]>
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 | |
7239ff4b NB |
248 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
249 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
250 | * |
251 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
252 | * The returned struct is not linked onto any lists and can be destroyed with | |
253 | * kfree() right away. | |
254 | */ | |
7239ff4b NB |
255 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
256 | const u8 *metadata_fsid) | |
2208a378 ID |
257 | { |
258 | struct btrfs_fs_devices *fs_devs; | |
259 | ||
78f2c9e6 | 260 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
261 | if (!fs_devs) |
262 | return ERR_PTR(-ENOMEM); | |
263 | ||
264 | mutex_init(&fs_devs->device_list_mutex); | |
265 | ||
266 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 267 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 | 268 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
c4babc5e | 269 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
270 | if (fsid) |
271 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 272 | |
7239ff4b NB |
273 | if (metadata_fsid) |
274 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
275 | else if (fsid) | |
276 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
277 | ||
2208a378 ID |
278 | return fs_devs; |
279 | } | |
280 | ||
a425f9d4 | 281 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf DS |
282 | { |
283 | rcu_string_free(device->name); | |
284 | bio_put(device->flush_bio); | |
285 | kfree(device); | |
286 | } | |
287 | ||
e4404d6e YZ |
288 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
289 | { | |
290 | struct btrfs_device *device; | |
291 | WARN_ON(fs_devices->opened); | |
292 | while (!list_empty(&fs_devices->devices)) { | |
293 | device = list_entry(fs_devices->devices.next, | |
294 | struct btrfs_device, dev_list); | |
295 | list_del(&device->dev_list); | |
a425f9d4 | 296 | btrfs_free_device(device); |
e4404d6e YZ |
297 | } |
298 | kfree(fs_devices); | |
299 | } | |
300 | ||
b8b8ff59 LC |
301 | static void btrfs_kobject_uevent(struct block_device *bdev, |
302 | enum kobject_action action) | |
303 | { | |
304 | int ret; | |
305 | ||
306 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
307 | if (ret) | |
efe120a0 | 308 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
309 | action, |
310 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
311 | &disk_to_dev(bdev->bd_disk)->kobj); | |
312 | } | |
313 | ||
ffc5a379 | 314 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
315 | { |
316 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 317 | |
2b82032c YZ |
318 | while (!list_empty(&fs_uuids)) { |
319 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
320 | struct btrfs_fs_devices, fs_list); |
321 | list_del(&fs_devices->fs_list); | |
e4404d6e | 322 | free_fs_devices(fs_devices); |
8a4b83cc | 323 | } |
8a4b83cc CM |
324 | } |
325 | ||
48dae9cf DS |
326 | /* |
327 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
328 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 329 | * btrfs_free_device. |
48dae9cf | 330 | */ |
12bd2fc0 ID |
331 | static struct btrfs_device *__alloc_device(void) |
332 | { | |
333 | struct btrfs_device *dev; | |
334 | ||
78f2c9e6 | 335 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
336 | if (!dev) |
337 | return ERR_PTR(-ENOMEM); | |
338 | ||
e0ae9994 DS |
339 | /* |
340 | * Preallocate a bio that's always going to be used for flushing device | |
341 | * barriers and matches the device lifespan | |
342 | */ | |
343 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
344 | if (!dev->flush_bio) { | |
345 | kfree(dev); | |
346 | return ERR_PTR(-ENOMEM); | |
347 | } | |
e0ae9994 | 348 | |
12bd2fc0 ID |
349 | INIT_LIST_HEAD(&dev->dev_list); |
350 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 351 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
352 | |
353 | spin_lock_init(&dev->io_lock); | |
354 | ||
12bd2fc0 | 355 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 356 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 357 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 358 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 359 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
360 | |
361 | return dev; | |
362 | } | |
363 | ||
35c70103 DS |
364 | /* |
365 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
366 | * return NULL. | |
367 | * | |
368 | * If devid and uuid are both specified, the match must be exact, otherwise | |
369 | * only devid is used. | |
370 | */ | |
371 | static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices, | |
372 | u64 devid, const u8 *uuid) | |
8a4b83cc CM |
373 | { |
374 | struct btrfs_device *dev; | |
8a4b83cc | 375 | |
636d2c9d | 376 | list_for_each_entry(dev, &fs_devices->devices, dev_list) { |
a443755f | 377 | if (dev->devid == devid && |
8f18cf13 | 378 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 379 | return dev; |
a443755f | 380 | } |
8a4b83cc CM |
381 | } |
382 | return NULL; | |
383 | } | |
384 | ||
7239ff4b NB |
385 | static noinline struct btrfs_fs_devices *find_fsid( |
386 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 387 | { |
8a4b83cc CM |
388 | struct btrfs_fs_devices *fs_devices; |
389 | ||
7239ff4b NB |
390 | ASSERT(fsid); |
391 | ||
7a62d0f0 NB |
392 | if (metadata_fsid) { |
393 | /* | |
394 | * Handle scanned device having completed its fsid change but | |
395 | * belonging to a fs_devices that was created by first scanning | |
396 | * a device which didn't have its fsid/metadata_uuid changed | |
397 | * at all and the CHANGING_FSID_V2 flag set. | |
398 | */ | |
399 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
400 | if (fs_devices->fsid_change && | |
401 | memcmp(metadata_fsid, fs_devices->fsid, | |
402 | BTRFS_FSID_SIZE) == 0 && | |
403 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
404 | BTRFS_FSID_SIZE) == 0) { | |
405 | return fs_devices; | |
406 | } | |
407 | } | |
cc5de4e7 NB |
408 | /* |
409 | * Handle scanned device having completed its fsid change but | |
410 | * belonging to a fs_devices that was created by a device that | |
411 | * has an outdated pair of fsid/metadata_uuid and | |
412 | * CHANGING_FSID_V2 flag set. | |
413 | */ | |
414 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
415 | if (fs_devices->fsid_change && | |
416 | memcmp(fs_devices->metadata_uuid, | |
417 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
418 | memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
419 | BTRFS_FSID_SIZE) == 0) { | |
420 | return fs_devices; | |
421 | } | |
422 | } | |
7a62d0f0 NB |
423 | } |
424 | ||
425 | /* Handle non-split brain cases */ | |
c4babc5e | 426 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
427 | if (metadata_fsid) { |
428 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
429 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
430 | BTRFS_FSID_SIZE) == 0) | |
431 | return fs_devices; | |
432 | } else { | |
433 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
434 | return fs_devices; | |
435 | } | |
8a4b83cc CM |
436 | } |
437 | return NULL; | |
438 | } | |
439 | ||
beaf8ab3 SB |
440 | static int |
441 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
442 | int flush, struct block_device **bdev, | |
443 | struct buffer_head **bh) | |
444 | { | |
445 | int ret; | |
446 | ||
447 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
448 | ||
449 | if (IS_ERR(*bdev)) { | |
450 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
451 | goto error; |
452 | } | |
453 | ||
454 | if (flush) | |
455 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 456 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
457 | if (ret) { |
458 | blkdev_put(*bdev, flags); | |
459 | goto error; | |
460 | } | |
461 | invalidate_bdev(*bdev); | |
462 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
463 | if (IS_ERR(*bh)) { |
464 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
465 | blkdev_put(*bdev, flags); |
466 | goto error; | |
467 | } | |
468 | ||
469 | return 0; | |
470 | ||
471 | error: | |
472 | *bdev = NULL; | |
473 | *bh = NULL; | |
474 | return ret; | |
475 | } | |
476 | ||
ffbd517d CM |
477 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
478 | struct bio *head, struct bio *tail) | |
479 | { | |
480 | ||
481 | struct bio *old_head; | |
482 | ||
483 | old_head = pending_bios->head; | |
484 | pending_bios->head = head; | |
485 | if (pending_bios->tail) | |
486 | tail->bi_next = old_head; | |
487 | else | |
488 | pending_bios->tail = tail; | |
489 | } | |
490 | ||
8b712842 CM |
491 | /* |
492 | * we try to collect pending bios for a device so we don't get a large | |
493 | * number of procs sending bios down to the same device. This greatly | |
494 | * improves the schedulers ability to collect and merge the bios. | |
495 | * | |
496 | * But, it also turns into a long list of bios to process and that is sure | |
497 | * to eventually make the worker thread block. The solution here is to | |
498 | * make some progress and then put this work struct back at the end of | |
499 | * the list if the block device is congested. This way, multiple devices | |
500 | * can make progress from a single worker thread. | |
501 | */ | |
143bede5 | 502 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 503 | { |
0b246afa | 504 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
505 | struct bio *pending; |
506 | struct backing_dev_info *bdi; | |
ffbd517d | 507 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
508 | struct bio *tail; |
509 | struct bio *cur; | |
510 | int again = 0; | |
ffbd517d | 511 | unsigned long num_run; |
d644d8a1 | 512 | unsigned long batch_run = 0; |
b765ead5 | 513 | unsigned long last_waited = 0; |
d84275c9 | 514 | int force_reg = 0; |
0e588859 | 515 | int sync_pending = 0; |
211588ad CM |
516 | struct blk_plug plug; |
517 | ||
518 | /* | |
519 | * this function runs all the bios we've collected for | |
520 | * a particular device. We don't want to wander off to | |
521 | * another device without first sending all of these down. | |
522 | * So, setup a plug here and finish it off before we return | |
523 | */ | |
524 | blk_start_plug(&plug); | |
8b712842 | 525 | |
efa7c9f9 | 526 | bdi = device->bdev->bd_bdi; |
b64a2851 | 527 | |
8b712842 CM |
528 | loop: |
529 | spin_lock(&device->io_lock); | |
530 | ||
a6837051 | 531 | loop_lock: |
d84275c9 | 532 | num_run = 0; |
ffbd517d | 533 | |
8b712842 CM |
534 | /* take all the bios off the list at once and process them |
535 | * later on (without the lock held). But, remember the | |
536 | * tail and other pointers so the bios can be properly reinserted | |
537 | * into the list if we hit congestion | |
538 | */ | |
d84275c9 | 539 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 540 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
541 | force_reg = 1; |
542 | } else { | |
ffbd517d | 543 | pending_bios = &device->pending_bios; |
d84275c9 CM |
544 | force_reg = 0; |
545 | } | |
ffbd517d CM |
546 | |
547 | pending = pending_bios->head; | |
548 | tail = pending_bios->tail; | |
8b712842 | 549 | WARN_ON(pending && !tail); |
8b712842 CM |
550 | |
551 | /* | |
552 | * if pending was null this time around, no bios need processing | |
553 | * at all and we can stop. Otherwise it'll loop back up again | |
554 | * and do an additional check so no bios are missed. | |
555 | * | |
556 | * device->running_pending is used to synchronize with the | |
557 | * schedule_bio code. | |
558 | */ | |
ffbd517d CM |
559 | if (device->pending_sync_bios.head == NULL && |
560 | device->pending_bios.head == NULL) { | |
8b712842 CM |
561 | again = 0; |
562 | device->running_pending = 0; | |
ffbd517d CM |
563 | } else { |
564 | again = 1; | |
565 | device->running_pending = 1; | |
8b712842 | 566 | } |
ffbd517d CM |
567 | |
568 | pending_bios->head = NULL; | |
569 | pending_bios->tail = NULL; | |
570 | ||
8b712842 CM |
571 | spin_unlock(&device->io_lock); |
572 | ||
d397712b | 573 | while (pending) { |
ffbd517d CM |
574 | |
575 | rmb(); | |
d84275c9 CM |
576 | /* we want to work on both lists, but do more bios on the |
577 | * sync list than the regular list | |
578 | */ | |
579 | if ((num_run > 32 && | |
580 | pending_bios != &device->pending_sync_bios && | |
581 | device->pending_sync_bios.head) || | |
582 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
583 | device->pending_bios.head)) { | |
ffbd517d CM |
584 | spin_lock(&device->io_lock); |
585 | requeue_list(pending_bios, pending, tail); | |
586 | goto loop_lock; | |
587 | } | |
588 | ||
8b712842 CM |
589 | cur = pending; |
590 | pending = pending->bi_next; | |
591 | cur->bi_next = NULL; | |
b64a2851 | 592 | |
dac56212 | 593 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 594 | |
2ab1ba68 CM |
595 | /* |
596 | * if we're doing the sync list, record that our | |
597 | * plug has some sync requests on it | |
598 | * | |
599 | * If we're doing the regular list and there are | |
600 | * sync requests sitting around, unplug before | |
601 | * we add more | |
602 | */ | |
603 | if (pending_bios == &device->pending_sync_bios) { | |
604 | sync_pending = 1; | |
605 | } else if (sync_pending) { | |
606 | blk_finish_plug(&plug); | |
607 | blk_start_plug(&plug); | |
608 | sync_pending = 0; | |
609 | } | |
610 | ||
4e49ea4a | 611 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
612 | num_run++; |
613 | batch_run++; | |
853d8ec4 DS |
614 | |
615 | cond_resched(); | |
8b712842 CM |
616 | |
617 | /* | |
618 | * we made progress, there is more work to do and the bdi | |
619 | * is now congested. Back off and let other work structs | |
620 | * run instead | |
621 | */ | |
57fd5a5f | 622 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 623 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 624 | struct io_context *ioc; |
8b712842 | 625 | |
b765ead5 CM |
626 | ioc = current->io_context; |
627 | ||
628 | /* | |
629 | * the main goal here is that we don't want to | |
630 | * block if we're going to be able to submit | |
631 | * more requests without blocking. | |
632 | * | |
633 | * This code does two great things, it pokes into | |
634 | * the elevator code from a filesystem _and_ | |
635 | * it makes assumptions about how batching works. | |
636 | */ | |
637 | if (ioc && ioc->nr_batch_requests > 0 && | |
638 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
639 | (last_waited == 0 || | |
640 | ioc->last_waited == last_waited)) { | |
641 | /* | |
642 | * we want to go through our batch of | |
643 | * requests and stop. So, we copy out | |
644 | * the ioc->last_waited time and test | |
645 | * against it before looping | |
646 | */ | |
647 | last_waited = ioc->last_waited; | |
853d8ec4 | 648 | cond_resched(); |
b765ead5 CM |
649 | continue; |
650 | } | |
8b712842 | 651 | spin_lock(&device->io_lock); |
ffbd517d | 652 | requeue_list(pending_bios, pending, tail); |
a6837051 | 653 | device->running_pending = 1; |
8b712842 CM |
654 | |
655 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
656 | btrfs_queue_work(fs_info->submit_workers, |
657 | &device->work); | |
8b712842 CM |
658 | goto done; |
659 | } | |
660 | } | |
ffbd517d | 661 | |
51684082 CM |
662 | cond_resched(); |
663 | if (again) | |
664 | goto loop; | |
665 | ||
666 | spin_lock(&device->io_lock); | |
667 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
668 | goto loop_lock; | |
669 | spin_unlock(&device->io_lock); | |
670 | ||
8b712842 | 671 | done: |
211588ad | 672 | blk_finish_plug(&plug); |
8b712842 CM |
673 | } |
674 | ||
b2950863 | 675 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
676 | { |
677 | struct btrfs_device *device; | |
678 | ||
679 | device = container_of(work, struct btrfs_device, work); | |
680 | run_scheduled_bios(device); | |
681 | } | |
682 | ||
d8367db3 AJ |
683 | /* |
684 | * Search and remove all stale (devices which are not mounted) devices. | |
685 | * When both inputs are NULL, it will search and release all stale devices. | |
686 | * path: Optional. When provided will it release all unmounted devices | |
687 | * matching this path only. | |
688 | * skip_dev: Optional. Will skip this device when searching for the stale | |
689 | * devices. | |
690 | */ | |
691 | static void btrfs_free_stale_devices(const char *path, | |
fa6d2ae5 | 692 | struct btrfs_device *skip_device) |
4fde46f0 | 693 | { |
fa6d2ae5 AJ |
694 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
695 | struct btrfs_device *device, *tmp_device; | |
4fde46f0 | 696 | |
fa6d2ae5 | 697 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
7bcb8164 AJ |
698 | mutex_lock(&fs_devices->device_list_mutex); |
699 | if (fs_devices->opened) { | |
700 | mutex_unlock(&fs_devices->device_list_mutex); | |
4fde46f0 | 701 | continue; |
7bcb8164 | 702 | } |
4fde46f0 | 703 | |
fa6d2ae5 AJ |
704 | list_for_each_entry_safe(device, tmp_device, |
705 | &fs_devices->devices, dev_list) { | |
522f1b45 | 706 | int not_found = 0; |
4fde46f0 | 707 | |
fa6d2ae5 | 708 | if (skip_device && skip_device == device) |
d8367db3 | 709 | continue; |
fa6d2ae5 | 710 | if (path && !device->name) |
4fde46f0 AJ |
711 | continue; |
712 | ||
4fde46f0 | 713 | rcu_read_lock(); |
d8367db3 | 714 | if (path) |
fa6d2ae5 | 715 | not_found = strcmp(rcu_str_deref(device->name), |
d8367db3 | 716 | path); |
4fde46f0 | 717 | rcu_read_unlock(); |
38cf665d AJ |
718 | if (not_found) |
719 | continue; | |
4fde46f0 | 720 | |
4fde46f0 | 721 | /* delete the stale device */ |
7bcb8164 AJ |
722 | fs_devices->num_devices--; |
723 | list_del(&device->dev_list); | |
724 | btrfs_free_device(device); | |
725 | ||
726 | if (fs_devices->num_devices == 0) | |
fd649f10 | 727 | break; |
7bcb8164 AJ |
728 | } |
729 | mutex_unlock(&fs_devices->device_list_mutex); | |
730 | if (fs_devices->num_devices == 0) { | |
731 | btrfs_sysfs_remove_fsid(fs_devices); | |
732 | list_del(&fs_devices->fs_list); | |
733 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
734 | } |
735 | } | |
736 | } | |
737 | ||
0fb08bcc AJ |
738 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
739 | struct btrfs_device *device, fmode_t flags, | |
740 | void *holder) | |
741 | { | |
742 | struct request_queue *q; | |
743 | struct block_device *bdev; | |
744 | struct buffer_head *bh; | |
745 | struct btrfs_super_block *disk_super; | |
746 | u64 devid; | |
747 | int ret; | |
748 | ||
749 | if (device->bdev) | |
750 | return -EINVAL; | |
751 | if (!device->name) | |
752 | return -EINVAL; | |
753 | ||
754 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
755 | &bdev, &bh); | |
756 | if (ret) | |
757 | return ret; | |
758 | ||
759 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
760 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
761 | if (devid != device->devid) | |
762 | goto error_brelse; | |
763 | ||
764 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
765 | goto error_brelse; | |
766 | ||
767 | device->generation = btrfs_super_generation(disk_super); | |
768 | ||
769 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
770 | if (btrfs_super_incompat_flags(disk_super) & |
771 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
772 | pr_err( | |
773 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
774 | goto error_brelse; | |
775 | } | |
776 | ||
ebbede42 | 777 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
778 | fs_devices->seeding = 1; |
779 | } else { | |
ebbede42 AJ |
780 | if (bdev_read_only(bdev)) |
781 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
782 | else | |
783 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
784 | } |
785 | ||
786 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
787 | if (!blk_queue_nonrot(q)) |
788 | fs_devices->rotating = 1; | |
789 | ||
790 | device->bdev = bdev; | |
e12c9621 | 791 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
792 | device->mode = flags; |
793 | ||
794 | fs_devices->open_devices++; | |
ebbede42 AJ |
795 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
796 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 797 | fs_devices->rw_devices++; |
b1b8e386 | 798 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
799 | } |
800 | brelse(bh); | |
801 | ||
802 | return 0; | |
803 | ||
804 | error_brelse: | |
805 | brelse(bh); | |
806 | blkdev_put(bdev, flags); | |
807 | ||
808 | return -EINVAL; | |
809 | } | |
810 | ||
7a62d0f0 NB |
811 | /* |
812 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
813 | * being created with a disk that has already completed its fsid change. | |
814 | */ | |
815 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
816 | struct btrfs_super_block *disk_super) | |
817 | { | |
818 | struct btrfs_fs_devices *fs_devices; | |
819 | ||
820 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
821 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
822 | BTRFS_FSID_SIZE) != 0 && | |
823 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
824 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
825 | return fs_devices; | |
826 | } | |
827 | } | |
828 | ||
829 | return NULL; | |
830 | } | |
831 | ||
cc5de4e7 NB |
832 | |
833 | static struct btrfs_fs_devices *find_fsid_changed( | |
834 | struct btrfs_super_block *disk_super) | |
835 | { | |
836 | struct btrfs_fs_devices *fs_devices; | |
837 | ||
838 | /* | |
839 | * Handles the case where scanned device is part of an fs that had | |
840 | * multiple successful changes of FSID but curently device didn't | |
841 | * observe it. Meaning our fsid will be different than theirs. | |
842 | */ | |
843 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
844 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
845 | BTRFS_FSID_SIZE) != 0 && | |
846 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
847 | BTRFS_FSID_SIZE) == 0 && | |
848 | memcmp(fs_devices->fsid, disk_super->fsid, | |
849 | BTRFS_FSID_SIZE) != 0) { | |
850 | return fs_devices; | |
851 | } | |
852 | } | |
853 | ||
854 | return NULL; | |
855 | } | |
60999ca4 DS |
856 | /* |
857 | * Add new device to list of registered devices | |
858 | * | |
859 | * Returns: | |
e124ece5 AJ |
860 | * device pointer which was just added or updated when successful |
861 | * error pointer when failed | |
60999ca4 | 862 | */ |
e124ece5 | 863 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
864 | struct btrfs_super_block *disk_super, |
865 | bool *new_device_added) | |
8a4b83cc CM |
866 | { |
867 | struct btrfs_device *device; | |
7a62d0f0 | 868 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 869 | struct rcu_string *name; |
8a4b83cc | 870 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 871 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
872 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
873 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
874 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
875 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 876 | |
cc5de4e7 NB |
877 | if (fsid_change_in_progress) { |
878 | if (!has_metadata_uuid) { | |
879 | /* | |
880 | * When we have an image which has CHANGING_FSID_V2 set | |
881 | * it might belong to either a filesystem which has | |
882 | * disks with completed fsid change or it might belong | |
883 | * to fs with no UUID changes in effect, handle both. | |
884 | */ | |
885 | fs_devices = find_fsid_inprogress(disk_super); | |
886 | if (!fs_devices) | |
887 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
888 | } else { | |
889 | fs_devices = find_fsid_changed(disk_super); | |
890 | } | |
7a62d0f0 NB |
891 | } else if (has_metadata_uuid) { |
892 | fs_devices = find_fsid(disk_super->fsid, | |
893 | disk_super->metadata_uuid); | |
894 | } else { | |
7239ff4b | 895 | fs_devices = find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
896 | } |
897 | ||
8a4b83cc | 898 | |
8a4b83cc | 899 | if (!fs_devices) { |
7239ff4b NB |
900 | if (has_metadata_uuid) |
901 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
902 | disk_super->metadata_uuid); | |
903 | else | |
904 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
905 | ||
d1a63002 NB |
906 | fs_devices->fsid_change = fsid_change_in_progress; |
907 | ||
2208a378 | 908 | if (IS_ERR(fs_devices)) |
e124ece5 | 909 | return ERR_CAST(fs_devices); |
2208a378 | 910 | |
9c6d173e | 911 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 912 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 913 | |
8a4b83cc CM |
914 | device = NULL; |
915 | } else { | |
9c6d173e | 916 | mutex_lock(&fs_devices->device_list_mutex); |
35c70103 DS |
917 | device = find_device(fs_devices, devid, |
918 | disk_super->dev_item.uuid); | |
7a62d0f0 NB |
919 | |
920 | /* | |
921 | * If this disk has been pulled into an fs devices created by | |
922 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
923 | * metadata_uuid/fsid values of the fs_devices. | |
924 | */ | |
925 | if (has_metadata_uuid && fs_devices->fsid_change && | |
926 | found_transid > fs_devices->latest_generation) { | |
927 | memcpy(fs_devices->fsid, disk_super->fsid, | |
928 | BTRFS_FSID_SIZE); | |
929 | memcpy(fs_devices->metadata_uuid, | |
930 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
931 | ||
932 | fs_devices->fsid_change = false; | |
933 | } | |
8a4b83cc | 934 | } |
443f24fe | 935 | |
8a4b83cc | 936 | if (!device) { |
9c6d173e AJ |
937 | if (fs_devices->opened) { |
938 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 939 | return ERR_PTR(-EBUSY); |
9c6d173e | 940 | } |
2b82032c | 941 | |
12bd2fc0 ID |
942 | device = btrfs_alloc_device(NULL, &devid, |
943 | disk_super->dev_item.uuid); | |
944 | if (IS_ERR(device)) { | |
9c6d173e | 945 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 946 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 947 | return device; |
8a4b83cc | 948 | } |
606686ee JB |
949 | |
950 | name = rcu_string_strdup(path, GFP_NOFS); | |
951 | if (!name) { | |
a425f9d4 | 952 | btrfs_free_device(device); |
9c6d173e | 953 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 954 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 955 | } |
606686ee | 956 | rcu_assign_pointer(device->name, name); |
90519d66 | 957 | |
1f78160c | 958 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 959 | fs_devices->num_devices++; |
e5e9a520 | 960 | |
2b82032c | 961 | device->fs_devices = fs_devices; |
4306a974 | 962 | *new_device_added = true; |
327f18cc AJ |
963 | |
964 | if (disk_super->label[0]) | |
965 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
966 | disk_super->label, devid, found_transid, path); | |
967 | else | |
968 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
969 | disk_super->fsid, devid, found_transid, path); | |
970 | ||
606686ee | 971 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
972 | /* |
973 | * When FS is already mounted. | |
974 | * 1. If you are here and if the device->name is NULL that | |
975 | * means this device was missing at time of FS mount. | |
976 | * 2. If you are here and if the device->name is different | |
977 | * from 'path' that means either | |
978 | * a. The same device disappeared and reappeared with | |
979 | * different name. or | |
980 | * b. The missing-disk-which-was-replaced, has | |
981 | * reappeared now. | |
982 | * | |
983 | * We must allow 1 and 2a above. But 2b would be a spurious | |
984 | * and unintentional. | |
985 | * | |
986 | * Further in case of 1 and 2a above, the disk at 'path' | |
987 | * would have missed some transaction when it was away and | |
988 | * in case of 2a the stale bdev has to be updated as well. | |
989 | * 2b must not be allowed at all time. | |
990 | */ | |
991 | ||
992 | /* | |
0f23ae74 CM |
993 | * For now, we do allow update to btrfs_fs_device through the |
994 | * btrfs dev scan cli after FS has been mounted. We're still | |
995 | * tracking a problem where systems fail mount by subvolume id | |
996 | * when we reject replacement on a mounted FS. | |
b96de000 | 997 | */ |
0f23ae74 | 998 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
999 | /* |
1000 | * That is if the FS is _not_ mounted and if you | |
1001 | * are here, that means there is more than one | |
1002 | * disk with same uuid and devid.We keep the one | |
1003 | * with larger generation number or the last-in if | |
1004 | * generation are equal. | |
1005 | */ | |
9c6d173e | 1006 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1007 | return ERR_PTR(-EEXIST); |
77bdae4d | 1008 | } |
b96de000 | 1009 | |
a9261d41 AJ |
1010 | /* |
1011 | * We are going to replace the device path for a given devid, | |
1012 | * make sure it's the same device if the device is mounted | |
1013 | */ | |
1014 | if (device->bdev) { | |
1015 | struct block_device *path_bdev; | |
1016 | ||
1017 | path_bdev = lookup_bdev(path); | |
1018 | if (IS_ERR(path_bdev)) { | |
1019 | mutex_unlock(&fs_devices->device_list_mutex); | |
1020 | return ERR_CAST(path_bdev); | |
1021 | } | |
1022 | ||
1023 | if (device->bdev != path_bdev) { | |
1024 | bdput(path_bdev); | |
1025 | mutex_unlock(&fs_devices->device_list_mutex); | |
1026 | btrfs_warn_in_rcu(device->fs_info, | |
1027 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
1028 | disk_super->fsid, devid, | |
1029 | rcu_str_deref(device->name), path); | |
1030 | return ERR_PTR(-EEXIST); | |
1031 | } | |
1032 | bdput(path_bdev); | |
1033 | btrfs_info_in_rcu(device->fs_info, | |
1034 | "device fsid %pU devid %llu moved old:%s new:%s", | |
1035 | disk_super->fsid, devid, | |
1036 | rcu_str_deref(device->name), path); | |
1037 | } | |
1038 | ||
606686ee | 1039 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
1040 | if (!name) { |
1041 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 1042 | return ERR_PTR(-ENOMEM); |
9c6d173e | 1043 | } |
606686ee JB |
1044 | rcu_string_free(device->name); |
1045 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 1046 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 1047 | fs_devices->missing_devices--; |
e6e674bd | 1048 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 1049 | } |
8a4b83cc CM |
1050 | } |
1051 | ||
77bdae4d AJ |
1052 | /* |
1053 | * Unmount does not free the btrfs_device struct but would zero | |
1054 | * generation along with most of the other members. So just update | |
1055 | * it back. We need it to pick the disk with largest generation | |
1056 | * (as above). | |
1057 | */ | |
d1a63002 | 1058 | if (!fs_devices->opened) { |
77bdae4d | 1059 | device->generation = found_transid; |
d1a63002 NB |
1060 | fs_devices->latest_generation = max_t(u64, found_transid, |
1061 | fs_devices->latest_generation); | |
1062 | } | |
77bdae4d | 1063 | |
f2788d2f AJ |
1064 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
1065 | ||
9c6d173e | 1066 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1067 | return device; |
8a4b83cc CM |
1068 | } |
1069 | ||
e4404d6e YZ |
1070 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
1071 | { | |
1072 | struct btrfs_fs_devices *fs_devices; | |
1073 | struct btrfs_device *device; | |
1074 | struct btrfs_device *orig_dev; | |
1075 | ||
7239ff4b | 1076 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1077 | if (IS_ERR(fs_devices)) |
1078 | return fs_devices; | |
e4404d6e | 1079 | |
adbbb863 | 1080 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1081 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 1082 | |
46224705 | 1083 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 1084 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
1085 | struct rcu_string *name; |
1086 | ||
12bd2fc0 ID |
1087 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1088 | orig_dev->uuid); | |
1089 | if (IS_ERR(device)) | |
e4404d6e YZ |
1090 | goto error; |
1091 | ||
606686ee JB |
1092 | /* |
1093 | * This is ok to do without rcu read locked because we hold the | |
1094 | * uuid mutex so nothing we touch in here is going to disappear. | |
1095 | */ | |
e755f780 | 1096 | if (orig_dev->name) { |
78f2c9e6 DS |
1097 | name = rcu_string_strdup(orig_dev->name->str, |
1098 | GFP_KERNEL); | |
e755f780 | 1099 | if (!name) { |
a425f9d4 | 1100 | btrfs_free_device(device); |
e755f780 AJ |
1101 | goto error; |
1102 | } | |
1103 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1104 | } |
e4404d6e | 1105 | |
e4404d6e YZ |
1106 | list_add(&device->dev_list, &fs_devices->devices); |
1107 | device->fs_devices = fs_devices; | |
1108 | fs_devices->num_devices++; | |
1109 | } | |
adbbb863 | 1110 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1111 | return fs_devices; |
1112 | error: | |
adbbb863 | 1113 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1114 | free_fs_devices(fs_devices); |
1115 | return ERR_PTR(-ENOMEM); | |
1116 | } | |
1117 | ||
9b99b115 AJ |
1118 | /* |
1119 | * After we have read the system tree and know devids belonging to | |
1120 | * this filesystem, remove the device which does not belong there. | |
1121 | */ | |
1122 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1123 | { |
c6e30871 | 1124 | struct btrfs_device *device, *next; |
443f24fe | 1125 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1126 | |
dfe25020 CM |
1127 | mutex_lock(&uuid_mutex); |
1128 | again: | |
46224705 | 1129 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1130 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1131 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1132 | &device->dev_state)) { | |
401e29c1 AJ |
1133 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1134 | &device->dev_state) && | |
1135 | (!latest_dev || | |
1136 | device->generation > latest_dev->generation)) { | |
443f24fe | 1137 | latest_dev = device; |
a6b0d5c8 | 1138 | } |
2b82032c | 1139 | continue; |
a6b0d5c8 | 1140 | } |
2b82032c | 1141 | |
8dabb742 SB |
1142 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1143 | /* | |
1144 | * In the first step, keep the device which has | |
1145 | * the correct fsid and the devid that is used | |
1146 | * for the dev_replace procedure. | |
1147 | * In the second step, the dev_replace state is | |
1148 | * read from the device tree and it is known | |
1149 | * whether the procedure is really active or | |
1150 | * not, which means whether this device is | |
1151 | * used or whether it should be removed. | |
1152 | */ | |
401e29c1 AJ |
1153 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1154 | &device->dev_state)) { | |
8dabb742 SB |
1155 | continue; |
1156 | } | |
1157 | } | |
2b82032c | 1158 | if (device->bdev) { |
d4d77629 | 1159 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1160 | device->bdev = NULL; |
1161 | fs_devices->open_devices--; | |
1162 | } | |
ebbede42 | 1163 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1164 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1165 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1166 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1167 | &device->dev_state)) | |
8dabb742 | 1168 | fs_devices->rw_devices--; |
2b82032c | 1169 | } |
e4404d6e YZ |
1170 | list_del_init(&device->dev_list); |
1171 | fs_devices->num_devices--; | |
a425f9d4 | 1172 | btrfs_free_device(device); |
dfe25020 | 1173 | } |
2b82032c YZ |
1174 | |
1175 | if (fs_devices->seed) { | |
1176 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1177 | goto again; |
1178 | } | |
1179 | ||
443f24fe | 1180 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1181 | |
dfe25020 | 1182 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1183 | } |
a0af469b | 1184 | |
f06c5965 | 1185 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
1186 | { |
1187 | struct btrfs_device *device; | |
1188 | ||
9f5316c1 | 1189 | device = container_of(head, struct btrfs_device, rcu); |
a425f9d4 | 1190 | btrfs_free_device(device); |
1f78160c XG |
1191 | } |
1192 | ||
14238819 AJ |
1193 | static void btrfs_close_bdev(struct btrfs_device *device) |
1194 | { | |
08ffcae8 DS |
1195 | if (!device->bdev) |
1196 | return; | |
1197 | ||
