fs_info->fs_devices, devid, |
6560 | uuid); | |
adfb69af | 6561 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6562 | free_extent_map(em); |
adfb69af AJ |
6563 | btrfs_err(fs_info, |
6564 | "failed to init missing dev %llu: %ld", | |
6565 | devid, PTR_ERR(map->stripes[i].dev)); | |
6566 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6567 | } |
2b902dfc | 6568 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6569 | } |
e12c9621 AJ |
6570 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6571 | &(map->stripes[i].dev->dev_state)); | |
6572 | ||
0b86a832 CM |
6573 | } |
6574 | ||
890871be | 6575 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6576 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6577 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6578 | if (ret < 0) { |
6579 | btrfs_err(fs_info, | |
6580 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6581 | em->start, em->len, ret); | |
6582 | } | |
0b86a832 CM |
6583 | free_extent_map(em); |
6584 | ||
64f64f43 | 6585 | return ret; |
0b86a832 CM |
6586 | } |
6587 | ||
143bede5 | 6588 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6589 | struct btrfs_dev_item *dev_item, |
6590 | struct btrfs_device *device) | |
6591 | { | |
6592 | unsigned long ptr; | |
0b86a832 CM |
6593 | |
6594 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6595 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6596 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6597 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6598 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6599 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6600 | device->type = btrfs_device_type(leaf, dev_item); |
6601 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6602 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6603 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6604 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6605 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6606 | |
410ba3a2 | 6607 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6608 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6609 | } |
6610 | ||
2ff7e61e | 6611 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6612 | u8 *fsid) |
2b82032c YZ |
6613 | { |
6614 | struct btrfs_fs_devices *fs_devices; | |
6615 | int ret; | |
6616 | ||
a32bf9a3 | 6617 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6618 | ASSERT(fsid); |
2b82032c | 6619 | |
0b246afa | 6620 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6621 | while (fs_devices) { |
44880fdc | 6622 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6623 | return fs_devices; |
6624 | ||
2b82032c YZ |
6625 | fs_devices = fs_devices->seed; |
6626 | } | |
6627 | ||
6628 | fs_devices = find_fsid(fsid); | |
6629 | if (!fs_devices) { | |
0b246afa | 6630 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6631 | return ERR_PTR(-ENOENT); |
6632 | ||
6633 | fs_devices = alloc_fs_devices(fsid); | |
6634 | if (IS_ERR(fs_devices)) | |
6635 | return fs_devices; | |
6636 | ||
6637 | fs_devices->seeding = 1; | |
6638 | fs_devices->opened = 1; | |
6639 | return fs_devices; | |
2b82032c | 6640 | } |
e4404d6e YZ |
6641 | |
6642 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6643 | if (IS_ERR(fs_devices)) |
6644 | return fs_devices; | |
2b82032c | 6645 | |
897fb573 | 6646 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6647 | if (ret) { |
6648 | free_fs_devices(fs_devices); | |
5f375835 | 6649 | fs_devices = ERR_PTR(ret); |
2b82032c | 6650 | goto out; |
48d28232 | 6651 | } |
2b82032c YZ |
6652 | |
6653 | if (!fs_devices->seeding) { | |
0226e0eb | 6654 | close_fs_devices(fs_devices); |
e4404d6e | 6655 | free_fs_devices(fs_devices); |
5f375835 | 6656 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6657 | goto out; |
6658 | } | |
6659 | ||
0b246afa JM |
6660 | fs_devices->seed = fs_info->fs_devices->seed; |
6661 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6662 | out: |
5f375835 | 6663 | return fs_devices; |
2b82032c YZ |
6664 | } |
6665 | ||
2ff7e61e | 6666 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6667 | struct extent_buffer *leaf, |
6668 | struct btrfs_dev_item *dev_item) | |
6669 | { | |
0b246afa | 6670 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6671 | struct btrfs_device *device; |
6672 | u64 devid; | |
6673 | int ret; | |
44880fdc | 6674 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6675 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6676 | ||
0b86a832 | 6677 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6678 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6679 | BTRFS_UUID_SIZE); |
1473b24e | 6680 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6681 | BTRFS_FSID_SIZE); |
2b82032c | 6682 | |
44880fdc | 6683 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6684 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6685 | if (IS_ERR(fs_devices)) |
6686 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6687 | } |
6688 | ||