ebbede42 | 1198 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1199 | sync_blockdev(device->bdev); |
1200 | invalidate_bdev(device->bdev); | |
1201 | } | |
1202 | ||
08ffcae8 | 1203 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1204 | } |
1205 | ||
959b1c04 | 1206 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1207 | { |
1208 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1209 | struct btrfs_device *new_device; | |
1210 | struct rcu_string *name; | |
1211 | ||
1212 | if (device->bdev) | |
1213 | fs_devices->open_devices--; | |
1214 | ||
ebbede42 | 1215 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1216 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1217 | list_del_init(&device->dev_alloc_list); | |
1218 | fs_devices->rw_devices--; | |
1219 | } | |
1220 | ||
e6e674bd | 1221 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1222 | fs_devices->missing_devices--; |
1223 | ||
959b1c04 NB |
1224 | btrfs_close_bdev(device); |
1225 | ||
f448341a AJ |
1226 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1227 | device->uuid); | |
1228 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1229 | ||
1230 | /* Safe because we are under uuid_mutex */ | |
1231 | if (device->name) { | |
1232 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1233 | BUG_ON(!name); /* -ENOMEM */ | |
1234 | rcu_assign_pointer(new_device->name, name); | |
1235 | } | |
1236 | ||
1237 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1238 | new_device->fs_devices = device->fs_devices; | |
959b1c04 NB |
1239 | |
1240 | call_rcu(&device->rcu, free_device_rcu); | |
f448341a AJ |
1241 | } |
1242 | ||
0226e0eb | 1243 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1244 | { |
2037a093 | 1245 | struct btrfs_device *device, *tmp; |
e4404d6e | 1246 | |
2b82032c YZ |
1247 | if (--fs_devices->opened > 0) |
1248 | return 0; | |
8a4b83cc | 1249 | |
c9513edb | 1250 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1251 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1252 | btrfs_close_one_device(device); |
8a4b83cc | 1253 | } |
c9513edb XG |
1254 | mutex_unlock(&fs_devices->device_list_mutex); |
1255 | ||
e4404d6e YZ |
1256 | WARN_ON(fs_devices->open_devices); |
1257 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1258 | fs_devices->opened = 0; |
1259 | fs_devices->seeding = 0; | |
2b82032c | 1260 | |
8a4b83cc CM |
1261 | return 0; |
1262 | } | |
1263 | ||
2b82032c YZ |
1264 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1265 | { | |
e4404d6e | 1266 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1267 | int ret; |
1268 | ||
1269 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1270 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1271 | if (!fs_devices->opened) { |
1272 | seed_devices = fs_devices->seed; | |
1273 | fs_devices->seed = NULL; | |
1274 | } | |
2b82032c | 1275 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1276 | |
1277 | while (seed_devices) { | |
1278 | fs_devices = seed_devices; | |
1279 | seed_devices = fs_devices->seed; | |
0226e0eb | 1280 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1281 | free_fs_devices(fs_devices); |
1282 | } | |
2b82032c YZ |
1283 | return ret; |
1284 | } | |
1285 | ||
897fb573 | 1286 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1287 | fmode_t flags, void *holder) |
8a4b83cc | 1288 | { |
8a4b83cc | 1289 | struct btrfs_device *device; |
443f24fe | 1290 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1291 | int ret = 0; |
8a4b83cc | 1292 | |
d4d77629 TH |
1293 | flags |= FMODE_EXCL; |
1294 | ||
f117e290 | 1295 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1296 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1297 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1298 | continue; |
a0af469b | 1299 | |
9f050db4 AJ |
1300 | if (!latest_dev || |
1301 | device->generation > latest_dev->generation) | |
1302 | latest_dev = device; | |
8a4b83cc | 1303 | } |
a0af469b | 1304 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1305 | ret = -EINVAL; |
a0af469b CM |
1306 | goto out; |
1307 | } | |
2b82032c | 1308 | fs_devices->opened = 1; |
443f24fe | 1309 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1310 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1311 | out: |
2b82032c YZ |
1312 | return ret; |
1313 | } | |
1314 | ||
f8e10cd3 AJ |
1315 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1316 | { | |
1317 | struct btrfs_device *dev1, *dev2; | |
1318 | ||
1319 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1320 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1321 | ||
1322 | if (dev1->devid < dev2->devid) | |
1323 | return -1; | |
1324 | else if (dev1->devid > dev2->devid) | |
1325 | return 1; | |
1326 | return 0; | |
1327 | } | |
1328 | ||
2b82032c | 1329 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1330 | fmode_t flags, void *holder) |
2b82032c YZ |
1331 | { |
1332 | int ret; | |
1333 | ||
f5194e34 DS |
1334 | lockdep_assert_held(&uuid_mutex); |
1335 | ||
542c5908 | 1336 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1337 | if (fs_devices->opened) { |
e4404d6e YZ |
1338 | fs_devices->opened++; |
1339 | ret = 0; | |
2b82032c | 1340 | } else { |
f8e10cd3 | 1341 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1342 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1343 | } |
542c5908 AJ |
1344 | mutex_unlock(&fs_devices->device_list_mutex); |
1345 | ||
8a4b83cc CM |
1346 | return ret; |
1347 | } | |
1348 | ||
c9162bdf | 1349 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1350 | { |
1351 | kunmap(page); | |
1352 | put_page(page); | |
1353 | } | |
1354 | ||
c9162bdf OS |
1355 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1356 | struct page **page, | |
1357 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1358 | { |
1359 | void *p; | |
1360 | pgoff_t index; | |
1361 | ||
1362 | /* make sure our super fits in the device */ | |
1363 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1364 | return 1; | |
1365 | ||
1366 | /* make sure our super fits in the page */ | |
1367 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1368 | return 1; | |
1369 | ||
1370 | /* make sure our super doesn't straddle pages on disk */ | |
1371 | index = bytenr >> PAGE_SHIFT; | |
1372 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1373 | return 1; | |
1374 | ||
1375 | /* pull in the page with our super */ | |
1376 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1377 | index, GFP_KERNEL); | |
1378 | ||
1379 | if (IS_ERR_OR_NULL(*page)) | |
1380 | return 1; | |
1381 | ||
1382 | p = kmap(*page); | |
1383 | ||
1384 | /* align our pointer to the offset of the super block */ | |
1385 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1386 | ||
1387 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1388 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1389 | btrfs_release_disk_super(*page); | |
1390 | return 1; | |
1391 | } | |
1392 | ||
1393 | if ((*disk_super)->label[0] && | |
1394 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1395 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1396 | ||
1397 | return 0; | |
1398 | } | |
1399 | ||
6f60cbd3 DS |
1400 | /* |
1401 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1402 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1403 | * is read via pagecache | |
1404 | */ | |
36350e95 GJ |
1405 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1406 | void *holder) | |
8a4b83cc CM |
1407 | { |
1408 | struct btrfs_super_block *disk_super; | |
4306a974 | 1409 | bool new_device_added = false; |
36350e95 | 1410 | struct btrfs_device *device = NULL; |
8a4b83cc | 1411 | struct block_device *bdev; |
6f60cbd3 | 1412 | struct page *page; |
6f60cbd3 | 1413 | u64 bytenr; |
8a4b83cc | 1414 | |
899f9307 DS |
1415 | lockdep_assert_held(&uuid_mutex); |
1416 | ||
6f60cbd3 DS |
1417 | /* |
1418 | * we would like to check all the supers, but that would make | |
1419 | * a btrfs mount succeed after a mkfs from a different FS. | |
1420 | * So, we need to add a special mount option to scan for | |
1421 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1422 | */ | |
1423 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1424 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1425 | |
1426 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1427 | if (IS_ERR(bdev)) |
36350e95 | 1428 | return ERR_CAST(bdev); |
6f60cbd3 | 1429 | |
05a5c55d | 1430 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1431 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1432 | goto error_bdev_put; |
05a5c55d | 1433 | } |
6f60cbd3 | 1434 | |
4306a974 | 1435 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1436 | if (!IS_ERR(device)) { |
4306a974 AJ |
1437 | if (new_device_added) |
1438 | btrfs_free_stale_devices(path, device); | |
1439 | } | |
6f60cbd3 | 1440 | |
6cf86a00 | 1441 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1442 | |
1443 | error_bdev_put: | |
d4d77629 | 1444 | blkdev_put(bdev, flags); |
b6ed73bc | 1445 | |
36350e95 | 1446 | return device; |
8a4b83cc | 1447 | } |
0b86a832 | 1448 | |
499f377f | 1449 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1450 | struct btrfs_device *device, |
1451 | u64 *start, u64 len) | |
1452 | { | |
fb456252 | 1453 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1454 | struct extent_map *em; |
499f377f | 1455 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1456 | int ret = 0; |
1b984508 | 1457 | u64 physical_start = *start; |
6df9a95e | 1458 | |
499f377f JM |
1459 | if (transaction) |
1460 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1461 | again: |
1462 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1463 | struct map_lookup *map; |
1464 | int i; | |
1465 | ||
95617d69 | 1466 | map = em->map_lookup; |
6df9a95e | 1467 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1468 | u64 end; |
1469 | ||
6df9a95e JB |
1470 | if (map->stripes[i].dev != device) |
1471 | continue; | |
1b984508 | 1472 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1473 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1474 | physical_start) |
6df9a95e | 1475 | continue; |
c152b63e FM |
1476 | /* |
1477 | * Make sure that while processing the pinned list we do | |
1478 | * not override our *start with a lower value, because | |
1479 | * we can have pinned chunks that fall within this | |
1480 | * device hole and that have lower physical addresses | |
1481 | * than the pending chunks we processed before. If we | |
1482 | * do not take this special care we can end up getting | |
1483 | * 2 pending chunks that start at the same physical | |
1484 | * device offsets because the end offset of a pinned | |
1485 | * chunk can be equal to the start offset of some | |
1486 | * pending chunk. | |
1487 | */ | |
1488 | end = map->stripes[i].physical + em->orig_block_len; | |
1489 | if (end > *start) { | |
1490 | *start = end; | |
1491 | ret = 1; | |
1492 | } | |
6df9a95e JB |
1493 | } |
1494 | } | |
499f377f JM |
1495 | if (search_list != &fs_info->pinned_chunks) { |
1496 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1497 | goto again; |
1498 | } | |
6df9a95e JB |
1499 | |
1500 | return ret; | |
1501 | } | |
1502 | ||
1503 | ||
0b86a832 | 1504 | /* |
499f377f JM |
1505 | * find_free_dev_extent_start - find free space in the specified device |
1506 | * @device: the device which we search the free space in | |
1507 | * @num_bytes: the size of the free space that we need | |
1508 | * @search_start: the position from which to begin the search | |
1509 | * @start: store the start of the free space. | |
1510 | * @len: the size of the free space. that we find, or the size | |
1511 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1512 | * |
0b86a832 CM |
1513 | * this uses a pretty simple search, the expectation is that it is |
1514 | * called very infrequently and that a given device has a small number | |
1515 | * of extents | |
7bfc837d MX |
1516 | * |
1517 | * @start is used to store the start of the free space if we find. But if we | |
1518 | * don't find suitable free space, it will be used to store the start position | |
1519 | * of the max free space. | |
1520 | * | |
1521 | * @len is used to store the size of the free space that we find. | |
1522 | * But if we don't find suitable free space, it is used to store the size of | |
1523 | * the max free space. | |
0b86a832 | 1524 | */ |
499f377f JM |
1525 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1526 | struct btrfs_device *device, u64 num_bytes, | |
1527 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1528 | { |
0b246afa JM |
1529 | struct btrfs_fs_info *fs_info = device->fs_info; |
1530 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1531 | struct btrfs_key key; |
7bfc837d | 1532 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1533 | struct btrfs_path *path; |
7bfc837d MX |
1534 | u64 hole_size; |
1535 | u64 max_hole_start; | |
1536 | u64 max_hole_size; | |
1537 | u64 extent_end; | |
0b86a832 CM |
1538 | u64 search_end = device->total_bytes; |
1539 | int ret; | |
7bfc837d | 1540 | int slot; |
0b86a832 | 1541 | struct extent_buffer *l; |
8cdc7c5b FM |
1542 | |
1543 | /* | |
1544 | * We don't want to overwrite the superblock on the drive nor any area | |
1545 | * used by the boot loader (grub for example), so we make sure to start | |
1546 | * at an offset of at least 1MB. | |
1547 | */ | |
0d0c71b3 | 1548 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1549 | |
6df9a95e JB |
1550 | path = btrfs_alloc_path(); |
1551 | if (!path) | |
1552 | return -ENOMEM; | |
f2ab7618 | 1553 | |
7bfc837d MX |
1554 | max_hole_start = search_start; |
1555 | max_hole_size = 0; | |
1556 | ||
f2ab7618 | 1557 | again: |
401e29c1 AJ |
1558 | if (search_start >= search_end || |
1559 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1560 | ret = -ENOSPC; |
6df9a95e | 1561 | goto out; |
7bfc837d MX |
1562 | } |
1563 | ||
e4058b54 | 1564 | path->reada = READA_FORWARD; |
6df9a95e JB |
1565 | path->search_commit_root = 1; |
1566 | path->skip_locking = 1; | |
7bfc837d | 1567 | |
0b86a832 CM |
1568 | key.objectid = device->devid; |
1569 | key.offset = search_start; | |
1570 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1571 | |
125ccb0a | 1572 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1573 | if (ret < 0) |
7bfc837d | 1574 | goto out; |
1fcbac58 YZ |
1575 | if (ret > 0) { |
1576 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1577 | if (ret < 0) | |
7bfc837d | 1578 | goto out; |
1fcbac58 | 1579 | } |
7bfc837d | 1580 | |
0b86a832 CM |
1581 | while (1) { |
1582 | l = path->nodes[0]; | |
1583 | slot = path->slots[0]; | |
1584 | if (slot >= btrfs_header_nritems(l)) { | |
1585 | ret = btrfs_next_leaf(root, path); | |
1586 | if (ret == 0) | |
1587 | continue; | |
1588 | if (ret < 0) | |
7bfc837d MX |
1589 | goto out; |
1590 | ||
1591 | break; | |
0b86a832 CM |
1592 | } |
1593 | btrfs_item_key_to_cpu(l, &key, slot); | |
1594 | ||
1595 | if (key.objectid < device->devid) | |
1596 | goto next; | |
1597 | ||
1598 | if (key.objectid > device->devid) | |
7bfc837d | 1599 | break; |
0b86a832 | 1600 | |
962a298f | 1601 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1602 | goto next; |
9779b72f | 1603 | |
7bfc837d MX |
1604 | if (key.offset > search_start) { |
1605 | hole_size = key.offset - search_start; | |
9779b72f | 1606 | |
6df9a95e JB |
1607 | /* |
1608 | * Have to check before we set max_hole_start, otherwise | |
1609 | * we could end up sending back this offset anyway. | |
1610 | */ | |
499f377f | 1611 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1612 | &search_start, |
1b984508 FL |
1613 | hole_size)) { |
1614 | if (key.offset >= search_start) { | |
1615 | hole_size = key.offset - search_start; | |
1616 | } else { | |
1617 | WARN_ON_ONCE(1); | |
1618 | hole_size = 0; | |
1619 | } | |
1620 | } | |
6df9a95e | 1621 | |
7bfc837d MX |
1622 | if (hole_size > max_hole_size) { |
1623 | max_hole_start = search_start; | |
1624 | max_hole_size = hole_size; | |
1625 | } | |
9779b72f | 1626 | |
7bfc837d MX |
1627 | /* |
1628 | * If this free space is greater than which we need, | |
1629 | * it must be the max free space that we have found | |
1630 | * until now, so max_hole_start must point to the start | |
1631 | * of this free space and the length of this free space | |
1632 | * is stored in max_hole_size. Thus, we return | |
1633 | * max_hole_start and max_hole_size and go back to the | |
1634 | * caller. | |
1635 | */ | |
1636 | if (hole_size >= num_bytes) { | |
1637 | ret = 0; | |
1638 | goto out; | |
0b86a832 CM |
1639 | } |
1640 | } | |
0b86a832 | 1641 | |
0b86a832 | 1642 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1643 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1644 | dev_extent); | |
1645 | if (extent_end > search_start) | |
1646 | search_start = extent_end; | |
0b86a832 CM |
1647 | next: |
1648 | path->slots[0]++; | |
1649 | cond_resched(); | |
1650 | } | |
0b86a832 | 1651 | |
38c01b96 | 1652 | /* |
1653 | * At this point, search_start should be the end of | |
1654 | * allocated dev extents, and when shrinking the device, | |
1655 | * search_end may be smaller than search_start. | |
1656 | */ | |
f2ab7618 | 1657 | if (search_end > search_start) { |
38c01b96 | 1658 | hole_size = search_end - search_start; |
1659 | ||
499f377f | 1660 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1661 | hole_size)) { |
1662 | btrfs_release_path(path); | |
1663 | goto again; | |
1664 | } | |
0b86a832 | 1665 | |
f2ab7618 ZL |
1666 | if (hole_size > max_hole_size) { |
1667 | max_hole_start = search_start; | |
1668 | max_hole_size = hole_size; | |
1669 | } | |
6df9a95e JB |
1670 | } |
1671 | ||
7bfc837d | 1672 | /* See above. */ |
f2ab7618 | 1673 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1674 | ret = -ENOSPC; |
1675 | else | |
1676 | ret = 0; | |
1677 | ||
1678 | out: | |
2b82032c | 1679 | btrfs_free_path(path); |
7bfc837d | 1680 | *start = max_hole_start; |
b2117a39 | 1681 | if (len) |
7bfc837d | 1682 | *len = max_hole_size; |
0b86a832 CM |
1683 | return ret; |
1684 | } | |
1685 | ||
499f377f JM |
1686 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1687 | struct btrfs_device *device, u64 num_bytes, | |
1688 | u64 *start, u64 *len) | |
1689 | { | |
499f377f | 1690 | /* FIXME use last free of some kind */ |
499f377f | 1691 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1692 | num_bytes, 0, start, len); |
499f377f JM |
1693 | } |
1694 | ||
b2950863 | 1695 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1696 | struct btrfs_device *device, |
2196d6e8 | 1697 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1698 | { |
0b246afa JM |
1699 | struct btrfs_fs_info *fs_info = device->fs_info; |
1700 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1701 | int ret; |
1702 | struct btrfs_path *path; | |
8f18cf13 | 1703 | struct btrfs_key key; |
a061fc8d CM |
1704 | struct btrfs_key found_key; |
1705 | struct extent_buffer *leaf = NULL; | |
1706 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1707 | |
1708 | path = btrfs_alloc_path(); | |
1709 | if (!path) | |
1710 | return -ENOMEM; | |
1711 | ||
1712 | key.objectid = device->devid; | |
1713 | key.offset = start; | |
1714 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1715 | again: |
8f18cf13 | 1716 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1717 | if (ret > 0) { |
1718 | ret = btrfs_previous_item(root, path, key.objectid, | |
1719 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1720 | if (ret) |
1721 | goto out; | |
a061fc8d CM |
1722 | leaf = path->nodes[0]; |
1723 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1724 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1725 | struct btrfs_dev_extent); | |
1726 | BUG_ON(found_key.offset > start || found_key.offset + | |
1727 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1728 | key = found_key; |
1729 | btrfs_release_path(path); | |
1730 | goto again; | |
a061fc8d CM |
1731 | } else if (ret == 0) { |
1732 | leaf = path->nodes[0]; | |
1733 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1734 | struct btrfs_dev_extent); | |
79787eaa | 1735 | } else { |
0b246afa | 1736 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1737 | goto out; |
a061fc8d | 1738 | } |
8f18cf13 | 1739 | |
2196d6e8 MX |
1740 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1741 | ||
8f18cf13 | 1742 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1743 | if (ret) { |
0b246afa JM |
1744 | btrfs_handle_fs_error(fs_info, ret, |
1745 | "Failed to remove dev extent item"); | |
13212b54 | 1746 | } else { |
3204d33c | 1747 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1748 | } |
b0b802d7 | 1749 | out: |
8f18cf13 CM |
1750 | btrfs_free_path(path); |
1751 | return ret; | |
1752 | } | |
1753 | ||
48a3b636 ES |
1754 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1755 | struct btrfs_device *device, | |
48a3b636 | 1756 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1757 | { |
1758 | int ret; | |
1759 | struct btrfs_path *path; | |
0b246afa JM |
1760 | struct btrfs_fs_info *fs_info = device->fs_info; |
1761 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1762 | struct btrfs_dev_extent *extent; |
1763 | struct extent_buffer *leaf; | |
1764 | struct btrfs_key key; | |
1765 | ||
e12c9621 | 1766 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1767 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1768 | path = btrfs_alloc_path(); |
1769 | if (!path) | |
1770 | return -ENOMEM; | |
1771 | ||
0b86a832 | 1772 | key.objectid = device->devid; |
2b82032c | 1773 | key.offset = start; |
0b86a832 CM |
1774 | key.type = BTRFS_DEV_EXTENT_KEY; |
1775 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1776 | sizeof(*extent)); | |
2cdcecbc MF |
1777 | if (ret) |
1778 | goto out; | |
0b86a832 CM |
1779 | |
1780 | leaf = path->nodes[0]; | |
1781 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1782 | struct btrfs_dev_extent); | |
b5d9071c NB |
1783 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1784 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1785 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1786 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1787 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1788 | ||
0b86a832 CM |
1789 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1790 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1791 | out: |
0b86a832 CM |
1792 | btrfs_free_path(path); |
1793 | return ret; | |
1794 | } | |
1795 | ||
6df9a95e | 1796 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1797 | { |
6df9a95e JB |
1798 | struct extent_map_tree *em_tree; |
1799 | struct extent_map *em; | |
1800 | struct rb_node *n; | |
1801 | u64 ret = 0; | |
0b86a832 | 1802 | |
6df9a95e JB |
1803 | em_tree = &fs_info->mapping_tree.map_tree; |
1804 | read_lock(&em_tree->lock); | |
07e1ce09 | 1805 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1806 | if (n) { |
1807 | em = rb_entry(n, struct extent_map, rb_node); | |
1808 | ret = em->start + em->len; | |
0b86a832 | 1809 | } |
6df9a95e JB |
1810 | read_unlock(&em_tree->lock); |
1811 | ||
0b86a832 CM |
1812 | return ret; |
1813 | } | |
1814 | ||
53f10659 ID |
1815 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1816 | u64 *devid_ret) | |
0b86a832 CM |
1817 | { |
1818 | int ret; | |
1819 | struct btrfs_key key; | |
1820 | struct btrfs_key found_key; | |
2b82032c YZ |
1821 | struct btrfs_path *path; |
1822 | ||
2b82032c YZ |
1823 | path = btrfs_alloc_path(); |
1824 | if (!path) | |
1825 | return -ENOMEM; | |
0b86a832 CM |
1826 | |
1827 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1828 | key.type = BTRFS_DEV_ITEM_KEY; | |
1829 | key.offset = (u64)-1; | |
1830 | ||
53f10659 | 1831 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1832 | if (ret < 0) |
1833 | goto error; | |
1834 | ||
79787eaa | 1835 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1836 | |
53f10659 ID |
1837 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1838 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1839 | BTRFS_DEV_ITEM_KEY); |
1840 | if (ret) { | |
53f10659 | 1841 | *devid_ret = 1; |
0b86a832 CM |
1842 | } else { |
1843 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1844 | path->slots[0]); | |
53f10659 | 1845 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1846 | } |
1847 | ret = 0; | |
1848 | error: | |
2b82032c | 1849 | btrfs_free_path(path); |
0b86a832 CM |
1850 | return ret; |
1851 | } | |
1852 | ||
1853 | /* | |
1854 | * the device information is stored in the chunk root | |
1855 | * the btrfs_device struct should be fully filled in | |
1856 | */ | |
c74a0b02 | 1857 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1858 | struct btrfs_device *device) |
0b86a832 CM |
1859 | { |
1860 | int ret; | |
1861 | struct btrfs_path *path; | |
1862 | struct btrfs_dev_item *dev_item; | |
1863 | struct extent_buffer *leaf; | |
1864 | struct btrfs_key key; | |
1865 | unsigned long ptr; | |
0b86a832 | 1866 | |
0b86a832 CM |
1867 | path = btrfs_alloc_path(); |
1868 | if (!path) | |
1869 | return -ENOMEM; | |
1870 | ||
0b86a832 CM |
1871 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1872 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1873 | key.offset = device->devid; |
0b86a832 | 1874 | |
8e87e856 NB |
1875 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1876 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1877 | if (ret) |
1878 | goto out; | |
1879 | ||
1880 | leaf = path->nodes[0]; | |
1881 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1882 | ||
1883 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1884 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1885 | btrfs_set_device_type(leaf, dev_item, device->type); |
1886 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1887 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1888 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1889 | btrfs_set_device_total_bytes(leaf, dev_item, |
1890 | btrfs_device_get_disk_total_bytes(device)); | |
1891 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1892 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1893 | btrfs_set_device_group(leaf, dev_item, 0); |
1894 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1895 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1896 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1897 | |
410ba3a2 | 1898 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1899 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1900 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1901 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1902 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1903 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1904 | |
2b82032c | 1905 | ret = 0; |
0b86a832 CM |
1906 | out: |
1907 | btrfs_free_path(path); | |
1908 | return ret; | |
1909 | } | |
8f18cf13 | 1910 | |
5a1972bd QW |
1911 | /* |
1912 | * Function to update ctime/mtime for a given device path. | |
1913 | * Mainly used for ctime/mtime based probe like libblkid. | |
1914 | */ | |
da353f6b | 1915 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1916 | { |
1917 | struct file *filp; | |
1918 | ||
1919 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1920 | if (IS_ERR(filp)) |
5a1972bd QW |
1921 | return; |
1922 | file_update_time(filp); | |
1923 | filp_close(filp, NULL); | |
5a1972bd QW |
1924 | } |
1925 | ||
5b4aacef | 1926 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1927 | struct btrfs_device *device) |
1928 | { | |
5b4aacef | 1929 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1930 | int ret; |
1931 | struct btrfs_path *path; | |
a061fc8d | 1932 | struct btrfs_key key; |
a061fc8d CM |
1933 | struct btrfs_trans_handle *trans; |
1934 | ||
a061fc8d CM |
1935 | path = btrfs_alloc_path(); |
1936 | if (!path) | |
1937 | return -ENOMEM; | |
1938 | ||
a22285a6 | 1939 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1940 | if (IS_ERR(trans)) { |
1941 | btrfs_free_path(path); | |
1942 | return PTR_ERR(trans); | |
1943 | } | |
a061fc8d CM |
1944 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1945 | key.type = BTRFS_DEV_ITEM_KEY; | |
1946 | key.offset = device->devid; | |
1947 | ||
1948 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1949 | if (ret) { |
1950 | if (ret > 0) | |
1951 | ret = -ENOENT; | |
1952 | btrfs_abort_transaction(trans, ret); | |
1953 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1954 | goto out; |
1955 | } | |
1956 | ||
1957 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1958 | if (ret) { |
1959 | btrfs_abort_transaction(trans, ret); | |
1960 | btrfs_end_transaction(trans); | |
1961 | } | |
1962 | ||
a061fc8d CM |
1963 | out: |
1964 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1965 | if (!ret) |
1966 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1967 | return ret; |
1968 | } | |
1969 | ||
3cc31a0d DS |
1970 | /* |
1971 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1972 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1973 | * replace. | |
1974 | */ | |
1975 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1976 | u64 num_devices) | |
a061fc8d | 1977 | { |
a061fc8d | 1978 | u64 all_avail; |
de98ced9 | 1979 | unsigned seq; |
418775a2 | 1980 | int i; |
a061fc8d | 1981 | |
de98ced9 | 1982 | do { |
bd45ffbc | 1983 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1984 | |
bd45ffbc AJ |
1985 | all_avail = fs_info->avail_data_alloc_bits | |
1986 | fs_info->avail_system_alloc_bits | | |
1987 | fs_info->avail_metadata_alloc_bits; | |
1988 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1989 | |
418775a2 | 1990 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1991 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1992 | continue; |
a061fc8d | 1993 | |
418775a2 | 1994 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 1995 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 1996 | |
418775a2 DS |
1997 | if (ret) |
1998 | return ret; | |
1999 | } | |
53b381b3 DW |
2000 | } |
2001 | ||
bd45ffbc | 2002 | return 0; |
f1fa7f26 AJ |
2003 | } |
2004 | ||
c9162bdf OS |
2005 | static struct btrfs_device * btrfs_find_next_active_device( |
2006 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2007 | { |
2b82032c | 2008 | struct btrfs_device *next_device; |
88acff64 AJ |
2009 | |
2010 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2011 | if (next_device != device && | |
e6e674bd AJ |
2012 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2013 | && next_device->bdev) | |
88acff64 AJ |
2014 | return next_device; |
2015 | } | |
2016 | ||
2017 | return NULL; | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2022 | * and replace it with the provided or the next active device, in the context | |
2023 | * where this function called, there should be always be another device (or | |
2024 | * this_dev) which is active. | |
2025 | */ | |
d6507cf1 NB |
2026 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
2027 | struct btrfs_device *this_dev) | |
88acff64 | 2028 | { |
d6507cf1 | 2029 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
2030 | struct btrfs_device *next_device; |
2031 | ||
2032 | if (this_dev) | |
2033 | next_device = this_dev; | |
2034 | else | |
2035 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
2036 | device); | |
2037 | ASSERT(next_device); | |
2038 | ||
2039 | if (fs_info->sb->s_bdev && | |
2040 | (fs_info->sb->s_bdev == device->bdev)) | |
2041 | fs_info->sb->s_bdev = next_device->bdev; | |
2042 | ||
2043 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2044 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2045 | } | |
2046 | ||
1da73967 AJ |
2047 | /* |
2048 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2049 | * currently replaced. | |
2050 | */ | |
2051 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2052 | { | |
2053 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2054 | ||
2055 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); | |
2056 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
2057 | ASSERT(num_devices > 1); | |
2058 | num_devices--; | |
2059 | } | |
2060 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); | |
2061 | ||
2062 | return num_devices; | |
2063 | } | |
2064 | ||
da353f6b DS |
2065 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2066 | u64 devid) | |
f1fa7f26 AJ |
2067 | { |
2068 | struct btrfs_device *device; | |
1f78160c | 2069 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2070 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2071 | u64 num_devices; |
a061fc8d CM |
2072 | int ret = 0; |
2073 | ||
a061fc8d CM |
2074 | mutex_lock(&uuid_mutex); |
2075 | ||
1da73967 | 2076 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2077 | |
0b246afa | 2078 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2079 | if (ret) |
a061fc8d | 2080 | goto out; |
a061fc8d | 2081 | |
a27a94c2 NB |
2082 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2083 | ||
2084 | if (IS_ERR(device)) { | |
2085 | if (PTR_ERR(device) == -ENOENT && | |
2086 | strcmp(device_path, "missing") == 0) | |
2087 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2088 | else | |
2089 | ret = PTR_ERR(device); | |
53b381b3 | 2090 | goto out; |
a27a94c2 | 2091 | } |
dfe25020 | 2092 | |
eede2bf3 OS |
2093 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2094 | btrfs_warn_in_rcu(fs_info, | |
2095 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2096 | rcu_str_deref(device->name), device->devid); | |
2097 | ret = -ETXTBSY; | |
2098 | goto out; | |
2099 | } | |
2100 | ||
401e29c1 | 2101 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2102 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2103 | goto out; |
63a212ab SB |
2104 | } |
2105 | ||
ebbede42 AJ |
2106 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2107 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2108 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2109 | goto out; |
2b82032c YZ |
2110 | } |
2111 | ||
ebbede42 | 2112 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2113 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2114 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2115 | device->fs_devices->rw_devices--; |
34441361 | 2116 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2117 | } |
a061fc8d | 2118 | |
d7901554 | 2119 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2120 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2121 | mutex_lock(&uuid_mutex); |
a061fc8d | 2122 | if (ret) |
9b3517e9 | 2123 | goto error_undo; |
a061fc8d | 2124 | |
63a212ab SB |
2125 | /* |
2126 | * TODO: the superblock still includes this device in its num_devices | |
2127 | * counter although write_all_supers() is not locked out. This | |
2128 | * could give a filesystem state which requires a degraded mount. | |
2129 | */ | |
0b246afa | 2130 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 2131 | if (ret) |
9b3517e9 | 2132 | goto error_undo; |
a061fc8d | 2133 | |
e12c9621 | 2134 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 2135 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
2136 | |
2137 | /* | |
2138 | * the device list mutex makes sure that we don't change | |
2139 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2140 | * the device supers. Whoever is writing all supers, should |
2141 | * lock the device list mutex before getting the number of | |
2142 | * devices in the super block (super_copy). Conversely, | |
2143 | * whoever updates the number of devices in the super block | |
2144 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2145 | */ |
1f78160c | 2146 | |
41a52a0f AJ |
2147 | /* |
2148 | * In normal cases the cur_devices == fs_devices. But in case | |
2149 | * of deleting a seed device, the cur_devices should point to | |
2150 | * its own fs_devices listed under the fs_devices->seed. | |
2151 | */ | |
1f78160c | 2152 | cur_devices = device->fs_devices; |
b5185197 | 2153 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2154 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2155 | |
41a52a0f AJ |
2156 | cur_devices->num_devices--; |
2157 | cur_devices->total_devices--; | |
b4993e64 AJ |
2158 | /* Update total_devices of the parent fs_devices if it's seed */ |
2159 | if (cur_devices != fs_devices) | |
2160 | fs_devices->total_devices--; | |
2b82032c | 2161 | |
e6e674bd | 2162 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2163 | cur_devices->missing_devices--; |
cd02dca5 | 2164 | |
d6507cf1 | 2165 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2166 | |
0bfaa9c5 | 2167 | if (device->bdev) { |
41a52a0f | 2168 | cur_devices->open_devices--; |
0bfaa9c5 | 2169 | /* remove sysfs entry */ |
b5185197 | 2170 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2171 | } |
99994cde | 2172 | |
0b246afa JM |
2173 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2174 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2175 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2176 | |
cea67ab9 JM |
2177 | /* |
2178 | * at this point, the device is zero sized and detached from | |
2179 | * the devices list. All that's left is to zero out the old | |
2180 | * supers and free the device. | |
2181 | */ | |
ebbede42 | 2182 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2183 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2184 | ||
2185 | btrfs_close_bdev(device); | |
f06c5965 | 2186 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2187 | |
1f78160c | 2188 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2189 | while (fs_devices) { |
8321cf25 RS |
2190 | if (fs_devices->seed == cur_devices) { |
2191 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2192 | break; |
8321cf25 | 2193 | } |
e4404d6e | 2194 | fs_devices = fs_devices->seed; |
2b82032c | 2195 | } |
1f78160c | 2196 | cur_devices->seed = NULL; |
0226e0eb | 2197 | close_fs_devices(cur_devices); |
1f78160c | 2198 | free_fs_devices(cur_devices); |
2b82032c YZ |
2199 | } |
2200 | ||
a061fc8d CM |
2201 | out: |
2202 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2203 | return ret; |
24fc572f | 2204 | |
9b3517e9 | 2205 | error_undo: |
ebbede42 | 2206 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2207 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2208 | list_add(&device->dev_alloc_list, |
b5185197 | 2209 | &fs_devices->alloc_list); |
c3929c36 | 2210 | device->fs_devices->rw_devices++; |
34441361 | 2211 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2212 | } |
24fc572f | 2213 | goto out; |
a061fc8d CM |
2214 | } |
2215 | ||
68a9db5f | 2216 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2217 | { |
d51908ce AJ |
2218 | struct btrfs_fs_devices *fs_devices; |
2219 | ||
68a9db5f | 2220 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2221 | |
25e8e911 AJ |
2222 | /* |
2223 | * in case of fs with no seed, srcdev->fs_devices will point | |
2224 | * to fs_devices of fs_info. However when the dev being replaced is | |
2225 | * a seed dev it will point to the seed's local fs_devices. In short | |
2226 | * srcdev will have its correct fs_devices in both the cases. | |
2227 | */ | |
2228 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2229 | |
e93c89c1 | 2230 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2231 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2232 | fs_devices->num_devices--; |
e6e674bd | 2233 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2234 | fs_devices->missing_devices--; |
e93c89c1 | 2235 | |
ebbede42 | 2236 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2237 | fs_devices->rw_devices--; |
1357272f | 2238 | |
82372bc8 | 2239 | if (srcdev->bdev) |
d51908ce | 2240 | fs_devices->open_devices--; |
084b6e7c QW |
2241 | } |
2242 | ||
2243 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2244 | struct btrfs_device *srcdev) | |
2245 | { | |
2246 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2247 | |
ebbede42 | 2248 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2249 | /* zero out the old super if it is writable */ |
2250 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2251 | } | |
14238819 AJ |
2252 | |
2253 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2254 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2255 | |
94d5f0c2 AJ |
2256 | /* if this is no devs we rather delete the fs_devices */ |
2257 | if (!fs_devices->num_devices) { | |
2258 | struct btrfs_fs_devices *tmp_fs_devices; | |
2259 | ||
6dd38f81 AJ |
2260 | /* |
2261 | * On a mounted FS, num_devices can't be zero unless it's a | |
2262 | * seed. In case of a seed device being replaced, the replace | |
2263 | * target added to the sprout FS, so there will be no more | |
2264 | * device left under the seed FS. | |
2265 | */ | |
2266 | ASSERT(fs_devices->seeding); | |
2267 | ||
94d5f0c2 AJ |
2268 | tmp_fs_devices = fs_info->fs_devices; |
2269 | while (tmp_fs_devices) { | |
2270 | if (tmp_fs_devices->seed == fs_devices) { | |
2271 | tmp_fs_devices->seed = fs_devices->seed; | |
2272 | break; | |
2273 | } | |
2274 | tmp_fs_devices = tmp_fs_devices->seed; | |
2275 | } | |
2276 | fs_devices->seed = NULL; | |
0226e0eb | 2277 | close_fs_devices(fs_devices); |
8bef8401 | 2278 | free_fs_devices(fs_devices); |
94d5f0c2 | 2279 | } |
e93c89c1 SB |
2280 | } |
2281 | ||
4f5ad7bd | 2282 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2283 | { |
4f5ad7bd | 2284 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2285 | |
e93c89c1 | 2286 | WARN_ON(!tgtdev); |
d9a071f0 | 2287 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2288 | |
d9a071f0 | 2289 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2290 | |
779bf3fe | 2291 | if (tgtdev->bdev) |
d9a071f0 | 2292 | fs_devices->open_devices--; |
779bf3fe | 2293 | |
d9a071f0 | 2294 | fs_devices->num_devices--; |
e93c89c1 | 2295 | |
d6507cf1 | 2296 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2297 | |
e93c89c1 | 2298 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2299 | |
d9a071f0 | 2300 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2301 | |
2302 | /* | |
2303 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2304 | * may lead to a call to btrfs_show_devname() which will try | |
2305 | * to hold device_list_mutex. And here this device | |
2306 | * is already out of device list, so we don't have to hold | |
2307 | * the device_list_mutex lock. | |
2308 | */ | |
2309 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2310 | |
2311 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2312 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2313 | } |
2314 | ||
b444ad46 NB |
2315 | static struct btrfs_device *btrfs_find_device_by_path( |
2316 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2317 | { |
2318 | int ret = 0; | |
2319 | struct btrfs_super_block *disk_super; | |
2320 | u64 devid; | |
2321 | u8 *dev_uuid; | |
2322 | struct block_device *bdev; | |
2323 | struct buffer_head *bh; | |
b444ad46 | 2324 | struct btrfs_device *device; |
7ba15b7d | 2325 | |
7ba15b7d | 2326 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2327 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2328 | if (ret) |
b444ad46 | 2329 | return ERR_PTR(ret); |
7ba15b7d SB |
2330 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2331 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2332 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b NB |
2333 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
2334 | device = btrfs_find_device(fs_info, devid, dev_uuid, | |
2335 | disk_super->metadata_uuid); | |
2336 | else | |
2337 | device = btrfs_find_device(fs_info, devid, | |
2338 | dev_uuid, disk_super->fsid); | |
2339 | ||
7ba15b7d | 2340 | brelse(bh); |
b444ad46 NB |
2341 | if (!device) |
2342 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2343 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2344 | return device; |
7ba15b7d SB |
2345 | } |
2346 | ||
6c050407 NB |
2347 | static struct btrfs_device *btrfs_find_device_missing_or_by_path( |
2348 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d | 2349 | { |
6c050407 | 2350 | struct btrfs_device *device = NULL; |
7ba15b7d SB |
2351 | if (strcmp(device_path, "missing") == 0) { |
2352 | struct list_head *devices; | |
2353 | struct btrfs_device *tmp; | |
2354 | ||
0b246afa | 2355 | devices = &fs_info->fs_devices->devices; |
7ba15b7d | 2356 | list_for_each_entry(tmp, devices, dev_list) { |
e12c9621 AJ |
2357 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2358 | &tmp->dev_state) && !tmp->bdev) { | |
6c050407 | 2359 | device = tmp; |
7ba15b7d SB |
2360 | break; |
2361 | } | |
2362 | } | |
2363 | ||
6c050407 NB |
2364 | if (!device) |
2365 | return ERR_PTR(-ENOENT); | |
7ba15b7d | 2366 | } else { |
6c050407 | 2367 | device = btrfs_find_device_by_path(fs_info, device_path); |
7ba15b7d | 2368 | } |
b444ad46 | 2369 | |
6c050407 | 2370 | return device; |
7ba15b7d SB |
2371 | } |
2372 | ||
5c5c0df0 DS |
2373 | /* |
2374 | * Lookup a device given by device id, or the path if the id is 0. | |
2375 | */ | |
a27a94c2 NB |
2376 | struct btrfs_device *btrfs_find_device_by_devspec( |
2377 | struct btrfs_fs_info *fs_info, u64 devid, const char *devpath) | |
24e0474b | 2378 | { |
a27a94c2 | 2379 | struct btrfs_device *device; |
24e0474b | 2380 | |
5c5c0df0 | 2381 | if (devid) { |
a27a94c2 NB |
2382 | device = btrfs_find_device(fs_info, devid, NULL, NULL); |
2383 | if (!device) | |
2384 | return ERR_PTR(-ENOENT); | |
24e0474b | 2385 | } else { |
5c5c0df0 | 2386 | if (!devpath || !devpath[0]) |
a27a94c2 NB |
2387 | return ERR_PTR(-EINVAL); |
2388 | device = btrfs_find_device_missing_or_by_path(fs_info, devpath); | |
24e0474b | 2389 | } |
a27a94c2 | 2390 | return device; |
24e0474b AJ |
2391 | } |
2392 | ||
2b82032c YZ |
2393 | /* |
2394 | * does all the dirty work required for changing file system's UUID. | |
2395 | */ | |
2ff7e61e | 2396 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2397 | { |
0b246afa | 2398 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2399 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2400 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2401 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2402 | struct btrfs_device *device; |
2403 | u64 super_flags; | |
2404 | ||
a32bf9a3 | 2405 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2406 | if (!fs_devices->seeding) |
2b82032c YZ |
2407 | return -EINVAL; |
2408 | ||
7239ff4b | 2409 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2410 | if (IS_ERR(seed_devices)) |
2411 | return PTR_ERR(seed_devices); | |
2b82032c | 2412 | |
e4404d6e YZ |
2413 | old_devices = clone_fs_devices(fs_devices); |
2414 | if (IS_ERR(old_devices)) { | |
2415 | kfree(seed_devices); | |
2416 | return PTR_ERR(old_devices); | |
2b82032c | 2417 | } |
e4404d6e | 2418 | |
c4babc5e | 2419 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2420 | |
e4404d6e YZ |
2421 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2422 | seed_devices->opened = 1; | |
2423 | INIT_LIST_HEAD(&seed_devices->devices); | |
2424 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2425 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2426 | |
321a4bf7 | 2427 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2428 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2429 | synchronize_rcu); | |
2196d6e8 MX |
2430 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2431 | device->fs_devices = seed_devices; | |
c9513edb | 2432 | |
34441361 | 2433 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2434 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2435 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2436 | |
2b82032c YZ |
2437 | fs_devices->seeding = 0; |
2438 | fs_devices->num_devices = 0; | |
2439 | fs_devices->open_devices = 0; | |
69611ac8 | 2440 | fs_devices->missing_devices = 0; |
69611ac8 | 2441 | fs_devices->rotating = 0; |
e4404d6e | 2442 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2443 | |
2444 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2445 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2446 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2447 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2448 | |
2b82032c YZ |
2449 | super_flags = btrfs_super_flags(disk_super) & |
2450 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2451 | btrfs_set_super_flags(disk_super, super_flags); | |
2452 | ||
2453 | return 0; | |
2454 | } | |
2455 | ||
2456 | /* | |
01327610 | 2457 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2458 | */ |
2459 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2460 | struct btrfs_fs_info *fs_info) |
2b82032c | 2461 | { |
5b4aacef | 2462 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2463 | struct btrfs_path *path; |
2464 | struct extent_buffer *leaf; | |
2465 | struct btrfs_dev_item *dev_item; | |
2466 | struct btrfs_device *device; | |
2467 | struct btrfs_key key; | |
44880fdc | 2468 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2469 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2470 | u64 devid; | |
2471 | int ret; | |
2472 | ||
2473 | path = btrfs_alloc_path(); | |
2474 | if (!path) | |
2475 | return -ENOMEM; | |
2476 | ||
2b82032c YZ |
2477 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2478 | key.offset = 0; | |
2479 | key.type = BTRFS_DEV_ITEM_KEY; | |
2480 | ||
2481 | while (1) { | |
2482 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2483 | if (ret < 0) | |
2484 | goto error; | |
2485 | ||
2486 | leaf = path->nodes[0]; | |
2487 | next_slot: | |
2488 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2489 | ret = btrfs_next_leaf(root, path); | |
2490 | if (ret > 0) | |
2491 | break; | |
2492 | if (ret < 0) | |
2493 | goto error; | |
2494 | leaf = path->nodes[0]; | |
2495 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2496 | btrfs_release_path(path); |
2b82032c YZ |
2497 | continue; |
2498 | } | |
2499 | ||
2500 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2501 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2502 | key.type != BTRFS_DEV_ITEM_KEY) | |
2503 | break; | |
2504 | ||
2505 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2506 | struct btrfs_dev_item); | |
2507 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2508 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2509 | BTRFS_UUID_SIZE); |
1473b24e | 2510 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2511 | BTRFS_FSID_SIZE); |
0b246afa | 2512 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2513 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2514 | |
2515 | if (device->fs_devices->seeding) { | |
2516 | btrfs_set_device_generation(leaf, dev_item, | |
2517 | device->generation); | |
2518 | btrfs_mark_buffer_dirty(leaf); | |
2519 | } | |
2520 | ||
2521 | path->slots[0]++; | |
2522 | goto next_slot; | |
2523 | } | |
2524 | ret = 0; | |
2525 | error: | |
2526 | btrfs_free_path(path); | |
2527 | return ret; | |
2528 | } | |
2529 | ||
da353f6b | 2530 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2531 | { |
5112febb | 2532 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2533 | struct request_queue *q; |
788f20eb CM |
2534 | struct btrfs_trans_handle *trans; |
2535 | struct btrfs_device *device; | |
2536 | struct block_device *bdev; | |
0b246afa | 2537 | struct super_block *sb = fs_info->sb; |
606686ee | 2538 | struct rcu_string *name; |
5da54bc1 | 2539 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2540 | u64 orig_super_total_bytes; |
2541 | u64 orig_super_num_devices; | |
2b82032c | 2542 | int seeding_dev = 0; |
788f20eb | 2543 | int ret = 0; |
7132a262 | 2544 | bool unlocked = false; |
788f20eb | 2545 | |
5da54bc1 | 2546 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2547 | return -EROFS; |
788f20eb | 2548 | |
a5d16333 | 2549 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2550 | fs_info->bdev_holder); |
7f59203a JB |
2551 | if (IS_ERR(bdev)) |
2552 | return PTR_ERR(bdev); | |
a2135011 | 2553 | |
5da54bc1 | 2554 | if (fs_devices->seeding) { |
2b82032c YZ |
2555 | seeding_dev = 1; |
2556 | down_write(&sb->s_umount); | |
2557 | mutex_lock(&uuid_mutex); | |
2558 | } | |
2559 | ||
8c8bee1d | 2560 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2561 | |
5da54bc1 | 2562 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2563 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2564 | if (device->bdev == bdev) { |
2565 | ret = -EEXIST; | |
d25628bd | 2566 | mutex_unlock( |
5da54bc1 | 2567 | &fs_devices->device_list_mutex); |
2b82032c | 2568 | goto error; |
788f20eb CM |
2569 | } |
2570 | } | |
5da54bc1 | 2571 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2572 | |
0b246afa | 2573 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2574 | if (IS_ERR(device)) { |
788f20eb | 2575 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2576 | ret = PTR_ERR(device); |
2b82032c | 2577 | goto error; |
788f20eb CM |
2578 | } |
2579 | ||
78f2c9e6 | 2580 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2581 | if (!name) { |
2b82032c | 2582 | ret = -ENOMEM; |
5c4cf6c9 | 2583 | goto error_free_device; |
788f20eb | 2584 | } |
606686ee | 2585 | rcu_assign_pointer(device->name, name); |
2b82032c | 2586 | |
a22285a6 | 2587 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2588 | if (IS_ERR(trans)) { |
98d5dc13 | 2589 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2590 | goto error_free_device; |
98d5dc13 TI |
2591 | } |
2592 | ||
d5e2003c | 2593 | q = bdev_get_queue(bdev); |
ebbede42 | 2594 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2595 | device->generation = trans->transid; |
0b246afa JM |
2596 | device->io_width = fs_info->sectorsize; |
2597 | device->io_align = fs_info->sectorsize; | |
2598 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2599 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2600 | fs_info->sectorsize); | |
2cc3c559 | 2601 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2602 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2603 | device->fs_info = fs_info; |
788f20eb | 2604 | device->bdev = bdev; |
e12c9621 | 2605 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2606 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2607 | device->mode = FMODE_EXCL; |
27087f37 | 2608 | device->dev_stats_valid = 1; |
9f6d2510 | 2609 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2610 | |
2b82032c | 2611 | if (seeding_dev) { |
1751e8a6 | 2612 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2613 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2614 | if (ret) { |
2615 | btrfs_abort_transaction(trans, ret); | |
2616 | goto error_trans; | |
2617 | } | |
2b82032c | 2618 | } |
788f20eb | 2619 | |
5da54bc1 | 2620 | device->fs_devices = fs_devices; |
e5e9a520 | 2621 | |
5da54bc1 | 2622 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2623 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2624 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2625 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2626 | fs_devices->num_devices++; | |
2627 | fs_devices->open_devices++; | |
2628 | fs_devices->rw_devices++; | |
2629 | fs_devices->total_devices++; | |
2630 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2631 | |
a5ed45f8 | 2632 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2633 | |
e884f4f0 | 2634 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2635 | fs_devices->rotating = 1; |
c289811c | 2636 | |
39379faa | 2637 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2638 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2639 | round_down(orig_super_total_bytes + device->total_bytes, |
2640 | fs_info->sectorsize)); | |
788f20eb | 2641 | |
39379faa NA |
2642 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2643 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2644 | orig_super_num_devices + 1); | |
0d39376a AJ |
2645 | |
2646 | /* add sysfs device entry */ | |
5da54bc1 | 2647 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2648 | |
2196d6e8 MX |
2649 | /* |
2650 | * we've got more storage, clear any full flags on the space | |
2651 | * infos | |
2652 | */ | |
0b246afa | 2653 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2654 | |
34441361 | 2655 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2656 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2657 | |
2b82032c | 2658 | if (seeding_dev) { |
34441361 | 2659 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2660 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2661 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2662 | if (ret) { |
66642832 | 2663 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2664 | goto error_sysfs; |
005d6427 | 2665 | } |
2196d6e8 MX |
2666 | } |
2667 | ||
8e87e856 | 2668 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2669 | if (ret) { |
66642832 | 2670 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2671 | goto error_sysfs; |
2196d6e8 MX |
2672 | } |
2673 | ||
2674 | if (seeding_dev) { | |
2675 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2676 | ||
0b246afa | 2677 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2678 | if (ret) { |
66642832 | 2679 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2680 | goto error_sysfs; |
005d6427 | 2681 | } |
b2373f25 AJ |
2682 | |
2683 | /* Sprouting would change fsid of the mounted root, | |
2684 | * so rename the fsid on the sysfs | |
2685 | */ | |
2686 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
de37aa51 | 2687 | fs_info->fs_devices->fsid); |
5da54bc1 | 2688 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2689 | btrfs_warn(fs_info, |
2690 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2691 | } |
2692 | ||
3a45bb20 | 2693 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2694 | |
2b82032c YZ |
2695 | if (seeding_dev) { |
2696 | mutex_unlock(&uuid_mutex); | |
2697 | up_write(&sb->s_umount); | |
7132a262 | 2698 | unlocked = true; |
788f20eb | 2699 | |
79787eaa JM |
2700 | if (ret) /* transaction commit */ |
2701 | return ret; | |
2702 | ||
2ff7e61e | 2703 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2704 | if (ret < 0) |
0b246afa | 2705 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2706 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2707 | trans = btrfs_attach_transaction(root); |
2708 | if (IS_ERR(trans)) { | |
2709 | if (PTR_ERR(trans) == -ENOENT) | |
2710 | return 0; | |
7132a262 AJ |
2711 | ret = PTR_ERR(trans); |
2712 | trans = NULL; | |
2713 | goto error_sysfs; | |
671415b7 | 2714 | } |
3a45bb20 | 2715 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2716 | } |
c9e9f97b | 2717 | |
5a1972bd QW |
2718 | /* Update ctime/mtime for libblkid */ |
2719 | update_dev_time(device_path); | |
2b82032c | 2720 | return ret; |
79787eaa | 2721 | |
d31c32f6 | 2722 | error_sysfs: |
5da54bc1 | 2723 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2724 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2725 | mutex_lock(&fs_info->chunk_mutex); | |
2726 | list_del_rcu(&device->dev_list); | |
2727 | list_del(&device->dev_alloc_list); | |
2728 | fs_info->fs_devices->num_devices--; | |
2729 | fs_info->fs_devices->open_devices--; | |
2730 | fs_info->fs_devices->rw_devices--; | |
2731 | fs_info->fs_devices->total_devices--; | |
2732 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2733 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2734 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2735 | orig_super_total_bytes); | |
2736 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2737 | orig_super_num_devices); | |
2738 | mutex_unlock(&fs_info->chunk_mutex); | |
2739 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2740 | error_trans: |
0af2c4bf | 2741 | if (seeding_dev) |
1751e8a6 | 2742 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2743 | if (trans) |
2744 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2745 | error_free_device: |
a425f9d4 | 2746 | btrfs_free_device(device); |
2b82032c | 2747 | error: |
e525fd89 | 2748 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2749 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2750 | mutex_unlock(&uuid_mutex); |
2751 | up_write(&sb->s_umount); | |
2752 | } | |
c9e9f97b | 2753 | return ret; |
788f20eb CM |
2754 | } |
2755 | ||
d397712b CM |
2756 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2757 | struct btrfs_device *device) | |
0b86a832 CM |
2758 | { |
2759 | int ret; | |
2760 | struct btrfs_path *path; | |
0b246afa | 2761 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2762 | struct btrfs_dev_item *dev_item; |
2763 | struct extent_buffer *leaf; | |
2764 | struct btrfs_key key; | |
2765 | ||
0b86a832 CM |
2766 | path = btrfs_alloc_path(); |
2767 | if (!path) | |
2768 | return -ENOMEM; | |
2769 | ||
2770 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2771 | key.type = BTRFS_DEV_ITEM_KEY; | |
2772 | key.offset = device->devid; | |
2773 | ||
2774 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2775 | if (ret < 0) | |
2776 | goto out; | |
2777 | ||
2778 | if (ret > 0) { | |
2779 | ret = -ENOENT; | |
2780 | goto out; | |
2781 | } | |
2782 | ||
2783 | leaf = path->nodes[0]; | |
2784 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2785 | ||
2786 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2787 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2788 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2789 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2790 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2791 | btrfs_set_device_total_bytes(leaf, dev_item, |
2792 | btrfs_device_get_disk_total_bytes(device)); | |
2793 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2794 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2795 | btrfs_mark_buffer_dirty(leaf); |
2796 | ||
2797 | out: | |
2798 | btrfs_free_path(path); | |
2799 | return ret; | |
2800 | } | |
2801 | ||
2196d6e8 | 2802 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2803 | struct btrfs_device *device, u64 new_size) |
2804 | { | |
0b246afa JM |
2805 | struct btrfs_fs_info *fs_info = device->fs_info; |
2806 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2807 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2808 | u64 old_total; |
2809 | u64 diff; | |
8f18cf13 | 2810 | |
ebbede42 | 2811 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2812 | return -EACCES; |
2196d6e8 | 2813 | |
7dfb8be1 NB |
2814 | new_size = round_down(new_size, fs_info->sectorsize); |
2815 | ||
34441361 | 2816 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2817 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2818 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2819 | |
63a212ab | 2820 | if (new_size <= device->total_bytes || |
401e29c1 | 2821 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2822 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2823 | return -EINVAL; |
2196d6e8 | 2824 | } |
2b82032c | 2825 | |
0b246afa | 2826 | fs_devices = fs_info->fs_devices; |
2b82032c | 2827 | |
7dfb8be1 NB |
2828 | btrfs_set_super_total_bytes(super_copy, |
2829 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2830 | device->fs_devices->total_rw_bytes += diff; |
2831 | ||
7cc8e58d MX |
2832 | btrfs_device_set_total_bytes(device, new_size); |
2833 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2834 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2835 | if (list_empty(&device->resized_list)) |
2836 | list_add_tail(&device->resized_list, | |
2837 | &fs_devices->resized_devices); | |
34441361 | 2838 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2839 | |
8f18cf13 CM |
2840 | return btrfs_update_device(trans, device); |
2841 | } | |
2842 | ||
f4208794 | 2843 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2844 | { |
f4208794 | 2845 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2846 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2847 | int ret; |
2848 | struct btrfs_path *path; | |
2849 | struct btrfs_key key; | |
2850 | ||
8f18cf13 CM |
2851 | path = btrfs_alloc_path(); |
2852 | if (!path) | |
2853 | return -ENOMEM; | |
2854 | ||
408fbf19 | 2855 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2856 | key.offset = chunk_offset; |
2857 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2858 | ||
2859 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2860 | if (ret < 0) |
2861 | goto out; | |
2862 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2863 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2864 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2865 | ret = -ENOENT; |
2866 | goto out; | |
2867 | } | |
8f18cf13 CM |
2868 | |
2869 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2870 | if (ret < 0) |
0b246afa JM |
2871 | btrfs_handle_fs_error(fs_info, ret, |
2872 | "Failed to delete chunk item."); | |
79787eaa | 2873 | out: |
8f18cf13 | 2874 | btrfs_free_path(path); |
65a246c5 | 2875 | return ret; |
8f18cf13 CM |
2876 | } |
2877 | ||
408fbf19 | 2878 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2879 | { |
0b246afa | 2880 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2881 | struct btrfs_disk_key *disk_key; |
2882 | struct btrfs_chunk *chunk; | |
2883 | u8 *ptr; | |
2884 | int ret = 0; | |
2885 | u32 num_stripes; | |
2886 | u32 array_size; | |
2887 | u32 len = 0; | |
2888 | u32 cur; | |
2889 | struct btrfs_key key; | |
2890 | ||
34441361 | 2891 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2892 | array_size = btrfs_super_sys_array_size(super_copy); |
2893 | ||
2894 | ptr = super_copy->sys_chunk_array; | |
2895 | cur = 0; | |
2896 | ||
2897 | while (cur < array_size) { | |
2898 | disk_key = (struct btrfs_disk_key *)ptr; | |
2899 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2900 | ||
2901 | len = sizeof(*disk_key); | |
2902 | ||
2903 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2904 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2905 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2906 | len += btrfs_chunk_item_size(num_stripes); | |
2907 | } else { | |
2908 | ret = -EIO; | |
2909 | break; | |
2910 | } | |
408fbf19 | 2911 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2912 | key.offset == chunk_offset) { |
2913 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2914 | array_size -= len; | |
2915 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2916 | } else { | |
2917 | ptr += len; | |
2918 | cur += len; | |
2919 | } | |
2920 | } | |
34441361 | 2921 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2922 | return ret; |
2923 | } | |
2924 | ||
60ca842e OS |
2925 | /* |
2926 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2927 | * @logical: Logical block offset in bytes. | |
2928 | * @length: Length of extent in bytes. | |
2929 | * | |
2930 | * Return: Chunk mapping or ERR_PTR. | |
2931 | */ | |
2932 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2933 | u64 logical, u64 length) | |
592d92ee LB |
2934 | { |
2935 | struct extent_map_tree *em_tree; | |
2936 | struct extent_map *em; | |
2937 | ||
2938 | em_tree = &fs_info->mapping_tree.map_tree; | |
2939 | read_lock(&em_tree->lock); | |
2940 | em = lookup_extent_mapping(em_tree, logical, length); | |
2941 | read_unlock(&em_tree->lock); | |
2942 | ||
2943 | if (!em) { | |
2944 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2945 | logical, length); | |
2946 | return ERR_PTR(-EINVAL); | |
2947 | } | |
2948 | ||
2949 | if (em->start > logical || em->start + em->len < logical) { | |
2950 | btrfs_crit(fs_info, | |
2951 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2952 | logical, length, em->start, em->start + em->len); | |
2953 | free_extent_map(em); | |
2954 | return ERR_PTR(-EINVAL); | |
2955 | } | |
2956 | ||
2957 | /* callers are responsible for dropping em's ref. */ | |
2958 | return em; | |
2959 | } | |
2960 | ||
97aff912 | 2961 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2962 | { |
97aff912 | 2963 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2964 | struct extent_map *em; |
2965 | struct map_lookup *map; | |
2196d6e8 | 2966 | u64 dev_extent_len = 0; |
47ab2a6c | 2967 | int i, ret = 0; |
0b246afa | 2968 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2969 | |
60ca842e | 2970 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2971 | if (IS_ERR(em)) { |
47ab2a6c JB |
2972 | /* |
2973 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2974 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2975 | * do anything we still error out. |
2976 | */ | |
2977 | ASSERT(0); | |
592d92ee | 2978 | return PTR_ERR(em); |
47ab2a6c | 2979 | } |
95617d69 | 2980 | map = em->map_lookup; |
34441361 | 2981 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2982 | check_system_chunk(trans, map->type); |
34441361 | 2983 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2984 | |
57ba4cb8 FM |
2985 | /* |
2986 | * Take the device list mutex to prevent races with the final phase of | |
2987 | * a device replace operation that replaces the device object associated | |
2988 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2989 | */ | |
2990 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2991 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2992 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2993 | ret = btrfs_free_dev_extent(trans, device, |
2994 | map->stripes[i].physical, | |
2995 | &dev_extent_len); | |
47ab2a6c | 2996 | if (ret) { |
57ba4cb8 | 2997 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2998 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2999 | goto out; |
3000 | } | |
a061fc8d | 3001 | |
2196d6e8 | 3002 | if (device->bytes_used > 0) { |
34441361 | 3003 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3004 | btrfs_device_set_bytes_used(device, |
3005 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3006 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3007 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3008 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3009 | } |
a061fc8d | 3010 | |
64bc6c2a NB |
3011 | ret = btrfs_update_device(trans, device); |
3012 | if (ret) { | |
3013 | mutex_unlock(&fs_devices->device_list_mutex); | |
3014 | btrfs_abort_transaction(trans, ret); | |
3015 | goto out; | |
dfe25020 | 3016 | } |
8f18cf13 | 3017 | } |
57ba4cb8 FM |
3018 | mutex_unlock(&fs_devices->device_list_mutex); |
3019 | ||
f4208794 | 3020 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3021 | if (ret) { |
66642832 | 3022 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3023 | goto out; |
3024 | } | |
8f18cf13 | 3025 | |
6bccf3ab | 3026 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3027 | |
8f18cf13 | 3028 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3029 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3030 | if (ret) { |
66642832 | 3031 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3032 | goto out; |
3033 | } | |
8f18cf13 CM |
3034 | } |
3035 | ||
5a98ec01 | 3036 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3037 | if (ret) { |
66642832 | 3038 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3039 | goto out; |
3040 | } | |
2b82032c | 3041 | |
47ab2a6c | 3042 | out: |
2b82032c YZ |
3043 | /* once for us */ |
3044 | free_extent_map(em); | |
47ab2a6c JB |
3045 | return ret; |
3046 | } | |
2b82032c | 3047 | |
5b4aacef | 3048 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3049 | { |
5b4aacef | 3050 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3051 | struct btrfs_trans_handle *trans; |
47ab2a6c | 3052 | int ret; |
2b82032c | 3053 | |
67c5e7d4 FM |
3054 | /* |
3055 | * Prevent races with automatic removal of unused block groups. | |
3056 | * After we relocate and before we remove the chunk with offset | |
3057 | * chunk_offset, automatic removal of the block group can kick in, | |