0b246afa | 6689 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6690 | if (!device) { |
c5502451 | 6691 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6692 | btrfs_report_missing_device(fs_info, devid, |
6693 | dev_uuid, true); | |
45dbdbc9 | 6694 | return -ENOENT; |
c5502451 | 6695 | } |
2b82032c | 6696 | |
2ff7e61e | 6697 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6698 | if (IS_ERR(device)) { |
6699 | btrfs_err(fs_info, | |
6700 | "failed to add missing dev %llu: %ld", | |
6701 | devid, PTR_ERR(device)); | |
6702 | return PTR_ERR(device); | |
6703 | } | |
2b902dfc | 6704 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6705 | } else { |
c5502451 | 6706 | if (!device->bdev) { |
2b902dfc AJ |
6707 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6708 | btrfs_report_missing_device(fs_info, | |
6709 | devid, dev_uuid, true); | |
45dbdbc9 | 6710 | return -ENOENT; |
2b902dfc AJ |
6711 | } |
6712 | btrfs_report_missing_device(fs_info, devid, | |
6713 | dev_uuid, false); | |
c5502451 | 6714 | } |
5f375835 | 6715 | |
e6e674bd AJ |
6716 | if (!device->bdev && |
6717 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6718 | /* |
6719 | * this happens when a device that was properly setup | |
6720 | * in the device info lists suddenly goes bad. | |
6721 | * device->bdev is NULL, and so we have to set | |
6722 | * device->missing to one here | |
6723 | */ | |
5f375835 | 6724 | device->fs_devices->missing_devices++; |
e6e674bd | 6725 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6726 | } |
5f375835 MX |
6727 | |
6728 | /* Move the device to its own fs_devices */ | |
6729 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6730 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6731 | &device->dev_state)); | |
5f375835 MX |
6732 | |
6733 | list_move(&device->dev_list, &fs_devices->devices); | |
6734 | device->fs_devices->num_devices--; | |
6735 | fs_devices->num_devices++; | |
6736 | ||
6737 | device->fs_devices->missing_devices--; | |
6738 | fs_devices->missing_devices++; | |
6739 | ||
6740 | device->fs_devices = fs_devices; | |
6741 | } | |
2b82032c YZ |
6742 | } |
6743 | ||
0b246afa | 6744 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6745 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6746 | if (device->generation != |
6747 | btrfs_device_generation(leaf, dev_item)) | |
6748 | return -EINVAL; | |
6324fbf3 | 6749 | } |
0b86a832 CM |
6750 | |
6751 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6752 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6753 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6754 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6755 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6756 | atomic64_add(device->total_bytes - device->bytes_used, |
6757 | &fs_info->free_chunk_space); | |
2bf64758 | 6758 | } |
0b86a832 | 6759 | ret = 0; |
0b86a832 CM |
6760 | return ret; |
6761 | } | |
6762 | ||
6bccf3ab | 6763 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6764 | { |
6bccf3ab | 6765 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6766 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6767 | struct extent_buffer *sb; |
0b86a832 | 6768 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6769 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6770 | u8 *array_ptr; |
6771 | unsigned long sb_array_offset; | |
84eed90f | 6772 | int ret = 0; |
0b86a832 CM |
6773 | u32 num_stripes; |
6774 | u32 array_size; | |
6775 | u32 len = 0; | |
1ffb22cf | 6776 | u32 cur_offset; |
e06cd3dd | 6777 | u64 type; |
84eed90f | 6778 | struct btrfs_key key; |
0b86a832 | 6779 | |
0b246afa | 6780 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6781 | /* |
6782 | * This will create extent buffer of nodesize, superblock size is | |
6783 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6784 | * overallocate but we can keep it as-is, only the first page is used. | |
6785 | */ | |
2ff7e61e | 6786 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6787 | if (IS_ERR(sb)) |
6788 | return PTR_ERR(sb); | |
4db8c528 | 6789 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6790 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6791 | /* |
01327610 | 6792 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6793 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6794 | * pages up-to-date when the page is larger: extent does not cover the |
6795 | * whole page and consequently check_page_uptodate does not find all | |
6796 | * the page's extents up-to-date (the hole beyond sb), | |
6797 | * write_extent_buffer then triggers a WARN_ON. | |
6798 | * | |
6799 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6800 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6801 | * to silence the warning eg. on PowerPC 64. | |
6802 | */ | |
09cbfeaf | 6803 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6804 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6805 | |
a061fc8d | 6806 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6807 | array_size = btrfs_super_sys_array_size(super_copy); |
6808 | ||