3058 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3059 | * | |
3060 | * Make sure to acquire this mutex before doing a tree search (dev | |
3061 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3062 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3063 | * we release the path used to search the chunk/dev tree and before | |
3064 | * the current task acquires this mutex and calls us. | |
3065 | */ | |
a32bf9a3 | 3066 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3067 | |
0b246afa | 3068 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
3069 | if (ret) |
3070 | return -ENOSPC; | |
3071 | ||
3072 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 3073 | btrfs_scrub_pause(fs_info); |
0b246afa | 3074 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3075 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3076 | if (ret) |
3077 | return ret; | |
3078 | ||
75cb379d JM |
3079 | /* |
3080 | * We add the kobjects here (and after forcing data chunk creation) | |
3081 | * since relocation is the only place we'll create chunks of a new | |
3082 | * type at runtime. The only place where we'll remove the last | |
3083 | * chunk of a type is the call immediately below this one. Even | |
3084 | * so, we're protected against races with the cleaner thread since | |
3085 | * we're covered by the delete_unused_bgs_mutex. | |
3086 | */ | |
3087 | btrfs_add_raid_kobjects(fs_info); | |
3088 | ||
19c4d2f9 CM |
3089 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3090 | chunk_offset); | |
3091 | if (IS_ERR(trans)) { | |
3092 | ret = PTR_ERR(trans); | |
3093 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3094 | return ret; | |
3095 | } | |
3096 | ||
47ab2a6c | 3097 | /* |
19c4d2f9 CM |
3098 | * step two, delete the device extents and the |
3099 | * chunk tree entries | |
47ab2a6c | 3100 | */ |
97aff912 | 3101 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3102 | btrfs_end_transaction(trans); |
19c4d2f9 | 3103 | return ret; |
2b82032c YZ |
3104 | } |
3105 | ||
2ff7e61e | 3106 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3107 | { |
0b246afa | 3108 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3109 | struct btrfs_path *path; |
3110 | struct extent_buffer *leaf; | |
3111 | struct btrfs_chunk *chunk; | |
3112 | struct btrfs_key key; | |
3113 | struct btrfs_key found_key; | |
2b82032c | 3114 | u64 chunk_type; |
ba1bf481 JB |
3115 | bool retried = false; |
3116 | int failed = 0; | |
2b82032c YZ |
3117 | int ret; |
3118 | ||
3119 | path = btrfs_alloc_path(); | |
3120 | if (!path) | |
3121 | return -ENOMEM; | |
3122 | ||
ba1bf481 | 3123 | again: |
2b82032c YZ |
3124 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3125 | key.offset = (u64)-1; | |
3126 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3127 | ||
3128 | while (1) { | |
0b246afa | 3129 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3130 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3131 | if (ret < 0) { |
0b246afa | 3132 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3133 | goto error; |
67c5e7d4 | 3134 | } |
79787eaa | 3135 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3136 | |
3137 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3138 | key.type); | |
67c5e7d4 | 3139 | if (ret) |
0b246afa | 3140 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3141 | if (ret < 0) |
3142 | goto error; | |
3143 | if (ret > 0) | |
3144 | break; | |
1a40e23b | 3145 | |
2b82032c YZ |
3146 | leaf = path->nodes[0]; |
3147 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3148 | |
2b82032c YZ |
3149 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3150 | struct btrfs_chunk); | |
3151 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3152 | btrfs_release_path(path); |
8f18cf13 | 3153 | |
2b82032c | 3154 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3155 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3156 | if (ret == -ENOSPC) |
3157 | failed++; | |
14586651 HS |
3158 | else |
3159 | BUG_ON(ret); | |
2b82032c | 3160 | } |
0b246afa | 3161 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3162 | |
2b82032c YZ |
3163 | if (found_key.offset == 0) |
3164 | break; | |
3165 | key.offset = found_key.offset - 1; | |
3166 | } | |
3167 | ret = 0; | |
ba1bf481 JB |
3168 | if (failed && !retried) { |
3169 | failed = 0; | |
3170 | retried = true; | |
3171 | goto again; | |
fae7f21c | 3172 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3173 | ret = -ENOSPC; |
3174 | } | |
2b82032c YZ |
3175 | error: |
3176 | btrfs_free_path(path); | |
3177 | return ret; | |
8f18cf13 CM |
3178 | } |
3179 | ||
a6f93c71 LB |
3180 | /* |
3181 | * return 1 : allocate a data chunk successfully, | |
3182 | * return <0: errors during allocating a data chunk, | |
3183 | * return 0 : no need to allocate a data chunk. | |
3184 | */ | |
3185 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3186 | u64 chunk_offset) | |
3187 | { | |
3188 | struct btrfs_block_group_cache *cache; | |
3189 | u64 bytes_used; | |
3190 | u64 chunk_type; | |
3191 | ||
3192 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3193 | ASSERT(cache); | |
3194 | chunk_type = cache->flags; | |
3195 | btrfs_put_block_group(cache); | |
3196 | ||
3197 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3198 | spin_lock(&fs_info->data_sinfo->lock); | |
3199 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3200 | spin_unlock(&fs_info->data_sinfo->lock); | |
3201 | ||
3202 | if (!bytes_used) { | |
3203 | struct btrfs_trans_handle *trans; | |
3204 | int ret; | |
3205 | ||
3206 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3207 | if (IS_ERR(trans)) | |
3208 | return PTR_ERR(trans); | |
3209 | ||
43a7e99d | 3210 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3211 | BTRFS_BLOCK_GROUP_DATA); |
3212 | btrfs_end_transaction(trans); | |
3213 | if (ret < 0) | |
3214 | return ret; | |
3215 | ||
75cb379d JM |
3216 | btrfs_add_raid_kobjects(fs_info); |
3217 | ||
a6f93c71 LB |
3218 | return 1; |
3219 | } | |
3220 | } | |
3221 | return 0; | |
3222 | } | |
3223 | ||
6bccf3ab | 3224 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3225 | struct btrfs_balance_control *bctl) |
3226 | { | |
6bccf3ab | 3227 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3228 | struct btrfs_trans_handle *trans; |
3229 | struct btrfs_balance_item *item; | |
3230 | struct btrfs_disk_balance_args disk_bargs; | |
3231 | struct btrfs_path *path; | |
3232 | struct extent_buffer *leaf; | |
3233 | struct btrfs_key key; | |
3234 | int ret, err; | |
3235 | ||
3236 | path = btrfs_alloc_path(); | |
3237 | if (!path) | |
3238 | return -ENOMEM; | |
3239 | ||
3240 | trans = btrfs_start_transaction(root, 0); | |
3241 | if (IS_ERR(trans)) { | |
3242 | btrfs_free_path(path); | |
3243 | return PTR_ERR(trans); | |
3244 | } | |
3245 | ||
3246 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3247 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3248 | key.offset = 0; |
3249 | ||
3250 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3251 | sizeof(*item)); | |
3252 | if (ret) | |
3253 | goto out; | |
3254 | ||
3255 | leaf = path->nodes[0]; | |
3256 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3257 | ||
b159fa28 | 3258 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3259 | |
3260 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3261 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3262 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3263 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3264 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3265 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3266 | ||
3267 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3268 | ||
3269 | btrfs_mark_buffer_dirty(leaf); | |
3270 | out: | |
3271 | btrfs_free_path(path); | |
3a45bb20 | 3272 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3273 | if (err && !ret) |
3274 | ret = err; | |
3275 | return ret; | |
3276 | } | |
3277 | ||
6bccf3ab | 3278 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3279 | { |
6bccf3ab | 3280 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3281 | struct btrfs_trans_handle *trans; |
3282 | struct btrfs_path *path; | |
3283 | struct btrfs_key key; | |
3284 | int ret, err; | |
3285 | ||
3286 | path = btrfs_alloc_path(); | |
3287 | if (!path) | |
3288 | return -ENOMEM; | |
3289 | ||
3290 | trans = btrfs_start_transaction(root, 0); | |
3291 | if (IS_ERR(trans)) { | |
3292 | btrfs_free_path(path); | |
3293 | return PTR_ERR(trans); | |
3294 | } | |
3295 | ||
3296 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3297 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3298 | key.offset = 0; |
3299 | ||
3300 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3301 | if (ret < 0) | |
3302 | goto out; | |
3303 | if (ret > 0) { | |
3304 | ret = -ENOENT; | |
3305 | goto out; | |
3306 | } | |
3307 | ||
3308 | ret = btrfs_del_item(trans, root, path); | |
3309 | out: | |
3310 | btrfs_free_path(path); | |
3a45bb20 | 3311 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3312 | if (err && !ret) |
3313 | ret = err; | |
3314 | return ret; | |
3315 | } | |
3316 | ||
59641015 ID |
3317 | /* |
3318 | * This is a heuristic used to reduce the number of chunks balanced on | |
3319 | * resume after balance was interrupted. | |
3320 | */ | |
3321 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3322 | { | |
3323 | /* | |
3324 | * Turn on soft mode for chunk types that were being converted. | |
3325 | */ | |
3326 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3327 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3328 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3329 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3330 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3331 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3332 | ||
3333 | /* | |
3334 | * Turn on usage filter if is not already used. The idea is | |
3335 | * that chunks that we have already balanced should be | |
3336 | * reasonably full. Don't do it for chunks that are being | |
3337 | * converted - that will keep us from relocating unconverted | |
3338 | * (albeit full) chunks. | |
3339 | */ | |
3340 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3341 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3342 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3343 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3344 | bctl->data.usage = 90; | |
3345 | } | |
3346 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3347 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3348 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3349 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3350 | bctl->sys.usage = 90; | |
3351 | } | |
3352 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3353 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3354 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3355 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3356 | bctl->meta.usage = 90; | |
3357 | } | |
3358 | } | |
3359 | ||
149196a2 DS |
3360 | /* |
3361 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3362 | */ | |
3363 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3364 | { |
3365 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3366 | int ret; |
c9e9f97b ID |
3367 | |
3368 | BUG_ON(!fs_info->balance_ctl); | |
3369 | ||
3370 | spin_lock(&fs_info->balance_lock); | |
3371 | fs_info->balance_ctl = NULL; | |
3372 | spin_unlock(&fs_info->balance_lock); | |
3373 | ||
3374 | kfree(bctl); | |
149196a2 DS |
3375 | ret = del_balance_item(fs_info); |
3376 | if (ret) | |
3377 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3378 | } |
3379 | ||
ed25e9b2 ID |
3380 | /* |
3381 | * Balance filters. Return 1 if chunk should be filtered out | |
3382 | * (should not be balanced). | |
3383 | */ | |
899c81ea | 3384 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3385 | struct btrfs_balance_args *bargs) |
3386 | { | |
899c81ea ID |
3387 | chunk_type = chunk_to_extended(chunk_type) & |
3388 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3389 | |
899c81ea | 3390 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3391 | return 0; |
3392 | ||
3393 | return 1; | |
3394 | } | |
3395 | ||
dba72cb3 | 3396 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3397 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3398 | { |
3399 | struct btrfs_block_group_cache *cache; | |
3400 | u64 chunk_used; | |
3401 | u64 user_thresh_min; | |
3402 | u64 user_thresh_max; | |
3403 | int ret = 1; | |
3404 | ||
3405 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3406 | chunk_used = btrfs_block_group_used(&cache->item); | |
3407 | ||
3408 | if (bargs->usage_min == 0) | |
3409 | user_thresh_min = 0; | |
3410 | else | |
3411 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3412 | bargs->usage_min); | |
3413 | ||
3414 | if (bargs->usage_max == 0) | |
3415 | user_thresh_max = 1; | |
3416 | else if (bargs->usage_max > 100) | |
3417 | user_thresh_max = cache->key.offset; | |
3418 | else | |
3419 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3420 | bargs->usage_max); | |
3421 | ||
3422 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3423 | ret = 0; | |
3424 | ||
3425 | btrfs_put_block_group(cache); | |
3426 | return ret; | |
3427 | } | |
3428 | ||
dba72cb3 | 3429 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3430 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3431 | { |
3432 | struct btrfs_block_group_cache *cache; | |
3433 | u64 chunk_used, user_thresh; | |
3434 | int ret = 1; | |
3435 | ||
3436 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3437 | chunk_used = btrfs_block_group_used(&cache->item); | |
3438 | ||
bc309467 | 3439 | if (bargs->usage_min == 0) |
3e39cea6 | 3440 | user_thresh = 1; |
a105bb88 ID |
3441 | else if (bargs->usage > 100) |
3442 | user_thresh = cache->key.offset; | |
3443 | else | |
3444 | user_thresh = div_factor_fine(cache->key.offset, | |
3445 | bargs->usage); | |
3446 | ||
5ce5b3c0 ID |
3447 | if (chunk_used < user_thresh) |
3448 | ret = 0; | |
3449 | ||
3450 | btrfs_put_block_group(cache); | |
3451 | return ret; | |
3452 | } | |
3453 | ||
409d404b ID |
3454 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3455 | struct btrfs_chunk *chunk, | |
3456 | struct btrfs_balance_args *bargs) | |
3457 | { | |
3458 | struct btrfs_stripe *stripe; | |
3459 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3460 | int i; | |
3461 | ||
3462 | for (i = 0; i < num_stripes; i++) { | |
3463 | stripe = btrfs_stripe_nr(chunk, i); | |
3464 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3465 | return 0; | |
3466 | } | |
3467 | ||
3468 | return 1; | |
3469 | } | |
3470 | ||
94e60d5a ID |
3471 | /* [pstart, pend) */ |
3472 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3473 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3474 | struct btrfs_balance_args *bargs) |
3475 | { | |
3476 | struct btrfs_stripe *stripe; | |
3477 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3478 | u64 stripe_offset; | |
3479 | u64 stripe_length; | |
3480 | int factor; | |
3481 | int i; | |
3482 | ||
3483 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3484 | return 0; | |
3485 | ||
3486 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3487 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3488 | factor = num_stripes / 2; | |
3489 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3490 | factor = num_stripes - 1; | |
3491 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3492 | factor = num_stripes - 2; | |
3493 | } else { | |
3494 | factor = num_stripes; | |
3495 | } | |
94e60d5a ID |
3496 | |
3497 | for (i = 0; i < num_stripes; i++) { | |
3498 | stripe = btrfs_stripe_nr(chunk, i); | |
3499 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3500 | continue; | |
3501 | ||
3502 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3503 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3504 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3505 | |
3506 | if (stripe_offset < bargs->pend && | |
3507 | stripe_offset + stripe_length > bargs->pstart) | |
3508 | return 0; | |
3509 | } | |
3510 | ||
3511 | return 1; | |
3512 | } | |
3513 | ||
ea67176a ID |
3514 | /* [vstart, vend) */ |
3515 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3516 | struct btrfs_chunk *chunk, | |
3517 | u64 chunk_offset, | |
3518 | struct btrfs_balance_args *bargs) | |
3519 | { | |
3520 | if (chunk_offset < bargs->vend && | |
3521 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3522 | /* at least part of the chunk is inside this vrange */ | |
3523 | return 0; | |
3524 | ||
3525 | return 1; | |
3526 | } | |
3527 | ||
dee32d0a GAP |
3528 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3529 | struct btrfs_chunk *chunk, | |
3530 | struct btrfs_balance_args *bargs) | |
3531 | { | |
3532 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3533 | ||
3534 | if (bargs->stripes_min <= num_stripes | |
3535 | && num_stripes <= bargs->stripes_max) | |
3536 | return 0; | |
3537 | ||
3538 | return 1; | |
3539 | } | |
3540 | ||
899c81ea | 3541 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3542 | struct btrfs_balance_args *bargs) |
3543 | { | |
3544 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3545 | return 0; | |
3546 | ||
899c81ea ID |
3547 | chunk_type = chunk_to_extended(chunk_type) & |
3548 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3549 | |
899c81ea | 3550 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3551 | return 1; |
3552 | ||
3553 | return 0; | |
3554 | } | |
3555 | ||
2ff7e61e | 3556 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3557 | struct extent_buffer *leaf, |
3558 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3559 | { | |
0b246afa | 3560 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3561 | struct btrfs_balance_args *bargs = NULL; |
3562 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3563 | ||
3564 | /* type filter */ | |
3565 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3566 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3567 | return 0; | |
3568 | } | |
3569 | ||
3570 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3571 | bargs = &bctl->data; | |
3572 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3573 | bargs = &bctl->sys; | |
3574 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3575 | bargs = &bctl->meta; | |
3576 | ||
ed25e9b2 ID |
3577 | /* profiles filter */ |
3578 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3579 | chunk_profiles_filter(chunk_type, bargs)) { | |
3580 | return 0; | |
5ce5b3c0 ID |
3581 | } |
3582 | ||
3583 | /* usage filter */ | |
3584 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3585 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3586 | return 0; |
bc309467 | 3587 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3588 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3589 | return 0; |
409d404b ID |
3590 | } |
3591 | ||
3592 | /* devid filter */ | |
3593 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3594 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3595 | return 0; | |
94e60d5a ID |
3596 | } |
3597 | ||
3598 | /* drange filter, makes sense only with devid filter */ | |
3599 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3600 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3601 | return 0; |
ea67176a ID |
3602 | } |
3603 | ||
3604 | /* vrange filter */ | |
3605 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3606 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3607 | return 0; | |
ed25e9b2 ID |
3608 | } |
3609 | ||
dee32d0a GAP |
3610 | /* stripes filter */ |
3611 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3612 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3613 | return 0; | |
3614 | } | |
3615 | ||
cfa4c961 ID |
3616 | /* soft profile changing mode */ |
3617 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3618 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3619 | return 0; | |
3620 | } | |
3621 | ||
7d824b6f DS |
3622 | /* |
3623 | * limited by count, must be the last filter | |
3624 | */ | |
3625 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3626 | if (bargs->limit == 0) | |
3627 | return 0; | |
3628 | else | |
3629 | bargs->limit--; | |
12907fc7 DS |
3630 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3631 | /* | |
3632 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3633 | * determined here because we do not have the global information |
12907fc7 DS |
3634 | * about the count of all chunks that satisfy the filters. |
3635 | */ | |
3636 | if (bargs->limit_max == 0) | |
3637 | return 0; | |
3638 | else | |
3639 | bargs->limit_max--; | |
7d824b6f DS |
3640 | } |
3641 | ||
f43ffb60 ID |
3642 | return 1; |
3643 | } | |
3644 | ||
c9e9f97b | 3645 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3646 | { |
19a39dce | 3647 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3648 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3649 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3650 | struct list_head *devices; | |
ec44a35c CM |
3651 | struct btrfs_device *device; |
3652 | u64 old_size; | |
3653 | u64 size_to_free; | |
12907fc7 | 3654 | u64 chunk_type; |
f43ffb60 | 3655 | struct btrfs_chunk *chunk; |
5a488b9d | 3656 | struct btrfs_path *path = NULL; |
ec44a35c | 3657 | struct btrfs_key key; |
ec44a35c | 3658 | struct btrfs_key found_key; |
c9e9f97b | 3659 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3660 | struct extent_buffer *leaf; |
3661 | int slot; | |
c9e9f97b ID |
3662 | int ret; |
3663 | int enospc_errors = 0; | |
19a39dce | 3664 | bool counting = true; |
12907fc7 | 3665 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3666 | u64 limit_data = bctl->data.limit; |
3667 | u64 limit_meta = bctl->meta.limit; | |
3668 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3669 | u32 count_data = 0; |
3670 | u32 count_meta = 0; | |
3671 | u32 count_sys = 0; | |
2c9fe835 | 3672 | int chunk_reserved = 0; |
ec44a35c | 3673 | |
ec44a35c | 3674 | /* step one make some room on all the devices */ |
c9e9f97b | 3675 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3676 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3677 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3678 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3679 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3680 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3681 | btrfs_device_get_total_bytes(device) - |
3682 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3683 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3684 | continue; |
3685 | ||
3686 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3687 | if (ret == -ENOSPC) |
3688 | break; | |
5a488b9d LB |
3689 | if (ret) { |
3690 | /* btrfs_shrink_device never returns ret > 0 */ | |
3691 | WARN_ON(ret > 0); | |
3692 | goto error; | |
3693 | } | |
ec44a35c | 3694 | |
a22285a6 | 3695 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3696 | if (IS_ERR(trans)) { |
3697 | ret = PTR_ERR(trans); | |
3698 | btrfs_info_in_rcu(fs_info, | |
3699 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3700 | rcu_str_deref(device->name), ret, | |
3701 | old_size, old_size - size_to_free); | |
3702 | goto error; | |
3703 | } | |
ec44a35c CM |
3704 | |
3705 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3706 | if (ret) { |
3a45bb20 | 3707 | btrfs_end_transaction(trans); |
5a488b9d LB |
3708 | /* btrfs_grow_device never returns ret > 0 */ |
3709 | WARN_ON(ret > 0); | |
3710 | btrfs_info_in_rcu(fs_info, | |
3711 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3712 | rcu_str_deref(device->name), ret, | |
3713 | old_size, old_size - size_to_free); | |
3714 | goto error; | |
3715 | } | |
ec44a35c | 3716 | |
3a45bb20 | 3717 | btrfs_end_transaction(trans); |
ec44a35c CM |
3718 | } |
3719 | ||
3720 | /* step two, relocate all the chunks */ | |
3721 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3722 | if (!path) { |
3723 | ret = -ENOMEM; | |
3724 | goto error; | |
3725 | } | |
19a39dce ID |
3726 | |
3727 | /* zero out stat counters */ | |
3728 | spin_lock(&fs_info->balance_lock); | |
3729 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3730 | spin_unlock(&fs_info->balance_lock); | |
3731 | again: | |
7d824b6f | 3732 | if (!counting) { |
12907fc7 DS |
3733 | /* |
3734 | * The single value limit and min/max limits use the same bytes | |
3735 | * in the | |
3736 | */ | |
7d824b6f DS |
3737 | bctl->data.limit = limit_data; |
3738 | bctl->meta.limit = limit_meta; | |
3739 | bctl->sys.limit = limit_sys; | |
3740 | } | |
ec44a35c CM |
3741 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3742 | key.offset = (u64)-1; | |
3743 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3744 | ||
d397712b | 3745 | while (1) { |
19a39dce | 3746 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3747 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3748 | ret = -ECANCELED; |
3749 | goto error; | |
3750 | } | |
3751 | ||
67c5e7d4 | 3752 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3753 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3754 | if (ret < 0) { |
3755 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3756 | goto error; |
67c5e7d4 | 3757 | } |
ec44a35c CM |
3758 | |
3759 | /* | |
3760 | * this shouldn't happen, it means the last relocate | |
3761 | * failed | |
3762 | */ | |
3763 | if (ret == 0) | |
c9e9f97b | 3764 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3765 | |
3766 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3767 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3768 | if (ret) { |
67c5e7d4 | 3769 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3770 | ret = 0; |
ec44a35c | 3771 | break; |
c9e9f97b | 3772 | } |
7d9eb12c | 3773 | |
f43ffb60 ID |
3774 | leaf = path->nodes[0]; |
3775 | slot = path->slots[0]; | |
3776 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3777 | |
67c5e7d4 FM |
3778 | if (found_key.objectid != key.objectid) { |
3779 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3780 | break; |
67c5e7d4 | 3781 | } |
7d9eb12c | 3782 | |
f43ffb60 | 3783 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3784 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3785 | |
19a39dce ID |
3786 | if (!counting) { |
3787 | spin_lock(&fs_info->balance_lock); | |
3788 | bctl->stat.considered++; | |
3789 | spin_unlock(&fs_info->balance_lock); | |
3790 | } | |
3791 | ||
2ff7e61e | 3792 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3793 | found_key.offset); |
2c9fe835 | 3794 | |
b3b4aa74 | 3795 | btrfs_release_path(path); |
67c5e7d4 FM |
3796 | if (!ret) { |
3797 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3798 | goto loop; |
67c5e7d4 | 3799 | } |
f43ffb60 | 3800 | |
19a39dce | 3801 | if (counting) { |
67c5e7d4 | 3802 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3803 | spin_lock(&fs_info->balance_lock); |
3804 | bctl->stat.expected++; | |
3805 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3806 | |
3807 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3808 | count_data++; | |
3809 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3810 | count_sys++; | |
3811 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3812 | count_meta++; | |
3813 | ||
3814 | goto loop; | |
3815 | } | |
3816 | ||
3817 | /* | |
3818 | * Apply limit_min filter, no need to check if the LIMITS | |
3819 | * filter is used, limit_min is 0 by default | |
3820 | */ | |
3821 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3822 | count_data < bctl->data.limit_min) | |
3823 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3824 | count_meta < bctl->meta.limit_min) | |
3825 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3826 | count_sys < bctl->sys.limit_min)) { | |
3827 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3828 | goto loop; |
3829 | } | |
3830 | ||
a6f93c71 LB |
3831 | if (!chunk_reserved) { |
3832 | /* | |
3833 | * We may be relocating the only data chunk we have, | |
3834 | * which could potentially end up with losing data's | |
3835 | * raid profile, so lets allocate an empty one in | |
3836 | * advance. | |
3837 | */ | |
3838 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3839 | found_key.offset); | |
2c9fe835 ZL |
3840 | if (ret < 0) { |
3841 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3842 | goto error; | |
a6f93c71 LB |
3843 | } else if (ret == 1) { |
3844 | chunk_reserved = 1; | |
2c9fe835 | 3845 | } |
2c9fe835 ZL |
3846 | } |
3847 | ||
5b4aacef | 3848 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3849 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3850 | if (ret == -ENOSPC) { |
c9e9f97b | 3851 | enospc_errors++; |
eede2bf3 OS |
3852 | } else if (ret == -ETXTBSY) { |
3853 | btrfs_info(fs_info, | |
3854 | "skipping relocation of block group %llu due to active swapfile", | |
3855 | found_key.offset); | |
3856 | ret = 0; | |
3857 | } else if (ret) { | |
3858 | goto error; | |
19a39dce ID |
3859 | } else { |
3860 | spin_lock(&fs_info->balance_lock); | |
3861 | bctl->stat.completed++; | |
3862 | spin_unlock(&fs_info->balance_lock); | |
3863 | } | |
f43ffb60 | 3864 | loop: |
795a3321 ID |
3865 | if (found_key.offset == 0) |
3866 | break; | |
ba1bf481 | 3867 | key.offset = found_key.offset - 1; |
ec44a35c | 3868 | } |
c9e9f97b | 3869 | |
19a39dce ID |
3870 | if (counting) { |
3871 | btrfs_release_path(path); | |
3872 | counting = false; | |
3873 | goto again; | |
3874 | } | |
ec44a35c CM |
3875 | error: |
3876 | btrfs_free_path(path); | |
c9e9f97b | 3877 | if (enospc_errors) { |
efe120a0 | 3878 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3879 | enospc_errors); |
c9e9f97b ID |
3880 | if (!ret) |
3881 | ret = -ENOSPC; | |
3882 | } | |
3883 | ||
ec44a35c CM |
3884 | return ret; |
3885 | } | |
3886 | ||
0c460c0d ID |
3887 | /** |
3888 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3889 | * @flags: profile to validate | |
3890 | * @extended: if true @flags is treated as an extended profile | |
3891 | */ | |
3892 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3893 | { | |
3894 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3895 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3896 | ||
3897 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3898 | ||
3899 | /* 1) check that all other bits are zeroed */ | |
3900 | if (flags & ~mask) | |
3901 | return 0; | |
3902 | ||
3903 | /* 2) see if profile is reduced */ | |
3904 | if (flags == 0) | |
3905 | return !extended; /* "0" is valid for usual profiles */ | |
3906 | ||
3907 | /* true if exactly one bit set */ | |
818255fe | 3908 | return is_power_of_2(flags); |
0c460c0d ID |
3909 | } |
3910 | ||
837d5b6e ID |
3911 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3912 | { | |
a7e99c69 ID |
3913 | /* cancel requested || normal exit path */ |
3914 | return atomic_read(&fs_info->balance_cancel_req) || | |
3915 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3916 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3917 | } |
3918 | ||
bdcd3c97 AM |
3919 | /* Non-zero return value signifies invalidity */ |
3920 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3921 | u64 allowed) | |
3922 | { | |
3923 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3924 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3925 | (bctl_arg->target & ~allowed))); | |
3926 | } | |
3927 | ||
c9e9f97b | 3928 | /* |
dccdb07b | 3929 | * Should be called with balance mutexe held |
c9e9f97b | 3930 | */ |
6fcf6e2b DS |
3931 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
3932 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
3933 | struct btrfs_ioctl_balance_args *bargs) |
3934 | { | |
14506127 | 3935 | u64 meta_target, data_target; |
f43ffb60 | 3936 | u64 allowed; |
e4837f8f | 3937 | int mixed = 0; |
c9e9f97b | 3938 | int ret; |
8dabb742 | 3939 | u64 num_devices; |
de98ced9 | 3940 | unsigned seq; |
c9e9f97b | 3941 | |
837d5b6e | 3942 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3943 | atomic_read(&fs_info->balance_pause_req) || |
3944 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3945 | ret = -EINVAL; |
3946 | goto out; | |
3947 | } | |
3948 | ||
e4837f8f ID |
3949 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3950 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3951 | mixed = 1; | |
3952 | ||
f43ffb60 ID |
3953 | /* |
3954 | * In case of mixed groups both data and meta should be picked, | |
3955 | * and identical options should be given for both of them. | |
3956 | */ | |
e4837f8f ID |
3957 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3958 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3959 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3960 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3961 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 3962 | btrfs_err(fs_info, |
6dac13f8 | 3963 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
3964 | ret = -EINVAL; |
3965 | goto out; | |
3966 | } | |
3967 | } | |
3968 | ||
1da73967 AJ |
3969 | num_devices = btrfs_num_devices(fs_info); |
3970 | ||
88be159c AH |
3971 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3972 | if (num_devices > 1) | |
e4d8ec0f | 3973 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3974 | if (num_devices > 2) |
3975 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3976 | if (num_devices > 3) | |
3977 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3978 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3979 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
3980 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
3981 | ||
5d163e0e | 3982 | btrfs_err(fs_info, |
6dac13f8 AJ |
3983 | "balance: invalid convert data profile %s", |
3984 | get_raid_name(index)); | |
e4d8ec0f ID |
3985 | ret = -EINVAL; |
3986 | goto out; | |
3987 | } | |
bdcd3c97 | 3988 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
3989 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
3990 | ||
efe120a0 | 3991 | btrfs_err(fs_info, |
6dac13f8 AJ |
3992 | "balance: invalid convert metadata profile %s", |
3993 | get_raid_name(index)); | |
e4d8ec0f ID |
3994 | ret = -EINVAL; |
3995 | goto out; | |
3996 | } | |
bdcd3c97 | 3997 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
3998 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
3999 | ||
efe120a0 | 4000 | btrfs_err(fs_info, |
6dac13f8 AJ |
4001 | "balance: invalid convert system profile %s", |
4002 | get_raid_name(index)); | |
e4d8ec0f ID |
4003 | ret = -EINVAL; |
4004 | goto out; | |
4005 | } | |
4006 | ||
e4d8ec0f ID |
4007 | /* allow to reduce meta or sys integrity only if force set */ |
4008 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
4009 | BTRFS_BLOCK_GROUP_RAID10 | |
4010 | BTRFS_BLOCK_GROUP_RAID5 | | |
4011 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
4012 | do { |
4013 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4014 | ||
4015 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4016 | (fs_info->avail_system_alloc_bits & allowed) && | |
4017 | !(bctl->sys.target & allowed)) || | |
4018 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4019 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
4020 | !(bctl->meta.target & allowed))) { | |
4021 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e | 4022 | btrfs_info(fs_info, |
6dac13f8 | 4023 | "balance: force reducing metadata integrity"); |
de98ced9 | 4024 | } else { |
5d163e0e | 4025 | btrfs_err(fs_info, |
6dac13f8 | 4026 | "balance: reduces metadata integrity, use --force if you want this"); |
de98ced9 MX |
4027 | ret = -EINVAL; |
4028 | goto out; | |
4029 | } | |
e4d8ec0f | 4030 | } |
de98ced9 | 4031 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4032 | |
14506127 AB |
4033 | /* if we're not converting, the target field is uninitialized */ |
4034 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4035 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4036 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4037 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
4038 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < | |
4039 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
4040 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
4041 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
4042 | ||
ee592d07 | 4043 | btrfs_warn(fs_info, |
6dac13f8 AJ |
4044 | "balance: metadata profile %s has lower redundancy than data profile %s", |
4045 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
4046 | } |
4047 | ||
6bccf3ab | 4048 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4049 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4050 | goto out; |
4051 | ||
59641015 ID |
4052 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4053 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4054 | BUG_ON(fs_info->balance_ctl); |
4055 | spin_lock(&fs_info->balance_lock); | |
4056 | fs_info->balance_ctl = bctl; | |
4057 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4058 | } else { |
4059 | BUG_ON(ret != -EEXIST); | |
4060 | spin_lock(&fs_info->balance_lock); | |
4061 | update_balance_args(bctl); | |
4062 | spin_unlock(&fs_info->balance_lock); | |
4063 | } | |
c9e9f97b | 4064 | |
3009a62f DS |
4065 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4066 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
c9e9f97b ID |
4067 | mutex_unlock(&fs_info->balance_mutex); |
4068 | ||
4069 | ret = __btrfs_balance(fs_info); | |
4070 | ||
4071 | mutex_lock(&fs_info->balance_mutex); | |
3009a62f | 4072 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4073 | |
4074 | if (bargs) { | |
4075 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4076 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4077 | } |
4078 | ||
3a01aa7a ID |
4079 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4080 | balance_need_close(fs_info)) { | |
149196a2 | 4081 | reset_balance_state(fs_info); |
a17c95df | 4082 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4083 | } |
4084 | ||
837d5b6e | 4085 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4086 | |
4087 | return ret; | |
4088 | out: | |
59641015 | 4089 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4090 | reset_balance_state(fs_info); |
a17c95df | 4091 | else |
59641015 | 4092 | kfree(bctl); |
a17c95df DS |
4093 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4094 | ||
59641015 ID |
4095 | return ret; |
4096 | } | |
4097 | ||
4098 | static int balance_kthread(void *data) | |
4099 | { | |
2b6ba629 | 4100 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4101 | int ret = 0; |
59641015 | 4102 | |
59641015 | 4103 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4104 | if (fs_info->balance_ctl) { |
6dac13f8 | 4105 | btrfs_info(fs_info, "balance: resuming"); |
6fcf6e2b | 4106 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
9555c6c1 | 4107 | } |
59641015 | 4108 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4109 | |
59641015 ID |
4110 | return ret; |
4111 | } | |
4112 | ||
2b6ba629 ID |
4113 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4114 | { | |
4115 | struct task_struct *tsk; | |
4116 | ||
1354e1a1 | 4117 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4118 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4119 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4120 | return 0; |
4121 | } | |
1354e1a1 | 4122 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4123 | |
3cdde224 | 4124 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4125 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4126 | return 0; |
4127 | } | |
4128 | ||
02ee654d AJ |
4129 | /* |
4130 | * A ro->rw remount sequence should continue with the paused balance | |
4131 | * regardless of who pauses it, system or the user as of now, so set | |
4132 | * the resume flag. | |
4133 | */ | |
4134 | spin_lock(&fs_info->balance_lock); | |
4135 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4136 | spin_unlock(&fs_info->balance_lock); | |
4137 | ||
2b6ba629 | 4138 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4139 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4140 | } |
4141 | ||
68310a5e | 4142 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4143 | { |
59641015 ID |
4144 | struct btrfs_balance_control *bctl; |
4145 | struct btrfs_balance_item *item; | |
4146 | struct btrfs_disk_balance_args disk_bargs; | |
4147 | struct btrfs_path *path; | |
4148 | struct extent_buffer *leaf; | |
4149 | struct btrfs_key key; | |
4150 | int ret; | |
4151 | ||
4152 | path = btrfs_alloc_path(); | |
4153 | if (!path) | |
4154 | return -ENOMEM; | |
4155 | ||
59641015 | 4156 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4157 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4158 | key.offset = 0; |
4159 | ||
68310a5e | 4160 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4161 | if (ret < 0) |
68310a5e | 4162 | goto out; |
59641015 ID |
4163 | if (ret > 0) { /* ret = -ENOENT; */ |
4164 | ret = 0; | |
68310a5e ID |
4165 | goto out; |
4166 | } | |
4167 | ||
4168 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4169 | if (!bctl) { | |
4170 | ret = -ENOMEM; | |
4171 | goto out; | |
59641015 ID |
4172 | } |
4173 | ||
4174 | leaf = path->nodes[0]; | |
4175 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4176 | ||
68310a5e ID |
4177 | bctl->flags = btrfs_balance_flags(leaf, item); |
4178 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4179 | |
4180 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4181 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4182 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4183 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4184 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4185 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4186 | ||
eee95e3f DS |
4187 | /* |
4188 | * This should never happen, as the paused balance state is recovered | |
4189 | * during mount without any chance of other exclusive ops to collide. | |
4190 | * | |
4191 | * This gives the exclusive op status to balance and keeps in paused | |
4192 | * state until user intervention (cancel or umount). If the ownership | |
4193 | * cannot be assigned, show a message but do not fail. The balance | |
4194 | * is in a paused state and must have fs_info::balance_ctl properly | |
4195 | * set up. | |
4196 | */ | |
4197 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4198 | btrfs_warn(fs_info, | |
6dac13f8 | 4199 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4200 | |
68310a5e | 4201 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4202 | BUG_ON(fs_info->balance_ctl); |
4203 | spin_lock(&fs_info->balance_lock); | |
4204 | fs_info->balance_ctl = bctl; | |
4205 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4206 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4207 | out: |
4208 | btrfs_free_path(path); | |
ec44a35c CM |
4209 | return ret; |
4210 | } | |
4211 | ||
837d5b6e ID |
4212 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4213 | { | |
4214 | int ret = 0; | |
4215 | ||
4216 | mutex_lock(&fs_info->balance_mutex); | |
4217 | if (!fs_info->balance_ctl) { | |
4218 | mutex_unlock(&fs_info->balance_mutex); | |
4219 | return -ENOTCONN; | |
4220 | } | |
4221 | ||
3009a62f | 4222 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4223 | atomic_inc(&fs_info->balance_pause_req); |
4224 | mutex_unlock(&fs_info->balance_mutex); | |
4225 | ||
4226 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4227 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4228 | |
4229 | mutex_lock(&fs_info->balance_mutex); | |
4230 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4231 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4232 | atomic_dec(&fs_info->balance_pause_req); |
4233 | } else { | |
4234 | ret = -ENOTCONN; | |
4235 | } | |
4236 | ||
4237 | mutex_unlock(&fs_info->balance_mutex); | |
4238 | return ret; | |
4239 | } | |
4240 | ||
a7e99c69 ID |
4241 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4242 | { | |
4243 | mutex_lock(&fs_info->balance_mutex); | |
4244 | if (!fs_info->balance_ctl) { | |
4245 | mutex_unlock(&fs_info->balance_mutex); | |
4246 | return -ENOTCONN; | |
4247 | } | |
4248 | ||
cf7d20f4 DS |
4249 | /* |
4250 | * A paused balance with the item stored on disk can be resumed at | |
4251 | * mount time if the mount is read-write. Otherwise it's still paused | |
4252 | * and we must not allow cancelling as it deletes the item. | |
4253 | */ | |
4254 | if (sb_rdonly(fs_info->sb)) { | |
4255 | mutex_unlock(&fs_info->balance_mutex); | |
4256 | return -EROFS; | |
4257 | } | |
4258 | ||
a7e99c69 ID |
4259 | atomic_inc(&fs_info->balance_cancel_req); |
4260 | /* | |
4261 | * if we are running just wait and return, balance item is | |
4262 | * deleted in btrfs_balance in this case | |
4263 | */ | |
3009a62f | 4264 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4265 | mutex_unlock(&fs_info->balance_mutex); |
4266 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4267 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4268 | mutex_lock(&fs_info->balance_mutex); |
4269 | } else { | |
a7e99c69 | 4270 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4271 | /* |
4272 | * Lock released to allow other waiters to continue, we'll | |
4273 | * reexamine the status again. | |
4274 | */ | |
a7e99c69 ID |
4275 | mutex_lock(&fs_info->balance_mutex); |
4276 | ||
a17c95df | 4277 | if (fs_info->balance_ctl) { |
149196a2 | 4278 | reset_balance_state(fs_info); |
a17c95df | 4279 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4280 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4281 | } |
a7e99c69 ID |
4282 | } |
4283 | ||
3009a62f DS |
4284 | BUG_ON(fs_info->balance_ctl || |
4285 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4286 | atomic_dec(&fs_info->balance_cancel_req); |
4287 | mutex_unlock(&fs_info->balance_mutex); | |
4288 | return 0; | |
4289 | } | |
4290 | ||
803b2f54 SB |
4291 | static int btrfs_uuid_scan_kthread(void *data) |
4292 | { | |
4293 | struct btrfs_fs_info *fs_info = data; | |
4294 | struct btrfs_root *root = fs_info->tree_root; | |
4295 | struct btrfs_key key; | |
803b2f54 SB |
4296 | struct btrfs_path *path = NULL; |
4297 | int ret = 0; | |
4298 | struct extent_buffer *eb; | |
4299 | int slot; | |
4300 | struct btrfs_root_item root_item; | |
4301 | u32 item_size; | |
f45388f3 | 4302 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4303 | |
4304 | path = btrfs_alloc_path(); | |
4305 | if (!path) { | |
4306 | ret = -ENOMEM; | |
4307 | goto out; | |
4308 | } | |
4309 | ||
4310 | key.objectid = 0; | |
4311 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4312 | key.offset = 0; | |
4313 | ||
803b2f54 | 4314 | while (1) { |
7c829b72 AJ |
4315 | ret = btrfs_search_forward(root, &key, path, |
4316 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4317 | if (ret) { |
4318 | if (ret > 0) | |
4319 | ret = 0; | |
4320 | break; | |
4321 | } | |
4322 | ||
4323 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4324 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4325 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4326 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4327 | goto skip; | |
4328 | ||
4329 | eb = path->nodes[0]; | |
4330 | slot = path->slots[0]; | |
4331 | item_size = btrfs_item_size_nr(eb, slot); | |
4332 | if (item_size < sizeof(root_item)) | |
4333 | goto skip; | |
4334 | ||
803b2f54 SB |
4335 | read_extent_buffer(eb, &root_item, |
4336 | btrfs_item_ptr_offset(eb, slot), | |
4337 | (int)sizeof(root_item)); | |
4338 | if (btrfs_root_refs(&root_item) == 0) | |
4339 | goto skip; | |
f45388f3 FDBM |
4340 | |
4341 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4342 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4343 | if (trans) | |
4344 | goto update_tree; | |
4345 | ||
4346 | btrfs_release_path(path); | |
803b2f54 SB |
4347 | /* |
4348 | * 1 - subvol uuid item | |
4349 | * 1 - received_subvol uuid item | |
4350 | */ | |
4351 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4352 | if (IS_ERR(trans)) { | |
4353 | ret = PTR_ERR(trans); | |
4354 | break; | |
4355 | } | |
f45388f3 FDBM |
4356 | continue; |
4357 | } else { | |
4358 | goto skip; | |
4359 | } | |
4360 | update_tree: | |
4361 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4362 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4363 | BTRFS_UUID_KEY_SUBVOL, |
4364 | key.objectid); | |
4365 | if (ret < 0) { | |
efe120a0 | 4366 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4367 | ret); |
803b2f54 SB |
4368 | break; |
4369 | } | |
4370 | } | |
4371 | ||
4372 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4373 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4374 | root_item.received_uuid, |
4375 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4376 | key.objectid); | |
4377 | if (ret < 0) { | |
efe120a0 | 4378 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4379 | ret); |
803b2f54 SB |
4380 | break; |
4381 | } | |
4382 | } | |
4383 | ||
f45388f3 | 4384 | skip: |
803b2f54 | 4385 | if (trans) { |
3a45bb20 | 4386 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4387 | trans = NULL; |
803b2f54 SB |
4388 | if (ret) |
4389 | break; | |
4390 | } | |
4391 | ||
803b2f54 SB |
4392 | btrfs_release_path(path); |
4393 | if (key.offset < (u64)-1) { | |
4394 | key.offset++; | |
4395 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4396 | key.offset = 0; | |
4397 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4398 | } else if (key.objectid < (u64)-1) { | |
4399 | key.offset = 0; | |
4400 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4401 | key.objectid++; | |
4402 | } else { | |
4403 | break; | |
4404 | } | |
4405 | cond_resched(); | |
4406 | } | |
4407 | ||
4408 | out: | |
4409 | btrfs_free_path(path); | |
f45388f3 | 4410 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4411 | btrfs_end_transaction(trans); |
803b2f54 | 4412 | if (ret) |
efe120a0 | 4413 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4414 | else |
afcdd129 | 4415 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4416 | up(&fs_info->uuid_tree_rescan_sem); |
4417 | return 0; | |
4418 | } | |
4419 | ||
70f80175 SB |
4420 | /* |
4421 | * Callback for btrfs_uuid_tree_iterate(). | |
4422 | * returns: | |
4423 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4424 | * < 0 if an error occurred. |
70f80175 SB |
4425 | * > 0 if the check failed, which means the caller shall remove the entry. |
4426 | */ | |
4427 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4428 | u8 *uuid, u8 type, u64 subid) | |
4429 | { | |
4430 | struct btrfs_key key; | |
4431 | int ret = 0; | |
4432 | struct btrfs_root *subvol_root; | |
4433 | ||
4434 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4435 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4436 | goto out; | |
4437 | ||
4438 | key.objectid = subid; | |
4439 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4440 | key.offset = (u64)-1; | |
4441 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4442 | if (IS_ERR(subvol_root)) { | |
4443 | ret = PTR_ERR(subvol_root); | |
4444 | if (ret == -ENOENT) | |
4445 | ret = 1; | |
4446 | goto out; | |
4447 | } | |
4448 | ||
4449 | switch (type) { | |
4450 | case BTRFS_UUID_KEY_SUBVOL: | |
4451 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4452 | ret = 1; | |
4453 | break; | |
4454 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4455 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4456 | BTRFS_UUID_SIZE)) | |
4457 | ret = 1; | |
4458 | break; | |
4459 | } | |
4460 | ||
4461 | out: | |
4462 | return ret; | |
4463 | } | |
4464 | ||
4465 | static int btrfs_uuid_rescan_kthread(void *data) | |
4466 | { | |
4467 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4468 | int ret; | |
4469 | ||
4470 | /* | |
4471 | * 1st step is to iterate through the existing UUID tree and | |
4472 | * to delete all entries that contain outdated data. | |
4473 | * 2nd step is to add all missing entries to the UUID tree. | |
4474 | */ | |
4475 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4476 | if (ret < 0) { | |
efe120a0 | 4477 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4478 | up(&fs_info->uuid_tree_rescan_sem); |
4479 | return ret; | |
4480 | } | |
4481 | return btrfs_uuid_scan_kthread(data); | |
4482 | } | |
4483 | ||
f7a81ea4 SB |
4484 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4485 | { | |
4486 | struct btrfs_trans_handle *trans; | |
4487 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4488 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4489 | struct task_struct *task; |
4490 | int ret; | |
f7a81ea4 SB |
4491 | |
4492 | /* | |
4493 | * 1 - root node | |
4494 | * 1 - root item | |
4495 | */ | |
4496 | trans = btrfs_start_transaction(tree_root, 2); | |
4497 | if (IS_ERR(trans)) | |
4498 | return PTR_ERR(trans); | |
4499 | ||
4500 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4501 | BTRFS_UUID_TREE_OBJECTID); | |
4502 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4503 | ret = PTR_ERR(uuid_root); |
66642832 | 4504 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4505 | btrfs_end_transaction(trans); |
6d13f549 | 4506 | return ret; |
f7a81ea4 SB |
4507 | } |
4508 | ||
4509 | fs_info->uuid_root = uuid_root; | |
4510 | ||
3a45bb20 | 4511 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4512 | if (ret) |
4513 | return ret; | |
4514 | ||
4515 | down(&fs_info->uuid_tree_rescan_sem); | |
4516 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4517 | if (IS_ERR(task)) { | |
70f80175 | 4518 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4519 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4520 | up(&fs_info->uuid_tree_rescan_sem); |
4521 | return PTR_ERR(task); | |
4522 | } | |
4523 | ||
4524 | return 0; | |
f7a81ea4 | 4525 | } |
803b2f54 | 4526 | |
70f80175 SB |
4527 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4528 | { | |
4529 | struct task_struct *task; | |
4530 | ||
4531 | down(&fs_info->uuid_tree_rescan_sem); | |
4532 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4533 | if (IS_ERR(task)) { | |
4534 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4535 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4536 | up(&fs_info->uuid_tree_rescan_sem); |
4537 | return PTR_ERR(task); | |
4538 | } | |
4539 | ||
4540 | return 0; | |
4541 | } | |
4542 | ||
8f18cf13 CM |
4543 | /* |
4544 | * shrinking a device means finding all of the device extents past | |
4545 | * the new size, and then following the back refs to the chunks. | |
4546 | * The chunk relocation code actually frees the device extent | |
4547 | */ | |
4548 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4549 | { | |
0b246afa JM |
4550 | struct btrfs_fs_info *fs_info = device->fs_info; |
4551 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4552 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4553 | struct btrfs_dev_extent *dev_extent = NULL; |
4554 | struct btrfs_path *path; | |
4555 | u64 length; | |
8f18cf13 CM |
4556 | u64 chunk_offset; |
4557 | int ret; | |
4558 | int slot; | |
ba1bf481 JB |
4559 | int failed = 0; |
4560 | bool retried = false; | |
53e489bc | 4561 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4562 | struct extent_buffer *l; |
4563 | struct btrfs_key key; | |
0b246afa | 4564 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4565 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4566 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4567 | u64 diff; |
4568 | ||
4569 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4570 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4571 | |
401e29c1 | 4572 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4573 | return -EINVAL; |
4574 | ||
8f18cf13 CM |
4575 | path = btrfs_alloc_path(); |
4576 | if (!path) | |
4577 | return -ENOMEM; | |
4578 | ||
0338dff6 | 4579 | path->reada = READA_BACK; |
8f18cf13 | 4580 | |
34441361 | 4581 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4582 | |
7cc8e58d | 4583 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4584 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4585 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4586 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4587 | } |
34441361 | 4588 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4589 | |
ba1bf481 | 4590 | again: |
8f18cf13 CM |
4591 | key.objectid = device->devid; |
4592 | key.offset = (u64)-1; | |
4593 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4594 | ||
213e64da | 4595 | do { |
0b246afa | 4596 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4597 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4598 | if (ret < 0) { |
0b246afa | 4599 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4600 | goto done; |
67c5e7d4 | 4601 | } |
8f18cf13 CM |
4602 | |
4603 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4604 | if (ret) |
0b246afa | 4605 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4606 | if (ret < 0) |
4607 | goto done; | |
4608 | if (ret) { | |
4609 | ret = 0; | |
b3b4aa74 | 4610 | btrfs_release_path(path); |
bf1fb512 | 4611 | break; |
8f18cf13 CM |
4612 | } |
4613 | ||
4614 | l = path->nodes[0]; | |
4615 | slot = path->slots[0]; | |
4616 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4617 | ||
ba1bf481 | 4618 | if (key.objectid != device->devid) { |
0b246afa | 4619 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4620 | btrfs_release_path(path); |
bf1fb512 | 4621 | break; |
ba1bf481 | 4622 | } |
8f18cf13 CM |
4623 | |
4624 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4625 | length = btrfs_dev_extent_length(l, dev_extent); | |
4626 | ||
ba1bf481 | 4627 | if (key.offset + length <= new_size) { |
0b246afa | 4628 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4629 | btrfs_release_path(path); |
d6397bae | 4630 | break; |
ba1bf481 | 4631 | } |
8f18cf13 | 4632 | |
8f18cf13 | 4633 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4634 | btrfs_release_path(path); |
8f18cf13 | 4635 | |
a6f93c71 LB |
4636 | /* |
4637 | * We may be relocating the only data chunk we have, | |
4638 | * which could potentially end up with losing data's | |
4639 | * raid profile, so lets allocate an empty one in | |
4640 | * advance. | |
4641 | */ | |
4642 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4643 | if (ret < 0) { | |
4644 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4645 | goto done; | |
4646 | } | |
4647 | ||
0b246afa JM |
4648 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4649 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4650 | if (ret == -ENOSPC) { |
ba1bf481 | 4651 | failed++; |
eede2bf3 OS |
4652 | } else if (ret) { |
4653 | if (ret == -ETXTBSY) { | |
4654 | btrfs_warn(fs_info, | |
4655 | "could not shrink block group %llu due to active swapfile", | |
4656 | chunk_offset); | |
4657 | } | |
4658 | goto done; | |
4659 | } | |
213e64da | 4660 | } while (key.offset-- > 0); |
ba1bf481 JB |
4661 | |
4662 | if (failed && !retried) { | |
4663 | failed = 0; | |
4664 | retried = true; | |
4665 | goto again; | |
4666 | } else if (failed && retried) { | |
4667 | ret = -ENOSPC; | |
ba1bf481 | 4668 | goto done; |
8f18cf13 CM |
4669 | } |
4670 | ||
d6397bae | 4671 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4672 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4673 | if (IS_ERR(trans)) { |
4674 | ret = PTR_ERR(trans); | |
4675 | goto done; | |
4676 | } | |
4677 | ||
34441361 | 4678 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4679 | |
4680 | /* | |
4681 | * We checked in the above loop all device extents that were already in | |
4682 | * the device tree. However before we have updated the device's | |
4683 | * total_bytes to the new size, we might have had chunk allocations that | |
4684 | * have not complete yet (new block groups attached to transaction | |
4685 | * handles), and therefore their device extents were not yet in the | |
4686 | * device tree and we missed them in the loop above. So if we have any | |
4687 | * pending chunk using a device extent that overlaps the device range | |
4688 | * that we can not use anymore, commit the current transaction and | |
4689 | * repeat the search on the device tree - this way we guarantee we will | |
4690 | * not have chunks using device extents that end beyond 'new_size'. | |
4691 | */ | |
4692 | if (!checked_pending_chunks) { | |
4693 | u64 start = new_size; | |
4694 | u64 len = old_size - new_size; | |
4695 | ||
499f377f JM |
4696 | if (contains_pending_extent(trans->transaction, device, |
4697 | &start, len)) { | |
34441361 | 4698 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4699 | checked_pending_chunks = true; |
4700 | failed = 0; | |
4701 | retried = false; | |
3a45bb20 | 4702 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4703 | if (ret) |
4704 | goto done; | |
4705 | goto again; | |
4706 | } | |
4707 | } | |
4708 | ||
7cc8e58d | 4709 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4710 | if (list_empty(&device->resized_list)) |
4711 | list_add_tail(&device->resized_list, | |
0b246afa | 4712 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4713 | |
d6397bae | 4714 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4715 | btrfs_set_super_total_bytes(super_copy, |
4716 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4717 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4718 | |
4719 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4720 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4721 | if (ret < 0) { |
4722 | btrfs_abort_transaction(trans, ret); | |
4723 | btrfs_end_transaction(trans); | |
4724 | } else { | |
4725 | ret = btrfs_commit_transaction(trans); | |
4726 | } | |
8f18cf13 CM |
4727 | done: |
4728 | btrfs_free_path(path); | |
53e489bc | 4729 | if (ret) { |
34441361 | 4730 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4731 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4732 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4733 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4734 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4735 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4736 | } |
8f18cf13 CM |
4737 | return ret; |
4738 | } | |
4739 | ||
2ff7e61e | 4740 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4741 | struct btrfs_key *key, |
4742 | struct btrfs_chunk *chunk, int item_size) | |
4743 | { | |
0b246afa | 4744 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4745 | struct btrfs_disk_key disk_key; |
4746 | u32 array_size; | |
4747 | u8 *ptr; | |
4748 | ||
34441361 | 4749 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4750 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4751 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4752 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4753 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4754 | return -EFBIG; |
fe48a5c0 | 4755 | } |
0b86a832 CM |
4756 | |
4757 | ptr = super_copy->sys_chunk_array + array_size; | |
4758 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4759 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4760 | ptr += sizeof(disk_key); | |
4761 | memcpy(ptr, chunk, item_size); | |
4762 | item_size += sizeof(disk_key); | |
4763 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4764 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4765 | |
0b86a832 CM |
4766 | return 0; |
4767 | } | |
4768 | ||
73c5de00 AJ |
4769 | /* |
4770 | * sort the devices in descending order by max_avail, total_avail | |
4771 | */ | |
4772 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4773 | { |
73c5de00 AJ |
4774 | const struct btrfs_device_info *di_a = a; |
4775 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4776 | |
73c5de00 | 4777 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4778 | return -1; |
73c5de00 | 4779 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4780 | return 1; |
73c5de00 AJ |
4781 | if (di_a->total_avail > di_b->total_avail) |
4782 | return -1; | |
4783 | if (di_a->total_avail < di_b->total_avail) | |
4784 | return 1; | |
4785 | return 0; | |
b2117a39 | 4786 | } |
0b86a832 | 4787 | |
53b381b3 DW |
4788 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4789 | { | |
ffe2d203 | 4790 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4791 | return; |
4792 | ||
ceda0864 | 4793 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4794 | } |
4795 | ||
062d4d1f | 4796 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
4797 | - sizeof(struct btrfs_chunk)) \ |
4798 | / sizeof(struct btrfs_stripe) + 1) | |
4799 | ||
4800 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4801 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4802 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4803 | / sizeof(struct btrfs_stripe) + 1) | |
4804 | ||
73c5de00 | 4805 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4806 | u64 start, u64 type) |
b2117a39 | 4807 | { |
2ff7e61e | 4808 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4809 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4810 | struct btrfs_device *device; |
73c5de00 AJ |
4811 | struct map_lookup *map = NULL; |
4812 | struct extent_map_tree *em_tree; | |
4813 | struct extent_map *em; | |
4814 | struct btrfs_device_info *devices_info = NULL; | |
4815 | u64 total_avail; | |
4816 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4817 | int data_stripes; /* number of stripes that count for |
4818 | block group size */ | |
73c5de00 AJ |
4819 | int sub_stripes; /* sub_stripes info for map */ |
4820 | int dev_stripes; /* stripes per dev */ | |
4821 | int devs_max; /* max devs to use */ | |
4822 | int devs_min; /* min devs needed */ | |
4823 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4824 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4825 | int nparity; /* number of stripes worth of bytes to |
4826 | store parity information */ | |
73c5de00 AJ |
4827 | int ret; |
4828 | u64 max_stripe_size; | |
4829 | u64 max_chunk_size; | |
4830 | u64 stripe_size; | |
23f0ff1e | 4831 | u64 chunk_size; |
73c5de00 AJ |
4832 | int ndevs; |
4833 | int i; | |
4834 | int j; | |
31e50229 | 4835 | int index; |
593060d7 | 4836 | |
0c460c0d | 4837 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4838 | |
4117f207 QW |
4839 | if (list_empty(&fs_devices->alloc_list)) { |
4840 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4841 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4842 | return -ENOSPC; |
4117f207 | 4843 | } |
b2117a39 | 4844 | |
3e72ee88 | 4845 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4846 | |
31e50229 LB |
4847 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4848 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4849 | devs_max = btrfs_raid_array[index].devs_max; | |
4850 | devs_min = btrfs_raid_array[index].devs_min; | |
4851 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4852 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 4853 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 4854 | |
9b3f68b9 | 4855 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4856 | max_stripe_size = SZ_1G; |
fce466ea | 4857 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 4858 | if (!devs_max) |
062d4d1f | 4859 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 4860 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4861 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4862 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4863 | max_stripe_size = SZ_1G; | |
1100373f | 4864 | else |
ee22184b | 4865 | max_stripe_size = SZ_256M; |
73c5de00 | 4866 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 4867 | if (!devs_max) |
062d4d1f | 4868 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 4869 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4870 | max_stripe_size = SZ_32M; |
73c5de00 | 4871 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4872 | if (!devs_max) |
4873 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4874 | } else { |
351fd353 | 4875 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4876 | type); |
4877 | BUG_ON(1); | |
9b3f68b9 CM |
4878 | } |
4879 | ||
2b82032c YZ |
4880 | /* we don't want a chunk larger than 10% of writeable space */ |
4881 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4882 | max_chunk_size); | |
9b3f68b9 | 4883 | |
31e818fe | 4884 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4885 | GFP_NOFS); |
4886 | if (!devices_info) | |
4887 | return -ENOMEM; | |
0cad8a11 | 4888 | |
9f680ce0 | 4889 | /* |
73c5de00 AJ |
4890 | * in the first pass through the devices list, we gather information |
4891 | * about the available holes on each device. | |
9f680ce0 | 4892 | */ |
73c5de00 | 4893 | ndevs = 0; |
ebcc9301 | 4894 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4895 | u64 max_avail; |
4896 | u64 dev_offset; | |
b2117a39 | 4897 | |
ebbede42 | 4898 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4899 | WARN(1, KERN_ERR |
efe120a0 | 4900 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4901 | continue; |
4902 | } | |
b2117a39 | 4903 | |
e12c9621 AJ |
4904 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4905 | &device->dev_state) || | |
401e29c1 | 4906 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4907 | continue; |
b2117a39 | 4908 | |
73c5de00 AJ |
4909 | if (device->total_bytes > device->bytes_used) |
4910 | total_avail = device->total_bytes - device->bytes_used; | |
4911 | else | |
4912 | total_avail = 0; | |
38c01b96 | 4913 | |
4914 | /* If there is no space on this device, skip it. */ | |
4915 | if (total_avail == 0) | |
4916 | continue; | |
b2117a39 | 4917 | |
6df9a95e | 4918 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4919 | max_stripe_size * dev_stripes, |
4920 | &dev_offset, &max_avail); | |
4921 | if (ret && ret != -ENOSPC) | |
4922 | goto error; | |
b2117a39 | 4923 | |
73c5de00 AJ |
4924 | if (ret == 0) |
4925 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4926 | |
4117f207 QW |
4927 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4928 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4929 | btrfs_debug(info, | |
4930 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4931 | __func__, device->devid, max_avail, | |
4932 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4933 | continue; |
4117f207 | 4934 | } |
b2117a39 | 4935 | |
063d006f ES |
4936 | if (ndevs == fs_devices->rw_devices) { |
4937 | WARN(1, "%s: found more than %llu devices\n", | |
4938 | __func__, fs_devices->rw_devices); | |
4939 | break; | |
4940 | } | |
73c5de00 AJ |
4941 | devices_info[ndevs].dev_offset = dev_offset; |
4942 | devices_info[ndevs].max_avail = max_avail; | |
4943 | devices_info[ndevs].total_avail = total_avail; | |
4944 | devices_info[ndevs].dev = device; | |
4945 | ++ndevs; | |
4946 | } | |
b2117a39 | 4947 | |
73c5de00 AJ |
4948 | /* |
4949 | * now sort the devices by hole size / available space | |
4950 | */ | |
4951 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4952 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4953 | |
73c5de00 | 4954 | /* round down to number of usable stripes */ |
e5600fd6 | 4955 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 4956 | |
ba89b802 | 4957 | if (ndevs < devs_min) { |
73c5de00 | 4958 | ret = -ENOSPC; |
4117f207 QW |
4959 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4960 | btrfs_debug(info, | |
4961 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 4962 | __func__, ndevs, devs_min); |
4117f207 | 4963 | } |
73c5de00 | 4964 | goto error; |
b2117a39 | 4965 | } |
9f680ce0 | 4966 | |
f148ef4d NB |
4967 | ndevs = min(ndevs, devs_max); |
4968 | ||
73c5de00 | 4969 | /* |
92e222df HK |