1ffb22cf DS |
6809 | array_ptr = super_copy->sys_chunk_array; |
6810 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6811 | cur_offset = 0; | |
0b86a832 | 6812 | |
1ffb22cf DS |
6813 | while (cur_offset < array_size) { |
6814 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6815 | len = sizeof(*disk_key); |
6816 | if (cur_offset + len > array_size) | |
6817 | goto out_short_read; | |
6818 | ||
0b86a832 CM |
6819 | btrfs_disk_key_to_cpu(&key, disk_key); |
6820 | ||
1ffb22cf DS |
6821 | array_ptr += len; |
6822 | sb_array_offset += len; | |
6823 | cur_offset += len; | |
0b86a832 | 6824 | |
0d81ba5d | 6825 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6826 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6827 | /* |
6828 | * At least one btrfs_chunk with one stripe must be | |
6829 | * present, exact stripe count check comes afterwards | |
6830 | */ | |
6831 | len = btrfs_chunk_item_size(1); | |
6832 | if (cur_offset + len > array_size) | |
6833 | goto out_short_read; | |
6834 | ||
6835 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6836 | if (!num_stripes) { |
ab8d0fc4 JM |
6837 | btrfs_err(fs_info, |
6838 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6839 | num_stripes, cur_offset); |
6840 | ret = -EIO; | |
6841 | break; | |
6842 | } | |
6843 | ||
e06cd3dd LB |
6844 | type = btrfs_chunk_type(sb, chunk); |
6845 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6846 | btrfs_err(fs_info, |
e06cd3dd LB |
6847 | "invalid chunk type %llu in sys_array at offset %u", |
6848 | type, cur_offset); | |
6849 | ret = -EIO; | |
6850 | break; | |
6851 | } | |
6852 | ||
e3540eab DS |
6853 | len = btrfs_chunk_item_size(num_stripes); |
6854 | if (cur_offset + len > array_size) | |
6855 | goto out_short_read; | |
6856 | ||
2ff7e61e | 6857 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
6858 | if (ret) |
6859 | break; | |
0b86a832 | 6860 | } else { |
ab8d0fc4 JM |
6861 | btrfs_err(fs_info, |
6862 | "unexpected item type %u in sys_array at offset %u", | |
6863 | (u32)key.type, cur_offset); | |
84eed90f CM |
6864 | ret = -EIO; |
6865 | break; | |
0b86a832 | 6866 | } |
1ffb22cf DS |
6867 | array_ptr += len; |
6868 | sb_array_offset += len; | |
6869 | cur_offset += len; | |
0b86a832 | 6870 | } |
d865177a | 6871 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6872 | free_extent_buffer_stale(sb); |
84eed90f | 6873 | return ret; |
e3540eab DS |
6874 | |
6875 | out_short_read: | |
ab8d0fc4 | 6876 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6877 | len, cur_offset); |
d865177a | 6878 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6879 | free_extent_buffer_stale(sb); |
e3540eab | 6880 | return -EIO; |
0b86a832 CM |
6881 | } |
6882 | ||
21634a19 QW |
6883 | /* |
6884 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6885 | * | |
6528b99d AJ |
6886 | * If the @failing_dev is specified, it's accounted as missing. |
6887 | * | |
21634a19 QW |
6888 | * Return true if all chunks meet the minimal RW mount requirements. |
6889 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6890 | */ | |
6528b99d AJ |
6891 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6892 | struct btrfs_device *failing_dev) | |
21634a19 QW |
6893 | { |
6894 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
6895 | struct extent_map *em; | |
6896 | u64 next_start = 0; | |
6897 | bool ret = true; | |
6898 | ||
6899 | read_lock(&map_tree->map_tree.lock); | |
6900 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
6901 | read_unlock(&map_tree->map_tree.lock); | |
6902 | /* No chunk at all? Return false anyway */ | |
6903 | if (!em) { | |
6904 | ret = false; | |
6905 | goto out; | |
6906 | } | |
6907 | while (em) { | |
6908 | struct map_lookup *map; | |
6909 | int missing = 0; | |
6910 | int max_tolerated; | |
6911 | int i; | |
6912 | ||
6913 | map = em->map_lookup; | |
6914 | max_tolerated = | |
6915 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6916 | map->type); | |
6917 | for (i = 0; i < map->num_stripes; i++) { | |
6918 | struct btrfs_device *dev = map->stripes[i].dev; | |
6919 | ||
e6e674bd AJ |
6920 | if (!dev || !dev->bdev || |
6921 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6922 | dev->last_flush_error) |
6923 | missing++; | |
6528b99d AJ |
6924 | else if (failing_dev && failing_dev == dev) |
6925 | missing++; | |
21634a19 QW |
6926 | } |
6927 | if (missing > max_tolerated) { | |
6528b99d AJ |
6928 | if (!failing_dev) |
6929 | btrfs_warn(fs_info, | |
21634a19 QW |
6930 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
6931 | em->start, missing, max_tolerated); | |
6932 | free_extent_map(em); | |
6933 | ret = false; | |
6934 | goto out; | |
6935 | } | |
6936 | next_start = extent_map_end(em); | |
6937 | free_extent_map(em); | |
6938 | ||
6939 | read_lock(&map_tree->map_tree.lock); | |
6940 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
6941 | (u64)(-1) - next_start); | |
6942 | read_unlock(&map_tree->map_tree.lock); | |
6943 | } | |
6944 | out: | |
6945 | return ret; | |
6946 | } | |
6947 | ||