4970 | * The primary goal is to maximize the number of stripes, so use as |
4971 | * many devices as possible, even if the stripes are not maximum sized. | |
4972 | * | |
4973 | * The DUP profile stores more than one stripe per device, the | |
4974 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4975 | */ |
92e222df | 4976 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4977 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4978 | |
53b381b3 DW |
4979 | /* |
4980 | * this will have to be fixed for RAID1 and RAID10 over | |
4981 | * more drives | |
4982 | */ | |
b50836ed | 4983 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
4984 | |
4985 | /* | |
4986 | * Use the number of data stripes to figure out how big this chunk | |
4987 | * is really going to be in terms of logical address space, | |
baf92114 HK |
4988 | * and compare that answer with the max chunk size. If it's higher, |
4989 | * we try to reduce stripe_size. | |
86db2578 CM |
4990 | */ |
4991 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 4992 | /* |
baf92114 HK |
4993 | * Reduce stripe_size, round it up to a 16MB boundary again and |
4994 | * then use it, unless it ends up being even bigger than the | |
4995 | * previous value we had already. | |
86db2578 | 4996 | */ |
baf92114 HK |
4997 | stripe_size = min(round_up(div_u64(max_chunk_size, |
4998 | data_stripes), SZ_16M), | |
793ff2c8 | 4999 | stripe_size); |
86db2578 CM |
5000 | } |
5001 | ||
37db63a4 | 5002 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 5003 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
5004 | |
5005 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
5006 | if (!map) { | |
5007 | ret = -ENOMEM; | |
5008 | goto error; | |
5009 | } | |
5010 | map->num_stripes = num_stripes; | |
9b3f68b9 | 5011 | |
73c5de00 AJ |
5012 | for (i = 0; i < ndevs; ++i) { |
5013 | for (j = 0; j < dev_stripes; ++j) { | |
5014 | int s = i * dev_stripes + j; | |
5015 | map->stripes[s].dev = devices_info[i].dev; | |
5016 | map->stripes[s].physical = devices_info[i].dev_offset + | |
5017 | j * stripe_size; | |
6324fbf3 | 5018 | } |
6324fbf3 | 5019 | } |
500ceed8 NB |
5020 | map->stripe_len = BTRFS_STRIPE_LEN; |
5021 | map->io_align = BTRFS_STRIPE_LEN; | |
5022 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5023 | map->type = type; |
2b82032c | 5024 | map->sub_stripes = sub_stripes; |
0b86a832 | 5025 | |
23f0ff1e | 5026 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5027 | |
23f0ff1e | 5028 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5029 | |
172ddd60 | 5030 | em = alloc_extent_map(); |
2b82032c | 5031 | if (!em) { |
298a8f9c | 5032 | kfree(map); |
b2117a39 MX |
5033 | ret = -ENOMEM; |
5034 | goto error; | |
593060d7 | 5035 | } |
298a8f9c | 5036 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5037 | em->map_lookup = map; |
2b82032c | 5038 | em->start = start; |
23f0ff1e | 5039 | em->len = chunk_size; |
2b82032c YZ |
5040 | em->block_start = 0; |
5041 | em->block_len = em->len; | |
6df9a95e | 5042 | em->orig_block_len = stripe_size; |
593060d7 | 5043 | |
0b246afa | 5044 | em_tree = &info->mapping_tree.map_tree; |
890871be | 5045 | write_lock(&em_tree->lock); |
09a2a8f9 | 5046 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5047 | if (ret) { |
1efb72a3 | 5048 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5049 | free_extent_map(em); |
1dd4602f | 5050 | goto error; |
0f5d42b2 | 5051 | } |
0b86a832 | 5052 | |
1efb72a3 NB |
5053 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
5054 | refcount_inc(&em->refs); | |
5055 | write_unlock(&em_tree->lock); | |
5056 | ||
23f0ff1e | 5057 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5058 | if (ret) |
5059 | goto error_del_extent; | |
2b82032c | 5060 | |
2f29df4f HK |
5061 | for (i = 0; i < map->num_stripes; i++) |
5062 | btrfs_device_set_bytes_used(map->stripes[i].dev, | |
5063 | map->stripes[i].dev->bytes_used + stripe_size); | |
43530c46 | 5064 | |
a5ed45f8 | 5065 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5066 | |
0f5d42b2 | 5067 | free_extent_map(em); |
0b246afa | 5068 | check_raid56_incompat_flag(info, type); |
53b381b3 | 5069 | |
b2117a39 | 5070 | kfree(devices_info); |
2b82032c | 5071 | return 0; |
b2117a39 | 5072 | |
6df9a95e | 5073 | error_del_extent: |
0f5d42b2 JB |
5074 | write_lock(&em_tree->lock); |
5075 | remove_extent_mapping(em_tree, em); | |
5076 | write_unlock(&em_tree->lock); | |
5077 | ||
5078 | /* One for our allocation */ | |
5079 | free_extent_map(em); | |
5080 | /* One for the tree reference */ | |
5081 | free_extent_map(em); | |
495e64f4 FM |
5082 | /* One for the pending_chunks list reference */ |
5083 | free_extent_map(em); | |
b2117a39 | 5084 | error: |
b2117a39 MX |
5085 | kfree(devices_info); |
5086 | return ret; | |
2b82032c YZ |
5087 | } |
5088 | ||
6df9a95e | 5089 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5090 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5091 | { |
97aff912 | 5092 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5093 | struct btrfs_root *extent_root = fs_info->extent_root; |
5094 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5095 | struct btrfs_key key; |
2b82032c YZ |
5096 | struct btrfs_device *device; |
5097 | struct btrfs_chunk *chunk; | |
5098 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5099 | struct extent_map *em; |
5100 | struct map_lookup *map; | |
5101 | size_t item_size; | |
5102 | u64 dev_offset; | |
5103 | u64 stripe_size; | |
5104 | int i = 0; | |
140e639f | 5105 | int ret = 0; |
2b82032c | 5106 | |
60ca842e | 5107 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5108 | if (IS_ERR(em)) |
5109 | return PTR_ERR(em); | |
6df9a95e | 5110 | |
95617d69 | 5111 | map = em->map_lookup; |
6df9a95e JB |
5112 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5113 | stripe_size = em->orig_block_len; | |
5114 | ||
2b82032c | 5115 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5116 | if (!chunk) { |
5117 | ret = -ENOMEM; | |
5118 | goto out; | |
5119 | } | |
5120 | ||
50460e37 FM |
5121 | /* |
5122 | * Take the device list mutex to prevent races with the final phase of | |
5123 | * a device replace operation that replaces the device object associated | |
5124 | * with the map's stripes, because the device object's id can change | |
5125 | * at any time during that final phase of the device replace operation | |
5126 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5127 | */ | |
0b246afa | 5128 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5129 | for (i = 0; i < map->num_stripes; i++) { |
5130 | device = map->stripes[i].dev; | |
5131 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5132 | |
0b86a832 | 5133 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5134 | if (ret) |
50460e37 | 5135 | break; |
b5d9071c NB |
5136 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5137 | dev_offset, stripe_size); | |
6df9a95e | 5138 | if (ret) |
50460e37 FM |
5139 | break; |
5140 | } | |
5141 | if (ret) { | |
0b246afa | 5142 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5143 | goto out; |
2b82032c YZ |
5144 | } |
5145 | ||
2b82032c | 5146 | stripe = &chunk->stripe; |
6df9a95e JB |
5147 | for (i = 0; i < map->num_stripes; i++) { |
5148 | device = map->stripes[i].dev; | |
5149 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5150 | |
e17cade2 CM |
5151 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5152 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5153 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5154 | stripe++; |
0b86a832 | 5155 | } |
0b246afa | 5156 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5157 | |
2b82032c | 5158 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5159 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5160 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5161 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5162 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5163 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5164 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5165 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5166 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5167 | |
2b82032c YZ |
5168 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5169 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5170 | key.offset = chunk_offset; | |
0b86a832 | 5171 | |
2b82032c | 5172 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5173 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5174 | /* | |
5175 | * TODO: Cleanup of inserted chunk root in case of | |
5176 | * failure. | |
5177 | */ | |
2ff7e61e | 5178 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5179 | } |
1abe9b8a | 5180 | |
6df9a95e | 5181 | out: |
0b86a832 | 5182 | kfree(chunk); |
6df9a95e | 5183 | free_extent_map(em); |
4ed1d16e | 5184 | return ret; |
2b82032c | 5185 | } |
0b86a832 | 5186 | |
2b82032c YZ |
5187 | /* |
5188 | * Chunk allocation falls into two parts. The first part does works | |
5189 | * that make the new allocated chunk useable, but not do any operation | |
5190 | * that modifies the chunk tree. The second part does the works that | |
5191 | * require modifying the chunk tree. This division is important for the | |
5192 | * bootstrap process of adding storage to a seed btrfs. | |
5193 | */ | |
c216b203 | 5194 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5195 | { |
5196 | u64 chunk_offset; | |
2b82032c | 5197 | |
c216b203 NB |
5198 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5199 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5200 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5201 | } |
5202 | ||
d397712b | 5203 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5204 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5205 | { |
5206 | u64 chunk_offset; | |
5207 | u64 sys_chunk_offset; | |
2b82032c | 5208 | u64 alloc_profile; |
2b82032c YZ |
5209 | int ret; |
5210 | ||
6df9a95e | 5211 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5212 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5213 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5214 | if (ret) |
5215 | return ret; | |
2b82032c | 5216 | |
0b246afa | 5217 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5218 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5219 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5220 | return ret; |
2b82032c YZ |
5221 | } |
5222 | ||
d20983b4 MX |
5223 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5224 | { | |
5225 | int max_errors; | |
5226 | ||
5227 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5228 | BTRFS_BLOCK_GROUP_RAID10 | | |
5229 | BTRFS_BLOCK_GROUP_RAID5 | | |
5230 | BTRFS_BLOCK_GROUP_DUP)) { | |
5231 | max_errors = 1; | |
5232 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5233 | max_errors = 2; | |
5234 | } else { | |
5235 | max_errors = 0; | |
005d6427 | 5236 | } |
2b82032c | 5237 | |
d20983b4 | 5238 | return max_errors; |
2b82032c YZ |
5239 | } |
5240 | ||
2ff7e61e | 5241 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5242 | { |
5243 | struct extent_map *em; | |
5244 | struct map_lookup *map; | |
2b82032c | 5245 | int readonly = 0; |
d20983b4 | 5246 | int miss_ndevs = 0; |
2b82032c YZ |
5247 | int i; |
5248 | ||
60ca842e | 5249 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5250 | if (IS_ERR(em)) |
2b82032c YZ |
5251 | return 1; |
5252 | ||
95617d69 | 5253 | map = em->map_lookup; |
2b82032c | 5254 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5255 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5256 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5257 | miss_ndevs++; |
5258 | continue; | |
5259 | } | |
ebbede42 AJ |
5260 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5261 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5262 | readonly = 1; |
d20983b4 | 5263 | goto end; |
2b82032c YZ |
5264 | } |
5265 | } | |
d20983b4 MX |
5266 | |
5267 | /* | |
5268 | * If the number of missing devices is larger than max errors, | |
5269 | * we can not write the data into that chunk successfully, so | |
5270 | * set it readonly. | |
5271 | */ | |
5272 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5273 | readonly = 1; | |
5274 | end: | |
0b86a832 | 5275 | free_extent_map(em); |
2b82032c | 5276 | return readonly; |
0b86a832 CM |
5277 | } |
5278 | ||
5279 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5280 | { | |
a8067e02 | 5281 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5282 | } |
5283 | ||
5284 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5285 | { | |
5286 | struct extent_map *em; | |
5287 | ||
d397712b | 5288 | while (1) { |
890871be | 5289 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5290 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5291 | if (em) | |
5292 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5293 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5294 | if (!em) |
5295 | break; | |
0b86a832 CM |
5296 | /* once for us */ |
5297 | free_extent_map(em); | |
5298 | /* once for the tree */ | |
5299 | free_extent_map(em); | |
5300 | } | |
5301 | } | |
5302 | ||
5d964051 | 5303 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5304 | { |
5305 | struct extent_map *em; | |
5306 | struct map_lookup *map; | |
f188591e CM |
5307 | int ret; |
5308 | ||
60ca842e | 5309 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5310 | if (IS_ERR(em)) |
5311 | /* | |
5312 | * We could return errors for these cases, but that could get | |
5313 | * ugly and we'd probably do the same thing which is just not do | |
5314 | * anything else and exit, so return 1 so the callers don't try | |
5315 | * to use other copies. | |
5316 | */ | |
fb7669b5 | 5317 | return 1; |
fb7669b5 | 5318 | |
95617d69 | 5319 | map = em->map_lookup; |
f188591e CM |
5320 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5321 | ret = map->num_stripes; | |
321aecc6 CM |
5322 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5323 | ret = map->sub_stripes; | |
53b381b3 DW |
5324 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5325 | ret = 2; | |
5326 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5327 | /* |
5328 | * There could be two corrupted data stripes, we need | |
5329 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5330 | * |
8810f751 LB |
5331 | * Fail a stripe at a time on every retry except the |
5332 | * stripe under reconstruction. | |
5333 | */ | |
5334 | ret = map->num_stripes; | |
f188591e CM |
5335 | else |
5336 | ret = 1; | |
5337 | free_extent_map(em); | |
ad6d620e | 5338 | |
7e79cb86 | 5339 | btrfs_dev_replace_read_lock(&fs_info->dev_replace); |
6fad823f LB |
5340 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5341 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5342 | ret++; |
7e79cb86 | 5343 | btrfs_dev_replace_read_unlock(&fs_info->dev_replace); |
ad6d620e | 5344 | |
f188591e CM |
5345 | return ret; |
5346 | } | |
5347 | ||
2ff7e61e | 5348 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5349 | u64 logical) |
5350 | { | |
5351 | struct extent_map *em; | |
5352 | struct map_lookup *map; | |
0b246afa | 5353 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5354 | |
60ca842e | 5355 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5356 | |
69f03f13 NB |
5357 | if (!WARN_ON(IS_ERR(em))) { |
5358 | map = em->map_lookup; | |
5359 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5360 | len = map->stripe_len * nr_data_stripes(map); | |
5361 | free_extent_map(em); | |
5362 | } | |
53b381b3 DW |
5363 | return len; |
5364 | } | |
5365 | ||
e4ff5fb5 | 5366 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5367 | { |
5368 | struct extent_map *em; | |
5369 | struct map_lookup *map; | |
53b381b3 DW |
5370 | int ret = 0; |
5371 | ||
60ca842e | 5372 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5373 | |
69f03f13 NB |
5374 | if(!WARN_ON(IS_ERR(em))) { |
5375 | map = em->map_lookup; | |
5376 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5377 | ret = 1; | |
5378 | free_extent_map(em); | |
5379 | } | |
53b381b3 DW |
5380 | return ret; |
5381 | } | |
5382 | ||
30d9861f | 5383 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5384 | struct map_lookup *map, int first, |
8ba0ae78 | 5385 | int dev_replace_is_ongoing) |
dfe25020 CM |
5386 | { |
5387 | int i; | |
99f92a7c | 5388 | int num_stripes; |
8ba0ae78 | 5389 | int preferred_mirror; |
30d9861f SB |
5390 | int tolerance; |
5391 | struct btrfs_device *srcdev; | |
5392 | ||
99f92a7c AJ |
5393 | ASSERT((map->type & |
5394 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5395 | ||
5396 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5397 | num_stripes = map->sub_stripes; | |
5398 | else | |
5399 | num_stripes = map->num_stripes; | |
5400 | ||
8ba0ae78 AJ |
5401 | preferred_mirror = first + current->pid % num_stripes; |
5402 | ||
30d9861f SB |
5403 | if (dev_replace_is_ongoing && |
5404 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5405 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5406 | srcdev = fs_info->dev_replace.srcdev; | |
5407 | else | |
5408 | srcdev = NULL; | |
5409 | ||
5410 | /* | |
5411 | * try to avoid the drive that is the source drive for a | |
5412 | * dev-replace procedure, only choose it if no other non-missing | |
5413 | * mirror is available | |
5414 | */ | |
5415 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5416 | if (map->stripes[preferred_mirror].dev->bdev && |
5417 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5418 | return preferred_mirror; | |
99f92a7c | 5419 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5420 | if (map->stripes[i].dev->bdev && |
5421 | (tolerance || map->stripes[i].dev != srcdev)) | |
5422 | return i; | |
5423 | } | |
dfe25020 | 5424 | } |
30d9861f | 5425 | |
dfe25020 CM |
5426 | /* we couldn't find one that doesn't fail. Just return something |
5427 | * and the io error handling code will clean up eventually | |
5428 | */ | |
8ba0ae78 | 5429 | return preferred_mirror; |
dfe25020 CM |
5430 | } |
5431 | ||
53b381b3 DW |
5432 | static inline int parity_smaller(u64 a, u64 b) |
5433 | { | |
5434 | return a > b; | |
5435 | } | |
5436 | ||
5437 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5438 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5439 | { |
5440 | struct btrfs_bio_stripe s; | |
5441 | int i; | |
5442 | u64 l; | |
5443 | int again = 1; | |
5444 | ||
5445 | while (again) { | |
5446 | again = 0; | |
cc7539ed | 5447 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5448 | if (parity_smaller(bbio->raid_map[i], |
5449 | bbio->raid_map[i+1])) { | |
53b381b3 | 5450 | s = bbio->stripes[i]; |
8e5cfb55 | 5451 | l = bbio->raid_map[i]; |
53b381b3 | 5452 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5453 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5454 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5455 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5456 | |
53b381b3 DW |
5457 | again = 1; |
5458 | } | |
5459 | } | |
5460 | } | |
5461 | } | |
5462 | ||
6e9606d2 ZL |
5463 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5464 | { | |
5465 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5466 | /* the size of the btrfs_bio */ |
6e9606d2 | 5467 | sizeof(struct btrfs_bio) + |
e57cf21e | 5468 | /* plus the variable array for the stripes */ |
6e9606d2 | 5469 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5470 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5471 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5472 | /* |
5473 | * plus the raid_map, which includes both the tgt dev | |
5474 | * and the stripes | |
5475 | */ | |
5476 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5477 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5478 | |
5479 | atomic_set(&bbio->error, 0); | |
140475ae | 5480 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5481 | |
5482 | return bbio; | |
5483 | } | |
5484 | ||
5485 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5486 | { | |
140475ae ER |
5487 | WARN_ON(!refcount_read(&bbio->refs)); |
5488 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5489 | } |
5490 | ||
5491 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5492 | { | |
5493 | if (!bbio) | |
5494 | return; | |
140475ae | 5495 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5496 | kfree(bbio); |
5497 | } | |
5498 | ||
0b3d4cd3 LB |
5499 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5500 | /* | |
5501 | * Please note that, discard won't be sent to target device of device | |
5502 | * replace. | |
5503 | */ | |
5504 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5505 | u64 logical, u64 length, | |
5506 | struct btrfs_bio **bbio_ret) | |
5507 | { | |
5508 | struct extent_map *em; | |
5509 | struct map_lookup *map; | |
5510 | struct btrfs_bio *bbio; | |
5511 | u64 offset; | |
5512 | u64 stripe_nr; | |
5513 | u64 stripe_nr_end; | |
5514 | u64 stripe_end_offset; | |
5515 | u64 stripe_cnt; | |
5516 | u64 stripe_len; | |
5517 | u64 stripe_offset; | |
5518 | u64 num_stripes; | |
5519 | u32 stripe_index; | |
5520 | u32 factor = 0; | |
5521 | u32 sub_stripes = 0; | |
5522 | u64 stripes_per_dev = 0; | |
5523 | u32 remaining_stripes = 0; | |
5524 | u32 last_stripe = 0; | |
5525 | int ret = 0; | |
5526 | int i; | |
5527 | ||
5528 | /* discard always return a bbio */ | |
5529 | ASSERT(bbio_ret); | |
5530 | ||
60ca842e | 5531 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5532 | if (IS_ERR(em)) |
5533 | return PTR_ERR(em); | |
5534 | ||
5535 | map = em->map_lookup; | |
5536 | /* we don't discard raid56 yet */ | |
5537 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5538 | ret = -EOPNOTSUPP; | |
5539 | goto out; | |
5540 | } | |
5541 | ||
5542 | offset = logical - em->start; | |
5543 | length = min_t(u64, em->len - offset, length); | |
5544 | ||
5545 | stripe_len = map->stripe_len; | |
5546 | /* | |
5547 | * stripe_nr counts the total number of stripes we have to stride | |
5548 | * to get to this block | |
5549 | */ | |
5550 | stripe_nr = div64_u64(offset, stripe_len); | |
5551 | ||
5552 | /* stripe_offset is the offset of this block in its stripe */ | |
5553 | stripe_offset = offset - stripe_nr * stripe_len; | |
5554 | ||
5555 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5556 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5557 | stripe_cnt = stripe_nr_end - stripe_nr; |
5558 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5559 | (offset + length); | |
5560 | /* | |
5561 | * after this, stripe_nr is the number of stripes on this | |
5562 | * device we have to walk to find the data, and stripe_index is | |
5563 | * the number of our device in the stripe array | |
5564 | */ | |
5565 | num_stripes = 1; | |
5566 | stripe_index = 0; | |
5567 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5568 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5569 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5570 | sub_stripes = 1; | |
5571 | else | |
5572 | sub_stripes = map->sub_stripes; | |
5573 | ||
5574 | factor = map->num_stripes / sub_stripes; | |
5575 | num_stripes = min_t(u64, map->num_stripes, | |
5576 | sub_stripes * stripe_cnt); | |
5577 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5578 | stripe_index *= sub_stripes; | |
5579 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5580 | &remaining_stripes); | |
5581 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5582 | last_stripe *= sub_stripes; | |
5583 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5584 | BTRFS_BLOCK_GROUP_DUP)) { | |
5585 | num_stripes = map->num_stripes; | |
5586 | } else { | |
5587 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5588 | &stripe_index); | |
5589 | } | |
5590 | ||
5591 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5592 | if (!bbio) { | |
5593 | ret = -ENOMEM; | |
5594 | goto out; | |
5595 | } | |
5596 | ||
5597 | for (i = 0; i < num_stripes; i++) { | |
5598 | bbio->stripes[i].physical = | |
5599 | map->stripes[stripe_index].physical + | |
5600 | stripe_offset + stripe_nr * map->stripe_len; | |
5601 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5602 | ||
5603 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5604 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5605 | bbio->stripes[i].length = stripes_per_dev * | |
5606 | map->stripe_len; | |
5607 | ||
5608 | if (i / sub_stripes < remaining_stripes) | |
5609 | bbio->stripes[i].length += | |
5610 | map->stripe_len; | |
5611 | ||
5612 | /* | |
5613 | * Special for the first stripe and | |
5614 | * the last stripe: | |
5615 | * | |
5616 | * |-------|...|-------| | |
5617 | * |----------| | |
5618 | * off end_off | |
5619 | */ | |
5620 | if (i < sub_stripes) | |
5621 | bbio->stripes[i].length -= | |
5622 | stripe_offset; | |
5623 | ||
5624 | if (stripe_index >= last_stripe && | |
5625 | stripe_index <= (last_stripe + | |
5626 | sub_stripes - 1)) | |
5627 | bbio->stripes[i].length -= | |
5628 | stripe_end_offset; | |
5629 | ||
5630 | if (i == sub_stripes - 1) | |
5631 | stripe_offset = 0; | |
5632 | } else { | |
5633 | bbio->stripes[i].length = length; | |
5634 | } | |
5635 | ||
5636 | stripe_index++; | |
5637 | if (stripe_index == map->num_stripes) { | |
5638 | stripe_index = 0; | |
5639 | stripe_nr++; | |
5640 | } | |
5641 | } | |
5642 | ||
5643 | *bbio_ret = bbio; | |
5644 | bbio->map_type = map->type; | |
5645 | bbio->num_stripes = num_stripes; | |
5646 | out: | |
5647 | free_extent_map(em); | |
5648 | return ret; | |
5649 | } | |
5650 | ||
5ab56090 LB |
5651 | /* |
5652 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5653 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5654 | * left cursor can also be read from the target drive. | |
5655 | * | |
5656 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5657 | * array of stripes. | |
5658 | * For READ, it also needs to be supported using the same mirror number. | |
5659 | * | |
5660 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5661 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5662 | * case. | |
5663 | */ | |
5664 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5665 | u64 logical, u64 length, | |
5666 | u64 srcdev_devid, int *mirror_num, | |
5667 | u64 *physical) | |
5668 | { | |
5669 | struct btrfs_bio *bbio = NULL; | |
5670 | int num_stripes; | |
5671 | int index_srcdev = 0; | |
5672 | int found = 0; | |
5673 | u64 physical_of_found = 0; | |
5674 | int i; | |
5675 | int ret = 0; | |
5676 | ||
5677 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5678 | logical, &length, &bbio, 0, 0); | |
5679 | if (ret) { | |
5680 | ASSERT(bbio == NULL); | |
5681 | return ret; | |
5682 | } | |
5683 | ||
5684 | num_stripes = bbio->num_stripes; | |
5685 | if (*mirror_num > num_stripes) { | |
5686 | /* | |
5687 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5688 | * that means that the requested area is not left of the left | |
5689 | * cursor | |
5690 | */ | |
5691 | btrfs_put_bbio(bbio); | |
5692 | return -EIO; | |
5693 | } | |
5694 | ||
5695 | /* | |
5696 | * process the rest of the function using the mirror_num of the source | |
5697 | * drive. Therefore look it up first. At the end, patch the device | |
5698 | * pointer to the one of the target drive. | |
5699 | */ | |
5700 | for (i = 0; i < num_stripes; i++) { | |
5701 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5702 | continue; | |
5703 | ||
5704 | /* | |
5705 | * In case of DUP, in order to keep it simple, only add the | |
5706 | * mirror with the lowest physical address | |
5707 | */ | |
5708 | if (found && | |
5709 | physical_of_found <= bbio->stripes[i].physical) | |
5710 | continue; | |
5711 | ||
5712 | index_srcdev = i; | |
5713 | found = 1; | |
5714 | physical_of_found = bbio->stripes[i].physical; | |
5715 | } | |
5716 | ||
5717 | btrfs_put_bbio(bbio); | |
5718 | ||
5719 | ASSERT(found); | |
5720 | if (!found) | |
5721 | return -EIO; | |
5722 | ||
5723 | *mirror_num = index_srcdev + 1; | |
5724 | *physical = physical_of_found; | |
5725 | return ret; | |
5726 | } | |
5727 | ||
73c0f228 LB |
5728 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5729 | struct btrfs_bio **bbio_ret, | |
5730 | struct btrfs_dev_replace *dev_replace, | |
5731 | int *num_stripes_ret, int *max_errors_ret) | |
5732 | { | |
5733 | struct btrfs_bio *bbio = *bbio_ret; | |
5734 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5735 | int tgtdev_indexes = 0; | |
5736 | int num_stripes = *num_stripes_ret; | |
5737 | int max_errors = *max_errors_ret; | |
5738 | int i; | |
5739 | ||
5740 | if (op == BTRFS_MAP_WRITE) { | |
5741 | int index_where_to_add; | |
5742 | ||
5743 | /* | |
5744 | * duplicate the write operations while the dev replace | |
5745 | * procedure is running. Since the copying of the old disk to | |
5746 | * the new disk takes place at run time while the filesystem is | |
5747 | * mounted writable, the regular write operations to the old | |
5748 | * disk have to be duplicated to go to the new disk as well. | |
5749 | * | |
5750 | * Note that device->missing is handled by the caller, and that | |
5751 | * the write to the old disk is already set up in the stripes | |
5752 | * array. | |
5753 | */ | |
5754 | index_where_to_add = num_stripes; | |
5755 | for (i = 0; i < num_stripes; i++) { | |
5756 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5757 | /* write to new disk, too */ | |
5758 | struct btrfs_bio_stripe *new = | |
5759 | bbio->stripes + index_where_to_add; | |
5760 | struct btrfs_bio_stripe *old = | |
5761 | bbio->stripes + i; | |
5762 | ||
5763 | new->physical = old->physical; | |
5764 | new->length = old->length; | |
5765 | new->dev = dev_replace->tgtdev; | |
5766 | bbio->tgtdev_map[i] = index_where_to_add; | |
5767 | index_where_to_add++; | |
5768 | max_errors++; | |
5769 | tgtdev_indexes++; | |
5770 | } | |
5771 | } | |
5772 | num_stripes = index_where_to_add; | |
5773 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5774 | int index_srcdev = 0; | |
5775 | int found = 0; | |
5776 | u64 physical_of_found = 0; | |
5777 | ||
5778 | /* | |
5779 | * During the dev-replace procedure, the target drive can also | |
5780 | * be used to read data in case it is needed to repair a corrupt | |
5781 | * block elsewhere. This is possible if the requested area is | |
5782 | * left of the left cursor. In this area, the target drive is a | |
5783 | * full copy of the source drive. | |
5784 | */ | |
5785 | for (i = 0; i < num_stripes; i++) { | |
5786 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5787 | /* | |
5788 | * In case of DUP, in order to keep it simple, | |
5789 | * only add the mirror with the lowest physical | |
5790 | * address | |
5791 | */ | |
5792 | if (found && | |
5793 | physical_of_found <= | |
5794 | bbio->stripes[i].physical) | |
5795 | continue; | |
5796 | index_srcdev = i; | |
5797 | found = 1; | |
5798 | physical_of_found = bbio->stripes[i].physical; | |
5799 | } | |
5800 | } | |
5801 | if (found) { | |
5802 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5803 | bbio->stripes + num_stripes; | |
5804 | ||
5805 | tgtdev_stripe->physical = physical_of_found; | |
5806 | tgtdev_stripe->length = | |
5807 | bbio->stripes[index_srcdev].length; | |
5808 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5809 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5810 | ||
5811 | tgtdev_indexes++; | |
5812 | num_stripes++; | |
5813 | } | |
5814 | } | |
5815 | ||
5816 | *num_stripes_ret = num_stripes; | |
5817 | *max_errors_ret = max_errors; | |
5818 | bbio->num_tgtdevs = tgtdev_indexes; | |
5819 | *bbio_ret = bbio; | |
5820 | } | |
5821 | ||
2b19a1fe LB |
5822 | static bool need_full_stripe(enum btrfs_map_op op) |
5823 | { | |
5824 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5825 | } | |
5826 | ||
cf8cddd3 CH |
5827 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5828 | enum btrfs_map_op op, | |
f2d8d74d | 5829 | u64 logical, u64 *length, |
a1d3c478 | 5830 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5831 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5832 | { |
5833 | struct extent_map *em; | |
5834 | struct map_lookup *map; | |
0b86a832 | 5835 | u64 offset; |
593060d7 CM |
5836 | u64 stripe_offset; |
5837 | u64 stripe_nr; | |
53b381b3 | 5838 | u64 stripe_len; |
9d644a62 | 5839 | u32 stripe_index; |
cea9e445 | 5840 | int i; |
de11cc12 | 5841 | int ret = 0; |
f2d8d74d | 5842 | int num_stripes; |
a236aed1 | 5843 | int max_errors = 0; |
2c8cdd6e | 5844 | int tgtdev_indexes = 0; |
a1d3c478 | 5845 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5846 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5847 | int dev_replace_is_ongoing = 0; | |
5848 | int num_alloc_stripes; | |
ad6d620e SB |
5849 | int patch_the_first_stripe_for_dev_replace = 0; |
5850 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5851 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5852 | |
0b3d4cd3 LB |
5853 | if (op == BTRFS_MAP_DISCARD) |
5854 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5855 | *length, bbio_ret); | |
5856 | ||
60ca842e | 5857 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
592d92ee LB |
5858 | if (IS_ERR(em)) |
5859 | return PTR_ERR(em); | |
0b86a832 | 5860 | |
95617d69 | 5861 | map = em->map_lookup; |
0b86a832 | 5862 | offset = logical - em->start; |
593060d7 | 5863 | |
53b381b3 | 5864 | stripe_len = map->stripe_len; |
593060d7 CM |
5865 | stripe_nr = offset; |
5866 | /* | |
5867 | * stripe_nr counts the total number of stripes we have to stride | |
5868 | * to get to this block | |
5869 | */ | |
47c5713f | 5870 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5871 | |
53b381b3 | 5872 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5873 | if (offset < stripe_offset) { |
5d163e0e JM |
5874 | btrfs_crit(fs_info, |
5875 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5876 | stripe_offset, offset, em->start, logical, |
5877 | stripe_len); | |
5878 | free_extent_map(em); | |
5879 | return -EINVAL; | |
5880 | } | |
593060d7 CM |
5881 | |
5882 | /* stripe_offset is the offset of this block in its stripe*/ | |
5883 | stripe_offset = offset - stripe_offset; | |
5884 | ||
53b381b3 | 5885 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5886 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5887 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5888 | raid56_full_stripe_start = offset; | |
5889 | ||
5890 | /* allow a write of a full stripe, but make sure we don't | |
5891 | * allow straddling of stripes | |
5892 | */ | |
47c5713f DS |
5893 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5894 | full_stripe_len); | |
53b381b3 DW |
5895 | raid56_full_stripe_start *= full_stripe_len; |
5896 | } | |
5897 | ||
0b3d4cd3 | 5898 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
5899 | u64 max_len; |
5900 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5901 | For other RAID types and for RAID[56] reads, just allow a single | |
5902 | stripe (on a single disk). */ | |
ffe2d203 | 5903 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5904 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5905 | max_len = stripe_len * nr_data_stripes(map) - |
5906 | (offset - raid56_full_stripe_start); | |
5907 | } else { | |
5908 | /* we limit the length of each bio to what fits in a stripe */ | |
5909 | max_len = stripe_len - stripe_offset; | |
5910 | } | |
5911 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5912 | } else { |
5913 | *length = em->len - offset; | |
5914 | } | |
f2d8d74d | 5915 | |
53b381b3 DW |
5916 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5917 | it cares about is the length */ | |
a1d3c478 | 5918 | if (!bbio_ret) |
cea9e445 CM |
5919 | goto out; |
5920 | ||
7e79cb86 | 5921 | btrfs_dev_replace_read_lock(dev_replace); |
472262f3 SB |
5922 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5923 | if (!dev_replace_is_ongoing) | |
7e79cb86 | 5924 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 LB |
5925 | else |
5926 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5927 | |
ad6d620e | 5928 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5929 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5930 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5931 | dev_replace->srcdev->devid, | |
5932 | &mirror_num, | |
5933 | &physical_to_patch_in_first_stripe); | |
5934 | if (ret) | |
ad6d620e | 5935 | goto out; |
5ab56090 LB |
5936 | else |
5937 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5938 | } else if (mirror_num > map->num_stripes) { |
5939 | mirror_num = 0; | |
5940 | } | |
5941 | ||
f2d8d74d | 5942 | num_stripes = 1; |
cea9e445 | 5943 | stripe_index = 0; |
fce3bb9a | 5944 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5945 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5946 | &stripe_index); | |
de483734 | 5947 | if (!need_full_stripe(op)) |
28e1cc7d | 5948 | mirror_num = 1; |
fce3bb9a | 5949 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 5950 | if (need_full_stripe(op)) |
f2d8d74d | 5951 | num_stripes = map->num_stripes; |
2fff734f | 5952 | else if (mirror_num) |
f188591e | 5953 | stripe_index = mirror_num - 1; |
dfe25020 | 5954 | else { |
30d9861f | 5955 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 5956 | dev_replace_is_ongoing); |
a1d3c478 | 5957 | mirror_num = stripe_index + 1; |
dfe25020 | 5958 | } |
2fff734f | 5959 | |
611f0e00 | 5960 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5961 | if (need_full_stripe(op)) { |
f2d8d74d | 5962 | num_stripes = map->num_stripes; |
a1d3c478 | 5963 | } else if (mirror_num) { |
f188591e | 5964 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5965 | } else { |
5966 | mirror_num = 1; | |
5967 | } | |
2fff734f | 5968 | |
321aecc6 | 5969 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5970 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5971 | |
47c5713f | 5972 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5973 | stripe_index *= map->sub_stripes; |
5974 | ||
de483734 | 5975 | if (need_full_stripe(op)) |
f2d8d74d | 5976 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5977 | else if (mirror_num) |
5978 | stripe_index += mirror_num - 1; | |
dfe25020 | 5979 | else { |
3e74317a | 5980 | int old_stripe_index = stripe_index; |
30d9861f SB |
5981 | stripe_index = find_live_mirror(fs_info, map, |
5982 | stripe_index, | |
30d9861f | 5983 | dev_replace_is_ongoing); |
3e74317a | 5984 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5985 | } |
53b381b3 | 5986 | |
ffe2d203 | 5987 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5988 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5989 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5990 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 5991 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
5992 | |
5993 | /* RAID[56] write or recovery. Return all stripes */ | |
5994 | num_stripes = map->num_stripes; | |
5995 | max_errors = nr_parity_stripes(map); | |
5996 | ||
53b381b3 DW |
5997 | *length = map->stripe_len; |
5998 | stripe_index = 0; | |
5999 | stripe_offset = 0; | |
6000 | } else { | |
6001 | /* | |
6002 | * Mirror #0 or #1 means the original data block. | |
6003 | * Mirror #2 is RAID5 parity block. | |
6004 | * Mirror #3 is RAID6 Q block. | |
6005 | */ | |
47c5713f DS |
6006 | stripe_nr = div_u64_rem(stripe_nr, |
6007 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
6008 | if (mirror_num > 1) |
6009 | stripe_index = nr_data_stripes(map) + | |
6010 | mirror_num - 2; | |
6011 | ||
6012 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6013 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6014 | &stripe_index); | |
de483734 | 6015 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6016 | mirror_num = 1; |
53b381b3 | 6017 | } |
8790d502 CM |
6018 | } else { |
6019 | /* | |
47c5713f DS |
6020 | * after this, stripe_nr is the number of stripes on this |
6021 | * device we have to walk to find the data, and stripe_index is | |
6022 | * the number of our device in the stripe array | |
8790d502 | 6023 | */ |
47c5713f DS |
6024 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6025 | &stripe_index); | |
a1d3c478 | 6026 | mirror_num = stripe_index + 1; |
8790d502 | 6027 | } |
e042d1ec | 6028 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6029 | btrfs_crit(fs_info, |
6030 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6031 | stripe_index, map->num_stripes); |
6032 | ret = -EINVAL; | |
6033 | goto out; | |
6034 | } | |
cea9e445 | 6035 | |
472262f3 | 6036 | num_alloc_stripes = num_stripes; |
6fad823f | 6037 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6038 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6039 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6040 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6041 | num_alloc_stripes++; |
2c8cdd6e | 6042 | tgtdev_indexes = num_stripes; |
ad6d620e | 6043 | } |
2c8cdd6e | 6044 | |
6e9606d2 | 6045 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6046 | if (!bbio) { |
6047 | ret = -ENOMEM; | |
6048 | goto out; | |
6049 | } | |
6fad823f | 6050 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6051 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6052 | |
8e5cfb55 | 6053 | /* build raid_map */ |
2b19a1fe LB |
6054 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6055 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6056 | u64 tmp; |
9d644a62 | 6057 | unsigned rot; |
8e5cfb55 ZL |
6058 | |
6059 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6060 | sizeof(struct btrfs_bio_stripe) * | |
6061 | num_alloc_stripes + | |
6062 | sizeof(int) * tgtdev_indexes); | |
6063 | ||
6064 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6065 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6066 | |
6067 | /* Fill in the logical address of each stripe */ | |
6068 | tmp = stripe_nr * nr_data_stripes(map); | |
6069 | for (i = 0; i < nr_data_stripes(map); i++) | |
6070 | bbio->raid_map[(i+rot) % num_stripes] = | |
6071 | em->start + (tmp + i) * map->stripe_len; | |
6072 | ||
6073 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6074 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6075 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6076 | RAID6_Q_STRIPE; | |
6077 | } | |
6078 | ||
b89203f7 | 6079 | |
0b3d4cd3 LB |
6080 | for (i = 0; i < num_stripes; i++) { |
6081 | bbio->stripes[i].physical = | |
6082 | map->stripes[stripe_index].physical + | |
6083 | stripe_offset + | |
6084 | stripe_nr * map->stripe_len; | |
6085 | bbio->stripes[i].dev = | |
6086 | map->stripes[stripe_index].dev; | |
6087 | stripe_index++; | |
593060d7 | 6088 | } |
de11cc12 | 6089 | |
2b19a1fe | 6090 | if (need_full_stripe(op)) |
d20983b4 | 6091 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6092 | |
8e5cfb55 ZL |
6093 | if (bbio->raid_map) |
6094 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6095 | |
73c0f228 | 6096 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6097 | need_full_stripe(op)) { |
73c0f228 LB |
6098 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6099 | &max_errors); | |
472262f3 SB |
6100 | } |
6101 | ||
de11cc12 | 6102 | *bbio_ret = bbio; |
10f11900 | 6103 | bbio->map_type = map->type; |
de11cc12 LZ |
6104 | bbio->num_stripes = num_stripes; |
6105 | bbio->max_errors = max_errors; | |
6106 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6107 | |
6108 | /* | |
6109 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6110 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6111 | * available as a mirror | |
6112 | */ | |
6113 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6114 | WARN_ON(num_stripes > 1); | |
6115 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6116 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6117 | bbio->mirror_num = map->num_stripes + 1; | |
6118 | } | |
cea9e445 | 6119 | out: |
73beece9 | 6120 | if (dev_replace_is_ongoing) { |
7fb2eced DS |
6121 | ASSERT(atomic_read(&dev_replace->blocking_readers) > 0); |
6122 | btrfs_dev_replace_read_lock(dev_replace); | |
6123 | /* Barrier implied by atomic_dec_and_test */ | |
6124 | if (atomic_dec_and_test(&dev_replace->blocking_readers)) | |
6125 | cond_wake_up_nomb(&dev_replace->read_lock_wq); | |
7e79cb86 | 6126 | btrfs_dev_replace_read_unlock(dev_replace); |
73beece9 | 6127 | } |
0b86a832 | 6128 | free_extent_map(em); |
de11cc12 | 6129 | return ret; |
0b86a832 CM |
6130 | } |
6131 | ||
cf8cddd3 | 6132 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6133 | u64 logical, u64 *length, |
a1d3c478 | 6134 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6135 | { |
b3d3fa51 | 6136 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6137 | mirror_num, 0); |
f2d8d74d CM |
6138 | } |
6139 | ||
af8e2d1d | 6140 | /* For Scrub/replace */ |
cf8cddd3 | 6141 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6142 | u64 logical, u64 *length, |
825ad4c9 | 6143 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6144 | { |
825ad4c9 | 6145 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6146 | } |
6147 | ||
63a9c7b9 NB |
6148 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6149 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6150 | { |
a512bbf8 YZ |
6151 | struct extent_map *em; |
6152 | struct map_lookup *map; | |
6153 | u64 *buf; | |
6154 | u64 bytenr; | |
6155 | u64 length; | |
6156 | u64 stripe_nr; | |
53b381b3 | 6157 | u64 rmap_len; |
a512bbf8 YZ |
6158 | int i, j, nr = 0; |
6159 | ||
60ca842e | 6160 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6161 | if (IS_ERR(em)) |
835d974f | 6162 | return -EIO; |
835d974f | 6163 | |
95617d69 | 6164 | map = em->map_lookup; |
a512bbf8 | 6165 | length = em->len; |
53b381b3 DW |
6166 | rmap_len = map->stripe_len; |
6167 | ||
a512bbf8 | 6168 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6169 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6170 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6171 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6172 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6173 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6174 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6175 | } | |
a512bbf8 | 6176 | |
31e818fe | 6177 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6178 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6179 | |
6180 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6181 | if (map->stripes[i].physical > physical || |
6182 | map->stripes[i].physical + length <= physical) | |
6183 | continue; | |
6184 | ||
6185 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6186 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6187 | |
6188 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6189 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6190 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6191 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6192 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6193 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6194 | * Alternatively, just use rmap_len below instead of | |
6195 | * map->stripe_len */ | |
6196 | ||
6197 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6198 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6199 | for (j = 0; j < nr; j++) { |
6200 | if (buf[j] == bytenr) | |
6201 | break; | |
6202 | } | |
934d375b CM |
6203 | if (j == nr) { |
6204 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6205 | buf[nr++] = bytenr; |
934d375b | 6206 | } |
a512bbf8 YZ |
6207 | } |
6208 | ||
a512bbf8 YZ |
6209 | *logical = buf; |
6210 | *naddrs = nr; | |
53b381b3 | 6211 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6212 | |
6213 | free_extent_map(em); | |
6214 | return 0; | |
f2d8d74d CM |
6215 | } |
6216 | ||
4246a0b6 | 6217 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6218 | { |
326e1dbb MS |
6219 | bio->bi_private = bbio->private; |
6220 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6221 | bio_endio(bio); |
326e1dbb | 6222 | |
6e9606d2 | 6223 | btrfs_put_bbio(bbio); |
8408c716 MX |
6224 | } |
6225 | ||
4246a0b6 | 6226 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6227 | { |
9be3395b | 6228 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6229 | int is_orig_bio = 0; |
8790d502 | 6230 | |
4e4cbee9 | 6231 | if (bio->bi_status) { |
a1d3c478 | 6232 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6233 | if (bio->bi_status == BLK_STS_IOERR || |
6234 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6235 | unsigned int stripe_index = |
9be3395b | 6236 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6237 | struct btrfs_device *dev; |
442a4f63 SB |
6238 | |
6239 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6240 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6241 | if (dev->bdev) { |
37226b21 | 6242 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6243 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6244 | BTRFS_DEV_STAT_WRITE_ERRS); |
6245 | else | |
1cb34c8e | 6246 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6247 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6248 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6249 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6250 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6251 | } |
442a4f63 SB |
6252 | } |
6253 | } | |
8790d502 | 6254 | |
a1d3c478 | 6255 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6256 | is_orig_bio = 1; |
6257 | ||
c404e0dc MX |
6258 | btrfs_bio_counter_dec(bbio->fs_info); |
6259 | ||
a1d3c478 | 6260 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6261 | if (!is_orig_bio) { |
6262 | bio_put(bio); | |
a1d3c478 | 6263 | bio = bbio->orig_bio; |
7d2b4daa | 6264 | } |
c7b22bb1 | 6265 | |
9be3395b | 6266 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6267 | /* only send an error to the higher layers if it is |
53b381b3 | 6268 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6269 | */ |
a1d3c478 | 6270 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6271 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6272 | } else { |
1259ab75 CM |
6273 | /* |
6274 | * this bio is actually up to date, we didn't | |
6275 | * go over the max number of errors | |
6276 | */ | |
2dbe0c77 | 6277 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6278 | } |
c55f1396 | 6279 | |
4246a0b6 | 6280 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6281 | } else if (!is_orig_bio) { |
8790d502 CM |
6282 | bio_put(bio); |
6283 | } | |
8790d502 CM |
6284 | } |
6285 | ||
8b712842 CM |
6286 | /* |
6287 | * see run_scheduled_bios for a description of why bios are collected for | |
6288 | * async submit. | |
6289 | * | |
6290 | * This will add one bio to the pending list for a device and make sure | |
6291 | * the work struct is scheduled. | |
6292 | */ | |
2ff7e61e | 6293 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6294 | struct bio *bio) |
8b712842 | 6295 | { |
0b246afa | 6296 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6297 | int should_queue = 1; |
ffbd517d | 6298 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6299 | |
6300 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6301 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6302 | btrfsic_submit_bio(bio); |
143bede5 | 6303 | return; |
8b712842 CM |
6304 | } |
6305 | ||
492bb6de | 6306 | WARN_ON(bio->bi_next); |
8b712842 | 6307 | bio->bi_next = NULL; |
8b712842 CM |
6308 | |
6309 | spin_lock(&device->io_lock); | |
67f055c7 | 6310 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6311 | pending_bios = &device->pending_sync_bios; |
6312 | else | |
6313 | pending_bios = &device->pending_bios; | |
8b712842 | 6314 | |
ffbd517d CM |
6315 | if (pending_bios->tail) |
6316 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6317 | |
ffbd517d CM |
6318 | pending_bios->tail = bio; |
6319 | if (!pending_bios->head) | |
6320 | pending_bios->head = bio; | |
8b712842 CM |
6321 | if (device->running_pending) |
6322 | should_queue = 0; | |
6323 | ||
6324 | spin_unlock(&device->io_lock); | |
6325 | ||
6326 | if (should_queue) | |
0b246afa | 6327 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6328 | } |
6329 | ||
2ff7e61e JM |
6330 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6331 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6332 | { |
6333 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6334 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6335 | |
6336 | bio->bi_private = bbio; | |
9be3395b | 6337 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6338 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6339 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6340 | btrfs_debug_in_rcu(fs_info, |
6341 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6342 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6343 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6344 | bio->bi_iter.bi_size); | |
74d46992 | 6345 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6346 | |
2ff7e61e | 6347 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6348 | |
de1ee92a | 6349 | if (async) |
2ff7e61e | 6350 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6351 | else |
4e49ea4a | 6352 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6353 | } |
6354 | ||
de1ee92a JB |
6355 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6356 | { | |
6357 | atomic_inc(&bbio->error); | |
6358 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6359 | /* Should be the original bio. */ |
8408c716 MX |
6360 | WARN_ON(bio != bbio->orig_bio); |
6361 | ||
9be3395b | 6362 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6363 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6364 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6365 | bio->bi_status = BLK_STS_IOERR; | |
6366 | else | |
6367 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6368 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6369 | } |
6370 | } | |
6371 | ||
58efbc9f OS |
6372 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6373 | int mirror_num, int async_submit) | |
0b86a832 | 6374 | { |
0b86a832 | 6375 | struct btrfs_device *dev; |
8790d502 | 6376 | struct bio *first_bio = bio; |
4f024f37 | 6377 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6378 | u64 length = 0; |
6379 | u64 map_length; | |
0b86a832 | 6380 | int ret; |
08da757d ZL |
6381 | int dev_nr; |
6382 | int total_devs; | |
a1d3c478 | 6383 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6384 | |
4f024f37 | 6385 | length = bio->bi_iter.bi_size; |
0b86a832 | 6386 | map_length = length; |
cea9e445 | 6387 | |
0b246afa | 6388 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6389 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6390 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6391 | if (ret) { |
0b246afa | 6392 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6393 | return errno_to_blk_status(ret); |
c404e0dc | 6394 | } |
cea9e445 | 6395 | |
a1d3c478 | 6396 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6397 | bbio->orig_bio = first_bio; |
6398 | bbio->private = first_bio->bi_private; | |
6399 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6400 | bbio->fs_info = fs_info; |
53b381b3 DW |
6401 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6402 | ||
ad1ba2a0 | 6403 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6404 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6405 | /* In this case, map_length has been set to the length of |
6406 | a single stripe; not the whole write */ | |
37226b21 | 6407 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6408 | ret = raid56_parity_write(fs_info, bio, bbio, |
6409 | map_length); | |
53b381b3 | 6410 | } else { |
2ff7e61e JM |
6411 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6412 | map_length, mirror_num, 1); | |
53b381b3 | 6413 | } |
4245215d | 6414 | |
0b246afa | 6415 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6416 | return errno_to_blk_status(ret); |
53b381b3 DW |
6417 | } |
6418 | ||
cea9e445 | 6419 | if (map_length < length) { |
0b246afa | 6420 | btrfs_crit(fs_info, |
5d163e0e JM |
6421 | "mapping failed logical %llu bio len %llu len %llu", |
6422 | logical, length, map_length); | |
cea9e445 CM |
6423 | BUG(); |
6424 | } | |
a1d3c478 | 6425 | |
08da757d | 6426 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6427 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6428 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6429 | &dev->dev_state) || | |
ebbede42 AJ |
6430 | (bio_op(first_bio) == REQ_OP_WRITE && |
6431 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6432 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6433 | continue; |
6434 | } | |
6435 | ||
3aa8e074 | 6436 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6437 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6438 | else |
a1d3c478 | 6439 | bio = first_bio; |
de1ee92a | 6440 | |
2ff7e61e JM |
6441 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6442 | dev_nr, async_submit); | |
8790d502 | 6443 | } |
0b246afa | 6444 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6445 | return BLK_STS_OK; |
0b86a832 CM |
6446 | } |
6447 | ||
aa1b8cd4 | 6448 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6449 | u8 *uuid, u8 *fsid) |
0b86a832 | 6450 | { |
2b82032c YZ |
6451 | struct btrfs_device *device; |
6452 | struct btrfs_fs_devices *cur_devices; | |
6453 | ||
aa1b8cd4 | 6454 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6455 | while (cur_devices) { |
6456 | if (!fsid || | |
7239ff4b | 6457 | !memcmp(cur_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6458 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6459 | if (device) |
6460 | return device; | |
6461 | } | |
6462 | cur_devices = cur_devices->seed; | |
6463 | } | |
6464 | return NULL; | |
0b86a832 CM |
6465 | } |
6466 | ||
2ff7e61e | 6467 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6468 | u64 devid, u8 *dev_uuid) |
6469 | { | |
6470 | struct btrfs_device *device; | |
dfe25020 | 6471 | |
12bd2fc0 ID |
6472 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6473 | if (IS_ERR(device)) | |
adfb69af | 6474 | return device; |
12bd2fc0 ID |
6475 | |
6476 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6477 | device->fs_devices = fs_devices; |
dfe25020 | 6478 | fs_devices->num_devices++; |
12bd2fc0 | 6479 | |
e6e674bd | 6480 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6481 | fs_devices->missing_devices++; |
12bd2fc0 | 6482 | |
dfe25020 CM |
6483 | return device; |
6484 | } | |
6485 | ||
12bd2fc0 ID |
6486 | /** |
6487 | * btrfs_alloc_device - allocate struct btrfs_device | |
6488 | * @fs_info: used only for generating a new devid, can be NULL if | |
6489 | * devid is provided (i.e. @devid != NULL). | |
6490 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6491 | * is generated. | |
6492 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6493 | * is generated. | |
6494 | * | |
6495 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6496 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6497 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6498 | */ |
6499 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6500 | const u64 *devid, | |
6501 | const u8 *uuid) | |
6502 | { | |
6503 | struct btrfs_device *dev; | |
6504 | u64 tmp; | |
6505 | ||
fae7f21c | 6506 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6507 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6508 | |
6509 | dev = __alloc_device(); | |
6510 | if (IS_ERR(dev)) | |
6511 | return dev; | |
6512 | ||
6513 | if (devid) | |
6514 | tmp = *devid; | |
6515 | else { | |
6516 | int ret; | |
6517 | ||
6518 | ret = find_next_devid(fs_info, &tmp); | |
6519 | if (ret) { | |
a425f9d4 | 6520 | btrfs_free_device(dev); |
12bd2fc0 ID |
6521 | return ERR_PTR(ret); |
6522 | } | |
6523 | } | |
6524 | dev->devid = tmp; | |
6525 | ||
6526 | if (uuid) | |
6527 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6528 | else | |
6529 | generate_random_uuid(dev->uuid); | |
6530 | ||
9e0af237 LB |
6531 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6532 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6533 | |
6534 | return dev; | |
6535 | } | |
6536 | ||
e06cd3dd | 6537 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6538 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6539 | struct extent_buffer *leaf, |
6540 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6541 | { |
0b86a832 | 6542 | u64 length; |
f04b772b | 6543 | u64 stripe_len; |
e06cd3dd LB |
6544 | u16 num_stripes; |
6545 | u16 sub_stripes; | |
6546 | u64 type; | |
315409b0 GJ |
6547 | u64 features; |
6548 | bool mixed = false; | |
0b86a832 | 6549 | |
e17cade2 | 6550 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6551 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6552 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6553 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6554 | type = btrfs_chunk_type(leaf, chunk); | |
6555 | ||
f04b772b | 6556 | if (!num_stripes) { |
0b246afa | 6557 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6558 | num_stripes); |
6559 | return -EIO; | |
6560 | } | |
0b246afa JM |
6561 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6562 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6563 | return -EIO; |
6564 | } | |
0b246afa JM |
6565 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6566 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6567 | btrfs_chunk_sector_size(leaf, chunk)); |
6568 | return -EIO; | |
6569 | } | |
0b246afa JM |
6570 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6571 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6572 | return -EIO; |
6573 | } | |
3d8da678 | 6574 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6575 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6576 | stripe_len); |
6577 | return -EIO; | |
6578 | } | |
6579 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6580 | type) { |
0b246afa | 6581 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6582 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6583 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6584 | btrfs_chunk_type(leaf, chunk)); | |
6585 | return -EIO; | |
6586 | } | |
315409b0 GJ |
6587 | |
6588 | if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) { | |
6589 | btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type); | |
6590 | return -EIO; | |
6591 | } | |
6592 | ||
6593 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
6594 | (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) { | |
6595 | btrfs_err(fs_info, | |
6596 | "system chunk with data or metadata type: 0x%llx", type); | |
6597 | return -EIO; | |
6598 | } | |
6599 | ||
6600 | features = btrfs_super_incompat_flags(fs_info->super_copy); | |
6601 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
6602 | mixed = true; | |
6603 | ||
6604 | if (!mixed) { | |
6605 | if ((type & BTRFS_BLOCK_GROUP_METADATA) && | |
6606 | (type & BTRFS_BLOCK_GROUP_DATA)) { | |
6607 | btrfs_err(fs_info, | |
6608 | "mixed chunk type in non-mixed mode: 0x%llx", type); | |
6609 | return -EIO; | |
6610 | } | |
6611 | } | |
6612 | ||
e06cd3dd LB |
6613 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6614 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6615 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6616 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6617 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6618 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6619 | num_stripes != 1)) { | |
0b246afa | 6620 | btrfs_err(fs_info, |
e06cd3dd LB |
6621 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6622 | num_stripes, sub_stripes, | |
6623 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6624 | return -EIO; | |
6625 | } | |
6626 | ||
6627 | return 0; | |
6628 | } | |
6629 | ||
5a2b8e60 | 6630 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6631 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6632 | { |
2b902dfc AJ |
6633 | if (error) |
6634 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6635 | devid, uuid); | |
6636 | else | |
6637 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6638 | devid, uuid); | |
5a2b8e60 AJ |
6639 | } |
6640 | ||
2ff7e61e | 6641 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6642 | struct extent_buffer *leaf, |
6643 | struct btrfs_chunk *chunk) | |
6644 | { | |
0b246afa | 6645 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6646 | struct map_lookup *map; |
6647 | struct extent_map *em; | |
6648 | u64 logical; | |
6649 | u64 length; | |
e06cd3dd LB |
6650 | u64 devid; |
6651 | u8 uuid[BTRFS_UUID_SIZE]; | |
6652 | int num_stripes; | |
6653 | int ret; | |
6654 | int i; | |
6655 | ||
6656 | logical = key->offset; | |
6657 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6658 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6659 | ||
2ff7e61e | 6660 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6661 | if (ret) |
6662 | return ret; | |
a061fc8d | 6663 | |
890871be | 6664 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6665 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6666 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6667 | |
6668 | /* already mapped? */ | |
6669 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6670 | free_extent_map(em); | |
0b86a832 CM |
6671 | return 0; |
6672 | } else if (em) { | |
6673 | free_extent_map(em); | |
6674 | } | |
0b86a832 | 6675 | |
172ddd60 | 6676 | em = alloc_extent_map(); |
0b86a832 CM |
6677 | if (!em) |
6678 | return -ENOMEM; | |
593060d7 | 6679 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6680 | if (!map) { |
6681 | free_extent_map(em); | |
6682 | return -ENOMEM; | |
6683 | } | |
6684 | ||
298a8f9c | 6685 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6686 | em->map_lookup = map; |
0b86a832 CM |
6687 | em->start = logical; |
6688 | em->len = length; | |
70c8a91c | 6689 | em->orig_start = 0; |
0b86a832 | 6690 | em->block_start = 0; |
c8b97818 | 6691 | em->block_len = em->len; |
0b86a832 | 6692 | |
593060d7 CM |
6693 | map->num_stripes = num_stripes; |
6694 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6695 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6696 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6697 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6698 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6699 | map->verified_stripes = 0; |
593060d7 CM |
6700 | for (i = 0; i < num_stripes; i++) { |
6701 | map->stripes[i].physical = | |
6702 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6703 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6704 | read_extent_buffer(leaf, uuid, (unsigned long) |
6705 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6706 | BTRFS_UUID_SIZE); | |
0b246afa | 6707 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6708 | uuid, NULL); |
3cdde224 | 6709 | if (!map->stripes[i].dev && |
0b246afa | 6710 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6711 | free_extent_map(em); |
2b902dfc | 6712 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6713 | return -ENOENT; |
593060d7 | 6714 | } |
dfe25020 CM |
6715 | if (!map->stripes[i].dev) { |
6716 | map->stripes[i].dev = | |
2ff7e61e JM |
6717 | add_missing_dev(fs_info->fs_devices, devid, |
6718 | uuid); | |
adfb69af | 6719 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6720 | free_extent_map(em); |
adfb69af AJ |
6721 | btrfs_err(fs_info, |
6722 | "failed to init missing dev %llu: %ld", | |
6723 | devid, PTR_ERR(map->stripes[i].dev)); | |
6724 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6725 | } |
2b902dfc | 6726 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6727 | } |
e12c9621 AJ |
6728 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6729 | &(map->stripes[i].dev->dev_state)); | |
6730 | ||
0b86a832 CM |
6731 | } |
6732 | ||
890871be | 6733 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6734 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6735 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6736 | if (ret < 0) { |
6737 | btrfs_err(fs_info, | |
6738 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6739 | em->start, em->len, ret); | |
6740 | } | |
0b86a832 CM |
6741 | free_extent_map(em); |
6742 | ||
64f64f43 | 6743 | return ret; |
0b86a832 CM |
6744 | } |
6745 | ||
143bede5 | 6746 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6747 | struct btrfs_dev_item *dev_item, |
6748 | struct btrfs_device *device) | |
6749 | { | |
6750 | unsigned long ptr; | |
0b86a832 CM |
6751 | |
6752 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6753 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6754 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6755 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6756 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6757 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6758 | device->type = btrfs_device_type(leaf, dev_item); |
6759 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6760 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6761 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6762 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6763 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6764 | |
410ba3a2 | 6765 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6766 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6767 | } |
6768 | ||
2ff7e61e | 6769 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6770 | u8 *fsid) |
2b82032c YZ |
6771 | { |
6772 | struct btrfs_fs_devices *fs_devices; | |
6773 | int ret; | |
6774 | ||
a32bf9a3 | 6775 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6776 | ASSERT(fsid); |
2b82032c | 6777 | |
0b246afa | 6778 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6779 | while (fs_devices) { |
44880fdc | 6780 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6781 | return fs_devices; |
6782 | ||
2b82032c YZ |
6783 | fs_devices = fs_devices->seed; |
6784 | } | |
6785 | ||
7239ff4b | 6786 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6787 | if (!fs_devices) { |
0b246afa | 6788 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6789 | return ERR_PTR(-ENOENT); |
6790 | ||
7239ff4b | 6791 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6792 | if (IS_ERR(fs_devices)) |
6793 | return fs_devices; | |
6794 | ||
6795 | fs_devices->seeding = 1; | |
6796 | fs_devices->opened = 1; | |
6797 | return fs_devices; | |
2b82032c | 6798 | } |
e4404d6e YZ |
6799 | |
6800 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6801 | if (IS_ERR(fs_devices)) |
6802 | return fs_devices; | |
2b82032c | 6803 | |
897fb573 | 6804 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6805 | if (ret) { |
6806 | free_fs_devices(fs_devices); | |
5f375835 | 6807 | fs_devices = ERR_PTR(ret); |
2b82032c | 6808 | goto out; |
48d28232 | 6809 | } |
2b82032c YZ |
6810 | |
6811 | if (!fs_devices->seeding) { | |
0226e0eb | 6812 | close_fs_devices(fs_devices); |
e4404d6e | 6813 | free_fs_devices(fs_devices); |
5f375835 | 6814 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6815 | goto out; |
6816 | } | |
6817 | ||
0b246afa JM |
6818 | fs_devices->seed = fs_info->fs_devices->seed; |
6819 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6820 | out: |
5f375835 | 6821 | return fs_devices; |
2b82032c YZ |
6822 | } |
6823 | ||
2ff7e61e | 6824 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6825 | struct extent_buffer *leaf, |
6826 | struct btrfs_dev_item *dev_item) | |
6827 | { | |
0b246afa | 6828 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6829 | struct btrfs_device *device; |
6830 | u64 devid; | |
6831 | int ret; | |
44880fdc | 6832 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6833 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6834 | ||