5b4aacef | 6948 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6949 | { |
5b4aacef | 6950 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6951 | struct btrfs_path *path; |
6952 | struct extent_buffer *leaf; | |
6953 | struct btrfs_key key; | |
6954 | struct btrfs_key found_key; | |
6955 | int ret; | |
6956 | int slot; | |
99e3ecfc | 6957 | u64 total_dev = 0; |
0b86a832 | 6958 | |
0b86a832 CM |
6959 | path = btrfs_alloc_path(); |
6960 | if (!path) | |
6961 | return -ENOMEM; | |
6962 | ||
3dd0f7a3 AJ |
6963 | /* |
6964 | * uuid_mutex is needed only if we are mounting a sprout FS | |
6965 | * otherwise we don't need it. | |
6966 | */ | |
b367e47f | 6967 | mutex_lock(&uuid_mutex); |
34441361 | 6968 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6969 | |
395927a9 FDBM |
6970 | /* |
6971 | * Read all device items, and then all the chunk items. All | |
6972 | * device items are found before any chunk item (their object id | |
6973 | * is smaller than the lowest possible object id for a chunk | |
6974 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6975 | */ |
6976 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6977 | key.offset = 0; | |
6978 | key.type = 0; | |
0b86a832 | 6979 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6980 | if (ret < 0) |
6981 | goto error; | |
d397712b | 6982 | while (1) { |
0b86a832 CM |
6983 | leaf = path->nodes[0]; |
6984 | slot = path->slots[0]; | |
6985 | if (slot >= btrfs_header_nritems(leaf)) { | |
6986 | ret = btrfs_next_leaf(root, path); | |
6987 | if (ret == 0) | |
6988 | continue; | |
6989 | if (ret < 0) | |
6990 | goto error; | |
6991 | break; | |
6992 | } | |
6993 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6994 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6995 | struct btrfs_dev_item *dev_item; | |
6996 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6997 | struct btrfs_dev_item); |
2ff7e61e | 6998 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
6999 | if (ret) |
7000 | goto error; | |
99e3ecfc | 7001 | total_dev++; |
0b86a832 CM |
7002 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7003 | struct btrfs_chunk *chunk; | |
7004 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7005 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7006 | if (ret) |
7007 | goto error; | |
0b86a832 CM |
7008 | } |
7009 | path->slots[0]++; | |
7010 | } | |
99e3ecfc LB |
7011 | |
7012 | /* | |
7013 | * After loading chunk tree, we've got all device information, | |
7014 | * do another round of validation checks. | |
7015 | */ | |
0b246afa JM |
7016 | if (total_dev != fs_info->fs_devices->total_devices) { |
7017 | btrfs_err(fs_info, | |
99e3ecfc | 7018 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7019 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7020 | total_dev); |
7021 | ret = -EINVAL; | |
7022 | goto error; | |
7023 | } | |
0b246afa JM |
7024 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7025 | fs_info->fs_devices->total_rw_bytes) { | |
7026 | btrfs_err(fs_info, | |
99e3ecfc | 7027 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7028 | btrfs_super_total_bytes(fs_info->super_copy), |
7029 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7030 | ret = -EINVAL; |
7031 | goto error; | |
7032 | } | |
0b86a832 CM |
7033 | ret = 0; |
7034 | error: | |
34441361 | 7035 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7036 | mutex_unlock(&uuid_mutex); |
7037 | ||
2b82032c | 7038 | btrfs_free_path(path); |
0b86a832 CM |
7039 | return ret; |
7040 | } | |
442a4f63 | 7041 | |
cb517eab MX |
7042 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7043 | { | |
7044 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7045 | struct btrfs_device *device; | |
7046 | ||
29cc83f6 LB |
7047 | while (fs_devices) { |
7048 | mutex_lock(&fs_devices->device_list_mutex); | |
7049 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7050 | device->fs_info = fs_info; |
29cc83f6 LB |
7051 | mutex_unlock(&fs_devices->device_list_mutex); |
7052 | ||
7053 | fs_devices = fs_devices->seed; | |
7054 | } | |
cb517eab MX |
7055 | } |
7056 | ||
733f4fbb SB |
7057 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7058 | { | |
7059 | int i; | |
7060 | ||
7061 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7062 | btrfs_dev_stat_reset(dev, i); | |
7063 | } | |
7064 | ||
7065 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7066 | { | |
7067 | struct btrfs_key key; | |
7068 | struct btrfs_key found_key; | |
7069 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7070 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7071 | struct extent_buffer *eb; | |
7072 | int slot; | |
7073 | int ret = 0; | |
7074 | struct btrfs_device *device; | |
7075 | struct btrfs_path *path = NULL; | |
7076 | int i; | |
7077 | ||
7078 | path = btrfs_alloc_path(); | |
7079 | if (!path) { | |
7080 | ret = -ENOMEM; | |
7081 | goto out; | |
7082 | } | |
7083 | ||
7084 | mutex_lock(&fs_devices->device_list_mutex); | |
7085 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7086 | int item_size; | |
7087 | struct btrfs_dev_stats_item *ptr; | |
7088 | ||
242e2956 DS |
7089 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7090 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7091 | key.offset = device->devid; |
7092 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7093 | if (ret) { | |
733f4fbb SB |
7094 | __btrfs_reset_dev_stats(device); |
7095 | device->dev_stats_valid = 1; | |
7096 | btrfs_release_path(path); | |
7097 | continue; | |
7098 | } | |
7099 | slot = path->slots[0]; | |
7100 | eb = path->nodes[0]; | |
7101 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7102 | item_size = btrfs_item_size_nr(eb, slot); | |
7103 | ||
7104 | ptr = btrfs_item_ptr(eb, slot, | |
7105 | struct btrfs_dev_stats_item); | |
7106 | ||
7107 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7108 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7109 | btrfs_dev_stat_set(device, i, | |
7110 | btrfs_dev_stats_value(eb, ptr, i)); | |
7111 | else | |
7112 | btrfs_dev_stat_reset(device, i); | |
7113 | } | |
7114 | ||
7115 | device->dev_stats_valid = 1; | |
7116 | btrfs_dev_stat_print_on_load(device); | |
7117 | btrfs_release_path(path); | |
7118 | } | |
7119 | mutex_unlock(&fs_devices->device_list_mutex); | |
7120 | ||
7121 | out: | |
7122 | btrfs_free_path(path); | |
7123 | return ret < 0 ? ret : 0; | |
7124 | } | |
7125 | ||
7126 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7127 | struct btrfs_device *device) |
7128 | { | |
5495f195 | 7129 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7130 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7131 | struct btrfs_path *path; |
7132 | struct btrfs_key key; | |
7133 | struct extent_buffer *eb; | |
7134 | struct btrfs_dev_stats_item *ptr; | |
7135 | int ret; | |
7136 | int i; | |
7137 | ||
242e2956 DS |
7138 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7139 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7140 | key.offset = device->devid; |
7141 | ||
7142 | path = btrfs_alloc_path(); | |
fa252992 DS |
7143 | if (!path) |
7144 | return -ENOMEM; | |
733f4fbb SB |
7145 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7146 | if (ret < 0) { | |
0b246afa | 7147 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7148 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7149 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7150 | goto out; |
7151 | } | |
7152 | ||
7153 | if (ret == 0 && | |
7154 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7155 | /* need to delete old one and insert a new one */ | |
7156 | ret = btrfs_del_item(trans, dev_root, path); | |
7157 | if (ret != 0) { | |
0b246afa | 7158 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7159 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7160 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7161 | goto out; |
7162 | } | |
7163 | ret = 1; | |
7164 | } | |
7165 | ||
7166 | if (ret == 1) { | |
7167 | /* need to insert a new item */ | |
7168 | btrfs_release_path(path); | |
7169 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7170 | &key, sizeof(*ptr)); | |
7171 | if (ret < 0) { | |
0b246afa | 7172 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7173 | "insert dev_stats item for device %s failed %d", |
7174 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7175 | goto out; |
7176 | } | |
7177 | } | |
7178 | ||
7179 | eb = path->nodes[0]; | |
7180 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7181 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7182 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7183 | btrfs_dev_stat_read(device, i)); | |
7184 | btrfs_mark_buffer_dirty(eb); | |
7185 | ||
7186 | out: | |
7187 | btrfs_free_path(path); | |
7188 | return ret; | |
7189 | } | |
7190 | ||
7191 | /* | |
7192 | * called from commit_transaction. Writes all changed device stats to disk. | |
7193 | */ | |
7194 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7195 | struct btrfs_fs_info *fs_info) | |
7196 | { | |
733f4fbb SB |
7197 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7198 | struct btrfs_device *device; | |
addc3fa7 | 7199 | int stats_cnt; |
733f4fbb SB |
7200 | int ret = 0; |
7201 | ||
7202 | mutex_lock(&fs_devices->device_list_mutex); | |
7203 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7204 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7205 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7206 | continue; |
7207 | ||
9deae968 NB |
7208 | |
7209 | /* | |
7210 | * There is a LOAD-LOAD control dependency between the value of | |
7211 | * dev_stats_ccnt and updating the on-disk values which requires | |
7212 | * reading the in-memory counters. Such control dependencies | |
7213 | * require explicit read memory barriers. | |
7214 | * | |
7215 | * This memory barriers pairs with smp_mb__before_atomic in | |
7216 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7217 | * barrier implied by atomic_xchg in | |
7218 | * btrfs_dev_stats_read_and_reset | |
7219 | */ | |
7220 | smp_rmb(); | |
7221 | ||