0b86a832 | 6835 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6836 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6837 | BTRFS_UUID_SIZE); |
1473b24e | 6838 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6839 | BTRFS_FSID_SIZE); |
2b82032c | 6840 | |
de37aa51 | 6841 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6842 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6843 | if (IS_ERR(fs_devices)) |
6844 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6845 | } |
6846 | ||
0b246afa | 6847 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6848 | if (!device) { |
c5502451 | 6849 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6850 | btrfs_report_missing_device(fs_info, devid, |
6851 | dev_uuid, true); | |
45dbdbc9 | 6852 | return -ENOENT; |
c5502451 | 6853 | } |
2b82032c | 6854 | |
2ff7e61e | 6855 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6856 | if (IS_ERR(device)) { |
6857 | btrfs_err(fs_info, | |
6858 | "failed to add missing dev %llu: %ld", | |
6859 | devid, PTR_ERR(device)); | |
6860 | return PTR_ERR(device); | |
6861 | } | |
2b902dfc | 6862 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6863 | } else { |
c5502451 | 6864 | if (!device->bdev) { |
2b902dfc AJ |
6865 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6866 | btrfs_report_missing_device(fs_info, | |
6867 | devid, dev_uuid, true); | |
45dbdbc9 | 6868 | return -ENOENT; |
2b902dfc AJ |
6869 | } |
6870 | btrfs_report_missing_device(fs_info, devid, | |
6871 | dev_uuid, false); | |
c5502451 | 6872 | } |
5f375835 | 6873 | |
e6e674bd AJ |
6874 | if (!device->bdev && |
6875 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6876 | /* |
6877 | * this happens when a device that was properly setup | |
6878 | * in the device info lists suddenly goes bad. | |
6879 | * device->bdev is NULL, and so we have to set | |
6880 | * device->missing to one here | |
6881 | */ | |
5f375835 | 6882 | device->fs_devices->missing_devices++; |
e6e674bd | 6883 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6884 | } |
5f375835 MX |
6885 | |
6886 | /* Move the device to its own fs_devices */ | |
6887 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6888 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6889 | &device->dev_state)); | |
5f375835 MX |
6890 | |
6891 | list_move(&device->dev_list, &fs_devices->devices); | |
6892 | device->fs_devices->num_devices--; | |
6893 | fs_devices->num_devices++; | |
6894 | ||
6895 | device->fs_devices->missing_devices--; | |
6896 | fs_devices->missing_devices++; | |
6897 | ||
6898 | device->fs_devices = fs_devices; | |
6899 | } | |
2b82032c YZ |
6900 | } |
6901 | ||
0b246afa | 6902 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6903 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6904 | if (device->generation != |
6905 | btrfs_device_generation(leaf, dev_item)) | |
6906 | return -EINVAL; | |
6324fbf3 | 6907 | } |
0b86a832 CM |
6908 | |
6909 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6910 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6911 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6912 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6913 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6914 | atomic64_add(device->total_bytes - device->bytes_used, |
6915 | &fs_info->free_chunk_space); | |
2bf64758 | 6916 | } |
0b86a832 | 6917 | ret = 0; |
0b86a832 CM |
6918 | return ret; |
6919 | } | |
6920 | ||
6bccf3ab | 6921 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6922 | { |
6bccf3ab | 6923 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6924 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6925 | struct extent_buffer *sb; |
0b86a832 | 6926 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6927 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6928 | u8 *array_ptr; |
6929 | unsigned long sb_array_offset; | |
84eed90f | 6930 | int ret = 0; |
0b86a832 CM |
6931 | u32 num_stripes; |
6932 | u32 array_size; | |
6933 | u32 len = 0; | |
1ffb22cf | 6934 | u32 cur_offset; |
e06cd3dd | 6935 | u64 type; |
84eed90f | 6936 | struct btrfs_key key; |
0b86a832 | 6937 | |
0b246afa | 6938 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6939 | /* |
6940 | * This will create extent buffer of nodesize, superblock size is | |
6941 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6942 | * overallocate but we can keep it as-is, only the first page is used. | |
6943 | */ | |
2ff7e61e | 6944 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6945 | if (IS_ERR(sb)) |
6946 | return PTR_ERR(sb); | |
4db8c528 | 6947 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6948 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6949 | /* |
01327610 | 6950 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6951 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6952 | * pages up-to-date when the page is larger: extent does not cover the |
6953 | * whole page and consequently check_page_uptodate does not find all | |
6954 | * the page's extents up-to-date (the hole beyond sb), | |
6955 | * write_extent_buffer then triggers a WARN_ON. | |
6956 | * | |
6957 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6958 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6959 | * to silence the warning eg. on PowerPC 64. | |
6960 | */ | |
09cbfeaf | 6961 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6962 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6963 | |
a061fc8d | 6964 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6965 | array_size = btrfs_super_sys_array_size(super_copy); |
6966 | ||
1ffb22cf DS |
6967 | array_ptr = super_copy->sys_chunk_array; |
6968 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6969 | cur_offset = 0; | |
0b86a832 | 6970 | |
1ffb22cf DS |
6971 | while (cur_offset < array_size) { |
6972 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6973 | len = sizeof(*disk_key); |
6974 | if (cur_offset + len > array_size) | |
6975 | goto out_short_read; | |
6976 | ||
0b86a832 CM |
6977 | btrfs_disk_key_to_cpu(&key, disk_key); |
6978 | ||
1ffb22cf DS |
6979 | array_ptr += len; |
6980 | sb_array_offset += len; | |
6981 | cur_offset += len; | |
0b86a832 | 6982 | |
0d81ba5d | 6983 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6984 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6985 | /* |
6986 | * At least one btrfs_chunk with one stripe must be | |
6987 | * present, exact stripe count check comes afterwards | |
6988 | */ | |
6989 | len = btrfs_chunk_item_size(1); | |
6990 | if (cur_offset + len > array_size) | |
6991 | goto out_short_read; | |
6992 | ||
6993 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6994 | if (!num_stripes) { |
ab8d0fc4 JM |
6995 | btrfs_err(fs_info, |
6996 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6997 | num_stripes, cur_offset); |
6998 | ret = -EIO; | |
6999 | break; | |
7000 | } | |
7001 | ||
e06cd3dd LB |
7002 | type = btrfs_chunk_type(sb, chunk); |
7003 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7004 | btrfs_err(fs_info, |
e06cd3dd LB |
7005 | "invalid chunk type %llu in sys_array at offset %u", |
7006 | type, cur_offset); | |
7007 | ret = -EIO; | |
7008 | break; | |
7009 | } | |
7010 | ||
e3540eab DS |
7011 | len = btrfs_chunk_item_size(num_stripes); |
7012 | if (cur_offset + len > array_size) | |
7013 | goto out_short_read; | |
7014 | ||
2ff7e61e | 7015 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
7016 | if (ret) |
7017 | break; | |
0b86a832 | 7018 | } else { |
ab8d0fc4 JM |
7019 | btrfs_err(fs_info, |
7020 | "unexpected item type %u in sys_array at offset %u", | |
7021 | (u32)key.type, cur_offset); | |
84eed90f CM |
7022 | ret = -EIO; |
7023 | break; | |
0b86a832 | 7024 | } |
1ffb22cf DS |
7025 | array_ptr += len; |
7026 | sb_array_offset += len; | |
7027 | cur_offset += len; | |
0b86a832 | 7028 | } |
d865177a | 7029 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7030 | free_extent_buffer_stale(sb); |
84eed90f | 7031 | return ret; |
e3540eab DS |
7032 | |
7033 | out_short_read: | |
ab8d0fc4 | 7034 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7035 | len, cur_offset); |
d865177a | 7036 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7037 | free_extent_buffer_stale(sb); |
e3540eab | 7038 | return -EIO; |
0b86a832 CM |
7039 | } |
7040 | ||
21634a19 QW |
7041 | /* |
7042 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7043 | * | |
6528b99d AJ |
7044 | * If the @failing_dev is specified, it's accounted as missing. |
7045 | * | |
21634a19 QW |
7046 | * Return true if all chunks meet the minimal RW mount requirements. |
7047 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7048 | */ | |
6528b99d AJ |
7049 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7050 | struct btrfs_device *failing_dev) | |
21634a19 QW |
7051 | { |
7052 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
7053 | struct extent_map *em; | |
7054 | u64 next_start = 0; | |
7055 | bool ret = true; | |
7056 | ||
7057 | read_lock(&map_tree->map_tree.lock); | |
7058 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
7059 | read_unlock(&map_tree->map_tree.lock); | |
7060 | /* No chunk at all? Return false anyway */ | |
7061 | if (!em) { | |
7062 | ret = false; | |
7063 | goto out; | |
7064 | } | |
7065 | while (em) { | |
7066 | struct map_lookup *map; | |
7067 | int missing = 0; | |
7068 | int max_tolerated; | |
7069 | int i; | |
7070 | ||
7071 | map = em->map_lookup; | |
7072 | max_tolerated = | |
7073 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7074 | map->type); | |
7075 | for (i = 0; i < map->num_stripes; i++) { | |
7076 | struct btrfs_device *dev = map->stripes[i].dev; | |
7077 | ||
e6e674bd AJ |
7078 | if (!dev || !dev->bdev || |
7079 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7080 | dev->last_flush_error) |
7081 | missing++; | |
6528b99d AJ |
7082 | else if (failing_dev && failing_dev == dev) |
7083 | missing++; | |
21634a19 QW |
7084 | } |
7085 | if (missing > max_tolerated) { | |
6528b99d AJ |
7086 | if (!failing_dev) |
7087 | btrfs_warn(fs_info, | |
21634a19 QW |
7088 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
7089 | em->start, missing, max_tolerated); | |
7090 | free_extent_map(em); | |
7091 | ret = false; | |
7092 | goto out; | |
7093 | } | |
7094 | next_start = extent_map_end(em); | |
7095 | free_extent_map(em); | |
7096 | ||
7097 | read_lock(&map_tree->map_tree.lock); | |
7098 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
7099 | (u64)(-1) - next_start); | |
7100 | read_unlock(&map_tree->map_tree.lock); | |
7101 | } | |
7102 | out: | |
7103 | return ret; | |
7104 | } | |
7105 | ||
5b4aacef | 7106 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7107 | { |
5b4aacef | 7108 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7109 | struct btrfs_path *path; |
7110 | struct extent_buffer *leaf; | |
7111 | struct btrfs_key key; | |
7112 | struct btrfs_key found_key; | |
7113 | int ret; | |
7114 | int slot; | |
99e3ecfc | 7115 | u64 total_dev = 0; |
0b86a832 | 7116 | |
0b86a832 CM |
7117 | path = btrfs_alloc_path(); |
7118 | if (!path) | |
7119 | return -ENOMEM; | |
7120 | ||
3dd0f7a3 AJ |
7121 | /* |
7122 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7123 | * otherwise we don't need it. | |
7124 | */ | |
b367e47f | 7125 | mutex_lock(&uuid_mutex); |
34441361 | 7126 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7127 | |
395927a9 FDBM |
7128 | /* |
7129 | * Read all device items, and then all the chunk items. All | |
7130 | * device items are found before any chunk item (their object id | |
7131 | * is smaller than the lowest possible object id for a chunk | |
7132 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7133 | */ |
7134 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7135 | key.offset = 0; | |
7136 | key.type = 0; | |
0b86a832 | 7137 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7138 | if (ret < 0) |
7139 | goto error; | |
d397712b | 7140 | while (1) { |
0b86a832 CM |
7141 | leaf = path->nodes[0]; |
7142 | slot = path->slots[0]; | |
7143 | if (slot >= btrfs_header_nritems(leaf)) { | |
7144 | ret = btrfs_next_leaf(root, path); | |
7145 | if (ret == 0) | |
7146 | continue; | |
7147 | if (ret < 0) | |
7148 | goto error; | |
7149 | break; | |
7150 | } | |
7151 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7152 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7153 | struct btrfs_dev_item *dev_item; | |
7154 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7155 | struct btrfs_dev_item); |
2ff7e61e | 7156 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
7157 | if (ret) |
7158 | goto error; | |
99e3ecfc | 7159 | total_dev++; |
0b86a832 CM |
7160 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7161 | struct btrfs_chunk *chunk; | |
7162 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7163 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7164 | if (ret) |
7165 | goto error; | |
0b86a832 CM |
7166 | } |
7167 | path->slots[0]++; | |
7168 | } | |
99e3ecfc LB |
7169 | |
7170 | /* | |
7171 | * After loading chunk tree, we've got all device information, | |
7172 | * do another round of validation checks. | |
7173 | */ | |
0b246afa JM |
7174 | if (total_dev != fs_info->fs_devices->total_devices) { |
7175 | btrfs_err(fs_info, | |
99e3ecfc | 7176 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7177 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7178 | total_dev); |
7179 | ret = -EINVAL; | |
7180 | goto error; | |
7181 | } | |
0b246afa JM |
7182 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7183 | fs_info->fs_devices->total_rw_bytes) { | |
7184 | btrfs_err(fs_info, | |
99e3ecfc | 7185 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7186 | btrfs_super_total_bytes(fs_info->super_copy), |
7187 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7188 | ret = -EINVAL; |
7189 | goto error; | |
7190 | } | |
0b86a832 CM |
7191 | ret = 0; |
7192 | error: | |
34441361 | 7193 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7194 | mutex_unlock(&uuid_mutex); |
7195 | ||
2b82032c | 7196 | btrfs_free_path(path); |
0b86a832 CM |
7197 | return ret; |
7198 | } | |
442a4f63 | 7199 | |
cb517eab MX |
7200 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7201 | { | |
7202 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7203 | struct btrfs_device *device; | |
7204 | ||
29cc83f6 LB |
7205 | while (fs_devices) { |
7206 | mutex_lock(&fs_devices->device_list_mutex); | |
7207 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7208 | device->fs_info = fs_info; |
29cc83f6 LB |
7209 | mutex_unlock(&fs_devices->device_list_mutex); |
7210 | ||
7211 | fs_devices = fs_devices->seed; | |
7212 | } | |
cb517eab MX |
7213 | } |
7214 | ||
733f4fbb SB |
7215 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7216 | { | |
7217 | int i; | |
7218 | ||
7219 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7220 | btrfs_dev_stat_reset(dev, i); | |
7221 | } | |
7222 | ||
7223 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7224 | { | |
7225 | struct btrfs_key key; | |
7226 | struct btrfs_key found_key; | |
7227 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7228 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7229 | struct extent_buffer *eb; | |
7230 | int slot; | |
7231 | int ret = 0; | |
7232 | struct btrfs_device *device; | |
7233 | struct btrfs_path *path = NULL; | |
7234 | int i; | |
7235 | ||
7236 | path = btrfs_alloc_path(); | |
7237 | if (!path) { | |
7238 | ret = -ENOMEM; | |
7239 | goto out; | |
7240 | } | |
7241 | ||
7242 | mutex_lock(&fs_devices->device_list_mutex); | |
7243 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7244 | int item_size; | |
7245 | struct btrfs_dev_stats_item *ptr; | |
7246 | ||
242e2956 DS |
7247 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7248 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7249 | key.offset = device->devid; |
7250 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7251 | if (ret) { | |
733f4fbb SB |
7252 | __btrfs_reset_dev_stats(device); |
7253 | device->dev_stats_valid = 1; | |
7254 | btrfs_release_path(path); | |
7255 | continue; | |
7256 | } | |
7257 | slot = path->slots[0]; | |
7258 | eb = path->nodes[0]; | |
7259 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7260 | item_size = btrfs_item_size_nr(eb, slot); | |
7261 | ||
7262 | ptr = btrfs_item_ptr(eb, slot, | |
7263 | struct btrfs_dev_stats_item); | |
7264 | ||
7265 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7266 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7267 | btrfs_dev_stat_set(device, i, | |
7268 | btrfs_dev_stats_value(eb, ptr, i)); | |
7269 | else | |
7270 | btrfs_dev_stat_reset(device, i); | |
7271 | } | |
7272 | ||
7273 | device->dev_stats_valid = 1; | |
7274 | btrfs_dev_stat_print_on_load(device); | |
7275 | btrfs_release_path(path); | |
7276 | } | |
7277 | mutex_unlock(&fs_devices->device_list_mutex); | |
7278 | ||
7279 | out: | |
7280 | btrfs_free_path(path); | |
7281 | return ret < 0 ? ret : 0; | |
7282 | } | |
7283 | ||
7284 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7285 | struct btrfs_device *device) |
7286 | { | |
5495f195 | 7287 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7288 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7289 | struct btrfs_path *path; |
7290 | struct btrfs_key key; | |
7291 | struct extent_buffer *eb; | |
7292 | struct btrfs_dev_stats_item *ptr; | |
7293 | int ret; | |
7294 | int i; | |
7295 | ||
242e2956 DS |
7296 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7297 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7298 | key.offset = device->devid; |
7299 | ||
7300 | path = btrfs_alloc_path(); | |
fa252992 DS |
7301 | if (!path) |
7302 | return -ENOMEM; | |
733f4fbb SB |
7303 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7304 | if (ret < 0) { | |
0b246afa | 7305 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7306 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7307 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7308 | goto out; |
7309 | } | |
7310 | ||
7311 | if (ret == 0 && | |
7312 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7313 | /* need to delete old one and insert a new one */ | |
7314 | ret = btrfs_del_item(trans, dev_root, path); | |
7315 | if (ret != 0) { | |
0b246afa | 7316 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7317 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7318 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7319 | goto out; |
7320 | } | |
7321 | ret = 1; | |
7322 | } | |
7323 | ||
7324 | if (ret == 1) { | |
7325 | /* need to insert a new item */ | |
7326 | btrfs_release_path(path); | |
7327 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7328 | &key, sizeof(*ptr)); | |
7329 | if (ret < 0) { | |
0b246afa | 7330 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7331 | "insert dev_stats item for device %s failed %d", |
7332 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7333 | goto out; |
7334 | } | |
7335 | } | |
7336 | ||
7337 | eb = path->nodes[0]; | |
7338 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7339 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7340 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7341 | btrfs_dev_stat_read(device, i)); | |
7342 | btrfs_mark_buffer_dirty(eb); | |
7343 | ||
7344 | out: | |
7345 | btrfs_free_path(path); | |
7346 | return ret; | |
7347 | } | |
7348 | ||
7349 | /* | |
7350 | * called from commit_transaction. Writes all changed device stats to disk. | |
7351 | */ | |
7352 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7353 | struct btrfs_fs_info *fs_info) | |
7354 | { | |
733f4fbb SB |
7355 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7356 | struct btrfs_device *device; | |
addc3fa7 | 7357 | int stats_cnt; |
733f4fbb SB |
7358 | int ret = 0; |
7359 | ||
7360 | mutex_lock(&fs_devices->device_list_mutex); | |
7361 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7362 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7363 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7364 | continue; |
7365 | ||
9deae968 NB |
7366 | |
7367 | /* | |
7368 | * There is a LOAD-LOAD control dependency between the value of | |
7369 | * dev_stats_ccnt and updating the on-disk values which requires | |
7370 | * reading the in-memory counters. Such control dependencies | |
7371 | * require explicit read memory barriers. | |
7372 | * | |
7373 | * This memory barriers pairs with smp_mb__before_atomic in | |
7374 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7375 | * barrier implied by atomic_xchg in | |
7376 | * btrfs_dev_stats_read_and_reset | |
7377 | */ | |
7378 | smp_rmb(); | |
7379 | ||
5495f195 | 7380 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7381 | if (!ret) |
addc3fa7 | 7382 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7383 | } |
7384 | mutex_unlock(&fs_devices->device_list_mutex); | |
7385 | ||
7386 | return ret; | |
7387 | } | |
7388 | ||
442a4f63 SB |
7389 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7390 | { | |
7391 | btrfs_dev_stat_inc(dev, index); | |
7392 | btrfs_dev_stat_print_on_error(dev); | |
7393 | } | |
7394 | ||
48a3b636 | 7395 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7396 | { |
733f4fbb SB |
7397 | if (!dev->dev_stats_valid) |
7398 | return; | |
fb456252 | 7399 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7400 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7401 | rcu_str_deref(dev->name), |
442a4f63 SB |
7402 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7403 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7404 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7405 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7406 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7407 | } |
c11d2c23 | 7408 | |
733f4fbb SB |
7409 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7410 | { | |
a98cdb85 SB |
7411 | int i; |
7412 | ||
7413 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7414 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7415 | break; | |
7416 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7417 | return; /* all values == 0, suppress message */ | |
7418 | ||
fb456252 | 7419 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7420 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7421 | rcu_str_deref(dev->name), |
733f4fbb SB |
7422 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7423 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7424 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7425 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7426 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7427 | } | |
7428 | ||
2ff7e61e | 7429 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7430 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7431 | { |
7432 | struct btrfs_device *dev; | |
0b246afa | 7433 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7434 | int i; |
7435 | ||
7436 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7437 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7438 | mutex_unlock(&fs_devices->device_list_mutex); |
7439 | ||
7440 | if (!dev) { | |
0b246afa | 7441 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7442 | return -ENODEV; |
733f4fbb | 7443 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7444 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7445 | return -ENODEV; |
b27f7c0c | 7446 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7447 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7448 | if (stats->nr_items > i) | |
7449 | stats->values[i] = | |
7450 | btrfs_dev_stat_read_and_reset(dev, i); | |
7451 | else | |
7452 | btrfs_dev_stat_reset(dev, i); | |
7453 | } | |
7454 | } else { | |
7455 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7456 | if (stats->nr_items > i) | |
7457 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7458 | } | |
7459 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7460 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7461 | return 0; | |
7462 | } | |
a8a6dab7 | 7463 | |
da353f6b | 7464 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7465 | { |
7466 | struct buffer_head *bh; | |
7467 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7468 | int copy_num; |
a8a6dab7 | 7469 | |
12b1c263 AJ |
7470 | if (!bdev) |
7471 | return; | |
a8a6dab7 | 7472 | |
12b1c263 AJ |
7473 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7474 | copy_num++) { | |
a8a6dab7 | 7475 | |
12b1c263 AJ |
7476 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7477 | continue; | |
7478 | ||
7479 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7480 | ||
7481 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7482 | set_buffer_dirty(bh); | |
7483 | sync_dirty_buffer(bh); | |
7484 | brelse(bh); | |
7485 | } | |
7486 | ||
7487 | /* Notify udev that device has changed */ | |
7488 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7489 | ||
7490 | /* Update ctime/mtime for device path for libblkid */ | |
7491 | update_dev_time(device_path); | |
a8a6dab7 | 7492 | } |
935e5cc9 MX |
7493 | |
7494 | /* | |
7495 | * Update the size of all devices, which is used for writing out the | |
7496 | * super blocks. | |
7497 | */ | |
7498 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7499 | { | |
7500 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7501 | struct btrfs_device *curr, *next; | |
7502 | ||
7503 | if (list_empty(&fs_devices->resized_devices)) | |
7504 | return; | |
7505 | ||
7506 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7507 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7508 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7509 | resized_list) { | |
7510 | list_del_init(&curr->resized_list); | |
7511 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7512 | } | |
34441361 | 7513 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7514 | mutex_unlock(&fs_devices->device_list_mutex); |
7515 | } | |
ce7213c7 MX |
7516 | |
7517 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7518 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7519 | { |
e9b919b1 | 7520 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7521 | struct extent_map *em; |
7522 | struct map_lookup *map; | |
7523 | struct btrfs_device *dev; | |
7524 | int i; | |
7525 | ||
e9b919b1 | 7526 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7527 | return; |
7528 | ||
7529 | /* In order to kick the device replace finish process */ | |
34441361 | 7530 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7531 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7532 | map = em->map_lookup; |
ce7213c7 MX |
7533 | |
7534 | for (i = 0; i < map->num_stripes; i++) { | |
7535 | dev = map->stripes[i].dev; | |
7536 | dev->commit_bytes_used = dev->bytes_used; | |
7537 | } | |
7538 | } | |
34441361 | 7539 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7540 | } |
5a13f430 AJ |
7541 | |
7542 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7543 | { | |
7544 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7545 | while (fs_devices) { | |
7546 | fs_devices->fs_info = fs_info; | |
7547 | fs_devices = fs_devices->seed; | |
7548 | } | |
7549 | } | |
7550 | ||
7551 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7552 | { | |
7553 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7554 | while (fs_devices) { | |
7555 | fs_devices->fs_info = NULL; | |
7556 | fs_devices = fs_devices->seed; | |
7557 | } | |
7558 | } | |
46df06b8 DS |
7559 | |
7560 | /* | |
7561 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7562 | */ | |
7563 | int btrfs_bg_type_to_factor(u64 flags) | |
7564 | { | |
7565 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7566 | BTRFS_BLOCK_GROUP_RAID10)) | |
7567 | return 2; | |
7568 | return 1; | |
7569 | } | |
cf90d884 QW |
7570 | |
7571 | ||
7572 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) | |
7573 | { | |
7574 | int index = btrfs_bg_flags_to_raid_index(type); | |
7575 | int ncopies = btrfs_raid_array[index].ncopies; | |
7576 | int data_stripes; | |
7577 | ||
7578 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
7579 | case BTRFS_BLOCK_GROUP_RAID5: | |
7580 | data_stripes = num_stripes - 1; | |
7581 | break; | |
7582 | case BTRFS_BLOCK_GROUP_RAID6: | |
7583 | data_stripes = num_stripes - 2; | |
7584 | break; | |
7585 | default: | |
7586 | data_stripes = num_stripes / ncopies; | |
7587 | break; | |
7588 | } | |
7589 | return div_u64(chunk_len, data_stripes); | |
7590 | } | |
7591 | ||
7592 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7593 | u64 chunk_offset, u64 devid, | |
7594 | u64 physical_offset, u64 physical_len) | |
7595 | { | |
7596 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7597 | struct extent_map *em; | |
7598 | struct map_lookup *map; | |
05a37c48 | 7599 | struct btrfs_device *dev; |
cf90d884 QW |
7600 | u64 stripe_len; |
7601 | bool found = false; | |
7602 | int ret = 0; | |
7603 | int i; | |
7604 | ||
7605 | read_lock(&em_tree->lock); | |
7606 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7607 | read_unlock(&em_tree->lock); | |
7608 | ||
7609 | if (!em) { | |
7610 | btrfs_err(fs_info, | |
7611 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7612 | physical_offset, devid); | |
7613 | ret = -EUCLEAN; | |
7614 | goto out; | |
7615 | } | |
7616 | ||
7617 | map = em->map_lookup; | |
7618 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7619 | if (physical_len != stripe_len) { | |
7620 | btrfs_err(fs_info, | |
7621 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7622 | physical_offset, devid, em->start, physical_len, | |
7623 | stripe_len); | |
7624 | ret = -EUCLEAN; | |
7625 | goto out; | |
7626 | } | |
7627 | ||
7628 | for (i = 0; i < map->num_stripes; i++) { | |
7629 | if (map->stripes[i].dev->devid == devid && | |
7630 | map->stripes[i].physical == physical_offset) { | |
7631 | found = true; | |
7632 | if (map->verified_stripes >= map->num_stripes) { | |
7633 | btrfs_err(fs_info, | |
7634 | "too many dev extents for chunk %llu found", | |
7635 | em->start); | |
7636 | ret = -EUCLEAN; | |
7637 | goto out; | |
7638 | } | |
7639 | map->verified_stripes++; | |
7640 | break; | |
7641 | } | |
7642 | } | |
7643 | if (!found) { | |
7644 | btrfs_err(fs_info, | |
7645 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7646 | physical_offset, devid); | |
7647 | ret = -EUCLEAN; | |
7648 | } | |
05a37c48 QW |
7649 | |
7650 | /* Make sure no dev extent is beyond device bondary */ | |
7651 | dev = btrfs_find_device(fs_info, devid, NULL, NULL); | |
7652 | if (!dev) { | |
7653 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7654 | ret = -EUCLEAN; | |
7655 | goto out; | |
7656 | } | |
7657 | if (physical_offset + physical_len > dev->disk_total_bytes) { | |
7658 | btrfs_err(fs_info, | |
7659 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7660 | devid, physical_offset, physical_len, | |
7661 | dev->disk_total_bytes); | |
7662 | ret = -EUCLEAN; | |
7663 | goto out; | |
7664 | } | |
cf90d884 QW |
7665 | out: |
7666 | free_extent_map(em); | |
7667 | return ret; | |
7668 | } | |
7669 | ||
7670 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7671 | { | |
7672 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7673 | struct extent_map *em; | |
7674 | struct rb_node *node; | |
7675 | int ret = 0; | |
7676 | ||
7677 | read_lock(&em_tree->lock); | |
07e1ce09 | 7678 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7679 | em = rb_entry(node, struct extent_map, rb_node); |
7680 | if (em->map_lookup->num_stripes != | |
7681 | em->map_lookup->verified_stripes) { | |
7682 | btrfs_err(fs_info, | |
7683 | "chunk %llu has missing dev extent, have %d expect %d", | |
7684 | em->start, em->map_lookup->verified_stripes, | |
7685 | em->map_lookup->num_stripes); | |
7686 | ret = -EUCLEAN; | |
7687 | goto out; | |
7688 | } | |
7689 | } | |
7690 | out: | |
7691 | read_unlock(&em_tree->lock); | |
7692 | return ret; | |
7693 | } | |
7694 | ||
7695 | /* | |
7696 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7697 | * later chunk allocation/free would cause unexpected behavior. | |
7698 | * | |
7699 | * NOTE: This will iterate through the whole device tree, which should be of | |
7700 | * the same size level as the chunk tree. This slightly increases mount time. | |
7701 | */ | |
7702 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7703 | { | |
7704 | struct btrfs_path *path; | |
7705 | struct btrfs_root *root = fs_info->dev_root; | |
7706 | struct btrfs_key key; | |
5eb19381 QW |
7707 | u64 prev_devid = 0; |
7708 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7709 | int ret = 0; |
7710 | ||
7711 | key.objectid = 1; | |
7712 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7713 | key.offset = 0; | |
7714 | ||
7715 | path = btrfs_alloc_path(); | |
7716 | if (!path) | |
7717 | return -ENOMEM; | |
7718 | ||
7719 | path->reada = READA_FORWARD; | |
7720 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7721 | if (ret < 0) | |
7722 | goto out; | |
7723 | ||
7724 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7725 | ret = btrfs_next_item(root, path); | |
7726 | if (ret < 0) | |
7727 | goto out; | |
7728 | /* No dev extents at all? Not good */ | |
7729 | if (ret > 0) { | |
7730 | ret = -EUCLEAN; | |
7731 | goto out; | |
7732 | } | |
7733 | } | |
7734 | while (1) { | |
7735 | struct extent_buffer *leaf = path->nodes[0]; | |
7736 | struct btrfs_dev_extent *dext; | |
7737 | int slot = path->slots[0]; | |
7738 | u64 chunk_offset; | |
7739 | u64 physical_offset; | |
7740 | u64 physical_len; | |
7741 | u64 devid; | |
7742 | ||
7743 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7744 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7745 | break; | |
7746 | devid = key.objectid; | |
7747 | physical_offset = key.offset; | |
7748 | ||
7749 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7750 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7751 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7752 | ||
5eb19381 QW |
7753 | /* Check if this dev extent overlaps with the previous one */ |
7754 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7755 | btrfs_err(fs_info, | |
7756 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7757 | devid, physical_offset, prev_dev_ext_end); | |
7758 | ret = -EUCLEAN; | |
7759 | goto out; | |
7760 | } | |
7761 | ||
cf90d884 QW |
7762 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7763 | physical_offset, physical_len); | |
7764 | if (ret < 0) | |
7765 | goto out; | |
5eb19381 QW |
7766 | prev_devid = devid; |
7767 | prev_dev_ext_end = physical_offset + physical_len; | |
7768 | ||
cf90d884 QW |
7769 | ret = btrfs_next_item(root, path); |
7770 | if (ret < 0) | |
7771 | goto out; | |
7772 | if (ret > 0) { | |
7773 | ret = 0; | |
7774 | break; | |
7775 | } | |
7776 | } | |
7777 | ||
7778 | /* Ensure all chunks have corresponding dev extents */ | |
7779 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7780 | out: | |
7781 | btrfs_free_path(path); | |
7782 | return ret; | |
7783 | } | |
eede2bf3 OS |
7784 | |
7785 | /* | |
7786 | * Check whether the given block group or device is pinned by any inode being | |
7787 | * used as a swapfile. | |
7788 | */ | |
7789 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7790 | { | |
7791 | struct btrfs_swapfile_pin *sp; | |
7792 | struct rb_node *node; | |
7793 | ||
7794 | spin_lock(&fs_info->swapfile_pins_lock); | |
7795 | node = fs_info->swapfile_pins.rb_node; | |
7796 | while (node) { | |
7797 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7798 | if (ptr < sp->ptr) | |
7799 | node = node->rb_left; | |
7800 | else if (ptr > sp->ptr) | |
7801 | node = node->rb_right; | |
7802 | else | |
7803 | break; | |
7804 | } | |
7805 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7806 | return node != NULL; | |
7807 | } |