5495f195 | 7222 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7223 | if (!ret) |
addc3fa7 | 7224 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7225 | } |
7226 | mutex_unlock(&fs_devices->device_list_mutex); | |
7227 | ||
7228 | return ret; | |
7229 | } | |
7230 | ||
442a4f63 SB |
7231 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7232 | { | |
7233 | btrfs_dev_stat_inc(dev, index); | |
7234 | btrfs_dev_stat_print_on_error(dev); | |
7235 | } | |
7236 | ||
48a3b636 | 7237 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7238 | { |
733f4fbb SB |
7239 | if (!dev->dev_stats_valid) |
7240 | return; | |
fb456252 | 7241 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7242 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7243 | rcu_str_deref(dev->name), |
442a4f63 SB |
7244 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7245 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7246 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7247 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7248 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7249 | } |
c11d2c23 | 7250 | |
733f4fbb SB |
7251 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7252 | { | |
a98cdb85 SB |
7253 | int i; |
7254 | ||
7255 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7256 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7257 | break; | |
7258 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7259 | return; /* all values == 0, suppress message */ | |
7260 | ||
fb456252 | 7261 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7262 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7263 | rcu_str_deref(dev->name), |
733f4fbb SB |
7264 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7265 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7266 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7267 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7268 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7269 | } | |
7270 | ||
2ff7e61e | 7271 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7272 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7273 | { |
7274 | struct btrfs_device *dev; | |
0b246afa | 7275 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7276 | int i; |
7277 | ||
7278 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7279 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7280 | mutex_unlock(&fs_devices->device_list_mutex); |
7281 | ||
7282 | if (!dev) { | |
0b246afa | 7283 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7284 | return -ENODEV; |
733f4fbb | 7285 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7286 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7287 | return -ENODEV; |
b27f7c0c | 7288 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7289 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7290 | if (stats->nr_items > i) | |
7291 | stats->values[i] = | |
7292 | btrfs_dev_stat_read_and_reset(dev, i); | |
7293 | else | |
7294 | btrfs_dev_stat_reset(dev, i); | |
7295 | } | |
7296 | } else { | |
7297 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7298 | if (stats->nr_items > i) | |
7299 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7300 | } | |
7301 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7302 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7303 | return 0; | |
7304 | } | |
a8a6dab7 | 7305 | |
da353f6b | 7306 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7307 | { |
7308 | struct buffer_head *bh; | |
7309 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7310 | int copy_num; |
a8a6dab7 | 7311 | |
12b1c263 AJ |
7312 | if (!bdev) |
7313 | return; | |
a8a6dab7 | 7314 | |
12b1c263 AJ |
7315 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7316 | copy_num++) { | |
a8a6dab7 | 7317 | |
12b1c263 AJ |
7318 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7319 | continue; | |
7320 | ||
7321 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7322 | ||
7323 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7324 | set_buffer_dirty(bh); | |
7325 | sync_dirty_buffer(bh); | |
7326 | brelse(bh); | |
7327 | } | |
7328 | ||
7329 | /* Notify udev that device has changed */ | |
7330 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7331 | ||
7332 | /* Update ctime/mtime for device path for libblkid */ | |
7333 | update_dev_time(device_path); | |
a8a6dab7 | 7334 | } |
935e5cc9 MX |
7335 | |
7336 | /* | |
7337 | * Update the size of all devices, which is used for writing out the | |
7338 | * super blocks. | |
7339 | */ | |
7340 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7341 | { | |
7342 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7343 | struct btrfs_device *curr, *next; | |
7344 | ||
7345 | if (list_empty(&fs_devices->resized_devices)) | |
7346 | return; | |
7347 | ||
7348 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7349 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7350 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7351 | resized_list) { | |
7352 | list_del_init(&curr->resized_list); | |
7353 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7354 | } | |
34441361 | 7355 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7356 | mutex_unlock(&fs_devices->device_list_mutex); |
7357 | } | |
ce7213c7 MX |
7358 | |
7359 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7360 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7361 | { |
e9b919b1 | 7362 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7363 | struct extent_map *em; |
7364 | struct map_lookup *map; | |
7365 | struct btrfs_device *dev; | |
7366 | int i; | |
7367 | ||
e9b919b1 | 7368 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7369 | return; |
7370 | ||
7371 | /* In order to kick the device replace finish process */ | |
34441361 | 7372 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7373 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7374 | map = em->map_lookup; |
ce7213c7 MX |
7375 | |
7376 | for (i = 0; i < map->num_stripes; i++) { | |
7377 | dev = map->stripes[i].dev; | |
7378 | dev->commit_bytes_used = dev->bytes_used; | |
7379 | } | |
7380 | } | |
34441361 | 7381 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7382 | } |
5a13f430 AJ |
7383 | |
7384 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7385 | { | |
7386 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7387 | while (fs_devices) { | |
7388 | fs_devices->fs_info = fs_info; | |
7389 | fs_devices = fs_devices->seed; | |
7390 | } | |
7391 | } | |
7392 | ||
7393 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7394 | { | |
7395 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7396 | while (fs_devices) { | |
7397 | fs_devices->fs_info = NULL; | |
7398 | fs_devices = fs_devices->seed; | |
7399 | } | |
7400 | } | |
46df06b8 DS |
7401 | |
7402 | /* | |
7403 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7404 | */ | |
7405 | int btrfs_bg_type_to_factor(u64 flags) | |
7406 | { | |
7407 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7408 | BTRFS_BLOCK_GROUP_RAID10)) | |
7409 | return 2; | |
7410 | return 1; | |
7411 | } | |
cf90d884 QW |
7412 | |
7413 | ||
7414 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) | |
7415 | { | |
7416 | int index = btrfs_bg_flags_to_raid_index(type); | |
7417 | int ncopies = btrfs_raid_array[index].ncopies; | |
7418 | int data_stripes; | |
7419 | ||
7420 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
7421 | case BTRFS_BLOCK_GROUP_RAID5: | |
7422 | data_stripes = num_stripes - 1; | |
7423 | break; | |
7424 | case BTRFS_BLOCK_GROUP_RAID6: | |
7425 | data_stripes = num_stripes - 2; | |
7426 | break; | |
7427 | default: | |
7428 | data_stripes = num_stripes / ncopies; | |
7429 | break; | |
7430 | } | |
7431 | return div_u64(chunk_len, data_stripes); | |
7432 | } | |
7433 | ||
7434 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7435 | u64 chunk_offset, u64 devid, | |
7436 | u64 physical_offset, u64 physical_len) | |
7437 | { | |
7438 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7439 | struct extent_map *em; | |
7440 | struct map_lookup *map; | |
05a37c48 | 7441 | struct btrfs_device *dev; |
cf90d884 QW |
7442 | u64 stripe_len; |
7443 | bool found = false; | |
7444 | int ret = 0; | |
7445 | int i; | |
7446 | ||
7447 | read_lock(&em_tree->lock); | |
7448 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7449 | read_unlock(&em_tree->lock); | |
7450 | ||
7451 | if (!em) { | |
7452 | btrfs_err(fs_info, | |
7453 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7454 | physical_offset, devid); | |
7455 | ret = -EUCLEAN; | |
7456 | goto out; | |
7457 | } | |
7458 | ||
7459 | map = em->map_lookup; | |
7460 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7461 | if (physical_len != stripe_len) { | |
7462 | btrfs_err(fs_info, | |
7463 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7464 | physical_offset, devid, em->start, physical_len, | |
7465 | stripe_len); | |
7466 | ret = -EUCLEAN; | |
7467 | goto out; | |
7468 | } | |
7469 | ||
7470 | for (i = 0; i < map->num_stripes; i++) { | |
7471 | if (map->stripes[i].dev->devid == devid && | |
7472 | map->stripes[i].physical == physical_offset) { | |
7473 | found = true; | |
7474 | if (map->verified_stripes >= map->num_stripes) { | |
7475 | btrfs_err(fs_info, | |
7476 | "too many dev extents for chunk %llu found", | |
7477 | em->start); | |
7478 | ret = -EUCLEAN; | |
7479 | goto out; | |
7480 | } | |
7481 | map->verified_stripes++; | |
7482 | break; | |
7483 | } | |
7484 | } | |
7485 | if (!found) { | |
7486 | btrfs_err(fs_info, | |
7487 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7488 | physical_offset, devid); | |
7489 | ret = -EUCLEAN; | |
7490 | } | |
05a37c48 QW |
7491 | |
7492 | /* Make sure no dev extent is beyond device bondary */ | |
7493 | dev = btrfs_find_device(fs_info, devid, NULL, NULL); | |
7494 | if (!dev) { | |
7495 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7496 | ret = -EUCLEAN; | |
7497 | goto out; | |
7498 | } | |
7499 | if (physical_offset + physical_len > dev->disk_total_bytes) { | |
7500 | btrfs_err(fs_info, | |
7501 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7502 | devid, physical_offset, physical_len, | |
7503 | dev->disk_total_bytes); | |
7504 | ret = -EUCLEAN; | |
7505 | goto out; | |
7506 | } | |
cf90d884 QW |
7507 | out: |
7508 | free_extent_map(em); | |
7509 | return ret; | |
7510 | } | |
7511 | ||
7512 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7513 | { | |
7514 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7515 | struct extent_map *em; | |
7516 | struct rb_node *node; | |
7517 | int ret = 0; | |
7518 | ||
7519 | read_lock(&em_tree->lock); | |
07e1ce09 | 7520 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7521 | em = rb_entry(node, struct extent_map, rb_node); |
7522 | if (em->map_lookup->num_stripes != | |
7523 | em->map_lookup->verified_stripes) { | |
7524 | btrfs_err(fs_info, | |
7525 | "chunk %llu has missing dev extent, have %d expect %d", | |
7526 | em->start, em->map_lookup->verified_stripes, | |
7527 | em->map_lookup->num_stripes); | |
7528 | ret = -EUCLEAN; | |
7529 | goto out; | |
7530 | } | |
7531 | } | |
7532 | out: | |
7533 | read_unlock(&em_tree->lock); | |
7534 | return ret; | |
7535 | } | |
7536 | ||
7537 | /* | |
7538 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7539 | * later chunk allocation/free would cause unexpected behavior. | |
7540 | * | |
7541 | * NOTE: This will iterate through the whole device tree, which should be of | |
7542 | * the same size level as the chunk tree. This slightly increases mount time. | |
7543 | */ | |
7544 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7545 | { | |
7546 | struct btrfs_path *path; | |
7547 | struct btrfs_root *root = fs_info->dev_root; | |
7548 | struct btrfs_key key; | |
5eb19381 QW |
7549 | u64 prev_devid = 0; |
7550 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7551 | int ret = 0; |
7552 | ||
7553 | key.objectid = 1; | |
7554 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7555 | key.offset = 0; | |
7556 | ||
7557 | path = btrfs_alloc_path(); | |
7558 | if (!path) | |
7559 | return -ENOMEM; | |
7560 | ||
7561 | path->reada = READA_FORWARD; | |
7562 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7563 | if (ret < 0) | |
7564 | goto out; | |
7565 | ||
7566 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7567 | ret = btrfs_next_item(root, path); | |
7568 | if (ret < 0) | |
7569 | goto out; | |
7570 | /* No dev extents at all? Not good */ | |
7571 | if (ret > 0) { | |
7572 | ret = -EUCLEAN; | |
7573 | goto out; | |
7574 | } | |
7575 | } | |
7576 | while (1) { | |
7577 | struct extent_buffer *leaf = path->nodes[0]; | |
7578 | struct btrfs_dev_extent *dext; | |
7579 | int slot = path->slots[0]; | |
7580 | u64 chunk_offset; | |
7581 | u64 physical_offset; | |
7582 | u64 physical_len; | |
7583 | u64 devid; | |
7584 | ||
7585 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7586 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7587 | break; | |
7588 | devid = key.objectid; | |
7589 | physical_offset = key.offset; | |
7590 | ||
7591 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7592 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7593 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7594 | ||
5eb19381 QW |
7595 | /* Check if this dev extent overlaps with the previous one */ |
7596 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7597 | btrfs_err(fs_info, | |
7598 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7599 | devid, physical_offset, prev_dev_ext_end); | |
7600 | ret = -EUCLEAN; | |
7601 | goto out; | |
7602 | } | |
7603 | ||
cf90d884 QW |
7604 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7605 | physical_offset, physical_len); | |
7606 | if (ret < 0) | |
7607 | goto out; | |
5eb19381 QW |
7608 | prev_devid = devid; |
7609 | prev_dev_ext_end = physical_offset + physical_len; | |
7610 | ||
cf90d884 QW |
7611 | ret = btrfs_next_item(root, path); |
7612 | if (ret < 0) | |
7613 | goto out; | |
7614 | if (ret > 0) { | |
7615 | ret = 0; | |
7616 | break; | |
7617 | } | |
7618 | } | |
7619 | ||
7620 | /* Ensure all chunks have corresponding dev extents */ | |
7621 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7622 | out: | |
7623 | btrfs_free_path(path); | |
7624 | return ret; | |
7625 | } | |
eede2bf3 OS |
7626 | |
7627 | /* | |
7628 | * Check whether the given block group or device is pinned by any inode being | |
7629 | * used as a swapfile. | |
7630 | */ | |
7631 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7632 | { | |
7633 | struct btrfs_swapfile_pin *sp; | |
7634 | struct rb_node *node; | |
7635 | ||
7636 | spin_lock(&fs_info->swapfile_pins_lock); | |
7637 | node = fs_info->swapfile_pins.rb_node; | |
7638 | while (node) { | |
7639 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7640 | if (ptr < sp->ptr) | |
7641 | node = node->rb_left; | |
7642 | else if (ptr > sp->ptr) | |
7643 | node = node->rb_right; | |
7644 | else | |
7645 | break; | |
7646 | } | |
7647 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7648 | return node != NULL; | |
7649 | } |